SAFE public projects git trees. - safe/jmp/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  100     /* every second */
  41 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  42
  43 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  44                                      printk(x, ## args); } while (0);
  45
  46 #if 0
  47 #define UDEBUG(x, args ...)     printk(KERN_DEBUG "%s(%d):%s(): " x,       \
  48                                         __FILE__, __LINE__, __FUNCTION__,  \
  49                                         ## args)
  50 #else
  51 #define UDEBUG(x, ...)
  52 #endif
  53
  54 struct nfulnl_instance {
  55         struct hlist_node hlist;        /* global list of instances */
  56         spinlock_t lock;
  57         atomic_t use;                   /* use count */
  58
  59         unsigned int qlen;              /* number of nlmsgs in skb */
  60         struct sk_buff *skb;            /* pre-allocatd skb */
  61         struct timer_list timer;
  62         int peer_pid;                   /* PID of the peer process */
  63
  64         /* configurable parameters */
  65         unsigned int flushtimeout;      /* timeout until queue flush */
  66         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  67         unsigned int qthreshold;        /* threshold of the queue */
  68         u_int32_t copy_range;
  69         u_int32_t seq;                  /* instance-local sequential counter */
  70         u_int16_t group_num;            /* number of this queue */
  71         u_int16_t flags;
  72         u_int8_t copy_mode;
  73 };
  74
  75 static DEFINE_RWLOCK(instances_lock);
  76 static atomic_t global_seq;
  77
  78 #define INSTANCE_BUCKETS        16
  79 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  80 static unsigned int hash_init;
  81
  82 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  83 {
  84         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  85 }
  86
  87 static struct nfulnl_instance *
  88 __instance_lookup(u_int16_t group_num)
  89 {
  90         struct hlist_head *head;
  91         struct hlist_node *pos;
  92         struct nfulnl_instance *inst;
  93
  94         UDEBUG("entering (group_num=%u)\n", group_num);
  95
  96         head = &instance_table[instance_hashfn(group_num)];
  97         hlist_for_each_entry(inst, pos, head, hlist) {
  98                 if (inst->group_num == group_num)
  99                         return inst;
 100         }
 101         return NULL;
 102 }
 103
 104 static inline void
 105 instance_get(struct nfulnl_instance *inst)
 106 {
 107         atomic_inc(&inst->use);
 108 }
 109
 110 static struct nfulnl_instance *
 111 instance_lookup_get(u_int16_t group_num)
 112 {
 113         struct nfulnl_instance *inst;
 114
 115         read_lock_bh(&instances_lock);
 116         inst = __instance_lookup(group_num);
 117         if (inst)
 118                 instance_get(inst);
 119         read_unlock_bh(&instances_lock);
 120
 121         return inst;
 122 }
 123
 124 static void
 125 instance_put(struct nfulnl_instance *inst)
 126 {
 127         if (inst && atomic_dec_and_test(&inst->use)) {
 128                 UDEBUG("kfree(inst=%p)\n", inst);
 129                 kfree(inst);
 130                 module_put(THIS_MODULE);
 131         }
 132 }
 133
 134 static void nfulnl_timer(unsigned long data);
 135
 136 static struct nfulnl_instance *
 137 instance_create(u_int16_t group_num, int pid)
 138 {
 139         struct nfulnl_instance *inst;
 140
 141         UDEBUG("entering (group_num=%u, pid=%d)\n", group_num,
 142                 pid);
 143
 144         write_lock_bh(&instances_lock);
 145         if (__instance_lookup(group_num)) {
 146                 inst = NULL;
 147                 UDEBUG("aborting, instance already exists\n");
 148                 goto out_unlock;
 149         }
 150
 151         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 152         if (!inst)
 153                 goto out_unlock;
 154
 155         if (!try_module_get(THIS_MODULE)) {
 156                 kfree(inst);
 157                 goto out_unlock;
 158         }
 159
 160         INIT_HLIST_NODE(&inst->hlist);
 161         spin_lock_init(&inst->lock);
 162         /* needs to be two, since we _put() after creation */
 163         atomic_set(&inst->use, 2);
 164
 165         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
 166
 167         inst->peer_pid = pid;
 168         inst->group_num = group_num;
 169
 170         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 171         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 172         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 173         inst->copy_mode         = NFULNL_COPY_PACKET;
 174         inst->copy_range        = 0xffff;
 175
 176         hlist_add_head(&inst->hlist,
 177                        &instance_table[instance_hashfn(group_num)]);
 178
 179         UDEBUG("newly added node: %p, next=%p\n", &inst->hlist,
 180                 inst->hlist.next);
 181
 182         write_unlock_bh(&instances_lock);
 183
 184         return inst;
 185
 186 out_unlock:
 187         write_unlock_bh(&instances_lock);
 188         return NULL;
 189 }
 190
 191 static void __nfulnl_flush(struct nfulnl_instance *inst);
 192
 193 static void
 194 __instance_destroy(struct nfulnl_instance *inst)
 195 {
 196         /* first pull it out of the global list */
 197         UDEBUG("removing instance %p (queuenum=%u) from hash\n",
 198                 inst, inst->group_num);
 199
 200         hlist_del(&inst->hlist);
 201
 202         /* then flush all pending packets from skb */
 203
 204         spin_lock_bh(&inst->lock);
 205         if (inst->skb)
 206                 __nfulnl_flush(inst);
 207         spin_unlock_bh(&inst->lock);
 208
 209         /* and finally put the refcount */
 210         instance_put(inst);
 211 }
 212
 213 static inline void
 214 instance_destroy(struct nfulnl_instance *inst)
 215 {
 216         write_lock_bh(&instances_lock);
 217         __instance_destroy(inst);
 218         write_unlock_bh(&instances_lock);
 219 }
 220
 221 static int
 222 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 223                   unsigned int range)
 224 {
 225         int status = 0;
 226
 227         spin_lock_bh(&inst->lock);
 228
 229         switch (mode) {
 230         case NFULNL_COPY_NONE:
 231         case NFULNL_COPY_META:
 232                 inst->copy_mode = mode;
 233                 inst->copy_range = 0;
 234                 break;
 235
 236         case NFULNL_COPY_PACKET:
 237                 inst->copy_mode = mode;
 238                 /* we're using struct nlattr which has 16bit nfa_len */
 239                 if (range > 0xffff)
 240                         inst->copy_range = 0xffff;
 241                 else
 242                         inst->copy_range = range;
 243                 break;
 244
 245         default:
 246                 status = -EINVAL;
 247                 break;
 248         }
 249
 250         spin_unlock_bh(&inst->lock);
 251
 252         return status;
 253 }
 254
 255 static int
 256 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 257 {
 258         int status;
 259
 260         spin_lock_bh(&inst->lock);
 261         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 262                 status = -ERANGE;
 263         else if (nlbufsiz > 131072)
 264                 status = -ERANGE;
 265         else {
 266                 inst->nlbufsiz = nlbufsiz;
 267                 status = 0;
 268         }
 269         spin_unlock_bh(&inst->lock);
 270
 271         return status;
 272 }
 273
 274 static int
 275 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 276 {
 277         spin_lock_bh(&inst->lock);
 278         inst->flushtimeout = timeout;
 279         spin_unlock_bh(&inst->lock);
 280
 281         return 0;
 282 }
 283
 284 static int
 285 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 286 {
 287         spin_lock_bh(&inst->lock);
 288         inst->qthreshold = qthresh;
 289         spin_unlock_bh(&inst->lock);
 290
 291         return 0;
 292 }
 293
 294 static int
 295 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 296 {
 297         spin_lock_bh(&inst->lock);
 298         inst->flags = flags;
 299         spin_unlock_bh(&inst->lock);
 300
 301         return 0;
 302 }
 303
 304 static struct sk_buff *
 305 nfulnl_alloc_skb(unsigned int inst_size, unsigned int pkt_size)
 306 {
 307         struct sk_buff *skb;
 308         unsigned int n;
 309
 310         UDEBUG("entered (%u, %u)\n", inst_size, pkt_size);
 311
 312         /* alloc skb which should be big enough for a whole multipart
 313          * message.  WARNING: has to be <= 128k due to slab restrictions */
 314
 315         n = max(inst_size, pkt_size);
 316         skb = alloc_skb(n, GFP_ATOMIC);
 317         if (!skb) {
 318                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
 319                         inst_size);
 320
 321                 if (n > pkt_size) {
 322                         /* try to allocate only as much as we need for current
 323                          * packet */
 324
 325                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 326                         if (!skb)
 327                                 PRINTR("nfnetlink_log: can't even alloc %u "
 328                                        "bytes\n", pkt_size);
 329                 }
 330         }
 331
 332         return skb;
 333 }
 334
 335 static int
 336 __nfulnl_send(struct nfulnl_instance *inst)
 337 {
 338         int status = -1;
 339
 340         if (inst->qlen > 1)
 341                 NLMSG_PUT(inst->skb, 0, 0,
 342                           NLMSG_DONE,
 343                           sizeof(struct nfgenmsg));
 344
 345         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
 346         if (status < 0) {
 347                 UDEBUG("netlink_unicast() failed\n");
 348                 /* FIXME: statistics */
 349         }
 350
 351         inst->qlen = 0;
 352         inst->skb = NULL;
 353
 354 nlmsg_failure:
 355         return status;
 356 }
 357
 358 static void
 359 __nfulnl_flush(struct nfulnl_instance *inst)
 360 {
 361         /* timer holds a reference */
 362         if (del_timer(&inst->timer))
 363                 instance_put(inst);
 364         if (inst->skb)
 365                 __nfulnl_send(inst);
 366 }
 367
 368 static void
 369 nfulnl_timer(unsigned long data)
 370 {
 371         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 372
 373         UDEBUG("timer function called, flushing buffer\n");
 374
 375         spin_lock_bh(&inst->lock);
 376         if (inst->skb)
 377                 __nfulnl_send(inst);
 378         spin_unlock_bh(&inst->lock);
 379         instance_put(inst);
 380 }
 381
 382 /* This is an inline function, we don't really care about a long
 383  * list of arguments */
 384 static inline int
 385 __build_packet_message(struct nfulnl_instance *inst,
 386                         const struct sk_buff *skb,
 387                         unsigned int data_len,
 388                         unsigned int pf,
 389                         unsigned int hooknum,
 390                         const struct net_device *indev,
 391                         const struct net_device *outdev,
 392                         const struct nf_loginfo *li,
 393                         const char *prefix, unsigned int plen)
 394 {
 395         struct nfulnl_msg_packet_hdr pmsg;
 396         struct nlmsghdr *nlh;
 397         struct nfgenmsg *nfmsg;
 398         __be32 tmp_uint;
 399         sk_buff_data_t old_tail = inst->skb->tail;
 400
 401         UDEBUG("entered\n");
 402
 403         nlh = NLMSG_PUT(inst->skb, 0, 0,
 404                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 405                         sizeof(struct nfgenmsg));
 406         nfmsg = NLMSG_DATA(nlh);
 407         nfmsg->nfgen_family = pf;
 408         nfmsg->version = NFNETLINK_V0;
 409         nfmsg->res_id = htons(inst->group_num);
 410
 411         pmsg.hw_protocol        = skb->protocol;
 412         pmsg.hook               = hooknum;
 413
 414         NLA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
 415
 416         if (prefix)
 417                 NLA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
 418
 419         if (indev) {
 420                 tmp_uint = htonl(indev->ifindex);
 421 #ifndef CONFIG_BRIDGE_NETFILTER
 422                 NLA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
 423                         &tmp_uint);
 424 #else
 425                 if (pf == PF_BRIDGE) {
 426                         /* Case 1: outdev is physical input device, we need to
 427                          * look for bridge group (when called from
 428                          * netfilter_bridge) */
 429                         NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 430                                 sizeof(tmp_uint), &tmp_uint);
 431                         /* this is the bridge group "brX" */
 432                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
 433                         NLA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 434                                 sizeof(tmp_uint), &tmp_uint);
 435                 } else {
 436                         /* Case 2: indev is bridge group, we need to look for
 437                          * physical device (when called from ipv4) */
 438                         NLA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 439                                 sizeof(tmp_uint), &tmp_uint);
 440                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
 441                                 tmp_uint =
 442                                     htonl(skb->nf_bridge->physindev->ifindex);
 443                                 NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 444                                         sizeof(tmp_uint), &tmp_uint);
 445                         }
 446                 }
 447 #endif
 448         }
 449
 450         if (outdev) {
 451                 tmp_uint = htonl(outdev->ifindex);
 452 #ifndef CONFIG_BRIDGE_NETFILTER
 453                 NLA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 454                         &tmp_uint);
 455 #else
 456                 if (pf == PF_BRIDGE) {
 457                         /* Case 1: outdev is physical output device, we need to
 458                          * look for bridge group (when called from
 459                          * netfilter_bridge) */
 460                         NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 461                                 sizeof(tmp_uint), &tmp_uint);
 462                         /* this is the bridge group "brX" */
 463                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
 464                         NLA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 465                                 sizeof(tmp_uint), &tmp_uint);
 466                 } else {
 467                         /* Case 2: indev is a bridge group, we need to look
 468                          * for physical device (when called from ipv4) */
 469                         NLA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 470                                 sizeof(tmp_uint), &tmp_uint);
 471                         if (skb->nf_bridge && skb->nf_bridge->physoutdev) {
 472                                 tmp_uint =
 473                                     htonl(skb->nf_bridge->physoutdev->ifindex);
 474                                 NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 475                                         sizeof(tmp_uint), &tmp_uint);
 476                         }
 477                 }
 478 #endif
 479         }
 480
 481         if (skb->mark) {
 482                 tmp_uint = htonl(skb->mark);
 483                 NLA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
 484         }
 485
 486         if (indev && skb->dev) {
 487                 struct nfulnl_msg_packet_hw phw;
 488                 int len = dev_parse_header(skb, phw.hw_addr);
 489                 if (len > 0) {
 490                         phw.hw_addrlen = htons(len);
 491                         NLA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
 492                 }
 493         }
 494
 495         if (skb->tstamp.tv64) {
 496                 struct nfulnl_msg_packet_timestamp ts;
 497                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 498                 ts.sec = cpu_to_be64(tv.tv_sec);
 499                 ts.usec = cpu_to_be64(tv.tv_usec);
 500
 501                 NLA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 502         }
 503
 504         /* UID */
 505         if (skb->sk) {
 506                 read_lock_bh(&skb->sk->sk_callback_lock);
 507                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 508                         __be32 uid = htonl(skb->sk->sk_socket->file->f_uid);
 509                         /* need to unlock here since NLA_PUT may goto */
 510                         read_unlock_bh(&skb->sk->sk_callback_lock);
 511                         NLA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
 512                 } else
 513                         read_unlock_bh(&skb->sk->sk_callback_lock);
 514         }
 515
 516         /* local sequence number */
 517         if (inst->flags & NFULNL_CFG_F_SEQ) {
 518                 tmp_uint = htonl(inst->seq++);
 519                 NLA_PUT(inst->skb, NFULA_SEQ, sizeof(tmp_uint), &tmp_uint);
 520         }
 521         /* global sequence number */
 522         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) {
 523                 tmp_uint = htonl(atomic_inc_return(&global_seq));
 524                 NLA_PUT(inst->skb, NFULA_SEQ_GLOBAL, sizeof(tmp_uint), &tmp_uint);
 525         }
 526
 527         if (data_len) {
 528                 struct nlattr *nla;
 529                 int size = nla_attr_size(data_len);
 530
 531                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
 532                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 533                         goto nlmsg_failure;
 534                 }
 535
 536                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 537                 nla->nla_type = NFULA_PAYLOAD;
 538                 nla->nla_len = size;
 539
 540                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 541                         BUG();
 542         }
 543
 544         nlh->nlmsg_len = inst->skb->tail - old_tail;
 545         return 0;
 546
 547 nlmsg_failure:
 548         UDEBUG("nlmsg_failure\n");
 549 nla_put_failure:
 550         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 551         return -1;
 552 }
 553
 554 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 555
 556 static struct nf_loginfo default_loginfo = {
 557         .type =         NF_LOG_TYPE_ULOG,
 558         .u = {
 559                 .ulog = {
 560                         .copy_len       = 0xffff,
 561                         .group          = 0,
 562                         .qthreshold     = 1,
 563                 },
 564         },
 565 };
 566
 567 /* log handler for internal netfilter logging api */
 568 static void
 569 nfulnl_log_packet(unsigned int pf,
 570                   unsigned int hooknum,
 571                   const struct sk_buff *skb,
 572                   const struct net_device *in,
 573                   const struct net_device *out,
 574                   const struct nf_loginfo *li_user,
 575                   const char *prefix)
 576 {
 577         unsigned int size, data_len;
 578         struct nfulnl_instance *inst;
 579         const struct nf_loginfo *li;
 580         unsigned int qthreshold;
 581         unsigned int plen;
 582
 583         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 584                 li = li_user;
 585         else
 586                 li = &default_loginfo;
 587
 588         inst = instance_lookup_get(li->u.ulog.group);
 589         if (!inst)
 590                 return;
 591
 592         plen = 0;
 593         if (prefix)
 594                 plen = strlen(prefix) + 1;
 595
 596         /* FIXME: do we want to make the size calculation conditional based on
 597          * what is actually present?  way more branches and checks, but more
 598          * memory efficient... */
 599         size =    NLMSG_ALIGN(sizeof(struct nfgenmsg))
 600                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 601                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 602                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 603 #ifdef CONFIG_BRIDGE_NETFILTER
 604                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 605                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 606 #endif
 607                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 608                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 609                 + nla_total_size(plen)                  /* prefix */
 610                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 611                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
 612
 613         UDEBUG("initial size=%u\n", size);
 614
 615         spin_lock_bh(&inst->lock);
 616
 617         if (inst->flags & NFULNL_CFG_F_SEQ)
 618                 size += nla_total_size(sizeof(u_int32_t));
 619         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 620                 size += nla_total_size(sizeof(u_int32_t));
 621
 622         qthreshold = inst->qthreshold;
 623         /* per-rule qthreshold overrides per-instance */
 624         if (qthreshold > li->u.ulog.qthreshold)
 625                 qthreshold = li->u.ulog.qthreshold;
 626
 627         switch (inst->copy_mode) {
 628         case NFULNL_COPY_META:
 629         case NFULNL_COPY_NONE:
 630                 data_len = 0;
 631                 break;
 632
 633         case NFULNL_COPY_PACKET:
 634                 if (inst->copy_range == 0
 635                     || inst->copy_range > skb->len)
 636                         data_len = skb->len;
 637                 else
 638                         data_len = inst->copy_range;
 639
 640                 size += nla_total_size(data_len);
 641                 UDEBUG("copy_packet, therefore size now %u\n", size);
 642                 break;
 643
 644         default:
 645                 goto unlock_and_release;
 646         }
 647
 648         if (inst->skb &&
 649             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
 650                 /* either the queue len is too high or we don't have
 651                  * enough room in the skb left. flush to userspace. */
 652                 UDEBUG("flushing old skb\n");
 653
 654                 __nfulnl_flush(inst);
 655         }
 656
 657         if (!inst->skb) {
 658                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
 659                 if (!inst->skb)
 660                         goto alloc_failure;
 661         }
 662
 663         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
 664         inst->qlen++;
 665
 666         __build_packet_message(inst, skb, data_len, pf,
 667                                 hooknum, in, out, li, prefix, plen);
 668
 669         if (inst->qlen >= qthreshold)
 670                 __nfulnl_flush(inst);
 671         /* timer_pending always called within inst->lock, so there
 672          * is no chance of a race here */
 673         else if (!timer_pending(&inst->timer)) {
 674                 instance_get(inst);
 675                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 676                 add_timer(&inst->timer);
 677         }
 678
 679 unlock_and_release:
 680         spin_unlock_bh(&inst->lock);
 681         instance_put(inst);
 682         return;
 683
 684 alloc_failure:
 685         UDEBUG("error allocating skb\n");
 686         /* FIXME: statistics */
 687         goto unlock_and_release;
 688 }
 689
 690 static int
 691 nfulnl_rcv_nl_event(struct notifier_block *this,
 692                    unsigned long event, void *ptr)
 693 {
 694         struct netlink_notify *n = ptr;
 695
 696         if (event == NETLINK_URELEASE &&
 697             n->protocol == NETLINK_NETFILTER && n->pid) {
 698                 int i;
 699
 700                 /* destroy all instances for this pid */
 701                 write_lock_bh(&instances_lock);
 702                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 703                         struct hlist_node *tmp, *t2;
 704                         struct nfulnl_instance *inst;
 705                         struct hlist_head *head = &instance_table[i];
 706
 707                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 708                                 UDEBUG("node = %p\n", inst);
 709                                 if ((n->net == &init_net) &&
 710                                     (n->pid == inst->peer_pid))
 711                                         __instance_destroy(inst);
 712                         }
 713                 }
 714                 write_unlock_bh(&instances_lock);
 715         }
 716         return NOTIFY_DONE;
 717 }
 718
 719 static struct notifier_block nfulnl_rtnl_notifier = {
 720         .notifier_call  = nfulnl_rcv_nl_event,
 721 };
 722
 723 static int
 724 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 725                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
 726 {
 727         return -ENOTSUPP;
 728 }
 729
 730 static struct nf_logger nfulnl_logger = {
 731         .name   = "nfnetlink_log",
 732         .logfn  = &nfulnl_log_packet,
 733         .me     = THIS_MODULE,
 734 };
 735
 736 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 737         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 738         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 739         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 740         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 741         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 742         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 743 };
 744
 745 static int
 746 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 747                    struct nlmsghdr *nlh, struct nlattr *nfula[])
 748 {
 749         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 750         u_int16_t group_num = ntohs(nfmsg->res_id);
 751         struct nfulnl_instance *inst;
 752         int ret = 0;
 753
 754         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
 755
 756         inst = instance_lookup_get(group_num);
 757         if (nfula[NFULA_CFG_CMD]) {
 758                 u_int8_t pf = nfmsg->nfgen_family;
 759                 struct nfulnl_msg_config_cmd *cmd;
 760                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 761                 UDEBUG("found CFG_CMD for\n");
 762
 763                 switch (cmd->command) {
 764                 case NFULNL_CFG_CMD_BIND:
 765                         if (inst) {
 766                                 ret = -EBUSY;
 767                                 goto out_put;
 768                         }
 769
 770                         inst = instance_create(group_num,
 771                                                NETLINK_CB(skb).pid);
 772                         if (!inst) {
 773                                 ret = -EINVAL;
 774                                 goto out;
 775                         }
 776                         break;
 777                 case NFULNL_CFG_CMD_UNBIND:
 778                         if (!inst) {
 779                                 ret = -ENODEV;
 780                                 goto out;
 781                         }
 782
 783                         if (inst->peer_pid != NETLINK_CB(skb).pid) {
 784                                 ret = -EPERM;
 785                                 goto out_put;
 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                 if (!inst)
 806                         goto out;
 807         } else {
 808                 if (!inst) {
 809                         UDEBUG("no config command, and no instance for "
 810                                 "group=%u pid=%u =>ENOENT\n",
 811                                 group_num, NETLINK_CB(skb).pid);
 812                         ret = -ENOENT;
 813                         goto out;
 814                 }
 815
 816                 if (inst->peer_pid != NETLINK_CB(skb).pid) {
 817                         UDEBUG("no config command, and wrong pid\n");
 818                         ret = -EPERM;
 819                         goto out_put;
 820                 }
 821         }
 822
 823         if (nfula[NFULA_CFG_MODE]) {
 824                 struct nfulnl_msg_config_mode *params;
 825                 params = nla_data(nfula[NFULA_CFG_MODE]);
 826
 827                 nfulnl_set_mode(inst, params->copy_mode,
 828                                 ntohl(params->copy_range));
 829         }
 830
 831         if (nfula[NFULA_CFG_TIMEOUT]) {
 832                 __be32 timeout =
 833                         *(__be32 *)nla_data(nfula[NFULA_CFG_TIMEOUT]);
 834
 835                 nfulnl_set_timeout(inst, ntohl(timeout));
 836         }
 837
 838         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 839                 __be32 nlbufsiz =
 840                         *(__be32 *)nla_data(nfula[NFULA_CFG_NLBUFSIZ]);
 841
 842                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 843         }
 844
 845         if (nfula[NFULA_CFG_QTHRESH]) {
 846                 __be32 qthresh =
 847                         *(__be32 *)nla_data(nfula[NFULA_CFG_QTHRESH]);
 848
 849                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 850         }
 851
 852         if (nfula[NFULA_CFG_FLAGS]) {
 853                 __be16 flags =
 854                         *(__be16 *)nla_data(nfula[NFULA_CFG_FLAGS]);
 855                 nfulnl_set_flags(inst, ntohs(flags));
 856         }
 857
 858 out_put:
 859         instance_put(inst);
 860 out:
 861         return ret;
 862 }
 863
 864 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 865         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 866                                     .attr_count = NFULA_MAX, },
 867         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 868                                     .attr_count = NFULA_CFG_MAX,
 869                                     .policy = nfula_cfg_policy },
 870 };
 871
 872 static const struct nfnetlink_subsystem nfulnl_subsys = {
 873         .name           = "log",
 874         .subsys_id      = NFNL_SUBSYS_ULOG,
 875         .cb_count       = NFULNL_MSG_MAX,
 876         .cb             = nfulnl_cb,
 877 };
 878
 879 #ifdef CONFIG_PROC_FS
 880 struct iter_state {
 881         unsigned int bucket;
 882 };
 883
 884 static struct hlist_node *get_first(struct iter_state *st)
 885 {
 886         if (!st)
 887                 return NULL;
 888
 889         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 890                 if (!hlist_empty(&instance_table[st->bucket]))
 891                         return instance_table[st->bucket].first;
 892         }
 893         return NULL;
 894 }
 895
 896 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
 897 {
 898         h = h->next;
 899         while (!h) {
 900                 if (++st->bucket >= INSTANCE_BUCKETS)
 901                         return NULL;
 902
 903                 h = instance_table[st->bucket].first;
 904         }
 905         return h;
 906 }
 907
 908 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
 909 {
 910         struct hlist_node *head;
 911         head = get_first(st);
 912
 913         if (head)
 914                 while (pos && (head = get_next(st, head)))
 915                         pos--;
 916         return pos ? NULL : head;
 917 }
 918
 919 static void *seq_start(struct seq_file *seq, loff_t *pos)
 920 {
 921         read_lock_bh(&instances_lock);
 922         return get_idx(seq->private, *pos);
 923 }
 924
 925 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 926 {
 927         (*pos)++;
 928         return get_next(s->private, v);
 929 }
 930
 931 static void seq_stop(struct seq_file *s, void *v)
 932 {
 933         read_unlock_bh(&instances_lock);
 934 }
 935
 936 static int seq_show(struct seq_file *s, void *v)
 937 {
 938         const struct nfulnl_instance *inst = v;
 939
 940         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 941                           inst->group_num,
 942                           inst->peer_pid, inst->qlen,
 943                           inst->copy_mode, inst->copy_range,
 944                           inst->flushtimeout, atomic_read(&inst->use));
 945 }
 946
 947 static const struct seq_operations nful_seq_ops = {
 948         .start  = seq_start,
 949         .next   = seq_next,
 950         .stop   = seq_stop,
 951         .show   = seq_show,
 952 };
 953
 954 static int nful_open(struct inode *inode, struct file *file)
 955 {
 956         struct seq_file *seq;
 957         struct iter_state *is;
 958         int ret;
 959
 960         is = kzalloc(sizeof(*is), GFP_KERNEL);
 961         if (!is)
 962                 return -ENOMEM;
 963         ret = seq_open(file, &nful_seq_ops);
 964         if (ret < 0)
 965                 goto out_free;
 966         seq = file->private_data;
 967         seq->private = is;
 968         return ret;
 969 out_free:
 970         kfree(is);
 971         return ret;
 972 }
 973
 974 static const struct file_operations nful_file_ops = {
 975         .owner   = THIS_MODULE,
 976         .open    = nful_open,
 977         .read    = seq_read,
 978         .llseek  = seq_lseek,
 979         .release = seq_release_private,
 980 };
 981
 982 #endif /* PROC_FS */
 983
 984 static int __init nfnetlink_log_init(void)
 985 {
 986         int i, status = -ENOMEM;
 987 #ifdef CONFIG_PROC_FS
 988         struct proc_dir_entry *proc_nful;
 989 #endif
 990
 991         for (i = 0; i < INSTANCE_BUCKETS; i++)
 992                 INIT_HLIST_HEAD(&instance_table[i]);
 993
 994         /* it's not really all that important to have a random value, so
 995          * we can do this from the init function, even if there hasn't
 996          * been that much entropy yet */
 997         get_random_bytes(&hash_init, sizeof(hash_init));
 998
 999         netlink_register_notifier(&nfulnl_rtnl_notifier);
1000         status = nfnetlink_subsys_register(&nfulnl_subsys);
1001         if (status < 0) {
1002                 printk(KERN_ERR "log: failed to create netlink socket\n");
1003                 goto cleanup_netlink_notifier;
1004         }
1005
1006 #ifdef CONFIG_PROC_FS
1007         proc_nful = create_proc_entry("nfnetlink_log", 0440,
1008                                       proc_net_netfilter);
1009         if (!proc_nful)
1010                 goto cleanup_subsys;
1011         proc_nful->proc_fops = &nful_file_ops;
1012 #endif
1013         return status;
1014
1015 #ifdef CONFIG_PROC_FS
1016 cleanup_subsys:
1017         nfnetlink_subsys_unregister(&nfulnl_subsys);
1018 #endif
1019 cleanup_netlink_notifier:
1020         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1021         return status;
1022 }
1023
1024 static void __exit nfnetlink_log_fini(void)
1025 {
1026         nf_log_unregister(&nfulnl_logger);
1027 #ifdef CONFIG_PROC_FS
1028         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1029 #endif
1030         nfnetlink_subsys_unregister(&nfulnl_subsys);
1031         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1032 }
1033
1034 MODULE_DESCRIPTION("netfilter userspace logging");
1035 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1036 MODULE_LICENSE("GPL");
1037 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1038
1039 module_init(nfnetlink_log_init);
1040 module_exit(nfnetlink_log_fini);