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 nlmsghdr *lastnlh;       /* netlink header of last msg in 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         INIT_HLIST_NODE(&inst->hlist);
 157         spin_lock_init(&inst->lock);
 158         /* needs to be two, since we _put() after creation */
 159         atomic_set(&inst->use, 2);
 160
 161         init_timer(&inst->timer);
 162         inst->timer.function = nfulnl_timer;
 163         inst->timer.data = (unsigned long)inst;
 164         /* don't start timer yet. (re)start it  with every packet */
 165
 166         inst->peer_pid = pid;
 167         inst->group_num = group_num;
 168
 169         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 170         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 171         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 172         inst->copy_mode         = NFULNL_COPY_PACKET;
 173         inst->copy_range        = 0xffff;
 174
 175         if (!try_module_get(THIS_MODULE))
 176                 goto out_free;
 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_free:
 189         instance_put(inst);
 190 out_unlock:
 191         write_unlock_bh(&instances_lock);
 192         return NULL;
 193 }
 194
 195 static int __nfulnl_send(struct nfulnl_instance *inst);
 196
 197 static void
 198 _instance_destroy2(struct nfulnl_instance *inst, int lock)
 199 {
 200         /* first pull it out of the global list */
 201         if (lock)
 202                 write_lock_bh(&instances_lock);
 203
 204         UDEBUG("removing instance %p (queuenum=%u) from hash\n",
 205                 inst, inst->group_num);
 206
 207         hlist_del(&inst->hlist);
 208
 209         if (lock)
 210                 write_unlock_bh(&instances_lock);
 211
 212         /* then flush all pending packets from skb */
 213
 214         spin_lock_bh(&inst->lock);
 215         if (inst->skb) {
 216                 /* timer "holds" one reference (we have one more) */
 217                 if (del_timer(&inst->timer))
 218                         instance_put(inst);
 219                 if (inst->qlen)
 220                         __nfulnl_send(inst);
 221                 if (inst->skb) {
 222                         kfree_skb(inst->skb);
 223                         inst->skb = NULL;
 224                 }
 225         }
 226         spin_unlock_bh(&inst->lock);
 227
 228         /* and finally put the refcount */
 229         instance_put(inst);
 230 }
 231
 232 static inline void
 233 __instance_destroy(struct nfulnl_instance *inst)
 234 {
 235         _instance_destroy2(inst, 0);
 236 }
 237
 238 static inline void
 239 instance_destroy(struct nfulnl_instance *inst)
 240 {
 241         _instance_destroy2(inst, 1);
 242 }
 243
 244 static int
 245 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 246                   unsigned int range)
 247 {
 248         int status = 0;
 249
 250         spin_lock_bh(&inst->lock);
 251
 252         switch (mode) {
 253         case NFULNL_COPY_NONE:
 254         case NFULNL_COPY_META:
 255                 inst->copy_mode = mode;
 256                 inst->copy_range = 0;
 257                 break;
 258
 259         case NFULNL_COPY_PACKET:
 260                 inst->copy_mode = mode;
 261                 /* we're using struct nfattr which has 16bit nfa_len */
 262                 if (range > 0xffff)
 263                         inst->copy_range = 0xffff;
 264                 else
 265                         inst->copy_range = range;
 266                 break;
 267
 268         default:
 269                 status = -EINVAL;
 270                 break;
 271         }
 272
 273         spin_unlock_bh(&inst->lock);
 274
 275         return status;
 276 }
 277
 278 static int
 279 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 280 {
 281         int status;
 282
 283         spin_lock_bh(&inst->lock);
 284         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 285                 status = -ERANGE;
 286         else if (nlbufsiz > 131072)
 287                 status = -ERANGE;
 288         else {
 289                 inst->nlbufsiz = nlbufsiz;
 290                 status = 0;
 291         }
 292         spin_unlock_bh(&inst->lock);
 293
 294         return status;
 295 }
 296
 297 static int
 298 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 299 {
 300         spin_lock_bh(&inst->lock);
 301         inst->flushtimeout = timeout;
 302         spin_unlock_bh(&inst->lock);
 303
 304         return 0;
 305 }
 306
 307 static int
 308 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 309 {
 310         spin_lock_bh(&inst->lock);
 311         inst->qthreshold = qthresh;
 312         spin_unlock_bh(&inst->lock);
 313
 314         return 0;
 315 }
 316
 317 static int
 318 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 319 {
 320         spin_lock_bh(&inst->lock);
 321         inst->flags = flags;
 322         spin_unlock_bh(&inst->lock);
 323
 324         return 0;
 325 }
 326
 327 static struct sk_buff *nfulnl_alloc_skb(unsigned int inst_size,
 328                                         unsigned int pkt_size)
 329 {
 330         struct sk_buff *skb;
 331         unsigned int n;
 332
 333         UDEBUG("entered (%u, %u)\n", inst_size, pkt_size);
 334
 335         /* alloc skb which should be big enough for a whole multipart
 336          * message.  WARNING: has to be <= 128k due to slab restrictions */
 337
 338         n = max(inst_size, pkt_size);
 339         skb = alloc_skb(n, GFP_ATOMIC);
 340         if (!skb) {
 341                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
 342                         inst_size);
 343
 344                 if (n > pkt_size) {
 345                         /* try to allocate only as much as we need for current
 346                          * packet */
 347
 348                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 349                         if (!skb)
 350                                 PRINTR("nfnetlink_log: can't even alloc %u "
 351                                        "bytes\n", pkt_size);
 352                 }
 353         }
 354
 355         return skb;
 356 }
 357
 358 static int
 359 __nfulnl_send(struct nfulnl_instance *inst)
 360 {
 361         int status;
 362
 363         if (inst->qlen > 1)
 364                 inst->lastnlh->nlmsg_type = NLMSG_DONE;
 365
 366         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
 367         if (status < 0) {
 368                 UDEBUG("netlink_unicast() failed\n");
 369                 /* FIXME: statistics */
 370         }
 371
 372         inst->qlen = 0;
 373         inst->skb = NULL;
 374         inst->lastnlh = NULL;
 375
 376         return status;
 377 }
 378
 379 static void nfulnl_timer(unsigned long data)
 380 {
 381         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 382
 383         UDEBUG("timer function called, flushing buffer\n");
 384
 385         spin_lock_bh(&inst->lock);
 386         if (inst->skb)
 387                 __nfulnl_send(inst);
 388         spin_unlock_bh(&inst->lock);
 389         instance_put(inst);
 390 }
 391
 392 /* This is an inline function, we don't really care about a long
 393  * list of arguments */
 394 static inline int
 395 __build_packet_message(struct nfulnl_instance *inst,
 396                         const struct sk_buff *skb,
 397                         unsigned int data_len,
 398                         unsigned int pf,
 399                         unsigned int hooknum,
 400                         const struct net_device *indev,
 401                         const struct net_device *outdev,
 402                         const struct nf_loginfo *li,
 403                         const char *prefix, unsigned int plen)
 404 {
 405         struct nfulnl_msg_packet_hdr pmsg;
 406         struct nlmsghdr *nlh;
 407         struct nfgenmsg *nfmsg;
 408         __be32 tmp_uint;
 409         sk_buff_data_t old_tail = inst->skb->tail;
 410
 411         UDEBUG("entered\n");
 412
 413         nlh = NLMSG_PUT(inst->skb, 0, 0,
 414                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 415                         sizeof(struct nfgenmsg));
 416         nfmsg = NLMSG_DATA(nlh);
 417         nfmsg->nfgen_family = pf;
 418         nfmsg->version = NFNETLINK_V0;
 419         nfmsg->res_id = htons(inst->group_num);
 420
 421         pmsg.hw_protocol        = skb->protocol;
 422         pmsg.hook               = hooknum;
 423
 424         NFA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
 425
 426         if (prefix)
 427                 NFA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
 428
 429         if (indev) {
 430                 tmp_uint = htonl(indev->ifindex);
 431 #ifndef CONFIG_BRIDGE_NETFILTER
 432                 NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
 433                         &tmp_uint);
 434 #else
 435                 if (pf == PF_BRIDGE) {
 436                         /* Case 1: outdev is physical input device, we need to
 437                          * look for bridge group (when called from
 438                          * netfilter_bridge) */
 439                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 440                                 sizeof(tmp_uint), &tmp_uint);
 441                         /* this is the bridge group "brX" */
 442                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
 443                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 444                                 sizeof(tmp_uint), &tmp_uint);
 445                 } else {
 446                         /* Case 2: indev is bridge group, we need to look for
 447                          * physical device (when called from ipv4) */
 448                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 449                                 sizeof(tmp_uint), &tmp_uint);
 450                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
 451                                 tmp_uint =
 452                                     htonl(skb->nf_bridge->physindev->ifindex);
 453                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 454                                         sizeof(tmp_uint), &tmp_uint);
 455                         }
 456                 }
 457 #endif
 458         }
 459
 460         if (outdev) {
 461                 tmp_uint = htonl(outdev->ifindex);
 462 #ifndef CONFIG_BRIDGE_NETFILTER
 463                 NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 464                         &tmp_uint);
 465 #else
 466                 if (pf == PF_BRIDGE) {
 467                         /* Case 1: outdev is physical output device, we need to
 468                          * look for bridge group (when called from
 469                          * netfilter_bridge) */
 470                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 471                                 sizeof(tmp_uint), &tmp_uint);
 472                         /* this is the bridge group "brX" */
 473                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
 474                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 475                                 sizeof(tmp_uint), &tmp_uint);
 476                 } else {
 477                         /* Case 2: indev is a bridge group, we need to look
 478                          * for physical device (when called from ipv4) */
 479                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 480                                 sizeof(tmp_uint), &tmp_uint);
 481                         if (skb->nf_bridge && skb->nf_bridge->physoutdev) {
 482                                 tmp_uint =
 483                                     htonl(skb->nf_bridge->physoutdev->ifindex);
 484                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 485                                         sizeof(tmp_uint), &tmp_uint);
 486                         }
 487                 }
 488 #endif
 489         }
 490
 491         if (skb->mark) {
 492                 tmp_uint = htonl(skb->mark);
 493                 NFA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
 494         }
 495
 496         if (indev && skb->dev && skb->dev->hard_header_parse) {
 497                 struct nfulnl_msg_packet_hw phw;
 498                 int len = skb->dev->hard_header_parse((struct sk_buff *)skb,
 499                                                     phw.hw_addr);
 500                 phw.hw_addrlen = htons(len);
 501                 NFA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
 502         }
 503
 504         if (skb->tstamp.tv64) {
 505                 struct nfulnl_msg_packet_timestamp ts;
 506                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 507                 ts.sec = cpu_to_be64(tv.tv_sec);
 508                 ts.usec = cpu_to_be64(tv.tv_usec);
 509
 510                 NFA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 511         }
 512
 513         /* UID */
 514         if (skb->sk) {
 515                 read_lock_bh(&skb->sk->sk_callback_lock);
 516                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 517                         __be32 uid = htonl(skb->sk->sk_socket->file->f_uid);
 518                         /* need to unlock here since NFA_PUT may goto */
 519                         read_unlock_bh(&skb->sk->sk_callback_lock);
 520                         NFA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
 521                 } else
 522                         read_unlock_bh(&skb->sk->sk_callback_lock);
 523         }
 524
 525         /* local sequence number */
 526         if (inst->flags & NFULNL_CFG_F_SEQ) {
 527                 tmp_uint = htonl(inst->seq++);
 528                 NFA_PUT(inst->skb, NFULA_SEQ, sizeof(tmp_uint), &tmp_uint);
 529         }
 530         /* global sequence number */
 531         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) {
 532                 tmp_uint = htonl(atomic_inc_return(&global_seq));
 533                 NFA_PUT(inst->skb, NFULA_SEQ_GLOBAL, sizeof(tmp_uint), &tmp_uint);
 534         }
 535
 536         if (data_len) {
 537                 struct nfattr *nfa;
 538                 int size = NFA_LENGTH(data_len);
 539
 540                 if (skb_tailroom(inst->skb) < (int)NFA_SPACE(data_len)) {
 541                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 542                         goto nlmsg_failure;
 543                 }
 544
 545                 nfa = (struct nfattr *)skb_put(inst->skb, NFA_ALIGN(size));
 546                 nfa->nfa_type = NFULA_PAYLOAD;
 547                 nfa->nfa_len = size;
 548
 549                 if (skb_copy_bits(skb, 0, NFA_DATA(nfa), data_len))
 550                         BUG();
 551         }
 552
 553         nlh->nlmsg_len = inst->skb->tail - old_tail;
 554         inst->lastnlh = nlh;
 555         return 0;
 556
 557 nlmsg_failure:
 558         UDEBUG("nlmsg_failure\n");
 559 nfattr_failure:
 560         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 561         return -1;
 562 }
 563
 564 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 565
 566 static struct nf_loginfo default_loginfo = {
 567         .type =         NF_LOG_TYPE_ULOG,
 568         .u = {
 569                 .ulog = {
 570                         .copy_len       = 0xffff,
 571                         .group          = 0,
 572                         .qthreshold     = 1,
 573                 },
 574         },
 575 };
 576
 577 /* log handler for internal netfilter logging api */
 578 static void
 579 nfulnl_log_packet(unsigned int pf,
 580                   unsigned int hooknum,
 581                   const struct sk_buff *skb,
 582                   const struct net_device *in,
 583                   const struct net_device *out,
 584                   const struct nf_loginfo *li_user,
 585                   const char *prefix)
 586 {
 587         unsigned int size, data_len;
 588         struct nfulnl_instance *inst;
 589         const struct nf_loginfo *li;
 590         unsigned int qthreshold;
 591         unsigned int plen;
 592
 593         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 594                 li = li_user;
 595         else
 596                 li = &default_loginfo;
 597
 598         inst = instance_lookup_get(li->u.ulog.group);
 599         if (!inst)
 600                 inst = instance_lookup_get(0);
 601         if (!inst) {
 602                 PRINTR("nfnetlink_log: trying to log packet, "
 603                         "but no instance for group %u\n", li->u.ulog.group);
 604                 return;
 605         }
 606
 607         plen = 0;
 608         if (prefix)
 609                 plen = strlen(prefix) + 1;
 610
 611         /* all macros expand to constant values at compile time */
 612         /* FIXME: do we want to make the size calculation conditional based on
 613          * what is actually present?  way more branches and checks, but more
 614          * memory efficient... */
 615         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 616                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hdr))
 617                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 618                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 619 #ifdef CONFIG_BRIDGE_NETFILTER
 620                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 621                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 622 #endif
 623                 + NFA_SPACE(sizeof(u_int32_t))  /* mark */
 624                 + NFA_SPACE(sizeof(u_int32_t))  /* uid */
 625                 + NFA_SPACE(plen)               /* prefix */
 626                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hw))
 627                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_timestamp));
 628
 629         UDEBUG("initial size=%u\n", size);
 630
 631         spin_lock_bh(&inst->lock);
 632
 633         if (inst->flags & NFULNL_CFG_F_SEQ)
 634                 size += NFA_SPACE(sizeof(u_int32_t));
 635         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 636                 size += NFA_SPACE(sizeof(u_int32_t));
 637
 638         qthreshold = inst->qthreshold;
 639         /* per-rule qthreshold overrides per-instance */
 640         if (qthreshold > li->u.ulog.qthreshold)
 641                 qthreshold = li->u.ulog.qthreshold;
 642
 643         switch (inst->copy_mode) {
 644         case NFULNL_COPY_META:
 645         case NFULNL_COPY_NONE:
 646                 data_len = 0;
 647                 break;
 648
 649         case NFULNL_COPY_PACKET:
 650                 if (inst->copy_range == 0
 651                     || inst->copy_range > skb->len)
 652                         data_len = skb->len;
 653                 else
 654                         data_len = inst->copy_range;
 655
 656                 size += NFA_SPACE(data_len);
 657                 UDEBUG("copy_packet, therefore size now %u\n", size);
 658                 break;
 659
 660         default:
 661                 goto unlock_and_release;
 662         }
 663
 664         if (inst->qlen >= qthreshold ||
 665             (inst->skb && size > skb_tailroom(inst->skb))) {
 666                 /* either the queue len is too high or we don't have
 667                  * enough room in the skb left. flush to userspace. */
 668                 UDEBUG("flushing old skb\n");
 669
 670                 /* timer "holds" one reference (we have another one) */
 671                 if (del_timer(&inst->timer))
 672                         instance_put(inst);
 673                 __nfulnl_send(inst);
 674         }
 675
 676         if (!inst->skb) {
 677                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
 678                 if (!inst->skb)
 679                         goto alloc_failure;
 680         }
 681
 682         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
 683         inst->qlen++;
 684
 685         __build_packet_message(inst, skb, data_len, pf,
 686                                 hooknum, in, out, li, prefix, plen);
 687
 688         /* timer_pending always called within inst->lock, so there
 689          * is no chance of a race here */
 690         if (!timer_pending(&inst->timer)) {
 691                 instance_get(inst);
 692                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 693                 add_timer(&inst->timer);
 694         }
 695
 696 unlock_and_release:
 697         spin_unlock_bh(&inst->lock);
 698         instance_put(inst);
 699         return;
 700
 701 alloc_failure:
 702         UDEBUG("error allocating skb\n");
 703         /* FIXME: statistics */
 704         goto unlock_and_release;
 705 }
 706
 707 static int
 708 nfulnl_rcv_nl_event(struct notifier_block *this,
 709                    unsigned long event, void *ptr)
 710 {
 711         struct netlink_notify *n = ptr;
 712
 713         if (event == NETLINK_URELEASE &&
 714             n->protocol == NETLINK_NETFILTER && n->pid) {
 715                 int i;
 716
 717                 /* destroy all instances for this pid */
 718                 write_lock_bh(&instances_lock);
 719                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 720                         struct hlist_node *tmp, *t2;
 721                         struct nfulnl_instance *inst;
 722                         struct hlist_head *head = &instance_table[i];
 723
 724                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 725                                 UDEBUG("node = %p\n", inst);
 726                                 if (n->pid == inst->peer_pid)
 727                                         __instance_destroy(inst);
 728                         }
 729                 }
 730                 write_unlock_bh(&instances_lock);
 731         }
 732         return NOTIFY_DONE;
 733 }
 734
 735 static struct notifier_block nfulnl_rtnl_notifier = {
 736         .notifier_call  = nfulnl_rcv_nl_event,
 737 };
 738
 739 static int
 740 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 741                   struct nlmsghdr *nlh, struct nfattr *nfqa[])
 742 {
 743         return -ENOTSUPP;
 744 }
 745
 746 static struct nf_logger nfulnl_logger = {
 747         .name   = "nfnetlink_log",
 748         .logfn  = &nfulnl_log_packet,
 749         .me     = THIS_MODULE,
 750 };
 751
 752 static const int nfula_min[NFULA_MAX] = {
 753         [NFULA_PACKET_HDR-1]    = sizeof(struct nfulnl_msg_packet_hdr),
 754         [NFULA_MARK-1]          = sizeof(u_int32_t),
 755         [NFULA_TIMESTAMP-1]     = sizeof(struct nfulnl_msg_packet_timestamp),
 756         [NFULA_IFINDEX_INDEV-1] = sizeof(u_int32_t),
 757         [NFULA_IFINDEX_OUTDEV-1]= sizeof(u_int32_t),
 758         [NFULA_IFINDEX_PHYSINDEV-1]     = sizeof(u_int32_t),
 759         [NFULA_IFINDEX_PHYSOUTDEV-1]    = sizeof(u_int32_t),
 760         [NFULA_HWADDR-1]        = sizeof(struct nfulnl_msg_packet_hw),
 761         [NFULA_PAYLOAD-1]       = 0,
 762         [NFULA_PREFIX-1]        = 0,
 763         [NFULA_UID-1]           = sizeof(u_int32_t),
 764         [NFULA_SEQ-1]           = sizeof(u_int32_t),
 765         [NFULA_SEQ_GLOBAL-1]    = sizeof(u_int32_t),
 766 };
 767
 768 static const int nfula_cfg_min[NFULA_CFG_MAX] = {
 769         [NFULA_CFG_CMD-1]       = sizeof(struct nfulnl_msg_config_cmd),
 770         [NFULA_CFG_MODE-1]      = sizeof(struct nfulnl_msg_config_mode),
 771         [NFULA_CFG_TIMEOUT-1]   = sizeof(u_int32_t),
 772         [NFULA_CFG_QTHRESH-1]   = sizeof(u_int32_t),
 773         [NFULA_CFG_NLBUFSIZ-1]  = sizeof(u_int32_t),
 774         [NFULA_CFG_FLAGS-1]     = sizeof(u_int16_t),
 775 };
 776
 777 static int
 778 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 779                    struct nlmsghdr *nlh, struct nfattr *nfula[])
 780 {
 781         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 782         u_int16_t group_num = ntohs(nfmsg->res_id);
 783         struct nfulnl_instance *inst;
 784         int ret = 0;
 785
 786         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
 787
 788         if (nfattr_bad_size(nfula, NFULA_CFG_MAX, nfula_cfg_min)) {
 789                 UDEBUG("bad attribute size\n");
 790                 return -EINVAL;
 791         }
 792
 793         inst = instance_lookup_get(group_num);
 794         if (nfula[NFULA_CFG_CMD-1]) {
 795                 u_int8_t pf = nfmsg->nfgen_family;
 796                 struct nfulnl_msg_config_cmd *cmd;
 797                 cmd = NFA_DATA(nfula[NFULA_CFG_CMD-1]);
 798                 UDEBUG("found CFG_CMD for\n");
 799
 800                 switch (cmd->command) {
 801                 case NFULNL_CFG_CMD_BIND:
 802                         if (inst) {
 803                                 ret = -EBUSY;
 804                                 goto out_put;
 805                         }
 806
 807                         inst = instance_create(group_num,
 808                                                NETLINK_CB(skb).pid);
 809                         if (!inst) {
 810                                 ret = -EINVAL;
 811                                 goto out;
 812                         }
 813                         break;
 814                 case NFULNL_CFG_CMD_UNBIND:
 815                         if (!inst) {
 816                                 ret = -ENODEV;
 817                                 goto out;
 818                         }
 819
 820                         if (inst->peer_pid != NETLINK_CB(skb).pid) {
 821                                 ret = -EPERM;
 822                                 goto out_put;
 823                         }
 824
 825                         instance_destroy(inst);
 826                         goto out;
 827                 case NFULNL_CFG_CMD_PF_BIND:
 828                         UDEBUG("registering log handler for pf=%u\n", pf);
 829                         ret = nf_log_register(pf, &nfulnl_logger);
 830                         break;
 831                 case NFULNL_CFG_CMD_PF_UNBIND:
 832                         UDEBUG("unregistering log handler for pf=%u\n", pf);
 833                         /* This is a bug and a feature.  We cannot unregister
 834                          * other handlers, like nfnetlink_inst can */
 835                         nf_log_unregister_pf(pf);
 836                         break;
 837                 default:
 838                         ret = -EINVAL;
 839                         break;
 840                 }
 841
 842                 if (!inst)
 843                         goto out;
 844         } else {
 845                 if (!inst) {
 846                         UDEBUG("no config command, and no instance for "
 847                                 "group=%u pid=%u =>ENOENT\n",
 848                                 group_num, NETLINK_CB(skb).pid);
 849                         ret = -ENOENT;
 850                         goto out;
 851                 }
 852
 853                 if (inst->peer_pid != NETLINK_CB(skb).pid) {
 854                         UDEBUG("no config command, and wrong pid\n");
 855                         ret = -EPERM;
 856                         goto out_put;
 857                 }
 858         }
 859
 860         if (nfula[NFULA_CFG_MODE-1]) {
 861                 struct nfulnl_msg_config_mode *params;
 862                 params = NFA_DATA(nfula[NFULA_CFG_MODE-1]);
 863
 864                 nfulnl_set_mode(inst, params->copy_mode,
 865                                 ntohl(params->copy_range));
 866         }
 867
 868         if (nfula[NFULA_CFG_TIMEOUT-1]) {
 869                 __be32 timeout =
 870                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_TIMEOUT-1]);
 871
 872                 nfulnl_set_timeout(inst, ntohl(timeout));
 873         }
 874
 875         if (nfula[NFULA_CFG_NLBUFSIZ-1]) {
 876                 __be32 nlbufsiz =
 877                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_NLBUFSIZ-1]);
 878
 879                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 880         }
 881
 882         if (nfula[NFULA_CFG_QTHRESH-1]) {
 883                 __be32 qthresh =
 884                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_QTHRESH-1]);
 885
 886                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 887         }
 888
 889         if (nfula[NFULA_CFG_FLAGS-1]) {
 890                 __be16 flags =
 891                         *(__be16 *)NFA_DATA(nfula[NFULA_CFG_FLAGS-1]);
 892                 nfulnl_set_flags(inst, ntohs(flags));
 893         }
 894
 895 out_put:
 896         instance_put(inst);
 897 out:
 898         return ret;
 899 }
 900
 901 static struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 902         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 903                                     .attr_count = NFULA_MAX, },
 904         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 905                                     .attr_count = NFULA_CFG_MAX, },
 906 };
 907
 908 static struct nfnetlink_subsystem nfulnl_subsys = {
 909         .name           = "log",
 910         .subsys_id      = NFNL_SUBSYS_ULOG,
 911         .cb_count       = NFULNL_MSG_MAX,
 912         .cb             = nfulnl_cb,
 913 };
 914
 915 #ifdef CONFIG_PROC_FS
 916 struct iter_state {
 917         unsigned int bucket;
 918 };
 919
 920 static struct hlist_node *get_first(struct iter_state *st)
 921 {
 922         if (!st)
 923                 return NULL;
 924
 925         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 926                 if (!hlist_empty(&instance_table[st->bucket]))
 927                         return instance_table[st->bucket].first;
 928         }
 929         return NULL;
 930 }
 931
 932 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
 933 {
 934         h = h->next;
 935         while (!h) {
 936                 if (++st->bucket >= INSTANCE_BUCKETS)
 937                         return NULL;
 938
 939                 h = instance_table[st->bucket].first;
 940         }
 941         return h;
 942 }
 943
 944 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
 945 {
 946         struct hlist_node *head;
 947         head = get_first(st);
 948
 949         if (head)
 950                 while (pos && (head = get_next(st, head)))
 951                         pos--;
 952         return pos ? NULL : head;
 953 }
 954
 955 static void *seq_start(struct seq_file *seq, loff_t *pos)
 956 {
 957         read_lock_bh(&instances_lock);
 958         return get_idx(seq->private, *pos);
 959 }
 960
 961 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 962 {
 963         (*pos)++;
 964         return get_next(s->private, v);
 965 }
 966
 967 static void seq_stop(struct seq_file *s, void *v)
 968 {
 969         read_unlock_bh(&instances_lock);
 970 }
 971
 972 static int seq_show(struct seq_file *s, void *v)
 973 {
 974         const struct nfulnl_instance *inst = v;
 975
 976         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 977                           inst->group_num,
 978                           inst->peer_pid, inst->qlen,
 979                           inst->copy_mode, inst->copy_range,
 980                           inst->flushtimeout, atomic_read(&inst->use));
 981 }
 982
 983 static struct seq_operations nful_seq_ops = {
 984         .start  = seq_start,
 985         .next   = seq_next,
 986         .stop   = seq_stop,
 987         .show   = seq_show,
 988 };
 989
 990 static int nful_open(struct inode *inode, struct file *file)
 991 {
 992         struct seq_file *seq;
 993         struct iter_state *is;
 994         int ret;
 995
 996         is = kzalloc(sizeof(*is), GFP_KERNEL);
 997         if (!is)
 998                 return -ENOMEM;
 999         ret = seq_open(file, &nful_seq_ops);
1000         if (ret < 0)
1001                 goto out_free;
1002         seq = file->private_data;
1003         seq->private = is;
1004         return ret;
1005 out_free:
1006         kfree(is);
1007         return ret;
1008 }
1009
1010 static const struct file_operations nful_file_ops = {
1011         .owner   = THIS_MODULE,
1012         .open    = nful_open,
1013         .read    = seq_read,
1014         .llseek  = seq_lseek,
1015         .release = seq_release_private,
1016 };
1017
1018 #endif /* PROC_FS */
1019
1020 static int __init nfnetlink_log_init(void)
1021 {
1022         int i, status = -ENOMEM;
1023 #ifdef CONFIG_PROC_FS
1024         struct proc_dir_entry *proc_nful;
1025 #endif
1026
1027         for (i = 0; i < INSTANCE_BUCKETS; i++)
1028                 INIT_HLIST_HEAD(&instance_table[i]);
1029
1030         /* it's not really all that important to have a random value, so
1031          * we can do this from the init function, even if there hasn't
1032          * been that much entropy yet */
1033         get_random_bytes(&hash_init, sizeof(hash_init));
1034
1035         netlink_register_notifier(&nfulnl_rtnl_notifier);
1036         status = nfnetlink_subsys_register(&nfulnl_subsys);
1037         if (status < 0) {
1038                 printk(KERN_ERR "log: failed to create netlink socket\n");
1039                 goto cleanup_netlink_notifier;
1040         }
1041
1042 #ifdef CONFIG_PROC_FS
1043         proc_nful = create_proc_entry("nfnetlink_log", 0440,
1044                                       proc_net_netfilter);
1045         if (!proc_nful)
1046                 goto cleanup_subsys;
1047         proc_nful->proc_fops = &nful_file_ops;
1048 #endif
1049         return status;
1050
1051 #ifdef CONFIG_PROC_FS
1052 cleanup_subsys:
1053         nfnetlink_subsys_unregister(&nfulnl_subsys);
1054 #endif
1055 cleanup_netlink_notifier:
1056         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1057         return status;
1058 }
1059
1060 static void __exit nfnetlink_log_fini(void)
1061 {
1062         nf_log_unregister(&nfulnl_logger);
1063 #ifdef CONFIG_PROC_FS
1064         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1065 #endif
1066         nfnetlink_subsys_unregister(&nfulnl_subsys);
1067         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1068 }
1069
1070 MODULE_DESCRIPTION("netfilter userspace logging");
1071 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1072 MODULE_LICENSE("GPL");
1073 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1074
1075 module_init(nfnetlink_log_init);
1076 module_exit(nfnetlink_log_fini);