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  */
  15 #include <linux/module.h>
  16 #include <linux/skbuff.h>
  17 #include <linux/init.h>
  18 #include <linux/ip.h>
  19 #include <linux/ipv6.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/netfilter.h>
  22 #include <linux/netlink.h>
  23 #include <linux/netfilter/nfnetlink.h>
  24 #include <linux/netfilter/nfnetlink_log.h>
  25 #include <linux/spinlock.h>
  26 #include <linux/sysctl.h>
  27 #include <linux/proc_fs.h>
  28 #include <linux/security.h>
  29 #include <linux/list.h>
  30 #include <linux/jhash.h>
  31 #include <linux/random.h>
  32 #include <net/sock.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  100     /* every second */
  42 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  43
  44 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  45                                      printk(x, ## args); } while (0);
  46
  47 #if 0
  48 #define UDEBUG(x, args ...)     printk(KERN_DEBUG "%s(%d):%s(): " x,       \
  49                                         __FILE__, __LINE__, __FUNCTION__,  \
  50                                         ## args)
  51 #else
  52 #define UDEBUG(x, ...)
  53 #endif
  54
  55 struct nfulnl_instance {
  56         struct hlist_node hlist;        /* global list of instances */
  57         spinlock_t lock;
  58         atomic_t use;                   /* use count */
  59
  60         unsigned int qlen;              /* number of nlmsgs in skb */
  61         struct sk_buff *skb;            /* pre-allocatd skb */
  62         struct nlmsghdr *lastnlh;       /* netlink header of last msg in 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_int16_t group_num;            /* number of this queue */
  72         u_int8_t copy_mode;
  73 };
  74
  75 static DEFINE_RWLOCK(instances_lock);
  76
  77 #define INSTANCE_BUCKETS        16
  78 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  79 static unsigned int hash_init;
  80
  81 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  82 {
  83         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  84 }
  85
  86 static struct nfulnl_instance *
  87 __instance_lookup(u_int16_t group_num)
  88 {
  89         struct hlist_head *head;
  90         struct hlist_node *pos;
  91         struct nfulnl_instance *inst;
  92
  93         UDEBUG("entering (group_num=%u)\n", group_num);
  94
  95         head = &instance_table[instance_hashfn(group_num)];
  96         hlist_for_each_entry(inst, pos, head, hlist) {
  97                 if (inst->group_num == group_num)
  98                         return inst;
  99         }
 100         return NULL;
 101 }
 102
 103 static inline void
 104 instance_get(struct nfulnl_instance *inst)
 105 {
 106         atomic_inc(&inst->use);
 107 }
 108
 109 static struct nfulnl_instance *
 110 instance_lookup_get(u_int16_t group_num)
 111 {
 112         struct nfulnl_instance *inst;
 113
 114         read_lock_bh(&instances_lock);
 115         inst = __instance_lookup(group_num);
 116         if (inst)
 117                 instance_get(inst);
 118         read_unlock_bh(&instances_lock);
 119
 120         return inst;
 121 }
 122
 123 static void
 124 instance_put(struct nfulnl_instance *inst)
 125 {
 126         if (inst && atomic_dec_and_test(&inst->use)) {
 127                 UDEBUG("kfree(inst=%p)\n", inst);
 128                 kfree(inst);
 129         }
 130 }
 131
 132 static void nfulnl_timer(unsigned long data);
 133
 134 static struct nfulnl_instance *
 135 instance_create(u_int16_t group_num, int pid)
 136 {
 137         struct nfulnl_instance *inst;
 138
 139         UDEBUG("entering (group_num=%u, pid=%d)\n", group_num,
 140                 pid);
 141
 142         write_lock_bh(&instances_lock);
 143         if (__instance_lookup(group_num)) {
 144                 inst = NULL;
 145                 UDEBUG("aborting, instance already exists\n");
 146                 goto out_unlock;
 147         }
 148
 149         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 150         if (!inst)
 151                 goto out_unlock;
 152
 153         INIT_HLIST_NODE(&inst->hlist);
 154         spin_lock_init(&inst->lock);
 155         /* needs to be two, since we _put() after creation */
 156         atomic_set(&inst->use, 2);
 157
 158         init_timer(&inst->timer);
 159         inst->timer.function = nfulnl_timer;
 160         inst->timer.data = (unsigned long)inst;
 161         /* don't start timer yet. (re)start it  with every packet */
 162
 163         inst->peer_pid = pid;
 164         inst->group_num = group_num;
 165
 166         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 167         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 168         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 169         inst->copy_mode         = NFULNL_COPY_PACKET;
 170         inst->copy_range        = 0xffff;
 171
 172         if (!try_module_get(THIS_MODULE))
 173                 goto out_free;
 174
 175         hlist_add_head(&inst->hlist,
 176                        &instance_table[instance_hashfn(group_num)]);
 177
 178         UDEBUG("newly added node: %p, next=%p\n", &inst->hlist,
 179                 inst->hlist.next);
 180
 181         write_unlock_bh(&instances_lock);
 182
 183         return inst;
 184
 185 out_free:
 186         instance_put(inst);
 187 out_unlock:
 188         write_unlock_bh(&instances_lock);
 189         return NULL;
 190 }
 191
 192 static int __nfulnl_send(struct nfulnl_instance *inst);
 193
 194 static void
 195 _instance_destroy2(struct nfulnl_instance *inst, int lock)
 196 {
 197         /* first pull it out of the global list */
 198         if (lock)
 199                 write_lock_bh(&instances_lock);
 200
 201         UDEBUG("removing instance %p (queuenum=%u) from hash\n",
 202                 inst, inst->group_num);
 203
 204         hlist_del(&inst->hlist);
 205
 206         if (lock)
 207                 write_unlock_bh(&instances_lock);
 208
 209         /* then flush all pending packets from skb */
 210
 211         spin_lock_bh(&inst->lock);
 212         if (inst->skb) {
 213                 if (inst->qlen)
 214                         __nfulnl_send(inst);
 215                 if (inst->skb) {
 216                         kfree_skb(inst->skb);
 217                         inst->skb = NULL;
 218                 }
 219         }
 220         spin_unlock_bh(&inst->lock);
 221
 222         /* and finally put the refcount */
 223         instance_put(inst);
 224
 225         module_put(THIS_MODULE);
 226 }
 227
 228 static inline void
 229 __instance_destroy(struct nfulnl_instance *inst)
 230 {
 231         _instance_destroy2(inst, 0);
 232 }
 233
 234 static inline void
 235 instance_destroy(struct nfulnl_instance *inst)
 236 {
 237         _instance_destroy2(inst, 1);
 238 }
 239
 240 static int
 241 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 242                   unsigned int range)
 243 {
 244         int status = 0;
 245
 246         spin_lock_bh(&inst->lock);
 247
 248         switch (mode) {
 249         case NFULNL_COPY_NONE:
 250         case NFULNL_COPY_META:
 251                 inst->copy_mode = mode;
 252                 inst->copy_range = 0;
 253                 break;
 254
 255         case NFULNL_COPY_PACKET:
 256                 inst->copy_mode = mode;
 257                 /* we're using struct nfattr which has 16bit nfa_len */
 258                 if (range > 0xffff)
 259                         inst->copy_range = 0xffff;
 260                 else
 261                         inst->copy_range = range;
 262                 break;
 263
 264         default:
 265                 status = -EINVAL;
 266                 break;
 267         }
 268
 269         spin_unlock_bh(&inst->lock);
 270
 271         return status;
 272 }
 273
 274 static int
 275 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 276 {
 277         int status;
 278
 279         spin_lock_bh(&inst->lock);
 280         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 281                 status = -ERANGE;
 282         else if (nlbufsiz > 131072)
 283                 status = -ERANGE;
 284         else {
 285                 inst->nlbufsiz = nlbufsiz;
 286                 status = 0;
 287         }
 288         spin_unlock_bh(&inst->lock);
 289
 290         return status;
 291 }
 292
 293 static int
 294 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 295 {
 296         spin_lock_bh(&inst->lock);
 297         inst->flushtimeout = timeout;
 298         spin_unlock_bh(&inst->lock);
 299
 300         return 0;
 301 }
 302
 303 static int
 304 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 305 {
 306         spin_lock_bh(&inst->lock);
 307         inst->qthreshold = qthresh;
 308         spin_unlock_bh(&inst->lock);
 309
 310         return 0;
 311 }
 312
 313 static struct sk_buff *nfulnl_alloc_skb(unsigned int inst_size,
 314                                         unsigned int pkt_size)
 315 {
 316         struct sk_buff *skb;
 317         unsigned int n;
 318
 319         UDEBUG("entered (%u, %u)\n", inst_size, pkt_size);
 320
 321         /* alloc skb which should be big enough for a whole multipart
 322          * message.  WARNING: has to be <= 128k due to slab restrictions */
 323
 324         n = max(inst_size, pkt_size);
 325         skb = alloc_skb(n, GFP_ATOMIC);
 326         if (!skb) {
 327                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
 328                         inst_size);
 329
 330                 if (n > pkt_size) {
 331                         /* try to allocate only as much as we need for current
 332                          * packet */
 333
 334                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 335                         if (!skb)
 336                                 PRINTR("nfnetlink_log: can't even alloc %u "
 337                                        "bytes\n", pkt_size);
 338                 }
 339         }
 340
 341         return skb;
 342 }
 343
 344 static int
 345 __nfulnl_send(struct nfulnl_instance *inst)
 346 {
 347         int status;
 348
 349         if (timer_pending(&inst->timer))
 350                 del_timer(&inst->timer);
 351
 352         if (inst->qlen > 1)
 353                 inst->lastnlh->nlmsg_type = NLMSG_DONE;
 354
 355         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
 356         if (status < 0) {
 357                 UDEBUG("netlink_unicast() failed\n");
 358                 /* FIXME: statistics */
 359         }
 360
 361         inst->qlen = 0;
 362         inst->skb = NULL;
 363         inst->lastnlh = NULL;
 364
 365         return status;
 366 }
 367
 368 static void 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         __nfulnl_send(inst);
 376         instance_put(inst);
 377         spin_unlock_bh(&inst->lock);
 378 }
 379
 380 static inline int
 381 __build_packet_message(struct nfulnl_instance *inst,
 382                         const struct sk_buff *skb,
 383                         unsigned int data_len,
 384                         unsigned int pf,
 385                         unsigned int hooknum,
 386                         const struct net_device *indev,
 387                         const struct net_device *outdev,
 388                         const struct nf_loginfo *li,
 389                         const char *prefix)
 390 {
 391         unsigned char *old_tail;
 392         struct nfulnl_msg_packet_hdr pmsg;
 393         struct nlmsghdr *nlh;
 394         struct nfgenmsg *nfmsg;
 395         u_int32_t tmp_uint;
 396
 397         UDEBUG("entered\n");
 398
 399         old_tail = inst->skb->tail;
 400         nlh = NLMSG_PUT(inst->skb, 0, 0,
 401                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 402                         sizeof(struct nfgenmsg));
 403         nfmsg = NLMSG_DATA(nlh);
 404         nfmsg->nfgen_family = pf;
 405         nfmsg->version = NFNETLINK_V0;
 406         nfmsg->res_id = htons(inst->group_num);
 407
 408         pmsg.hw_protocol        = htons(skb->protocol);
 409         pmsg.hook               = hooknum;
 410
 411         NFA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
 412
 413         if (prefix) {
 414                 int slen = strlen(prefix);
 415                 if (slen > NFULNL_PREFIXLEN)
 416                         slen = NFULNL_PREFIXLEN;
 417                 NFA_PUT(inst->skb, NFULA_PREFIX, slen, prefix);
 418         }
 419
 420         if (indev) {
 421                 tmp_uint = htonl(indev->ifindex);
 422 #ifndef CONFIG_BRIDGE_NETFILTER
 423                 NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
 424                         &tmp_uint);
 425 #else
 426                 if (pf == PF_BRIDGE) {
 427                         /* Case 1: outdev is physical input device, we need to
 428                          * look for bridge group (when called from
 429                          * netfilter_bridge) */
 430                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 431                                 sizeof(tmp_uint), &tmp_uint);
 432                         /* this is the bridge group "brX" */
 433                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
 434                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 435                                 sizeof(tmp_uint), &tmp_uint);
 436                 } else {
 437                         /* Case 2: indev is bridge group, we need to look for
 438                          * physical device (when called from ipv4) */
 439                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 440                                 sizeof(tmp_uint), &tmp_uint);
 441                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
 442                                 tmp_uint =
 443                                     htonl(skb->nf_bridge->physindev->ifindex);
 444                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 445                                         sizeof(tmp_uint), &tmp_uint);
 446                         }
 447                 }
 448 #endif
 449         }
 450
 451         if (outdev) {
 452                 tmp_uint = htonl(outdev->ifindex);
 453 #ifndef CONFIG_BRIDGE_NETFILTER
 454                 NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 455                         &tmp_uint);
 456 #else
 457                 if (pf == PF_BRIDGE) {
 458                         /* Case 1: outdev is physical output device, we need to
 459                          * look for bridge group (when called from
 460                          * netfilter_bridge) */
 461                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 462                                 sizeof(tmp_uint), &tmp_uint);
 463                         /* this is the bridge group "brX" */
 464                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
 465                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 466                                 sizeof(tmp_uint), &tmp_uint);
 467                 } else {
 468                         /* Case 2: indev is a bridge group, we need to look
 469                          * for physical device (when called from ipv4) */
 470                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 471                                 sizeof(tmp_uint), &tmp_uint);
 472                         if (skb->nf_bridge) {
 473                                 tmp_uint =
 474                                     htonl(skb->nf_bridge->physoutdev->ifindex);
 475                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 476                                         sizeof(tmp_uint), &tmp_uint);
 477                         }
 478                 }
 479 #endif
 480         }
 481
 482         if (skb->nfmark) {
 483                 tmp_uint = htonl(skb->nfmark);
 484                 NFA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
 485         }
 486
 487         if (indev && skb->dev && skb->dev->hard_header_parse) {
 488                 struct nfulnl_msg_packet_hw phw;
 489
 490                 phw.hw_addrlen =
 491                         skb->dev->hard_header_parse((struct sk_buff *)skb,
 492                                                     phw.hw_addr);
 493                 phw.hw_addrlen = htons(phw.hw_addrlen);
 494                 NFA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
 495         }
 496
 497         if (skb->tstamp.off_sec) {
 498                 struct nfulnl_msg_packet_timestamp ts;
 499
 500                 ts.sec = cpu_to_be64(skb->tstamp.off_sec);
 501                 ts.usec = cpu_to_be64(skb->tstamp.off_usec);
 502
 503                 NFA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 504         }
 505
 506         /* UID */
 507         if (skb->sk) {
 508                 read_lock_bh(&skb->sk->sk_callback_lock);
 509                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 510                         u_int32_t uid = htonl(skb->sk->sk_socket->file->f_uid);
 511                         /* need to unlock here since NFA_PUT may goto */
 512                         read_unlock_bh(&skb->sk->sk_callback_lock);
 513                         NFA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
 514                 } else
 515                         read_unlock_bh(&skb->sk->sk_callback_lock);
 516         }
 517
 518         if (data_len) {
 519                 struct nfattr *nfa;
 520                 int size = NFA_LENGTH(data_len);
 521
 522                 if (skb_tailroom(inst->skb) < (int)NFA_SPACE(data_len)) {
 523                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 524                         goto nlmsg_failure;
 525                 }
 526
 527                 nfa = (struct nfattr *)skb_put(inst->skb, NFA_ALIGN(size));
 528                 nfa->nfa_type = NFULA_PAYLOAD;
 529                 nfa->nfa_len = size;
 530
 531                 if (skb_copy_bits(skb, 0, NFA_DATA(nfa), data_len))
 532                         BUG();
 533         }
 534
 535         nlh->nlmsg_len = inst->skb->tail - old_tail;
 536         return 0;
 537
 538 nlmsg_failure:
 539         UDEBUG("nlmsg_failure\n");
 540 nfattr_failure:
 541         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 542         return -1;
 543 }
 544
 545 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 546
 547 static struct nf_loginfo default_loginfo = {
 548         .type =         NF_LOG_TYPE_ULOG,
 549         .u = {
 550                 .ulog = {
 551                         .copy_len       = 0xffff,
 552                         .group          = 0,
 553                         .qthreshold     = 1,
 554                 },
 555         },
 556 };
 557
 558 /* log handler for internal netfilter logging api */
 559 static void
 560 nfulnl_log_packet(unsigned int pf,
 561                   unsigned int hooknum,
 562                   const struct sk_buff *skb,
 563                   const struct net_device *in,
 564                   const struct net_device *out,
 565                   const struct nf_loginfo *li_user,
 566                   const char *prefix)
 567 {
 568         unsigned int size, data_len;
 569         struct nfulnl_instance *inst;
 570         const struct nf_loginfo *li;
 571         unsigned int qthreshold;
 572         unsigned int nlbufsiz;
 573
 574         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 575                 li = li_user;
 576         else
 577                 li = &default_loginfo;
 578
 579         inst = instance_lookup_get(li->u.ulog.group);
 580         if (!inst)
 581                 inst = instance_lookup_get(0);
 582         if (!inst) {
 583                 PRINTR("nfnetlink_log: trying to log packet, "
 584                         "but no instance for group %u\n", li->u.ulog.group);
 585                 return;
 586         }
 587
 588         /* all macros expand to constant values at compile time */
 589         /* FIXME: do we want to make the size calculation conditional based on
 590          * what is actually present?  way more branches and checks, but more
 591          * memory efficient... */
 592         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 593                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hdr))
 594                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 595                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 596 #ifdef CONFIG_BRIDGE_NETFILTER
 597                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 598                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 599 #endif
 600                 + NFA_SPACE(sizeof(u_int32_t))  /* mark */
 601                 + NFA_SPACE(sizeof(u_int32_t))  /* uid */
 602                 + NFA_SPACE(NFULNL_PREFIXLEN)   /* prefix */
 603                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hw))
 604                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_timestamp));
 605
 606         UDEBUG("initial size=%u\n", size);
 607
 608         spin_lock_bh(&inst->lock);
 609
 610         qthreshold = inst->qthreshold;
 611         /* per-rule qthreshold overrides per-instance */
 612         if (qthreshold > li->u.ulog.qthreshold)
 613                 qthreshold = li->u.ulog.qthreshold;
 614
 615         switch (inst->copy_mode) {
 616         case NFULNL_COPY_META:
 617         case NFULNL_COPY_NONE:
 618                 data_len = 0;
 619                 break;
 620
 621         case NFULNL_COPY_PACKET:
 622                 if (inst->copy_range == 0
 623                     || inst->copy_range > skb->len)
 624                         data_len = skb->len;
 625                 else
 626                         data_len = inst->copy_range;
 627
 628                 size += NFA_SPACE(data_len);
 629                 UDEBUG("copy_packet, therefore size now %u\n", size);
 630                 break;
 631
 632         default:
 633                 spin_unlock_bh(&inst->lock);
 634                 instance_put(inst);
 635                 return;
 636         }
 637
 638         if (size > inst->nlbufsiz)
 639                 nlbufsiz = size;
 640         else
 641                 nlbufsiz = inst->nlbufsiz;
 642
 643         if (!inst->skb) {
 644                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
 645                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
 646                                 inst->nlbufsiz, size);
 647                         goto alloc_failure;
 648                 }
 649         } else if (inst->qlen >= qthreshold ||
 650                    size > skb_tailroom(inst->skb)) {
 651                 /* either the queue len is too high or we don't have
 652                  * enough room in the skb left. flush to userspace. */
 653                 UDEBUG("flushing old skb\n");
 654
 655                 __nfulnl_send(inst);
 656
 657                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
 658                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
 659                                 inst->nlbufsiz, size);
 660                         goto alloc_failure;
 661                 }
 662         }
 663
 664         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
 665         inst->qlen++;
 666
 667         __build_packet_message(inst, skb, data_len, pf,
 668                                 hooknum, in, out, li, prefix);
 669
 670         /* timer_pending always called within inst->lock, so there
 671          * is no chance of a race here */
 672         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         spin_unlock_bh(&inst->lock);
 678
 679         return;
 680
 681 alloc_failure:
 682         spin_unlock_bh(&inst->lock);
 683         instance_put(inst);
 684         UDEBUG("error allocating skb\n");
 685         /* FIXME: statistics */
 686 }
 687
 688 static int
 689 nfulnl_rcv_nl_event(struct notifier_block *this,
 690                    unsigned long event, void *ptr)
 691 {
 692         struct netlink_notify *n = ptr;
 693
 694         if (event == NETLINK_URELEASE &&
 695             n->protocol == NETLINK_NETFILTER && n->pid) {
 696                 int i;
 697
 698                 /* destroy all instances for this pid */
 699                 write_lock_bh(&instances_lock);
 700                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 701                         struct hlist_node *tmp, *t2;
 702                         struct nfulnl_instance *inst;
 703                         struct hlist_head *head = &instance_table[i];
 704
 705                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 706                                 UDEBUG("node = %p\n", inst);
 707                                 if (n->pid == inst->peer_pid)
 708                                         __instance_destroy(inst);
 709                         }
 710                 }
 711                 write_unlock_bh(&instances_lock);
 712         }
 713         return NOTIFY_DONE;
 714 }
 715
 716 static struct notifier_block nfulnl_rtnl_notifier = {
 717         .notifier_call  = nfulnl_rcv_nl_event,
 718 };
 719
 720 static int
 721 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 722                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
 723 {
 724         return -ENOTSUPP;
 725 }
 726
 727 static struct nf_logger nfulnl_logger = {
 728         .name   = "nfnetlink_log",
 729         .logfn  = &nfulnl_log_packet,
 730         .me     = THIS_MODULE,
 731 };
 732
 733 static const int nfula_min[NFULA_MAX] = {
 734         [NFULA_PACKET_HDR-1]    = sizeof(struct nfulnl_msg_packet_hdr),
 735         [NFULA_MARK-1]          = sizeof(u_int32_t),
 736         [NFULA_TIMESTAMP-1]     = sizeof(struct nfulnl_msg_packet_timestamp),
 737         [NFULA_IFINDEX_INDEV-1] = sizeof(u_int32_t),
 738         [NFULA_IFINDEX_OUTDEV-1]= sizeof(u_int32_t),
 739         [NFULA_HWADDR-1]        = sizeof(struct nfulnl_msg_packet_hw),
 740         [NFULA_PAYLOAD-1]       = 0,
 741         [NFULA_PREFIX-1]        = 0,
 742         [NFULA_UID-1]           = sizeof(u_int32_t),
 743 };
 744
 745 static const int nfula_cfg_min[NFULA_CFG_MAX] = {
 746         [NFULA_CFG_CMD-1]       = sizeof(struct nfulnl_msg_config_cmd),
 747         [NFULA_CFG_MODE-1]      = sizeof(struct nfulnl_msg_config_mode),
 748         [NFULA_CFG_TIMEOUT-1]   = sizeof(u_int32_t),
 749         [NFULA_CFG_QTHRESH-1]   = sizeof(u_int32_t),
 750         [NFULA_CFG_NLBUFSIZ-1]  = sizeof(u_int32_t),
 751 };
 752
 753 static int
 754 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 755                    struct nlmsghdr *nlh, struct nfattr *nfula[], int *errp)
 756 {
 757         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 758         u_int16_t group_num = ntohs(nfmsg->res_id);
 759         struct nfulnl_instance *inst;
 760         int ret = 0;
 761
 762         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
 763
 764         if (nfattr_bad_size(nfula, NFULA_CFG_MAX, nfula_cfg_min)) {
 765                 UDEBUG("bad attribute size\n");
 766                 return -EINVAL;
 767         }
 768
 769         inst = instance_lookup_get(group_num);
 770         if (nfula[NFULA_CFG_CMD-1]) {
 771                 u_int8_t pf = nfmsg->nfgen_family;
 772                 struct nfulnl_msg_config_cmd *cmd;
 773                 cmd = NFA_DATA(nfula[NFULA_CFG_CMD-1]);
 774                 UDEBUG("found CFG_CMD for\n");
 775
 776                 switch (cmd->command) {
 777                 case NFULNL_CFG_CMD_BIND:
 778                         if (inst) {
 779                                 ret = -EBUSY;
 780                                 goto out_put;
 781                         }
 782
 783                         inst = instance_create(group_num,
 784                                                NETLINK_CB(skb).pid);
 785                         if (!inst) {
 786                                 ret = -EINVAL;
 787                                 goto out_put;
 788                         }
 789                         break;
 790                 case NFULNL_CFG_CMD_UNBIND:
 791                         if (!inst) {
 792                                 ret = -ENODEV;
 793                                 goto out_put;
 794                         }
 795
 796                         if (inst->peer_pid != NETLINK_CB(skb).pid) {
 797                                 ret = -EPERM;
 798                                 goto out_put;
 799                         }
 800
 801                         instance_destroy(inst);
 802                         break;
 803                 case NFULNL_CFG_CMD_PF_BIND:
 804                         UDEBUG("registering log handler for pf=%u\n", pf);
 805                         ret = nf_log_register(pf, &nfulnl_logger);
 806                         break;
 807                 case NFULNL_CFG_CMD_PF_UNBIND:
 808                         UDEBUG("unregistering log handler for pf=%u\n", pf);
 809                         /* This is a bug and a feature.  We cannot unregister
 810                          * other handlers, like nfnetlink_inst can */
 811                         nf_log_unregister_pf(pf);
 812                         break;
 813                 default:
 814                         ret = -EINVAL;
 815                         break;
 816                 }
 817         } else {
 818                 if (!inst) {
 819                         UDEBUG("no config command, and no instance for "
 820                                 "group=%u pid=%u =>ENOENT\n",
 821                                 group_num, NETLINK_CB(skb).pid);
 822                         ret = -ENOENT;
 823                         goto out_put;
 824                 }
 825
 826                 if (inst->peer_pid != NETLINK_CB(skb).pid) {
 827                         UDEBUG("no config command, and wrong pid\n");
 828                         ret = -EPERM;
 829                         goto out_put;
 830                 }
 831         }
 832
 833         if (nfula[NFULA_CFG_MODE-1]) {
 834                 struct nfulnl_msg_config_mode *params;
 835                 params = NFA_DATA(nfula[NFULA_CFG_MODE-1]);
 836
 837                 nfulnl_set_mode(inst, params->copy_mode,
 838                                 ntohs(params->copy_range));
 839         }
 840
 841         if (nfula[NFULA_CFG_TIMEOUT-1]) {
 842                 u_int32_t timeout =
 843                         *(u_int32_t *)NFA_DATA(nfula[NFULA_CFG_TIMEOUT-1]);
 844
 845                 nfulnl_set_timeout(inst, ntohl(timeout));
 846         }
 847
 848         if (nfula[NFULA_CFG_NLBUFSIZ-1]) {
 849                 u_int32_t nlbufsiz =
 850                         *(u_int32_t *)NFA_DATA(nfula[NFULA_CFG_NLBUFSIZ-1]);
 851
 852                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 853         }
 854
 855         if (nfula[NFULA_CFG_QTHRESH-1]) {
 856                 u_int32_t qthresh =
 857                         *(u_int16_t *)NFA_DATA(nfula[NFULA_CFG_QTHRESH-1]);
 858
 859                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 860         }
 861
 862 out_put:
 863         instance_put(inst);
 864         return ret;
 865 }
 866
 867 static struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 868         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 869                                     .attr_count = NFULA_MAX, },
 870         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 871                                     .attr_count = NFULA_CFG_MAX, },
 872 };
 873
 874 static struct nfnetlink_subsystem nfulnl_subsys = {
 875         .name           = "log",
 876         .subsys_id      = NFNL_SUBSYS_ULOG,
 877         .cb_count       = NFULNL_MSG_MAX,
 878         .cb             = nfulnl_cb,
 879 };
 880
 881 #ifdef CONFIG_PROC_FS
 882 struct iter_state {
 883         unsigned int bucket;
 884 };
 885
 886 static struct hlist_node *get_first(struct seq_file *seq)
 887 {
 888         struct iter_state *st = seq->private;
 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 seq_file *seq, struct hlist_node *h)
 901 {
 902         struct iter_state *st = seq->private;
 903
 904         h = h->next;
 905         while (!h) {
 906                 if (++st->bucket >= INSTANCE_BUCKETS)
 907                         return NULL;
 908
 909                 h = instance_table[st->bucket].first;
 910         }
 911         return h;
 912 }
 913
 914 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 915 {
 916         struct hlist_node *head;
 917         head = get_first(seq);
 918
 919         if (head)
 920                 while (pos && (head = get_next(seq, head)))
 921                         pos--;
 922         return pos ? NULL : head;
 923 }
 924
 925 static void *seq_start(struct seq_file *seq, loff_t *pos)
 926 {
 927         read_lock_bh(&instances_lock);
 928         return get_idx(seq, *pos);
 929 }
 930
 931 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 932 {
 933         (*pos)++;
 934         return get_next(s, v);
 935 }
 936
 937 static void seq_stop(struct seq_file *s, void *v)
 938 {
 939         read_unlock_bh(&instances_lock);
 940 }
 941
 942 static int seq_show(struct seq_file *s, void *v)
 943 {
 944         const struct nfulnl_instance *inst = v;
 945
 946         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 947                           inst->group_num,
 948                           inst->peer_pid, inst->qlen,
 949                           inst->copy_mode, inst->copy_range,
 950                           inst->flushtimeout, atomic_read(&inst->use));
 951 }
 952
 953 static struct seq_operations nful_seq_ops = {
 954         .start  = seq_start,
 955         .next   = seq_next,
 956         .stop   = seq_stop,
 957         .show   = seq_show,
 958 };
 959
 960 static int nful_open(struct inode *inode, struct file *file)
 961 {
 962         struct seq_file *seq;
 963         struct iter_state *is;
 964         int ret;
 965
 966         is = kzalloc(sizeof(*is), GFP_KERNEL);
 967         if (!is)
 968                 return -ENOMEM;
 969         ret = seq_open(file, &nful_seq_ops);
 970         if (ret < 0)
 971                 goto out_free;
 972         seq = file->private_data;
 973         seq->private = is;
 974         return ret;
 975 out_free:
 976         kfree(is);
 977         return ret;
 978 }
 979
 980 static struct file_operations nful_file_ops = {
 981         .owner   = THIS_MODULE,
 982         .open    = nful_open,
 983         .read    = seq_read,
 984         .llseek  = seq_lseek,
 985         .release = seq_release_private,
 986 };
 987
 988 #endif /* PROC_FS */
 989
 990 static int
 991 init_or_cleanup(int init)
 992 {
 993         int i, status = -ENOMEM;
 994 #ifdef CONFIG_PROC_FS
 995         struct proc_dir_entry *proc_nful;
 996 #endif
 997
 998         if (!init)
 999                 goto cleanup;
1000
1001         for (i = 0; i < INSTANCE_BUCKETS; i++)
1002                 INIT_HLIST_HEAD(&instance_table[i]);
1003
1004         /* it's not really all that important to have a random value, so
1005          * we can do this from the init function, even if there hasn't
1006          * been that much entropy yet */
1007         get_random_bytes(&hash_init, sizeof(hash_init));
1008
1009         netlink_register_notifier(&nfulnl_rtnl_notifier);
1010         status = nfnetlink_subsys_register(&nfulnl_subsys);
1011         if (status < 0) {
1012                 printk(KERN_ERR "log: failed to create netlink socket\n");
1013                 goto cleanup_netlink_notifier;
1014         }
1015
1016 #ifdef CONFIG_PROC_FS
1017         proc_nful = create_proc_entry("nfnetlink_log", 0440,
1018                                       proc_net_netfilter);
1019         if (!proc_nful)
1020                 goto cleanup_subsys;
1021         proc_nful->proc_fops = &nful_file_ops;
1022 #endif
1023
1024         return status;
1025
1026 cleanup:
1027         nf_log_unregister_logger(&nfulnl_logger);
1028 #ifdef CONFIG_PROC_FS
1029         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1030 cleanup_subsys:
1031 #endif
1032         nfnetlink_subsys_unregister(&nfulnl_subsys);
1033 cleanup_netlink_notifier:
1034         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1035         return status;
1036 }
1037
1038 static int __init init(void)
1039 {
1040
1041         return init_or_cleanup(1);
1042 }
1043
1044 static void __exit fini(void)
1045 {
1046         init_or_cleanup(0);
1047 }
1048
1049 MODULE_DESCRIPTION("netfilter userspace logging");
1050 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1051 MODULE_LICENSE("GPL");
1052 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1053
1054 module_init(init);
1055 module_exit(fini);