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->skb)
 353                 return 0;
 354
 355         if (inst->qlen > 1)
 356                 inst->lastnlh->nlmsg_type = NLMSG_DONE;
 357
 358         if (!inst->skb)
 359                 return 0;
 360
 361         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
 362         if (status < 0) {
 363                 UDEBUG("netlink_unicast() failed\n");
 364                 /* FIXME: statistics */
 365         }
 366
 367         inst->qlen = 0;
 368         inst->skb = NULL;
 369         inst->lastnlh = NULL;
 370
 371         return status;
 372 }
 373
 374 static void nfulnl_timer(unsigned long data)
 375 {
 376         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 377
 378         UDEBUG("timer function called, flushing buffer\n");
 379
 380         spin_lock_bh(&inst->lock);
 381         __nfulnl_send(inst);
 382         instance_put(inst);
 383         spin_unlock_bh(&inst->lock);
 384 }
 385
 386 static inline int
 387 __build_packet_message(struct nfulnl_instance *inst,
 388                         const struct sk_buff *skb,
 389                         unsigned int data_len,
 390                         unsigned int pf,
 391                         unsigned int hooknum,
 392                         const struct net_device *indev,
 393                         const struct net_device *outdev,
 394                         const struct nf_loginfo *li,
 395                         const char *prefix)
 396 {
 397         unsigned char *old_tail;
 398         struct nfulnl_msg_packet_hdr pmsg;
 399         struct nlmsghdr *nlh;
 400         struct nfgenmsg *nfmsg;
 401         u_int32_t tmp_uint;
 402
 403         UDEBUG("entered\n");
 404
 405         old_tail = inst->skb->tail;
 406         nlh = NLMSG_PUT(inst->skb, 0, 0,
 407                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 408                         sizeof(struct nfgenmsg));
 409         nfmsg = NLMSG_DATA(nlh);
 410         nfmsg->nfgen_family = pf;
 411         nfmsg->version = NFNETLINK_V0;
 412         nfmsg->res_id = htons(inst->group_num);
 413
 414         pmsg.hw_protocol        = htons(skb->protocol);
 415         pmsg.hook               = hooknum;
 416
 417         NFA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
 418
 419         if (prefix) {
 420                 int slen = strlen(prefix);
 421                 if (slen > NFULNL_PREFIXLEN)
 422                         slen = NFULNL_PREFIXLEN;
 423                 NFA_PUT(inst->skb, NFULA_PREFIX, slen, prefix);
 424         }
 425
 426         if (indev) {
 427                 tmp_uint = htonl(indev->ifindex);
 428 #ifndef CONFIG_BRIDGE_NETFILTER
 429                 NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
 430                         &tmp_uint);
 431 #else
 432                 if (pf == PF_BRIDGE) {
 433                         /* Case 1: outdev is physical input device, we need to
 434                          * look for bridge group (when called from
 435                          * netfilter_bridge) */
 436                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 437                                 sizeof(tmp_uint), &tmp_uint);
 438                         /* this is the bridge group "brX" */
 439                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
 440                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 441                                 sizeof(tmp_uint), &tmp_uint);
 442                 } else {
 443                         /* Case 2: indev is bridge group, we need to look for
 444                          * physical device (when called from ipv4) */
 445                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 446                                 sizeof(tmp_uint), &tmp_uint);
 447                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
 448                                 tmp_uint =
 449                                     htonl(skb->nf_bridge->physindev->ifindex);
 450                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 451                                         sizeof(tmp_uint), &tmp_uint);
 452                         }
 453                 }
 454 #endif
 455         }
 456
 457         if (outdev) {
 458                 tmp_uint = htonl(outdev->ifindex);
 459 #ifndef CONFIG_BRIDGE_NETFILTER
 460                 NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 461                         &tmp_uint);
 462 #else
 463                 if (pf == PF_BRIDGE) {
 464                         /* Case 1: outdev is physical output device, we need to
 465                          * look for bridge group (when called from
 466                          * netfilter_bridge) */
 467                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 468                                 sizeof(tmp_uint), &tmp_uint);
 469                         /* this is the bridge group "brX" */
 470                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
 471                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 472                                 sizeof(tmp_uint), &tmp_uint);
 473                 } else {
 474                         /* Case 2: indev is a bridge group, we need to look
 475                          * for physical device (when called from ipv4) */
 476                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 477                                 sizeof(tmp_uint), &tmp_uint);
 478                         if (skb->nf_bridge) {
 479                                 tmp_uint =
 480                                     htonl(skb->nf_bridge->physoutdev->ifindex);
 481                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 482                                         sizeof(tmp_uint), &tmp_uint);
 483                         }
 484                 }
 485 #endif
 486         }
 487
 488         if (skb->nfmark) {
 489                 tmp_uint = htonl(skb->nfmark);
 490                 NFA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
 491         }
 492
 493         if (indev && skb->dev && skb->dev->hard_header_parse) {
 494                 struct nfulnl_msg_packet_hw phw;
 495
 496                 phw.hw_addrlen =
 497                         skb->dev->hard_header_parse((struct sk_buff *)skb,
 498                                                     phw.hw_addr);
 499                 phw.hw_addrlen = htons(phw.hw_addrlen);
 500                 NFA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
 501         }
 502
 503         if (skb->tstamp.off_sec) {
 504                 struct nfulnl_msg_packet_timestamp ts;
 505
 506                 ts.sec = cpu_to_be64(skb->tstamp.off_sec);
 507                 ts.usec = cpu_to_be64(skb->tstamp.off_usec);
 508
 509                 NFA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 510         }
 511
 512         /* UID */
 513         if (skb->sk) {
 514                 read_lock_bh(&skb->sk->sk_callback_lock);
 515                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 516                         u_int32_t uid = htonl(skb->sk->sk_socket->file->f_uid);
 517                         /* need to unlock here since NFA_PUT may goto */
 518                         read_unlock_bh(&skb->sk->sk_callback_lock);
 519                         NFA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
 520                 } else
 521                         read_unlock_bh(&skb->sk->sk_callback_lock);
 522         }
 523
 524         if (data_len) {
 525                 struct nfattr *nfa;
 526                 int size = NFA_LENGTH(data_len);
 527
 528                 if (skb_tailroom(inst->skb) < (int)NFA_SPACE(data_len)) {
 529                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 530                         goto nlmsg_failure;
 531                 }
 532
 533                 nfa = (struct nfattr *)skb_put(inst->skb, NFA_ALIGN(size));
 534                 nfa->nfa_type = NFULA_PAYLOAD;
 535                 nfa->nfa_len = size;
 536
 537                 if (skb_copy_bits(skb, 0, NFA_DATA(nfa), data_len))
 538                         BUG();
 539         }
 540
 541         nlh->nlmsg_len = inst->skb->tail - old_tail;
 542         return 0;
 543
 544 nlmsg_failure:
 545         UDEBUG("nlmsg_failure\n");
 546 nfattr_failure:
 547         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 548         return -1;
 549 }
 550
 551 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 552
 553 static struct nf_loginfo default_loginfo = {
 554         .type =         NF_LOG_TYPE_ULOG,
 555         .u = {
 556                 .ulog = {
 557                         .copy_len       = 0xffff,
 558                         .group          = 0,
 559                         .qthreshold     = 1,
 560                 },
 561         },
 562 };
 563
 564 /* log handler for internal netfilter logging api */
 565 static void
 566 nfulnl_log_packet(unsigned int pf,
 567                   unsigned int hooknum,
 568                   const struct sk_buff *skb,
 569                   const struct net_device *in,
 570                   const struct net_device *out,
 571                   const struct nf_loginfo *li_user,
 572                   const char *prefix)
 573 {
 574         unsigned int size, data_len;
 575         struct nfulnl_instance *inst;
 576         const struct nf_loginfo *li;
 577         unsigned int qthreshold;
 578         unsigned int nlbufsiz;
 579
 580         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 581                 li = li_user;
 582         else
 583                 li = &default_loginfo;
 584
 585         inst = instance_lookup_get(li->u.ulog.group);
 586         if (!inst)
 587                 inst = instance_lookup_get(0);
 588         if (!inst) {
 589                 PRINTR("nfnetlink_log: trying to log packet, "
 590                         "but no instance for group %u\n", li->u.ulog.group);
 591                 return;
 592         }
 593
 594         /* all macros expand to constant values at compile time */
 595         /* FIXME: do we want to make the size calculation conditional based on
 596          * what is actually present?  way more branches and checks, but more
 597          * memory efficient... */
 598         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 599                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hdr))
 600                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 601                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 602 #ifdef CONFIG_BRIDGE_NETFILTER
 603                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 604                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 605 #endif
 606                 + NFA_SPACE(sizeof(u_int32_t))  /* mark */
 607                 + NFA_SPACE(sizeof(u_int32_t))  /* uid */
 608                 + NFA_SPACE(NFULNL_PREFIXLEN)   /* prefix */
 609                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hw))
 610                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_timestamp));
 611
 612         UDEBUG("initial size=%u\n", size);
 613
 614         spin_lock_bh(&inst->lock);
 615
 616         qthreshold = inst->qthreshold;
 617         /* per-rule qthreshold overrides per-instance */
 618         if (qthreshold > li->u.ulog.qthreshold)
 619                 qthreshold = li->u.ulog.qthreshold;
 620
 621         switch (inst->copy_mode) {
 622         case NFULNL_COPY_META:
 623         case NFULNL_COPY_NONE:
 624                 data_len = 0;
 625                 break;
 626
 627         case NFULNL_COPY_PACKET:
 628                 if (inst->copy_range == 0
 629                     || inst->copy_range > skb->len)
 630                         data_len = skb->len;
 631                 else
 632                         data_len = inst->copy_range;
 633
 634                 size += NFA_SPACE(data_len);
 635                 UDEBUG("copy_packet, therefore size now %u\n", size);
 636                 break;
 637
 638         default:
 639                 spin_unlock_bh(&inst->lock);
 640                 instance_put(inst);
 641                 return;
 642         }
 643
 644         if (size > inst->nlbufsiz)
 645                 nlbufsiz = size;
 646         else
 647                 nlbufsiz = inst->nlbufsiz;
 648
 649         if (!inst->skb) {
 650                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
 651                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
 652                                 inst->nlbufsiz, size);
 653                         goto alloc_failure;
 654                 }
 655         } else if (inst->qlen >= qthreshold ||
 656                    size > skb_tailroom(inst->skb)) {
 657                 /* either the queue len is too high or we don't have
 658                  * enough room in the skb left. flush to userspace. */
 659                 UDEBUG("flushing old skb\n");
 660
 661                 __nfulnl_send(inst);
 662
 663                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
 664                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
 665                                 inst->nlbufsiz, size);
 666                         goto alloc_failure;
 667                 }
 668         }
 669
 670         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
 671         inst->qlen++;
 672
 673         __build_packet_message(inst, skb, data_len, pf,
 674                                 hooknum, in, out, li, prefix);
 675
 676         /* timer_pending always called within inst->lock, so there
 677          * is no chance of a race here */
 678         if (!timer_pending(&inst->timer)) {
 679                 instance_get(inst);
 680                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 681                 add_timer(&inst->timer);
 682         }
 683         spin_unlock_bh(&inst->lock);
 684
 685         return;
 686
 687 alloc_failure:
 688         spin_unlock_bh(&inst->lock);
 689         instance_put(inst);
 690         UDEBUG("error allocating skb\n");
 691         /* FIXME: statistics */
 692 }
 693
 694 static int
 695 nfulnl_rcv_nl_event(struct notifier_block *this,
 696                    unsigned long event, void *ptr)
 697 {
 698         struct netlink_notify *n = ptr;
 699
 700         if (event == NETLINK_URELEASE &&
 701             n->protocol == NETLINK_NETFILTER && n->pid) {
 702                 int i;
 703
 704                 /* destroy all instances for this pid */
 705                 write_lock_bh(&instances_lock);
 706                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 707                         struct hlist_node *tmp, *t2;
 708                         struct nfulnl_instance *inst;
 709                         struct hlist_head *head = &instance_table[i];
 710
 711                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 712                                 UDEBUG("node = %p\n", inst);
 713                                 if (n->pid == inst->peer_pid)
 714                                         __instance_destroy(inst);
 715                         }
 716                 }
 717                 write_unlock_bh(&instances_lock);
 718         }
 719         return NOTIFY_DONE;
 720 }
 721
 722 static struct notifier_block nfulnl_rtnl_notifier = {
 723         .notifier_call  = nfulnl_rcv_nl_event,
 724 };
 725
 726 static int
 727 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 728                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
 729 {
 730         return -ENOTSUPP;
 731 }
 732
 733 static struct nf_logger nfulnl_logger = {
 734         .name   = "nfnetlink_log",
 735         .logfn  = &nfulnl_log_packet,
 736         .me     = THIS_MODULE,
 737 };
 738
 739 static const int nfula_min[NFULA_MAX] = {
 740         [NFULA_PACKET_HDR-1]    = sizeof(struct nfulnl_msg_packet_hdr),
 741         [NFULA_MARK-1]          = sizeof(u_int32_t),
 742         [NFULA_TIMESTAMP-1]     = sizeof(struct nfulnl_msg_packet_timestamp),
 743         [NFULA_IFINDEX_INDEV-1] = sizeof(u_int32_t),
 744         [NFULA_IFINDEX_OUTDEV-1]= sizeof(u_int32_t),
 745         [NFULA_HWADDR-1]        = sizeof(struct nfulnl_msg_packet_hw),
 746         [NFULA_PAYLOAD-1]       = 0,
 747         [NFULA_PREFIX-1]        = 0,
 748         [NFULA_UID-1]           = sizeof(u_int32_t),
 749 };
 750
 751 static const int nfula_cfg_min[NFULA_CFG_MAX] = {
 752         [NFULA_CFG_CMD-1]       = sizeof(struct nfulnl_msg_config_cmd),
 753         [NFULA_CFG_MODE-1]      = sizeof(struct nfulnl_msg_config_mode),
 754         [NFULA_CFG_TIMEOUT-1]   = sizeof(u_int32_t),
 755         [NFULA_CFG_QTHRESH-1]   = sizeof(u_int32_t),
 756         [NFULA_CFG_NLBUFSIZ-1]  = sizeof(u_int32_t),
 757 };
 758
 759 static int
 760 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 761                    struct nlmsghdr *nlh, struct nfattr *nfula[], int *errp)
 762 {
 763         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 764         u_int16_t group_num = ntohs(nfmsg->res_id);
 765         struct nfulnl_instance *inst;
 766         int ret = 0;
 767
 768         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
 769
 770         if (nfattr_bad_size(nfula, NFULA_CFG_MAX, nfula_cfg_min)) {
 771                 UDEBUG("bad attribute size\n");
 772                 return -EINVAL;
 773         }
 774
 775         inst = instance_lookup_get(group_num);
 776         if (nfula[NFULA_CFG_CMD-1]) {
 777                 u_int8_t pf = nfmsg->nfgen_family;
 778                 struct nfulnl_msg_config_cmd *cmd;
 779                 cmd = NFA_DATA(nfula[NFULA_CFG_CMD-1]);
 780                 UDEBUG("found CFG_CMD for\n");
 781
 782                 switch (cmd->command) {
 783                 case NFULNL_CFG_CMD_BIND:
 784                         if (inst) {
 785                                 ret = -EBUSY;
 786                                 goto out_put;
 787                         }
 788
 789                         inst = instance_create(group_num,
 790                                                NETLINK_CB(skb).pid);
 791                         if (!inst) {
 792                                 ret = -EINVAL;
 793                                 goto out_put;
 794                         }
 795                         break;
 796                 case NFULNL_CFG_CMD_UNBIND:
 797                         if (!inst) {
 798                                 ret = -ENODEV;
 799                                 goto out_put;
 800                         }
 801
 802                         if (inst->peer_pid != NETLINK_CB(skb).pid) {
 803                                 ret = -EPERM;
 804                                 goto out_put;
 805                         }
 806
 807                         instance_destroy(inst);
 808                         break;
 809                 case NFULNL_CFG_CMD_PF_BIND:
 810                         UDEBUG("registering log handler for pf=%u\n", pf);
 811                         ret = nf_log_register(pf, &nfulnl_logger);
 812                         break;
 813                 case NFULNL_CFG_CMD_PF_UNBIND:
 814                         UDEBUG("unregistering log handler for pf=%u\n", pf);
 815                         /* This is a bug and a feature.  We cannot unregister
 816                          * other handlers, like nfnetlink_inst can */
 817                         nf_log_unregister_pf(pf);
 818                         break;
 819                 default:
 820                         ret = -EINVAL;
 821                         break;
 822                 }
 823         } else {
 824                 if (!inst) {
 825                         UDEBUG("no config command, and no instance for "
 826                                 "group=%u pid=%u =>ENOENT\n",
 827                                 group_num, NETLINK_CB(skb).pid);
 828                         ret = -ENOENT;
 829                         goto out_put;
 830                 }
 831
 832                 if (inst->peer_pid != NETLINK_CB(skb).pid) {
 833                         UDEBUG("no config command, and wrong pid\n");
 834                         ret = -EPERM;
 835                         goto out_put;
 836                 }
 837         }
 838
 839         if (nfula[NFULA_CFG_MODE-1]) {
 840                 struct nfulnl_msg_config_mode *params;
 841                 params = NFA_DATA(nfula[NFULA_CFG_MODE-1]);
 842
 843                 nfulnl_set_mode(inst, params->copy_mode,
 844                                 ntohs(params->copy_range));
 845         }
 846
 847         if (nfula[NFULA_CFG_TIMEOUT-1]) {
 848                 u_int32_t timeout =
 849                         *(u_int32_t *)NFA_DATA(nfula[NFULA_CFG_TIMEOUT-1]);
 850
 851                 nfulnl_set_timeout(inst, ntohl(timeout));
 852         }
 853
 854         if (nfula[NFULA_CFG_NLBUFSIZ-1]) {
 855                 u_int32_t nlbufsiz =
 856                         *(u_int32_t *)NFA_DATA(nfula[NFULA_CFG_NLBUFSIZ-1]);
 857
 858                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 859         }
 860
 861         if (nfula[NFULA_CFG_QTHRESH-1]) {
 862                 u_int32_t qthresh =
 863                         *(u_int16_t *)NFA_DATA(nfula[NFULA_CFG_QTHRESH-1]);
 864
 865                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 866         }
 867
 868 out_put:
 869         instance_put(inst);
 870         return ret;
 871 }
 872
 873 static struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 874         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 875                                     .attr_count = NFULA_MAX, },
 876         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 877                                     .attr_count = NFULA_CFG_MAX, },
 878 };
 879
 880 static struct nfnetlink_subsystem nfulnl_subsys = {
 881         .name           = "log",
 882         .subsys_id      = NFNL_SUBSYS_ULOG,
 883         .cb_count       = NFULNL_MSG_MAX,
 884         .cb             = nfulnl_cb,
 885 };
 886
 887 #ifdef CONFIG_PROC_FS
 888 struct iter_state {
 889         unsigned int bucket;
 890 };
 891
 892 static struct hlist_node *get_first(struct seq_file *seq)
 893 {
 894         struct iter_state *st = seq->private;
 895
 896         if (!st)
 897                 return NULL;
 898
 899         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 900                 if (!hlist_empty(&instance_table[st->bucket]))
 901                         return instance_table[st->bucket].first;
 902         }
 903         return NULL;
 904 }
 905
 906 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
 907 {
 908         struct iter_state *st = seq->private;
 909
 910         h = h->next;
 911         while (!h) {
 912                 if (++st->bucket >= INSTANCE_BUCKETS)
 913                         return NULL;
 914
 915                 h = instance_table[st->bucket].first;
 916         }
 917         return h;
 918 }
 919
 920 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 921 {
 922         struct hlist_node *head;
 923         head = get_first(seq);
 924
 925         if (head)
 926                 while (pos && (head = get_next(seq, head)))
 927                         pos--;
 928         return pos ? NULL : head;
 929 }
 930
 931 static void *seq_start(struct seq_file *seq, loff_t *pos)
 932 {
 933         read_lock_bh(&instances_lock);
 934         return get_idx(seq, *pos);
 935 }
 936
 937 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 938 {
 939         (*pos)++;
 940         return get_next(s, v);
 941 }
 942
 943 static void seq_stop(struct seq_file *s, void *v)
 944 {
 945         read_unlock_bh(&instances_lock);
 946 }
 947
 948 static int seq_show(struct seq_file *s, void *v)
 949 {
 950         const struct nfulnl_instance *inst = v;
 951
 952         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 953                           inst->group_num,
 954                           inst->peer_pid, inst->qlen,
 955                           inst->copy_mode, inst->copy_range,
 956                           inst->flushtimeout, atomic_read(&inst->use));
 957 }
 958
 959 static struct seq_operations nful_seq_ops = {
 960         .start  = seq_start,
 961         .next   = seq_next,
 962         .stop   = seq_stop,
 963         .show   = seq_show,
 964 };
 965
 966 static int nful_open(struct inode *inode, struct file *file)
 967 {
 968         struct seq_file *seq;
 969         struct iter_state *is;
 970         int ret;
 971
 972         is = kzalloc(sizeof(*is), GFP_KERNEL);
 973         if (!is)
 974                 return -ENOMEM;
 975         ret = seq_open(file, &nful_seq_ops);
 976         if (ret < 0)
 977                 goto out_free;
 978         seq = file->private_data;
 979         seq->private = is;
 980         return ret;
 981 out_free:
 982         kfree(is);
 983         return ret;
 984 }
 985
 986 static struct file_operations nful_file_ops = {
 987         .owner   = THIS_MODULE,
 988         .open    = nful_open,
 989         .read    = seq_read,
 990         .llseek  = seq_lseek,
 991         .release = seq_release_private,
 992 };
 993
 994 #endif /* PROC_FS */
 995
 996 static int
 997 init_or_cleanup(int init)
 998 {
 999         int i, status = -ENOMEM;
1000 #ifdef CONFIG_PROC_FS
1001         struct proc_dir_entry *proc_nful;
1002 #endif
1003
1004         if (!init)
1005                 goto cleanup;
1006
1007         for (i = 0; i < INSTANCE_BUCKETS; i++)
1008                 INIT_HLIST_HEAD(&instance_table[i]);
1009
1010         /* it's not really all that important to have a random value, so
1011          * we can do this from the init function, even if there hasn't
1012          * been that much entropy yet */
1013         get_random_bytes(&hash_init, sizeof(hash_init));
1014
1015         netlink_register_notifier(&nfulnl_rtnl_notifier);
1016         status = nfnetlink_subsys_register(&nfulnl_subsys);
1017         if (status < 0) {
1018                 printk(KERN_ERR "log: failed to create netlink socket\n");
1019                 goto cleanup_netlink_notifier;
1020         }
1021
1022 #ifdef CONFIG_PROC_FS
1023         proc_nful = create_proc_entry("nfnetlink_log", 0440,
1024                                       proc_net_netfilter);
1025         if (!proc_nful)
1026                 goto cleanup_subsys;
1027         proc_nful->proc_fops = &nful_file_ops;
1028 #endif
1029
1030         return status;
1031
1032 cleanup:
1033         nf_log_unregister_logger(&nfulnl_logger);
1034 #ifdef CONFIG_PROC_FS
1035         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1036 cleanup_subsys:
1037 #endif
1038         nfnetlink_subsys_unregister(&nfulnl_subsys);
1039 cleanup_netlink_notifier:
1040         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1041         return status;
1042 }
1043
1044 static int __init init(void)
1045 {
1046
1047         return init_or_cleanup(1);
1048 }
1049
1050 static void __exit fini(void)
1051 {
1052         init_or_cleanup(0);
1053 }
1054
1055 MODULE_DESCRIPTION("netfilter userspace logging");
1056 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1057 MODULE_LICENSE("GPL");
1058 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1059
1060 module_init(init);
1061 module_exit(fini);