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 4096
  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
 318         UDEBUG("entered (%u, %u)\n", inst_size, pkt_size);
 319
 320         /* alloc skb which should be big enough for a whole multipart
 321          * message.  WARNING: has to be <= 128k due to slab restrictions */
 322
 323         skb = alloc_skb(inst_size, GFP_ATOMIC);
 324         if (!skb) {
 325                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
 326                         inst_size);
 327
 328                 /* try to allocate only as much as we need for current
 329                  * packet */
 330
 331                 skb = alloc_skb(pkt_size, GFP_ATOMIC);
 332                 if (!skb)
 333                         PRINTR("nfnetlink_log: can't even alloc %u bytes\n",
 334                                 pkt_size);
 335         }
 336
 337         return skb;
 338 }
 339
 340 static int
 341 __nfulnl_send(struct nfulnl_instance *inst)
 342 {
 343         int status;
 344
 345         if (timer_pending(&inst->timer))
 346                 del_timer(&inst->timer);
 347
 348         if (inst->qlen > 1)
 349                 inst->lastnlh->nlmsg_type = NLMSG_DONE;
 350
 351         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
 352         if (status < 0) {
 353                 UDEBUG("netlink_unicast() failed\n");
 354                 /* FIXME: statistics */
 355         }
 356
 357         inst->qlen = 0;
 358         inst->skb = NULL;
 359         inst->lastnlh = NULL;
 360
 361         return status;
 362 }
 363
 364 static void nfulnl_timer(unsigned long data)
 365 {
 366         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 367
 368         UDEBUG("timer function called, flushing buffer\n");
 369
 370         spin_lock_bh(&inst->lock);
 371         __nfulnl_send(inst);
 372         instance_put(inst);
 373         spin_unlock_bh(&inst->lock);
 374 }
 375
 376 static inline int
 377 __build_packet_message(struct nfulnl_instance *inst,
 378                         const struct sk_buff *skb,
 379                         unsigned int data_len,
 380                         unsigned int pf,
 381                         unsigned int hooknum,
 382                         const struct net_device *indev,
 383                         const struct net_device *outdev,
 384                         const struct nf_loginfo *li,
 385                         const char *prefix)
 386 {
 387         unsigned char *old_tail;
 388         struct nfulnl_msg_packet_hdr pmsg;
 389         struct nlmsghdr *nlh;
 390         struct nfgenmsg *nfmsg;
 391         u_int32_t tmp_uint;
 392
 393         UDEBUG("entered\n");
 394
 395         old_tail = inst->skb->tail;
 396         nlh = NLMSG_PUT(inst->skb, 0, 0,
 397                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 398                         sizeof(struct nfgenmsg));
 399         nfmsg = NLMSG_DATA(nlh);
 400         nfmsg->nfgen_family = pf;
 401         nfmsg->version = NFNETLINK_V0;
 402         nfmsg->res_id = htons(inst->group_num);
 403
 404         pmsg.hw_protocol        = htons(skb->protocol);
 405         pmsg.hook               = hooknum;
 406
 407         NFA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
 408
 409         if (prefix) {
 410                 int slen = strlen(prefix);
 411                 if (slen > NFULNL_PREFIXLEN)
 412                         slen = NFULNL_PREFIXLEN;
 413                 NFA_PUT(inst->skb, NFULA_PREFIX, slen, prefix);
 414         }
 415
 416         if (indev) {
 417                 tmp_uint = htonl(indev->ifindex);
 418 #ifndef CONFIG_BRIDGE_NETFILTER
 419                 NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
 420                         &tmp_uint);
 421 #else
 422                 if (pf == PF_BRIDGE) {
 423                         /* Case 1: outdev is physical input device, we need to
 424                          * look for bridge group (when called from
 425                          * netfilter_bridge) */
 426                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 427                                 sizeof(tmp_uint), &tmp_uint);
 428                         /* this is the bridge group "brX" */
 429                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
 430                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 431                                 sizeof(tmp_uint), &tmp_uint);
 432                 } else {
 433                         /* Case 2: indev is bridge group, we need to look for
 434                          * physical device (when called from ipv4) */
 435                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 436                                 sizeof(tmp_uint), &tmp_uint);
 437                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
 438                                 tmp_uint =
 439                                     htonl(skb->nf_bridge->physindev->ifindex);
 440                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 441                                         sizeof(tmp_uint), &tmp_uint);
 442                         }
 443                 }
 444 #endif
 445         }
 446
 447         if (outdev) {
 448                 tmp_uint = htonl(outdev->ifindex);
 449 #ifndef CONFIG_BRIDGE_NETFILTER
 450                 NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 451                         &tmp_uint);
 452 #else
 453                 if (pf == PF_BRIDGE) {
 454                         /* Case 1: outdev is physical output device, we need to
 455                          * look for bridge group (when called from
 456                          * netfilter_bridge) */
 457                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 458                                 sizeof(tmp_uint), &tmp_uint);
 459                         /* this is the bridge group "brX" */
 460                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
 461                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 462                                 sizeof(tmp_uint), &tmp_uint);
 463                 } else {
 464                         /* Case 2: indev is a bridge group, we need to look
 465                          * for physical device (when called from ipv4) */
 466                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 467                                 sizeof(tmp_uint), &tmp_uint);
 468                         if (skb->nf_bridge) {
 469                                 tmp_uint =
 470                                     htonl(skb->nf_bridge->physoutdev->ifindex);
 471                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 472                                         sizeof(tmp_uint), &tmp_uint);
 473                         }
 474                 }
 475 #endif
 476         }
 477
 478         if (skb->nfmark) {
 479                 tmp_uint = htonl(skb->nfmark);
 480                 NFA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
 481         }
 482
 483         if (indev && skb->dev && skb->dev->hard_header_parse) {
 484                 struct nfulnl_msg_packet_hw phw;
 485
 486                 phw.hw_addrlen =
 487                         skb->dev->hard_header_parse((struct sk_buff *)skb,
 488                                                     phw.hw_addr);
 489                 phw.hw_addrlen = htons(phw.hw_addrlen);
 490                 NFA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
 491         }
 492
 493         if (skb->tstamp.off_sec) {
 494                 struct nfulnl_msg_packet_timestamp ts;
 495
 496                 ts.sec = cpu_to_be64(skb->tstamp.off_sec);
 497                 ts.usec = cpu_to_be64(skb->tstamp.off_usec);
 498
 499                 NFA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 500         }
 501
 502         /* UID */
 503         if (skb->sk) {
 504                 read_lock_bh(&skb->sk->sk_callback_lock);
 505                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 506                         u_int32_t uid = htonl(skb->sk->sk_socket->file->f_uid);
 507                         /* need to unlock here since NFA_PUT may goto */
 508                         read_unlock_bh(&skb->sk->sk_callback_lock);
 509                         NFA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
 510                 } else
 511                         read_unlock_bh(&skb->sk->sk_callback_lock);
 512         }
 513
 514         if (data_len) {
 515                 struct nfattr *nfa;
 516                 int size = NFA_LENGTH(data_len);
 517
 518                 if (skb_tailroom(inst->skb) < (int)NFA_SPACE(data_len)) {
 519                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 520                         goto nlmsg_failure;
 521                 }
 522
 523                 nfa = (struct nfattr *)skb_put(inst->skb, NFA_ALIGN(size));
 524                 nfa->nfa_type = NFULA_PAYLOAD;
 525                 nfa->nfa_len = size;
 526
 527                 if (skb_copy_bits(skb, 0, NFA_DATA(nfa), data_len))
 528                         BUG();
 529         }
 530
 531         nlh->nlmsg_len = inst->skb->tail - old_tail;
 532         return 0;
 533
 534 nlmsg_failure:
 535         UDEBUG("nlmsg_failure\n");
 536 nfattr_failure:
 537         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 538         return -1;
 539 }
 540
 541 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 542
 543 static struct nf_loginfo default_loginfo = {
 544         .type =         NF_LOG_TYPE_ULOG,
 545         .u = {
 546                 .ulog = {
 547                         .copy_len       = 0xffff,
 548                         .group          = 0,
 549                         .qthreshold     = 1,
 550                 },
 551         },
 552 };
 553
 554 /* log handler for internal netfilter logging api */
 555 static void
 556 nfulnl_log_packet(unsigned int pf,
 557                   unsigned int hooknum,
 558                   const struct sk_buff *skb,
 559                   const struct net_device *in,
 560                   const struct net_device *out,
 561                   const struct nf_loginfo *li_user,
 562                   const char *prefix)
 563 {
 564         unsigned int size, data_len;
 565         struct nfulnl_instance *inst;
 566         const struct nf_loginfo *li;
 567         unsigned int qthreshold;
 568         unsigned int nlbufsiz;
 569
 570         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 571                 li = li_user;
 572         else
 573                 li = &default_loginfo;
 574
 575         inst = instance_lookup_get(li->u.ulog.group);
 576         if (!inst)
 577                 inst = instance_lookup_get(0);
 578         if (!inst) {
 579                 PRINTR("nfnetlink_log: trying to log packet, "
 580                         "but no instance for group %u\n", li->u.ulog.group);
 581                 return;
 582         }
 583
 584         /* all macros expand to constant values at compile time */
 585         /* FIXME: do we want to make the size calculation conditional based on
 586          * what is actually present?  way more branches and checks, but more
 587          * memory efficient... */
 588         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 589                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hdr))
 590                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 591                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 592 #ifdef CONFIG_BRIDGE_NETFILTER
 593                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 594                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 595 #endif
 596                 + NFA_SPACE(sizeof(u_int32_t))  /* mark */
 597                 + NFA_SPACE(sizeof(u_int32_t))  /* uid */
 598                 + NFA_SPACE(NFULNL_PREFIXLEN)   /* prefix */
 599                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hw))
 600                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_timestamp));
 601
 602         UDEBUG("initial size=%u\n", size);
 603
 604         spin_lock_bh(&inst->lock);
 605
 606         qthreshold = inst->qthreshold;
 607         /* per-rule qthreshold overrides per-instance */
 608         if (qthreshold > li->u.ulog.qthreshold)
 609                 qthreshold = li->u.ulog.qthreshold;
 610
 611         switch (inst->copy_mode) {
 612         case NFULNL_COPY_META:
 613         case NFULNL_COPY_NONE:
 614                 data_len = 0;
 615                 break;
 616
 617         case NFULNL_COPY_PACKET:
 618                 if (inst->copy_range == 0
 619                     || inst->copy_range > skb->len)
 620                         data_len = skb->len;
 621                 else
 622                         data_len = inst->copy_range;
 623
 624                 size += NFA_SPACE(data_len);
 625                 UDEBUG("copy_packet, therefore size now %u\n", size);
 626                 break;
 627
 628         default:
 629                 spin_unlock_bh(&inst->lock);
 630                 instance_put(inst);
 631                 return;
 632         }
 633
 634         if (size > inst->nlbufsiz)
 635                 nlbufsiz = size;
 636         else
 637                 nlbufsiz = inst->nlbufsiz;
 638
 639         if (!inst->skb) {
 640                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
 641                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
 642                                 inst->nlbufsiz, size);
 643                         goto alloc_failure;
 644                 }
 645         } else if (inst->qlen >= qthreshold ||
 646                    size > skb_tailroom(inst->skb)) {
 647                 /* either the queue len is too high or we don't have
 648                  * enough room in the skb left. flush to userspace. */
 649                 UDEBUG("flushing old skb\n");
 650
 651                 __nfulnl_send(inst);
 652
 653                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
 654                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
 655                                 inst->nlbufsiz, size);
 656                         goto alloc_failure;
 657                 }
 658         }
 659
 660         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
 661         inst->qlen++;
 662
 663         __build_packet_message(inst, skb, data_len, pf,
 664                                 hooknum, in, out, li, prefix);
 665
 666         /* timer_pending always called within inst->lock, so there
 667          * is no chance of a race here */
 668         if (!timer_pending(&inst->timer)) {
 669                 instance_get(inst);
 670                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 671                 add_timer(&inst->timer);
 672         }
 673         spin_unlock_bh(&inst->lock);
 674
 675         return;
 676
 677 alloc_failure:
 678         spin_unlock_bh(&inst->lock);
 679         instance_put(inst);
 680         UDEBUG("error allocating skb\n");
 681         /* FIXME: statistics */
 682 }
 683
 684 static int
 685 nfulnl_rcv_nl_event(struct notifier_block *this,
 686                    unsigned long event, void *ptr)
 687 {
 688         struct netlink_notify *n = ptr;
 689
 690         if (event == NETLINK_URELEASE &&
 691             n->protocol == NETLINK_NETFILTER && n->pid) {
 692                 int i;
 693
 694                 /* destroy all instances for this pid */
 695                 write_lock_bh(&instances_lock);
 696                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 697                         struct hlist_node *tmp, *t2;
 698                         struct nfulnl_instance *inst;
 699                         struct hlist_head *head = &instance_table[i];
 700
 701                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 702                                 UDEBUG("node = %p\n", inst);
 703                                 if (n->pid == inst->peer_pid)
 704                                         __instance_destroy(inst);
 705                         }
 706                 }
 707                 write_unlock_bh(&instances_lock);
 708         }
 709         return NOTIFY_DONE;
 710 }
 711
 712 static struct notifier_block nfulnl_rtnl_notifier = {
 713         .notifier_call  = nfulnl_rcv_nl_event,
 714 };
 715
 716 static int
 717 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 718                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
 719 {
 720         return -ENOTSUPP;
 721 }
 722
 723 static struct nf_logger nfulnl_logger = {
 724         .name   = "nfnetlink_log",
 725         .logfn  = &nfulnl_log_packet,
 726         .me     = THIS_MODULE,
 727 };
 728
 729 static const int nfula_min[NFULA_MAX] = {
 730         [NFULA_PACKET_HDR-1]    = sizeof(struct nfulnl_msg_packet_hdr),
 731         [NFULA_MARK-1]          = sizeof(u_int32_t),
 732         [NFULA_TIMESTAMP-1]     = sizeof(struct nfulnl_msg_packet_timestamp),
 733         [NFULA_IFINDEX_INDEV-1] = sizeof(u_int32_t),
 734         [NFULA_IFINDEX_OUTDEV-1]= sizeof(u_int32_t),
 735         [NFULA_HWADDR-1]        = sizeof(struct nfulnl_msg_packet_hw),
 736         [NFULA_PAYLOAD-1]       = 0,
 737         [NFULA_PREFIX-1]        = 0,
 738         [NFULA_UID-1]           = sizeof(u_int32_t),
 739 };
 740
 741 static const int nfula_cfg_min[NFULA_CFG_MAX] = {
 742         [NFULA_CFG_CMD-1]       = sizeof(struct nfulnl_msg_config_cmd),
 743         [NFULA_CFG_MODE-1]      = sizeof(struct nfulnl_msg_config_mode),
 744         [NFULA_CFG_TIMEOUT-1]   = sizeof(u_int32_t),
 745         [NFULA_CFG_QTHRESH-1]   = sizeof(u_int32_t),
 746         [NFULA_CFG_NLBUFSIZ-1]  = sizeof(u_int32_t),
 747 };
 748
 749 static int
 750 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 751                    struct nlmsghdr *nlh, struct nfattr *nfula[], int *errp)
 752 {
 753         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 754         u_int16_t group_num = ntohs(nfmsg->res_id);
 755         struct nfulnl_instance *inst;
 756         int ret = 0;
 757
 758         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
 759
 760         if (nfattr_bad_size(nfula, NFULA_CFG_MAX, nfula_cfg_min)) {
 761                 UDEBUG("bad attribute size\n");
 762                 return -EINVAL;
 763         }
 764
 765         inst = instance_lookup_get(group_num);
 766         if (nfula[NFULA_CFG_CMD-1]) {
 767                 u_int8_t pf = nfmsg->nfgen_family;
 768                 struct nfulnl_msg_config_cmd *cmd;
 769                 cmd = NFA_DATA(nfula[NFULA_CFG_CMD-1]);
 770                 UDEBUG("found CFG_CMD for\n");
 771
 772                 switch (cmd->command) {
 773                 case NFULNL_CFG_CMD_BIND:
 774                         if (inst) {
 775                                 ret = -EBUSY;
 776                                 goto out_put;
 777                         }
 778
 779                         inst = instance_create(group_num,
 780                                                NETLINK_CB(skb).pid);
 781                         if (!inst) {
 782                                 ret = -EINVAL;
 783                                 goto out_put;
 784                         }
 785                         break;
 786                 case NFULNL_CFG_CMD_UNBIND:
 787                         if (!inst) {
 788                                 ret = -ENODEV;
 789                                 goto out_put;
 790                         }
 791
 792                         if (inst->peer_pid != NETLINK_CB(skb).pid) {
 793                                 ret = -EPERM;
 794                                 goto out_put;
 795                         }
 796
 797                         instance_destroy(inst);
 798                         break;
 799                 case NFULNL_CFG_CMD_PF_BIND:
 800                         UDEBUG("registering log handler for pf=%u\n", pf);
 801                         ret = nf_log_register(pf, &nfulnl_logger);
 802                         break;
 803                 case NFULNL_CFG_CMD_PF_UNBIND:
 804                         UDEBUG("unregistering log handler for pf=%u\n", pf);
 805                         /* This is a bug and a feature.  We cannot unregister
 806                          * other handlers, like nfnetlink_inst can */
 807                         nf_log_unregister_pf(pf);
 808                         break;
 809                 default:
 810                         ret = -EINVAL;
 811                         break;
 812                 }
 813         } else {
 814                 if (!inst) {
 815                         UDEBUG("no config command, and no instance for "
 816                                 "group=%u pid=%u =>ENOENT\n",
 817                                 group_num, NETLINK_CB(skb).pid);
 818                         ret = -ENOENT;
 819                         goto out_put;
 820                 }
 821
 822                 if (inst->peer_pid != NETLINK_CB(skb).pid) {
 823                         UDEBUG("no config command, and wrong pid\n");
 824                         ret = -EPERM;
 825                         goto out_put;
 826                 }
 827         }
 828
 829         if (nfula[NFULA_CFG_MODE-1]) {
 830                 struct nfulnl_msg_config_mode *params;
 831                 params = NFA_DATA(nfula[NFULA_CFG_MODE-1]);
 832
 833                 nfulnl_set_mode(inst, params->copy_mode,
 834                                 ntohs(params->copy_range));
 835         }
 836
 837         if (nfula[NFULA_CFG_TIMEOUT-1]) {
 838                 u_int32_t timeout =
 839                         *(u_int32_t *)NFA_DATA(nfula[NFULA_CFG_TIMEOUT-1]);
 840
 841                 nfulnl_set_timeout(inst, ntohl(timeout));
 842         }
 843
 844         if (nfula[NFULA_CFG_NLBUFSIZ-1]) {
 845                 u_int32_t nlbufsiz =
 846                         *(u_int32_t *)NFA_DATA(nfula[NFULA_CFG_NLBUFSIZ-1]);
 847
 848                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 849         }
 850
 851         if (nfula[NFULA_CFG_QTHRESH-1]) {
 852                 u_int32_t qthresh =
 853                         *(u_int16_t *)NFA_DATA(nfula[NFULA_CFG_QTHRESH-1]);
 854
 855                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 856         }
 857
 858 out_put:
 859         instance_put(inst);
 860         return ret;
 861 }
 862
 863 static struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 864         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 865                                     .attr_count = NFULA_MAX, },
 866         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 867                                     .attr_count = NFULA_CFG_MAX, },
 868 };
 869
 870 static struct nfnetlink_subsystem nfulnl_subsys = {
 871         .name           = "log",
 872         .subsys_id      = NFNL_SUBSYS_ULOG,
 873         .cb_count       = NFULNL_MSG_MAX,
 874         .cb             = nfulnl_cb,
 875 };
 876
 877 #ifdef CONFIG_PROC_FS
 878 struct iter_state {
 879         unsigned int bucket;
 880 };
 881
 882 static struct hlist_node *get_first(struct seq_file *seq)
 883 {
 884         struct iter_state *st = seq->private;
 885
 886         if (!st)
 887                 return NULL;
 888
 889         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 890                 if (!hlist_empty(&instance_table[st->bucket]))
 891                         return instance_table[st->bucket].first;
 892         }
 893         return NULL;
 894 }
 895
 896 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
 897 {
 898         struct iter_state *st = seq->private;
 899
 900         h = h->next;
 901         while (!h) {
 902                 if (++st->bucket >= INSTANCE_BUCKETS)
 903                         return NULL;
 904
 905                 h = instance_table[st->bucket].first;
 906         }
 907         return h;
 908 }
 909
 910 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 911 {
 912         struct hlist_node *head;
 913         head = get_first(seq);
 914
 915         if (head)
 916                 while (pos && (head = get_next(seq, head)))
 917                         pos--;
 918         return pos ? NULL : head;
 919 }
 920
 921 static void *seq_start(struct seq_file *seq, loff_t *pos)
 922 {
 923         read_lock_bh(&instances_lock);
 924         return get_idx(seq, *pos);
 925 }
 926
 927 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 928 {
 929         (*pos)++;
 930         return get_next(s, v);
 931 }
 932
 933 static void seq_stop(struct seq_file *s, void *v)
 934 {
 935         read_unlock_bh(&instances_lock);
 936 }
 937
 938 static int seq_show(struct seq_file *s, void *v)
 939 {
 940         const struct nfulnl_instance *inst = v;
 941
 942         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 943                           inst->group_num,
 944                           inst->peer_pid, inst->qlen,
 945                           inst->copy_mode, inst->copy_range,
 946                           inst->flushtimeout, atomic_read(&inst->use));
 947 }
 948
 949 static struct seq_operations nful_seq_ops = {
 950         .start  = seq_start,
 951         .next   = seq_next,
 952         .stop   = seq_stop,
 953         .show   = seq_show,
 954 };
 955
 956 static int nful_open(struct inode *inode, struct file *file)
 957 {
 958         struct seq_file *seq;
 959         struct iter_state *is;
 960         int ret;
 961
 962         is = kzalloc(sizeof(*is), GFP_KERNEL);
 963         if (!is)
 964                 return -ENOMEM;
 965         ret = seq_open(file, &nful_seq_ops);
 966         if (ret < 0)
 967                 goto out_free;
 968         seq = file->private_data;
 969         seq->private = is;
 970         return ret;
 971 out_free:
 972         kfree(is);
 973         return ret;
 974 }
 975
 976 static struct file_operations nful_file_ops = {
 977         .owner   = THIS_MODULE,
 978         .open    = nful_open,
 979         .read    = seq_read,
 980         .llseek  = seq_lseek,
 981         .release = seq_release_private,
 982 };
 983
 984 #endif /* PROC_FS */
 985
 986 static int
 987 init_or_cleanup(int init)
 988 {
 989         int i, status = -ENOMEM;
 990 #ifdef CONFIG_PROC_FS
 991         struct proc_dir_entry *proc_nful;
 992 #endif
 993
 994         if (!init)
 995                 goto cleanup;
 996
 997         for (i = 0; i < INSTANCE_BUCKETS; i++)
 998                 INIT_HLIST_HEAD(&instance_table[i]);
 999
1000         /* it's not really all that important to have a random value, so
1001          * we can do this from the init function, even if there hasn't
1002          * been that much entropy yet */
1003         get_random_bytes(&hash_init, sizeof(hash_init));
1004
1005         netlink_register_notifier(&nfulnl_rtnl_notifier);
1006         status = nfnetlink_subsys_register(&nfulnl_subsys);
1007         if (status < 0) {
1008                 printk(KERN_ERR "log: failed to create netlink socket\n");
1009                 goto cleanup_netlink_notifier;
1010         }
1011
1012 #ifdef CONFIG_PROC_FS
1013         proc_nful = create_proc_entry("nfnetlink_log", 0440,
1014                                       proc_net_netfilter);
1015         if (!proc_nful)
1016                 goto cleanup_subsys;
1017         proc_nful->proc_fops = &nful_file_ops;
1018 #endif
1019
1020         return status;
1021
1022 cleanup:
1023         nf_log_unregister_logger(&nfulnl_logger);
1024 #ifdef CONFIG_PROC_FS
1025         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1026 cleanup_subsys:
1027 #endif
1028         nfnetlink_subsys_unregister(&nfulnl_subsys);
1029 cleanup_netlink_notifier:
1030         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1031         return status;
1032 }
1033
1034 static int __init init(void)
1035 {
1036
1037         return init_or_cleanup(1);
1038 }
1039
1040 static void __exit fini(void)
1041 {
1042         init_or_cleanup(0);
1043 }
1044
1045 MODULE_DESCRIPTION("netfilter userspace logging");
1046 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1047 MODULE_LICENSE("GPL");
1048 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1049
1050 module_init(init);
1051 module_exit(fini);