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  * 2006-01-26 Harald Welte <laforge@netfilter.org>
  15  *      - Add optional local and global sequence number to detect lost
  16  *        events from userspace
  17  *
  18  */
  19 #include <linux/module.h>
  20 #include <linux/skbuff.h>
  21 #include <linux/init.h>
  22 #include <linux/ip.h>
  23 #include <linux/ipv6.h>
  24 #include <linux/netdevice.h>
  25 #include <linux/netfilter.h>
  26 #include <linux/netlink.h>
  27 #include <linux/netfilter/nfnetlink.h>
  28 #include <linux/netfilter/nfnetlink_log.h>
  29 #include <linux/spinlock.h>
  30 #include <linux/sysctl.h>
  31 #include <linux/proc_fs.h>
  32 #include <linux/security.h>
  33 #include <linux/list.h>
  34 #include <linux/jhash.h>
  35 #include <linux/random.h>
  36 #include <net/sock.h>
  37
  38 #include <asm/atomic.h>
  39
  40 #ifdef CONFIG_BRIDGE_NETFILTER
  41 #include "../bridge/br_private.h"
  42 #endif
  43
  44 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
  45 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
  46 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  47
  48 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  49                                      printk(x, ## args); } while (0);
  50
  51 #if 0
  52 #define UDEBUG(x, args ...)     printk(KERN_DEBUG "%s(%d):%s(): " x,       \
  53                                         __FILE__, __LINE__, __FUNCTION__,  \
  54                                         ## args)
  55 #else
  56 #define UDEBUG(x, ...)
  57 #endif
  58
  59 struct nfulnl_instance {
  60         struct hlist_node hlist;        /* global list of instances */
  61         spinlock_t lock;
  62         atomic_t use;                   /* use count */
  63
  64         unsigned int qlen;              /* number of nlmsgs in skb */
  65         struct sk_buff *skb;            /* pre-allocatd skb */
  66         struct nlmsghdr *lastnlh;       /* netlink header of last msg in skb */
  67         struct timer_list timer;
  68         int peer_pid;                   /* PID of the peer process */
  69
  70         /* configurable parameters */
  71         unsigned int flushtimeout;      /* timeout until queue flush */
  72         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  73         unsigned int qthreshold;        /* threshold of the queue */
  74         u_int32_t copy_range;
  75         u_int32_t seq;                  /* instance-local sequential counter */
  76         u_int16_t group_num;            /* number of this queue */
  77         u_int16_t flags;
  78         u_int8_t copy_mode;
  79 };
  80
  81 static DEFINE_RWLOCK(instances_lock);
  82 static atomic_t global_seq;
  83
  84 #define INSTANCE_BUCKETS        16
  85 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  86 static unsigned int hash_init;
  87
  88 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  89 {
  90         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  91 }
  92
  93 static struct nfulnl_instance *
  94 __instance_lookup(u_int16_t group_num)
  95 {
  96         struct hlist_head *head;
  97         struct hlist_node *pos;
  98         struct nfulnl_instance *inst;
  99
 100         UDEBUG("entering (group_num=%u)\n", group_num);
 101
 102         head = &instance_table[instance_hashfn(group_num)];
 103         hlist_for_each_entry(inst, pos, head, hlist) {
 104                 if (inst->group_num == group_num)
 105                         return inst;
 106         }
 107         return NULL;
 108 }
 109
 110 static inline void
 111 instance_get(struct nfulnl_instance *inst)
 112 {
 113         atomic_inc(&inst->use);
 114 }
 115
 116 static struct nfulnl_instance *
 117 instance_lookup_get(u_int16_t group_num)
 118 {
 119         struct nfulnl_instance *inst;
 120
 121         read_lock_bh(&instances_lock);
 122         inst = __instance_lookup(group_num);
 123         if (inst)
 124                 instance_get(inst);
 125         read_unlock_bh(&instances_lock);
 126
 127         return inst;
 128 }
 129
 130 static void
 131 instance_put(struct nfulnl_instance *inst)
 132 {
 133         if (inst && atomic_dec_and_test(&inst->use)) {
 134                 UDEBUG("kfree(inst=%p)\n", inst);
 135                 kfree(inst);
 136         }
 137 }
 138
 139 static void nfulnl_timer(unsigned long data);
 140
 141 static struct nfulnl_instance *
 142 instance_create(u_int16_t group_num, int pid)
 143 {
 144         struct nfulnl_instance *inst;
 145
 146         UDEBUG("entering (group_num=%u, pid=%d)\n", group_num,
 147                 pid);
 148
 149         write_lock_bh(&instances_lock);
 150         if (__instance_lookup(group_num)) {
 151                 inst = NULL;
 152                 UDEBUG("aborting, instance already exists\n");
 153                 goto out_unlock;
 154         }
 155
 156         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 157         if (!inst)
 158                 goto out_unlock;
 159
 160         INIT_HLIST_NODE(&inst->hlist);
 161         spin_lock_init(&inst->lock);
 162         /* needs to be two, since we _put() after creation */
 163         atomic_set(&inst->use, 2);
 164
 165         init_timer(&inst->timer);
 166         inst->timer.function = nfulnl_timer;
 167         inst->timer.data = (unsigned long)inst;
 168         /* don't start timer yet. (re)start it  with every packet */
 169
 170         inst->peer_pid = pid;
 171         inst->group_num = group_num;
 172
 173         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 174         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 175         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 176         inst->copy_mode         = NFULNL_COPY_PACKET;
 177         inst->copy_range        = 0xffff;
 178
 179         if (!try_module_get(THIS_MODULE))
 180                 goto out_free;
 181
 182         hlist_add_head(&inst->hlist,
 183                        &instance_table[instance_hashfn(group_num)]);
 184
 185         UDEBUG("newly added node: %p, next=%p\n", &inst->hlist,
 186                 inst->hlist.next);
 187
 188         write_unlock_bh(&instances_lock);
 189
 190         return inst;
 191
 192 out_free:
 193         instance_put(inst);
 194 out_unlock:
 195         write_unlock_bh(&instances_lock);
 196         return NULL;
 197 }
 198
 199 static int __nfulnl_send(struct nfulnl_instance *inst);
 200
 201 static void
 202 _instance_destroy2(struct nfulnl_instance *inst, int lock)
 203 {
 204         /* first pull it out of the global list */
 205         if (lock)
 206                 write_lock_bh(&instances_lock);
 207
 208         UDEBUG("removing instance %p (queuenum=%u) from hash\n",
 209                 inst, inst->group_num);
 210
 211         hlist_del(&inst->hlist);
 212
 213         if (lock)
 214                 write_unlock_bh(&instances_lock);
 215
 216         /* then flush all pending packets from skb */
 217
 218         spin_lock_bh(&inst->lock);
 219         if (inst->skb) {
 220                 /* timer "holds" one reference (we have one more) */
 221                 if (del_timer(&inst->timer))
 222                         instance_put(inst);
 223                 if (inst->qlen)
 224                         __nfulnl_send(inst);
 225                 if (inst->skb) {
 226                         kfree_skb(inst->skb);
 227                         inst->skb = NULL;
 228                 }
 229         }
 230         spin_unlock_bh(&inst->lock);
 231
 232         /* and finally put the refcount */
 233         instance_put(inst);
 234
 235         module_put(THIS_MODULE);
 236 }
 237
 238 static inline void
 239 __instance_destroy(struct nfulnl_instance *inst)
 240 {
 241         _instance_destroy2(inst, 0);
 242 }
 243
 244 static inline void
 245 instance_destroy(struct nfulnl_instance *inst)
 246 {
 247         _instance_destroy2(inst, 1);
 248 }
 249
 250 static int
 251 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 252                   unsigned int range)
 253 {
 254         int status = 0;
 255
 256         spin_lock_bh(&inst->lock);
 257
 258         switch (mode) {
 259         case NFULNL_COPY_NONE:
 260         case NFULNL_COPY_META:
 261                 inst->copy_mode = mode;
 262                 inst->copy_range = 0;
 263                 break;
 264
 265         case NFULNL_COPY_PACKET:
 266                 inst->copy_mode = mode;
 267                 /* we're using struct nfattr which has 16bit nfa_len */
 268                 if (range > 0xffff)
 269                         inst->copy_range = 0xffff;
 270                 else
 271                         inst->copy_range = range;
 272                 break;
 273
 274         default:
 275                 status = -EINVAL;
 276                 break;
 277         }
 278
 279         spin_unlock_bh(&inst->lock);
 280
 281         return status;
 282 }
 283
 284 static int
 285 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 286 {
 287         int status;
 288
 289         spin_lock_bh(&inst->lock);
 290         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 291                 status = -ERANGE;
 292         else if (nlbufsiz > 131072)
 293                 status = -ERANGE;
 294         else {
 295                 inst->nlbufsiz = nlbufsiz;
 296                 status = 0;
 297         }
 298         spin_unlock_bh(&inst->lock);
 299
 300         return status;
 301 }
 302
 303 static int
 304 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 305 {
 306         spin_lock_bh(&inst->lock);
 307         inst->flushtimeout = timeout;
 308         spin_unlock_bh(&inst->lock);
 309
 310         return 0;
 311 }
 312
 313 static int
 314 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 315 {
 316         spin_lock_bh(&inst->lock);
 317         inst->qthreshold = qthresh;
 318         spin_unlock_bh(&inst->lock);
 319
 320         return 0;
 321 }
 322
 323 static int
 324 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 325 {
 326         spin_lock_bh(&inst->lock);
 327         inst->flags = flags;
 328         spin_unlock_bh(&inst->lock);
 329
 330         return 0;
 331 }
 332
 333 static struct sk_buff *nfulnl_alloc_skb(unsigned int inst_size,
 334                                         unsigned int pkt_size)
 335 {
 336         struct sk_buff *skb;
 337         unsigned int n;
 338
 339         UDEBUG("entered (%u, %u)\n", inst_size, pkt_size);
 340
 341         /* alloc skb which should be big enough for a whole multipart
 342          * message.  WARNING: has to be <= 128k due to slab restrictions */
 343
 344         n = max(inst_size, pkt_size);
 345         skb = alloc_skb(n, GFP_ATOMIC);
 346         if (!skb) {
 347                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
 348                         inst_size);
 349
 350                 if (n > pkt_size) {
 351                         /* try to allocate only as much as we need for current
 352                          * packet */
 353
 354                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 355                         if (!skb)
 356                                 PRINTR("nfnetlink_log: can't even alloc %u "
 357                                        "bytes\n", pkt_size);
 358                 }
 359         }
 360
 361         return skb;
 362 }
 363
 364 static int
 365 __nfulnl_send(struct nfulnl_instance *inst)
 366 {
 367         int status;
 368
 369         if (!inst->skb)
 370                 return 0;
 371
 372         if (inst->qlen > 1)
 373                 inst->lastnlh->nlmsg_type = NLMSG_DONE;
 374
 375         if (!inst->skb)
 376                 return 0;
 377
 378         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
 379         if (status < 0) {
 380                 UDEBUG("netlink_unicast() failed\n");
 381                 /* FIXME: statistics */
 382         }
 383
 384         inst->qlen = 0;
 385         inst->skb = NULL;
 386         inst->lastnlh = NULL;
 387
 388         return status;
 389 }
 390
 391 static void nfulnl_timer(unsigned long data)
 392 {
 393         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 394
 395         UDEBUG("timer function called, flushing buffer\n");
 396
 397         spin_lock_bh(&inst->lock);
 398         if (timer_pending(&inst->timer))        /* is it always true or false here? */
 399                 del_timer(&inst->timer);
 400         __nfulnl_send(inst);
 401         spin_unlock_bh(&inst->lock);
 402         instance_put(inst);
 403 }
 404
 405 /* This is an inline function, we don't really care about a long
 406  * list of arguments */
 407 static inline int
 408 __build_packet_message(struct nfulnl_instance *inst,
 409                         const struct sk_buff *skb,
 410                         unsigned int data_len,
 411                         unsigned int pf,
 412                         unsigned int hooknum,
 413                         const struct net_device *indev,
 414                         const struct net_device *outdev,
 415                         const struct nf_loginfo *li,
 416                         const char *prefix, unsigned int plen)
 417 {
 418         unsigned char *old_tail;
 419         struct nfulnl_msg_packet_hdr pmsg;
 420         struct nlmsghdr *nlh;
 421         struct nfgenmsg *nfmsg;
 422         __be32 tmp_uint;
 423
 424         UDEBUG("entered\n");
 425
 426         old_tail = inst->skb->tail;
 427         nlh = NLMSG_PUT(inst->skb, 0, 0,
 428                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 429                         sizeof(struct nfgenmsg));
 430         nfmsg = NLMSG_DATA(nlh);
 431         nfmsg->nfgen_family = pf;
 432         nfmsg->version = NFNETLINK_V0;
 433         nfmsg->res_id = htons(inst->group_num);
 434
 435         pmsg.hw_protocol        = skb->protocol;
 436         pmsg.hook               = hooknum;
 437
 438         NFA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
 439
 440         if (prefix)
 441                 NFA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
 442
 443         if (indev) {
 444                 tmp_uint = htonl(indev->ifindex);
 445 #ifndef CONFIG_BRIDGE_NETFILTER
 446                 NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
 447                         &tmp_uint);
 448 #else
 449                 if (pf == PF_BRIDGE) {
 450                         /* Case 1: outdev is physical input device, we need to
 451                          * look for bridge group (when called from
 452                          * netfilter_bridge) */
 453                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 454                                 sizeof(tmp_uint), &tmp_uint);
 455                         /* this is the bridge group "brX" */
 456                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
 457                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 458                                 sizeof(tmp_uint), &tmp_uint);
 459                 } else {
 460                         /* Case 2: indev is bridge group, we need to look for
 461                          * physical device (when called from ipv4) */
 462                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 463                                 sizeof(tmp_uint), &tmp_uint);
 464                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
 465                                 tmp_uint =
 466                                     htonl(skb->nf_bridge->physindev->ifindex);
 467                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 468                                         sizeof(tmp_uint), &tmp_uint);
 469                         }
 470                 }
 471 #endif
 472         }
 473
 474         if (outdev) {
 475                 tmp_uint = htonl(outdev->ifindex);
 476 #ifndef CONFIG_BRIDGE_NETFILTER
 477                 NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 478                         &tmp_uint);
 479 #else
 480                 if (pf == PF_BRIDGE) {
 481                         /* Case 1: outdev is physical output device, we need to
 482                          * look for bridge group (when called from
 483                          * netfilter_bridge) */
 484                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 485                                 sizeof(tmp_uint), &tmp_uint);
 486                         /* this is the bridge group "brX" */
 487                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
 488                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 489                                 sizeof(tmp_uint), &tmp_uint);
 490                 } else {
 491                         /* Case 2: indev is a bridge group, we need to look
 492                          * for physical device (when called from ipv4) */
 493                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 494                                 sizeof(tmp_uint), &tmp_uint);
 495                         if (skb->nf_bridge && skb->nf_bridge->physoutdev) {
 496                                 tmp_uint =
 497                                     htonl(skb->nf_bridge->physoutdev->ifindex);
 498                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 499                                         sizeof(tmp_uint), &tmp_uint);
 500                         }
 501                 }
 502 #endif
 503         }
 504
 505         if (skb->mark) {
 506                 tmp_uint = htonl(skb->mark);
 507                 NFA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
 508         }
 509
 510         if (indev && skb->dev && skb->dev->hard_header_parse) {
 511                 struct nfulnl_msg_packet_hw phw;
 512                 int len = skb->dev->hard_header_parse((struct sk_buff *)skb,
 513                                                     phw.hw_addr);
 514                 phw.hw_addrlen = htons(len);
 515                 NFA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
 516         }
 517
 518         if (skb->tstamp.off_sec) {
 519                 struct nfulnl_msg_packet_timestamp ts;
 520
 521                 ts.sec = cpu_to_be64(skb->tstamp.off_sec);
 522                 ts.usec = cpu_to_be64(skb->tstamp.off_usec);
 523
 524                 NFA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 525         }
 526
 527         /* UID */
 528         if (skb->sk) {
 529                 read_lock_bh(&skb->sk->sk_callback_lock);
 530                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 531                         __be32 uid = htonl(skb->sk->sk_socket->file->f_uid);
 532                         /* need to unlock here since NFA_PUT may goto */
 533                         read_unlock_bh(&skb->sk->sk_callback_lock);
 534                         NFA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
 535                 } else
 536                         read_unlock_bh(&skb->sk->sk_callback_lock);
 537         }
 538
 539         /* local sequence number */
 540         if (inst->flags & NFULNL_CFG_F_SEQ) {
 541                 tmp_uint = htonl(inst->seq++);
 542                 NFA_PUT(inst->skb, NFULA_SEQ, sizeof(tmp_uint), &tmp_uint);
 543         }
 544         /* global sequence number */
 545         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) {
 546                 tmp_uint = htonl(atomic_inc_return(&global_seq));
 547                 NFA_PUT(inst->skb, NFULA_SEQ_GLOBAL, sizeof(tmp_uint), &tmp_uint);
 548         }
 549
 550         if (data_len) {
 551                 struct nfattr *nfa;
 552                 int size = NFA_LENGTH(data_len);
 553
 554                 if (skb_tailroom(inst->skb) < (int)NFA_SPACE(data_len)) {
 555                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 556                         goto nlmsg_failure;
 557                 }
 558
 559                 nfa = (struct nfattr *)skb_put(inst->skb, NFA_ALIGN(size));
 560                 nfa->nfa_type = NFULA_PAYLOAD;
 561                 nfa->nfa_len = size;
 562
 563                 if (skb_copy_bits(skb, 0, NFA_DATA(nfa), data_len))
 564                         BUG();
 565         }
 566
 567         nlh->nlmsg_len = inst->skb->tail - old_tail;
 568         inst->lastnlh = nlh;
 569         return 0;
 570
 571 nlmsg_failure:
 572         UDEBUG("nlmsg_failure\n");
 573 nfattr_failure:
 574         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 575         return -1;
 576 }
 577
 578 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 579
 580 static struct nf_loginfo default_loginfo = {
 581         .type =         NF_LOG_TYPE_ULOG,
 582         .u = {
 583                 .ulog = {
 584                         .copy_len       = 0xffff,
 585                         .group          = 0,
 586                         .qthreshold     = 1,
 587                 },
 588         },
 589 };
 590
 591 /* log handler for internal netfilter logging api */
 592 static void
 593 nfulnl_log_packet(unsigned int pf,
 594                   unsigned int hooknum,
 595                   const struct sk_buff *skb,
 596                   const struct net_device *in,
 597                   const struct net_device *out,
 598                   const struct nf_loginfo *li_user,
 599                   const char *prefix)
 600 {
 601         unsigned int size, data_len;
 602         struct nfulnl_instance *inst;
 603         const struct nf_loginfo *li;
 604         unsigned int qthreshold;
 605         unsigned int nlbufsiz;
 606         unsigned int plen;
 607
 608         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 609                 li = li_user;
 610         else
 611                 li = &default_loginfo;
 612
 613         inst = instance_lookup_get(li->u.ulog.group);
 614         if (!inst)
 615                 inst = instance_lookup_get(0);
 616         if (!inst) {
 617                 PRINTR("nfnetlink_log: trying to log packet, "
 618                         "but no instance for group %u\n", li->u.ulog.group);
 619                 return;
 620         }
 621
 622         plen = 0;
 623         if (prefix)
 624                 plen = strlen(prefix) + 1;
 625
 626         /* all macros expand to constant values at compile time */
 627         /* FIXME: do we want to make the size calculation conditional based on
 628          * what is actually present?  way more branches and checks, but more
 629          * memory efficient... */
 630         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 631                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hdr))
 632                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 633                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 634 #ifdef CONFIG_BRIDGE_NETFILTER
 635                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 636                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 637 #endif
 638                 + NFA_SPACE(sizeof(u_int32_t))  /* mark */
 639                 + NFA_SPACE(sizeof(u_int32_t))  /* uid */
 640                 + NFA_SPACE(plen)               /* prefix */
 641                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hw))
 642                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_timestamp));
 643
 644         UDEBUG("initial size=%u\n", size);
 645
 646         spin_lock_bh(&inst->lock);
 647
 648         if (inst->flags & NFULNL_CFG_F_SEQ)
 649                 size += NFA_SPACE(sizeof(u_int32_t));
 650         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 651                 size += NFA_SPACE(sizeof(u_int32_t));
 652
 653         qthreshold = inst->qthreshold;
 654         /* per-rule qthreshold overrides per-instance */
 655         if (qthreshold > li->u.ulog.qthreshold)
 656                 qthreshold = li->u.ulog.qthreshold;
 657
 658         switch (inst->copy_mode) {
 659         case NFULNL_COPY_META:
 660         case NFULNL_COPY_NONE:
 661                 data_len = 0;
 662                 break;
 663
 664         case NFULNL_COPY_PACKET:
 665                 if (inst->copy_range == 0
 666                     || inst->copy_range > skb->len)
 667                         data_len = skb->len;
 668                 else
 669                         data_len = inst->copy_range;
 670
 671                 size += NFA_SPACE(data_len);
 672                 UDEBUG("copy_packet, therefore size now %u\n", size);
 673                 break;
 674
 675         default:
 676                 spin_unlock_bh(&inst->lock);
 677                 instance_put(inst);
 678                 return;
 679         }
 680
 681         if (size > inst->nlbufsiz)
 682                 nlbufsiz = size;
 683         else
 684                 nlbufsiz = inst->nlbufsiz;
 685
 686         if (!inst->skb) {
 687                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
 688                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
 689                                 inst->nlbufsiz, size);
 690                         goto alloc_failure;
 691                 }
 692         } else if (inst->qlen >= qthreshold ||
 693                    size > skb_tailroom(inst->skb)) {
 694                 /* either the queue len is too high or we don't have
 695                  * enough room in the skb left. flush to userspace. */
 696                 UDEBUG("flushing old skb\n");
 697
 698                 /* timer "holds" one reference (we have another one) */
 699                 if (del_timer(&inst->timer))
 700                         instance_put(inst);
 701                 __nfulnl_send(inst);
 702
 703                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
 704                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
 705                                 inst->nlbufsiz, size);
 706                         goto alloc_failure;
 707                 }
 708         }
 709
 710         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
 711         inst->qlen++;
 712
 713         __build_packet_message(inst, skb, data_len, pf,
 714                                 hooknum, in, out, li, prefix, plen);
 715
 716         /* timer_pending always called within inst->lock, so there
 717          * is no chance of a race here */
 718         if (!timer_pending(&inst->timer)) {
 719                 instance_get(inst);
 720                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 721                 add_timer(&inst->timer);
 722         }
 723
 724 unlock_and_release:
 725         spin_unlock_bh(&inst->lock);
 726         instance_put(inst);
 727         return;
 728
 729 alloc_failure:
 730         UDEBUG("error allocating skb\n");
 731         /* FIXME: statistics */
 732         goto unlock_and_release;
 733 }
 734
 735 static int
 736 nfulnl_rcv_nl_event(struct notifier_block *this,
 737                    unsigned long event, void *ptr)
 738 {
 739         struct netlink_notify *n = ptr;
 740
 741         if (event == NETLINK_URELEASE &&
 742             n->protocol == NETLINK_NETFILTER && n->pid) {
 743                 int i;
 744
 745                 /* destroy all instances for this pid */
 746                 write_lock_bh(&instances_lock);
 747                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 748                         struct hlist_node *tmp, *t2;
 749                         struct nfulnl_instance *inst;
 750                         struct hlist_head *head = &instance_table[i];
 751
 752                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 753                                 UDEBUG("node = %p\n", inst);
 754                                 if (n->pid == inst->peer_pid)
 755                                         __instance_destroy(inst);
 756                         }
 757                 }
 758                 write_unlock_bh(&instances_lock);
 759         }
 760         return NOTIFY_DONE;
 761 }
 762
 763 static struct notifier_block nfulnl_rtnl_notifier = {
 764         .notifier_call  = nfulnl_rcv_nl_event,
 765 };
 766
 767 static int
 768 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 769                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
 770 {
 771         return -ENOTSUPP;
 772 }
 773
 774 static struct nf_logger nfulnl_logger = {
 775         .name   = "nfnetlink_log",
 776         .logfn  = &nfulnl_log_packet,
 777         .me     = THIS_MODULE,
 778 };
 779
 780 static const int nfula_min[NFULA_MAX] = {
 781         [NFULA_PACKET_HDR-1]    = sizeof(struct nfulnl_msg_packet_hdr),
 782         [NFULA_MARK-1]          = sizeof(u_int32_t),
 783         [NFULA_TIMESTAMP-1]     = sizeof(struct nfulnl_msg_packet_timestamp),
 784         [NFULA_IFINDEX_INDEV-1] = sizeof(u_int32_t),
 785         [NFULA_IFINDEX_OUTDEV-1]= sizeof(u_int32_t),
 786         [NFULA_IFINDEX_PHYSINDEV-1]     = sizeof(u_int32_t),
 787         [NFULA_IFINDEX_PHYSOUTDEV-1]    = sizeof(u_int32_t),
 788         [NFULA_HWADDR-1]        = sizeof(struct nfulnl_msg_packet_hw),
 789         [NFULA_PAYLOAD-1]       = 0,
 790         [NFULA_PREFIX-1]        = 0,
 791         [NFULA_UID-1]           = sizeof(u_int32_t),
 792         [NFULA_SEQ-1]           = sizeof(u_int32_t),
 793         [NFULA_SEQ_GLOBAL-1]    = sizeof(u_int32_t),
 794 };
 795
 796 static const int nfula_cfg_min[NFULA_CFG_MAX] = {
 797         [NFULA_CFG_CMD-1]       = sizeof(struct nfulnl_msg_config_cmd),
 798         [NFULA_CFG_MODE-1]      = sizeof(struct nfulnl_msg_config_mode),
 799         [NFULA_CFG_TIMEOUT-1]   = sizeof(u_int32_t),
 800         [NFULA_CFG_QTHRESH-1]   = sizeof(u_int32_t),
 801         [NFULA_CFG_NLBUFSIZ-1]  = sizeof(u_int32_t),
 802         [NFULA_CFG_FLAGS-1]     = sizeof(u_int16_t),
 803 };
 804
 805 static int
 806 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 807                    struct nlmsghdr *nlh, struct nfattr *nfula[], int *errp)
 808 {
 809         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 810         u_int16_t group_num = ntohs(nfmsg->res_id);
 811         struct nfulnl_instance *inst;
 812         int ret = 0;
 813
 814         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
 815
 816         if (nfattr_bad_size(nfula, NFULA_CFG_MAX, nfula_cfg_min)) {
 817                 UDEBUG("bad attribute size\n");
 818                 return -EINVAL;
 819         }
 820
 821         inst = instance_lookup_get(group_num);
 822         if (nfula[NFULA_CFG_CMD-1]) {
 823                 u_int8_t pf = nfmsg->nfgen_family;
 824                 struct nfulnl_msg_config_cmd *cmd;
 825                 cmd = NFA_DATA(nfula[NFULA_CFG_CMD-1]);
 826                 UDEBUG("found CFG_CMD for\n");
 827
 828                 switch (cmd->command) {
 829                 case NFULNL_CFG_CMD_BIND:
 830                         if (inst) {
 831                                 ret = -EBUSY;
 832                                 goto out_put;
 833                         }
 834
 835                         inst = instance_create(group_num,
 836                                                NETLINK_CB(skb).pid);
 837                         if (!inst) {
 838                                 ret = -EINVAL;
 839                                 goto out_put;
 840                         }
 841                         break;
 842                 case NFULNL_CFG_CMD_UNBIND:
 843                         if (!inst) {
 844                                 ret = -ENODEV;
 845                                 goto out_put;
 846                         }
 847
 848                         if (inst->peer_pid != NETLINK_CB(skb).pid) {
 849                                 ret = -EPERM;
 850                                 goto out_put;
 851                         }
 852
 853                         instance_destroy(inst);
 854                         break;
 855                 case NFULNL_CFG_CMD_PF_BIND:
 856                         UDEBUG("registering log handler for pf=%u\n", pf);
 857                         ret = nf_log_register(pf, &nfulnl_logger);
 858                         break;
 859                 case NFULNL_CFG_CMD_PF_UNBIND:
 860                         UDEBUG("unregistering log handler for pf=%u\n", pf);
 861                         /* This is a bug and a feature.  We cannot unregister
 862                          * other handlers, like nfnetlink_inst can */
 863                         nf_log_unregister_pf(pf);
 864                         break;
 865                 default:
 866                         ret = -EINVAL;
 867                         break;
 868                 }
 869
 870                 if (!inst)
 871                         goto out;
 872         } else {
 873                 if (!inst) {
 874                         UDEBUG("no config command, and no instance for "
 875                                 "group=%u pid=%u =>ENOENT\n",
 876                                 group_num, NETLINK_CB(skb).pid);
 877                         ret = -ENOENT;
 878                         goto out_put;
 879                 }
 880
 881                 if (inst->peer_pid != NETLINK_CB(skb).pid) {
 882                         UDEBUG("no config command, and wrong pid\n");
 883                         ret = -EPERM;
 884                         goto out_put;
 885                 }
 886         }
 887
 888         if (nfula[NFULA_CFG_MODE-1]) {
 889                 struct nfulnl_msg_config_mode *params;
 890                 params = NFA_DATA(nfula[NFULA_CFG_MODE-1]);
 891
 892                 nfulnl_set_mode(inst, params->copy_mode,
 893                                 ntohl(params->copy_range));
 894         }
 895
 896         if (nfula[NFULA_CFG_TIMEOUT-1]) {
 897                 __be32 timeout =
 898                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_TIMEOUT-1]);
 899
 900                 nfulnl_set_timeout(inst, ntohl(timeout));
 901         }
 902
 903         if (nfula[NFULA_CFG_NLBUFSIZ-1]) {
 904                 __be32 nlbufsiz =
 905                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_NLBUFSIZ-1]);
 906
 907                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 908         }
 909
 910         if (nfula[NFULA_CFG_QTHRESH-1]) {
 911                 __be32 qthresh =
 912                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_QTHRESH-1]);
 913
 914                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 915         }
 916
 917         if (nfula[NFULA_CFG_FLAGS-1]) {
 918                 __be16 flags =
 919                         *(__be16 *)NFA_DATA(nfula[NFULA_CFG_FLAGS-1]);
 920                 nfulnl_set_flags(inst, ntohs(flags));
 921         }
 922
 923 out_put:
 924         instance_put(inst);
 925 out:
 926         return ret;
 927 }
 928
 929 static struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 930         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 931                                     .attr_count = NFULA_MAX, },
 932         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 933                                     .attr_count = NFULA_CFG_MAX, },
 934 };
 935
 936 static struct nfnetlink_subsystem nfulnl_subsys = {
 937         .name           = "log",
 938         .subsys_id      = NFNL_SUBSYS_ULOG,
 939         .cb_count       = NFULNL_MSG_MAX,
 940         .cb             = nfulnl_cb,
 941 };
 942
 943 #ifdef CONFIG_PROC_FS
 944 struct iter_state {
 945         unsigned int bucket;
 946 };
 947
 948 static struct hlist_node *get_first(struct seq_file *seq)
 949 {
 950         struct iter_state *st = seq->private;
 951
 952         if (!st)
 953                 return NULL;
 954
 955         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 956                 if (!hlist_empty(&instance_table[st->bucket]))
 957                         return instance_table[st->bucket].first;
 958         }
 959         return NULL;
 960 }
 961
 962 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
 963 {
 964         struct iter_state *st = seq->private;
 965
 966         h = h->next;
 967         while (!h) {
 968                 if (++st->bucket >= INSTANCE_BUCKETS)
 969                         return NULL;
 970
 971                 h = instance_table[st->bucket].first;
 972         }
 973         return h;
 974 }
 975
 976 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 977 {
 978         struct hlist_node *head;
 979         head = get_first(seq);
 980
 981         if (head)
 982                 while (pos && (head = get_next(seq, head)))
 983                         pos--;
 984         return pos ? NULL : head;
 985 }
 986
 987 static void *seq_start(struct seq_file *seq, loff_t *pos)
 988 {
 989         read_lock_bh(&instances_lock);
 990         return get_idx(seq, *pos);
 991 }
 992
 993 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 994 {
 995         (*pos)++;
 996         return get_next(s, v);
 997 }
 998
 999 static void seq_stop(struct seq_file *s, void *v)
1000 {
1001         read_unlock_bh(&instances_lock);
1002 }
1003
1004 static int seq_show(struct seq_file *s, void *v)
1005 {
1006         const struct nfulnl_instance *inst = v;
1007
1008         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
1009                           inst->group_num,
1010                           inst->peer_pid, inst->qlen,
1011                           inst->copy_mode, inst->copy_range,
1012                           inst->flushtimeout, atomic_read(&inst->use));
1013 }
1014
1015 static struct seq_operations nful_seq_ops = {
1016         .start  = seq_start,
1017         .next   = seq_next,
1018         .stop   = seq_stop,
1019         .show   = seq_show,
1020 };
1021
1022 static int nful_open(struct inode *inode, struct file *file)
1023 {
1024         struct seq_file *seq;
1025         struct iter_state *is;
1026         int ret;
1027
1028         is = kzalloc(sizeof(*is), GFP_KERNEL);
1029         if (!is)
1030                 return -ENOMEM;
1031         ret = seq_open(file, &nful_seq_ops);
1032         if (ret < 0)
1033                 goto out_free;
1034         seq = file->private_data;
1035         seq->private = is;
1036         return ret;
1037 out_free:
1038         kfree(is);
1039         return ret;
1040 }
1041
1042 static struct file_operations nful_file_ops = {
1043         .owner   = THIS_MODULE,
1044         .open    = nful_open,
1045         .read    = seq_read,
1046         .llseek  = seq_lseek,
1047         .release = seq_release_private,
1048 };
1049
1050 #endif /* PROC_FS */
1051
1052 static int __init nfnetlink_log_init(void)
1053 {
1054         int i, status = -ENOMEM;
1055 #ifdef CONFIG_PROC_FS
1056         struct proc_dir_entry *proc_nful;
1057 #endif
1058
1059         for (i = 0; i < INSTANCE_BUCKETS; i++)
1060                 INIT_HLIST_HEAD(&instance_table[i]);
1061
1062         /* it's not really all that important to have a random value, so
1063          * we can do this from the init function, even if there hasn't
1064          * been that much entropy yet */
1065         get_random_bytes(&hash_init, sizeof(hash_init));
1066
1067         netlink_register_notifier(&nfulnl_rtnl_notifier);
1068         status = nfnetlink_subsys_register(&nfulnl_subsys);
1069         if (status < 0) {
1070                 printk(KERN_ERR "log: failed to create netlink socket\n");
1071                 goto cleanup_netlink_notifier;
1072         }
1073
1074 #ifdef CONFIG_PROC_FS
1075         proc_nful = create_proc_entry("nfnetlink_log", 0440,
1076                                       proc_net_netfilter);
1077         if (!proc_nful)
1078                 goto cleanup_subsys;
1079         proc_nful->proc_fops = &nful_file_ops;
1080 #endif
1081         return status;
1082
1083 #ifdef CONFIG_PROC_FS
1084 cleanup_subsys:
1085         nfnetlink_subsys_unregister(&nfulnl_subsys);
1086 #endif
1087 cleanup_netlink_notifier:
1088         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1089         return status;
1090 }
1091
1092 static void __exit nfnetlink_log_fini(void)
1093 {
1094         nf_log_unregister_logger(&nfulnl_logger);
1095 #ifdef CONFIG_PROC_FS
1096         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1097 #endif
1098         nfnetlink_subsys_unregister(&nfulnl_subsys);
1099         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1100 }
1101
1102 MODULE_DESCRIPTION("netfilter userspace logging");
1103 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1104 MODULE_LICENSE("GPL");
1105 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1106
1107 module_init(nfnetlink_log_init);
1108 module_exit(nfnetlink_log_fini);