SAFE public projects git trees. - safe/jmp/linux-2.6/blob - net/netfilter/nfnetlink_log.c

   1 /*
   2  * This is a module which is used for logging packets to userspace via
   3  * nfetlink.
   4  *
   5  * (C) 2005 by Harald Welte <laforge@netfilter.org>
   6  *
   7  * Based on the old ipv4-only ipt_ULOG.c:
   8  * (C) 2000-2004 by Harald Welte <laforge@netfilter.org>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License version 2 as
  12  * published by the Free Software Foundation.
  13  *
  14  * 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                 module_put(THIS_MODULE);
 137         }
 138 }
 139
 140 static void nfulnl_timer(unsigned long data);
 141
 142 static struct nfulnl_instance *
 143 instance_create(u_int16_t group_num, int pid)
 144 {
 145         struct nfulnl_instance *inst;
 146
 147         UDEBUG("entering (group_num=%u, pid=%d)\n", group_num,
 148                 pid);
 149
 150         write_lock_bh(&instances_lock);
 151         if (__instance_lookup(group_num)) {
 152                 inst = NULL;
 153                 UDEBUG("aborting, instance already exists\n");
 154                 goto out_unlock;
 155         }
 156
 157         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 158         if (!inst)
 159                 goto out_unlock;
 160
 161         INIT_HLIST_NODE(&inst->hlist);
 162         spin_lock_init(&inst->lock);
 163         /* needs to be two, since we _put() after creation */
 164         atomic_set(&inst->use, 2);
 165
 166         init_timer(&inst->timer);
 167         inst->timer.function = nfulnl_timer;
 168         inst->timer.data = (unsigned long)inst;
 169         /* don't start timer yet. (re)start it  with every packet */
 170
 171         inst->peer_pid = pid;
 172         inst->group_num = group_num;
 173
 174         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 175         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 176         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 177         inst->copy_mode         = NFULNL_COPY_PACKET;
 178         inst->copy_range        = 0xffff;
 179
 180         if (!try_module_get(THIS_MODULE))
 181                 goto out_free;
 182
 183         hlist_add_head(&inst->hlist,
 184                        &instance_table[instance_hashfn(group_num)]);
 185
 186         UDEBUG("newly added node: %p, next=%p\n", &inst->hlist,
 187                 inst->hlist.next);
 188
 189         write_unlock_bh(&instances_lock);
 190
 191         return inst;
 192
 193 out_free:
 194         instance_put(inst);
 195 out_unlock:
 196         write_unlock_bh(&instances_lock);
 197         return NULL;
 198 }
 199
 200 static int __nfulnl_send(struct nfulnl_instance *inst);
 201
 202 static void
 203 _instance_destroy2(struct nfulnl_instance *inst, int lock)
 204 {
 205         /* first pull it out of the global list */
 206         if (lock)
 207                 write_lock_bh(&instances_lock);
 208
 209         UDEBUG("removing instance %p (queuenum=%u) from hash\n",
 210                 inst, inst->group_num);
 211
 212         hlist_del(&inst->hlist);
 213
 214         if (lock)
 215                 write_unlock_bh(&instances_lock);
 216
 217         /* then flush all pending packets from skb */
 218
 219         spin_lock_bh(&inst->lock);
 220         if (inst->skb) {
 221                 /* timer "holds" one reference (we have one more) */
 222                 if (del_timer(&inst->timer))
 223                         instance_put(inst);
 224                 if (inst->qlen)
 225                         __nfulnl_send(inst);
 226                 if (inst->skb) {
 227                         kfree_skb(inst->skb);
 228                         inst->skb = NULL;
 229                 }
 230         }
 231         spin_unlock_bh(&inst->lock);
 232
 233         /* and finally put the refcount */
 234         instance_put(inst);
 235 }
 236
 237 static inline void
 238 __instance_destroy(struct nfulnl_instance *inst)
 239 {
 240         _instance_destroy2(inst, 0);
 241 }
 242
 243 static inline void
 244 instance_destroy(struct nfulnl_instance *inst)
 245 {
 246         _instance_destroy2(inst, 1);
 247 }
 248
 249 static int
 250 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 251                   unsigned int range)
 252 {
 253         int status = 0;
 254
 255         spin_lock_bh(&inst->lock);
 256
 257         switch (mode) {
 258         case NFULNL_COPY_NONE:
 259         case NFULNL_COPY_META:
 260                 inst->copy_mode = mode;
 261                 inst->copy_range = 0;
 262                 break;
 263
 264         case NFULNL_COPY_PACKET:
 265                 inst->copy_mode = mode;
 266                 /* we're using struct nfattr which has 16bit nfa_len */
 267                 if (range > 0xffff)
 268                         inst->copy_range = 0xffff;
 269                 else
 270                         inst->copy_range = range;
 271                 break;
 272
 273         default:
 274                 status = -EINVAL;
 275                 break;
 276         }
 277
 278         spin_unlock_bh(&inst->lock);
 279
 280         return status;
 281 }
 282
 283 static int
 284 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 285 {
 286         int status;
 287
 288         spin_lock_bh(&inst->lock);
 289         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 290                 status = -ERANGE;
 291         else if (nlbufsiz > 131072)
 292                 status = -ERANGE;
 293         else {
 294                 inst->nlbufsiz = nlbufsiz;
 295                 status = 0;
 296         }
 297         spin_unlock_bh(&inst->lock);
 298
 299         return status;
 300 }
 301
 302 static int
 303 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 304 {
 305         spin_lock_bh(&inst->lock);
 306         inst->flushtimeout = timeout;
 307         spin_unlock_bh(&inst->lock);
 308
 309         return 0;
 310 }
 311
 312 static int
 313 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 314 {
 315         spin_lock_bh(&inst->lock);
 316         inst->qthreshold = qthresh;
 317         spin_unlock_bh(&inst->lock);
 318
 319         return 0;
 320 }
 321
 322 static int
 323 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 324 {
 325         spin_lock_bh(&inst->lock);
 326         inst->flags = flags;
 327         spin_unlock_bh(&inst->lock);
 328
 329         return 0;
 330 }
 331
 332 static struct sk_buff *nfulnl_alloc_skb(unsigned int inst_size,
 333                                         unsigned int pkt_size)
 334 {
 335         struct sk_buff *skb;
 336         unsigned int n;
 337
 338         UDEBUG("entered (%u, %u)\n", inst_size, pkt_size);
 339
 340         /* alloc skb which should be big enough for a whole multipart
 341          * message.  WARNING: has to be <= 128k due to slab restrictions */
 342
 343         n = max(inst_size, pkt_size);
 344         skb = alloc_skb(n, GFP_ATOMIC);
 345         if (!skb) {
 346                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
 347                         inst_size);
 348
 349                 if (n > pkt_size) {
 350                         /* try to allocate only as much as we need for current
 351                          * packet */
 352
 353                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 354                         if (!skb)
 355                                 PRINTR("nfnetlink_log: can't even alloc %u "
 356                                        "bytes\n", pkt_size);
 357                 }
 358         }
 359
 360         return skb;
 361 }
 362
 363 static int
 364 __nfulnl_send(struct nfulnl_instance *inst)
 365 {
 366         int status;
 367
 368         if (!inst->skb)
 369                 return 0;
 370
 371         if (inst->qlen > 1)
 372                 inst->lastnlh->nlmsg_type = NLMSG_DONE;
 373
 374         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
 375         if (status < 0) {
 376                 UDEBUG("netlink_unicast() failed\n");
 377                 /* FIXME: statistics */
 378         }
 379
 380         inst->qlen = 0;
 381         inst->skb = NULL;
 382         inst->lastnlh = NULL;
 383
 384         return status;
 385 }
 386
 387 static void nfulnl_timer(unsigned long data)
 388 {
 389         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 390
 391         UDEBUG("timer function called, flushing buffer\n");
 392
 393         spin_lock_bh(&inst->lock);
 394         __nfulnl_send(inst);
 395         spin_unlock_bh(&inst->lock);
 396         instance_put(inst);
 397 }
 398
 399 /* This is an inline function, we don't really care about a long
 400  * list of arguments */
 401 static inline int
 402 __build_packet_message(struct nfulnl_instance *inst,
 403                         const struct sk_buff *skb,
 404                         unsigned int data_len,
 405                         unsigned int pf,
 406                         unsigned int hooknum,
 407                         const struct net_device *indev,
 408                         const struct net_device *outdev,
 409                         const struct nf_loginfo *li,
 410                         const char *prefix, unsigned int plen)
 411 {
 412         struct nfulnl_msg_packet_hdr pmsg;
 413         struct nlmsghdr *nlh;
 414         struct nfgenmsg *nfmsg;
 415         __be32 tmp_uint;
 416         sk_buff_data_t old_tail = inst->skb->tail;
 417
 418         UDEBUG("entered\n");
 419
 420         nlh = NLMSG_PUT(inst->skb, 0, 0,
 421                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 422                         sizeof(struct nfgenmsg));
 423         nfmsg = NLMSG_DATA(nlh);
 424         nfmsg->nfgen_family = pf;
 425         nfmsg->version = NFNETLINK_V0;
 426         nfmsg->res_id = htons(inst->group_num);
 427
 428         pmsg.hw_protocol        = skb->protocol;
 429         pmsg.hook               = hooknum;
 430
 431         NFA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
 432
 433         if (prefix)
 434                 NFA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
 435
 436         if (indev) {
 437                 tmp_uint = htonl(indev->ifindex);
 438 #ifndef CONFIG_BRIDGE_NETFILTER
 439                 NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
 440                         &tmp_uint);
 441 #else
 442                 if (pf == PF_BRIDGE) {
 443                         /* Case 1: outdev is physical input device, we need to
 444                          * look for bridge group (when called from
 445                          * netfilter_bridge) */
 446                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 447                                 sizeof(tmp_uint), &tmp_uint);
 448                         /* this is the bridge group "brX" */
 449                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
 450                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 451                                 sizeof(tmp_uint), &tmp_uint);
 452                 } else {
 453                         /* Case 2: indev is bridge group, we need to look for
 454                          * physical device (when called from ipv4) */
 455                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 456                                 sizeof(tmp_uint), &tmp_uint);
 457                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
 458                                 tmp_uint =
 459                                     htonl(skb->nf_bridge->physindev->ifindex);
 460                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 461                                         sizeof(tmp_uint), &tmp_uint);
 462                         }
 463                 }
 464 #endif
 465         }
 466
 467         if (outdev) {
 468                 tmp_uint = htonl(outdev->ifindex);
 469 #ifndef CONFIG_BRIDGE_NETFILTER
 470                 NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 471                         &tmp_uint);
 472 #else
 473                 if (pf == PF_BRIDGE) {
 474                         /* Case 1: outdev is physical output device, we need to
 475                          * look for bridge group (when called from
 476                          * netfilter_bridge) */
 477                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 478                                 sizeof(tmp_uint), &tmp_uint);
 479                         /* this is the bridge group "brX" */
 480                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
 481                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 482                                 sizeof(tmp_uint), &tmp_uint);
 483                 } else {
 484                         /* Case 2: indev is a bridge group, we need to look
 485                          * for physical device (when called from ipv4) */
 486                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 487                                 sizeof(tmp_uint), &tmp_uint);
 488                         if (skb->nf_bridge && skb->nf_bridge->physoutdev) {
 489                                 tmp_uint =
 490                                     htonl(skb->nf_bridge->physoutdev->ifindex);
 491                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 492                                         sizeof(tmp_uint), &tmp_uint);
 493                         }
 494                 }
 495 #endif
 496         }
 497
 498         if (skb->mark) {
 499                 tmp_uint = htonl(skb->mark);
 500                 NFA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
 501         }
 502
 503         if (indev && skb->dev && skb->dev->hard_header_parse) {
 504                 struct nfulnl_msg_packet_hw phw;
 505                 int len = skb->dev->hard_header_parse((struct sk_buff *)skb,
 506                                                     phw.hw_addr);
 507                 phw.hw_addrlen = htons(len);
 508                 NFA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
 509         }
 510
 511         if (skb->tstamp.tv64) {
 512                 struct nfulnl_msg_packet_timestamp ts;
 513                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 514                 ts.sec = cpu_to_be64(tv.tv_sec);
 515                 ts.usec = cpu_to_be64(tv.tv_usec);
 516
 517                 NFA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 518         }
 519
 520         /* UID */
 521         if (skb->sk) {
 522                 read_lock_bh(&skb->sk->sk_callback_lock);
 523                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 524                         __be32 uid = htonl(skb->sk->sk_socket->file->f_uid);
 525                         /* need to unlock here since NFA_PUT may goto */
 526                         read_unlock_bh(&skb->sk->sk_callback_lock);
 527                         NFA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
 528                 } else
 529                         read_unlock_bh(&skb->sk->sk_callback_lock);
 530         }
 531
 532         /* local sequence number */
 533         if (inst->flags & NFULNL_CFG_F_SEQ) {
 534                 tmp_uint = htonl(inst->seq++);
 535                 NFA_PUT(inst->skb, NFULA_SEQ, sizeof(tmp_uint), &tmp_uint);
 536         }
 537         /* global sequence number */
 538         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) {
 539                 tmp_uint = htonl(atomic_inc_return(&global_seq));
 540                 NFA_PUT(inst->skb, NFULA_SEQ_GLOBAL, sizeof(tmp_uint), &tmp_uint);
 541         }
 542
 543         if (data_len) {
 544                 struct nfattr *nfa;
 545                 int size = NFA_LENGTH(data_len);
 546
 547                 if (skb_tailroom(inst->skb) < (int)NFA_SPACE(data_len)) {
 548                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 549                         goto nlmsg_failure;
 550                 }
 551
 552                 nfa = (struct nfattr *)skb_put(inst->skb, NFA_ALIGN(size));
 553                 nfa->nfa_type = NFULA_PAYLOAD;
 554                 nfa->nfa_len = size;
 555
 556                 if (skb_copy_bits(skb, 0, NFA_DATA(nfa), data_len))
 557                         BUG();
 558         }
 559
 560         nlh->nlmsg_len = inst->skb->tail - old_tail;
 561         inst->lastnlh = nlh;
 562         return 0;
 563
 564 nlmsg_failure:
 565         UDEBUG("nlmsg_failure\n");
 566 nfattr_failure:
 567         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 568         return -1;
 569 }
 570
 571 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 572
 573 static struct nf_loginfo default_loginfo = {
 574         .type =         NF_LOG_TYPE_ULOG,
 575         .u = {
 576                 .ulog = {
 577                         .copy_len       = 0xffff,
 578                         .group          = 0,
 579                         .qthreshold     = 1,
 580                 },
 581         },
 582 };
 583
 584 /* log handler for internal netfilter logging api */
 585 static void
 586 nfulnl_log_packet(unsigned int pf,
 587                   unsigned int hooknum,
 588                   const struct sk_buff *skb,
 589                   const struct net_device *in,
 590                   const struct net_device *out,
 591                   const struct nf_loginfo *li_user,
 592                   const char *prefix)
 593 {
 594         unsigned int size, data_len;
 595         struct nfulnl_instance *inst;
 596         const struct nf_loginfo *li;
 597         unsigned int qthreshold;
 598         unsigned int nlbufsiz;
 599         unsigned int plen;
 600
 601         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 602                 li = li_user;
 603         else
 604                 li = &default_loginfo;
 605
 606         inst = instance_lookup_get(li->u.ulog.group);
 607         if (!inst)
 608                 inst = instance_lookup_get(0);
 609         if (!inst) {
 610                 PRINTR("nfnetlink_log: trying to log packet, "
 611                         "but no instance for group %u\n", li->u.ulog.group);
 612                 return;
 613         }
 614
 615         plen = 0;
 616         if (prefix)
 617                 plen = strlen(prefix) + 1;
 618
 619         /* all macros expand to constant values at compile time */
 620         /* FIXME: do we want to make the size calculation conditional based on
 621          * what is actually present?  way more branches and checks, but more
 622          * memory efficient... */
 623         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 624                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hdr))
 625                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 626                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 627 #ifdef CONFIG_BRIDGE_NETFILTER
 628                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 629                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 630 #endif
 631                 + NFA_SPACE(sizeof(u_int32_t))  /* mark */
 632                 + NFA_SPACE(sizeof(u_int32_t))  /* uid */
 633                 + NFA_SPACE(plen)               /* prefix */
 634                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hw))
 635                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_timestamp));
 636
 637         UDEBUG("initial size=%u\n", size);
 638
 639         spin_lock_bh(&inst->lock);
 640
 641         if (inst->flags & NFULNL_CFG_F_SEQ)
 642                 size += NFA_SPACE(sizeof(u_int32_t));
 643         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 644                 size += NFA_SPACE(sizeof(u_int32_t));
 645
 646         qthreshold = inst->qthreshold;
 647         /* per-rule qthreshold overrides per-instance */
 648         if (qthreshold > li->u.ulog.qthreshold)
 649                 qthreshold = li->u.ulog.qthreshold;
 650
 651         switch (inst->copy_mode) {
 652         case NFULNL_COPY_META:
 653         case NFULNL_COPY_NONE:
 654                 data_len = 0;
 655                 break;
 656
 657         case NFULNL_COPY_PACKET:
 658                 if (inst->copy_range == 0
 659                     || inst->copy_range > skb->len)
 660                         data_len = skb->len;
 661                 else
 662                         data_len = inst->copy_range;
 663
 664                 size += NFA_SPACE(data_len);
 665                 UDEBUG("copy_packet, therefore size now %u\n", size);
 666                 break;
 667
 668         default:
 669                 spin_unlock_bh(&inst->lock);
 670                 instance_put(inst);
 671                 return;
 672         }
 673
 674         if (size > inst->nlbufsiz)
 675                 nlbufsiz = size;
 676         else
 677                 nlbufsiz = inst->nlbufsiz;
 678
 679         if (!inst->skb) {
 680                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
 681                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
 682                                 inst->nlbufsiz, size);
 683                         goto alloc_failure;
 684                 }
 685         } else if (inst->qlen >= qthreshold ||
 686                    size > skb_tailroom(inst->skb)) {
 687                 /* either the queue len is too high or we don't have
 688                  * enough room in the skb left. flush to userspace. */
 689                 UDEBUG("flushing old skb\n");
 690
 691                 /* timer "holds" one reference (we have another one) */
 692                 if (del_timer(&inst->timer))
 693                         instance_put(inst);
 694                 __nfulnl_send(inst);
 695
 696                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
 697                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
 698                                 inst->nlbufsiz, size);
 699                         goto alloc_failure;
 700                 }
 701         }
 702
 703         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
 704         inst->qlen++;
 705
 706         __build_packet_message(inst, skb, data_len, pf,
 707                                 hooknum, in, out, li, prefix, plen);
 708
 709         /* timer_pending always called within inst->lock, so there
 710          * is no chance of a race here */
 711         if (!timer_pending(&inst->timer)) {
 712                 instance_get(inst);
 713                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 714                 add_timer(&inst->timer);
 715         }
 716
 717 unlock_and_release:
 718         spin_unlock_bh(&inst->lock);
 719         instance_put(inst);
 720         return;
 721
 722 alloc_failure:
 723         UDEBUG("error allocating skb\n");
 724         /* FIXME: statistics */
 725         goto unlock_and_release;
 726 }
 727
 728 static int
 729 nfulnl_rcv_nl_event(struct notifier_block *this,
 730                    unsigned long event, void *ptr)
 731 {
 732         struct netlink_notify *n = ptr;
 733
 734         if (event == NETLINK_URELEASE &&
 735             n->protocol == NETLINK_NETFILTER && n->pid) {
 736                 int i;
 737
 738                 /* destroy all instances for this pid */
 739                 write_lock_bh(&instances_lock);
 740                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 741                         struct hlist_node *tmp, *t2;
 742                         struct nfulnl_instance *inst;
 743                         struct hlist_head *head = &instance_table[i];
 744
 745                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 746                                 UDEBUG("node = %p\n", inst);
 747                                 if (n->pid == inst->peer_pid)
 748                                         __instance_destroy(inst);
 749                         }
 750                 }
 751                 write_unlock_bh(&instances_lock);
 752         }
 753         return NOTIFY_DONE;
 754 }
 755
 756 static struct notifier_block nfulnl_rtnl_notifier = {
 757         .notifier_call  = nfulnl_rcv_nl_event,
 758 };
 759
 760 static int
 761 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 762                   struct nlmsghdr *nlh, struct nfattr *nfqa[])
 763 {
 764         return -ENOTSUPP;
 765 }
 766
 767 static struct nf_logger nfulnl_logger = {
 768         .name   = "nfnetlink_log",
 769         .logfn  = &nfulnl_log_packet,
 770         .me     = THIS_MODULE,
 771 };
 772
 773 static const int nfula_min[NFULA_MAX] = {
 774         [NFULA_PACKET_HDR-1]    = sizeof(struct nfulnl_msg_packet_hdr),
 775         [NFULA_MARK-1]          = sizeof(u_int32_t),
 776         [NFULA_TIMESTAMP-1]     = sizeof(struct nfulnl_msg_packet_timestamp),
 777         [NFULA_IFINDEX_INDEV-1] = sizeof(u_int32_t),
 778         [NFULA_IFINDEX_OUTDEV-1]= sizeof(u_int32_t),
 779         [NFULA_IFINDEX_PHYSINDEV-1]     = sizeof(u_int32_t),
 780         [NFULA_IFINDEX_PHYSOUTDEV-1]    = sizeof(u_int32_t),
 781         [NFULA_HWADDR-1]        = sizeof(struct nfulnl_msg_packet_hw),
 782         [NFULA_PAYLOAD-1]       = 0,
 783         [NFULA_PREFIX-1]        = 0,
 784         [NFULA_UID-1]           = sizeof(u_int32_t),
 785         [NFULA_SEQ-1]           = sizeof(u_int32_t),
 786         [NFULA_SEQ_GLOBAL-1]    = sizeof(u_int32_t),
 787 };
 788
 789 static const int nfula_cfg_min[NFULA_CFG_MAX] = {
 790         [NFULA_CFG_CMD-1]       = sizeof(struct nfulnl_msg_config_cmd),
 791         [NFULA_CFG_MODE-1]      = sizeof(struct nfulnl_msg_config_mode),
 792         [NFULA_CFG_TIMEOUT-1]   = sizeof(u_int32_t),
 793         [NFULA_CFG_QTHRESH-1]   = sizeof(u_int32_t),
 794         [NFULA_CFG_NLBUFSIZ-1]  = sizeof(u_int32_t),
 795         [NFULA_CFG_FLAGS-1]     = sizeof(u_int16_t),
 796 };
 797
 798 static int
 799 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 800                    struct nlmsghdr *nlh, struct nfattr *nfula[])
 801 {
 802         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 803         u_int16_t group_num = ntohs(nfmsg->res_id);
 804         struct nfulnl_instance *inst;
 805         int ret = 0;
 806
 807         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
 808
 809         if (nfattr_bad_size(nfula, NFULA_CFG_MAX, nfula_cfg_min)) {
 810                 UDEBUG("bad attribute size\n");
 811                 return -EINVAL;
 812         }
 813
 814         inst = instance_lookup_get(group_num);
 815         if (nfula[NFULA_CFG_CMD-1]) {
 816                 u_int8_t pf = nfmsg->nfgen_family;
 817                 struct nfulnl_msg_config_cmd *cmd;
 818                 cmd = NFA_DATA(nfula[NFULA_CFG_CMD-1]);
 819                 UDEBUG("found CFG_CMD for\n");
 820
 821                 switch (cmd->command) {
 822                 case NFULNL_CFG_CMD_BIND:
 823                         if (inst) {
 824                                 ret = -EBUSY;
 825                                 goto out_put;
 826                         }
 827
 828                         inst = instance_create(group_num,
 829                                                NETLINK_CB(skb).pid);
 830                         if (!inst) {
 831                                 ret = -EINVAL;
 832                                 goto out_put;
 833                         }
 834                         break;
 835                 case NFULNL_CFG_CMD_UNBIND:
 836                         if (!inst) {
 837                                 ret = -ENODEV;
 838                                 goto out_put;
 839                         }
 840
 841                         if (inst->peer_pid != NETLINK_CB(skb).pid) {
 842                                 ret = -EPERM;
 843                                 goto out_put;
 844                         }
 845
 846                         instance_destroy(inst);
 847                         break;
 848                 case NFULNL_CFG_CMD_PF_BIND:
 849                         UDEBUG("registering log handler for pf=%u\n", pf);
 850                         ret = nf_log_register(pf, &nfulnl_logger);
 851                         break;
 852                 case NFULNL_CFG_CMD_PF_UNBIND:
 853                         UDEBUG("unregistering log handler for pf=%u\n", pf);
 854                         /* This is a bug and a feature.  We cannot unregister
 855                          * other handlers, like nfnetlink_inst can */
 856                         nf_log_unregister_pf(pf);
 857                         break;
 858                 default:
 859                         ret = -EINVAL;
 860                         break;
 861                 }
 862
 863                 if (!inst)
 864                         goto out;
 865         } else {
 866                 if (!inst) {
 867                         UDEBUG("no config command, and no instance for "
 868                                 "group=%u pid=%u =>ENOENT\n",
 869                                 group_num, NETLINK_CB(skb).pid);
 870                         ret = -ENOENT;
 871                         goto out_put;
 872                 }
 873
 874                 if (inst->peer_pid != NETLINK_CB(skb).pid) {
 875                         UDEBUG("no config command, and wrong pid\n");
 876                         ret = -EPERM;
 877                         goto out_put;
 878                 }
 879         }
 880
 881         if (nfula[NFULA_CFG_MODE-1]) {
 882                 struct nfulnl_msg_config_mode *params;
 883                 params = NFA_DATA(nfula[NFULA_CFG_MODE-1]);
 884
 885                 nfulnl_set_mode(inst, params->copy_mode,
 886                                 ntohl(params->copy_range));
 887         }
 888
 889         if (nfula[NFULA_CFG_TIMEOUT-1]) {
 890                 __be32 timeout =
 891                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_TIMEOUT-1]);
 892
 893                 nfulnl_set_timeout(inst, ntohl(timeout));
 894         }
 895
 896         if (nfula[NFULA_CFG_NLBUFSIZ-1]) {
 897                 __be32 nlbufsiz =
 898                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_NLBUFSIZ-1]);
 899
 900                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 901         }
 902
 903         if (nfula[NFULA_CFG_QTHRESH-1]) {
 904                 __be32 qthresh =
 905                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_QTHRESH-1]);
 906
 907                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 908         }
 909
 910         if (nfula[NFULA_CFG_FLAGS-1]) {
 911                 __be16 flags =
 912                         *(__be16 *)NFA_DATA(nfula[NFULA_CFG_FLAGS-1]);
 913                 nfulnl_set_flags(inst, ntohs(flags));
 914         }
 915
 916 out_put:
 917         instance_put(inst);
 918 out:
 919         return ret;
 920 }
 921
 922 static struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 923         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 924                                     .attr_count = NFULA_MAX, },
 925         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 926                                     .attr_count = NFULA_CFG_MAX, },
 927 };
 928
 929 static struct nfnetlink_subsystem nfulnl_subsys = {
 930         .name           = "log",
 931         .subsys_id      = NFNL_SUBSYS_ULOG,
 932         .cb_count       = NFULNL_MSG_MAX,
 933         .cb             = nfulnl_cb,
 934 };
 935
 936 #ifdef CONFIG_PROC_FS
 937 struct iter_state {
 938         unsigned int bucket;
 939 };
 940
 941 static struct hlist_node *get_first(struct seq_file *seq)
 942 {
 943         struct iter_state *st = seq->private;
 944
 945         if (!st)
 946                 return NULL;
 947
 948         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 949                 if (!hlist_empty(&instance_table[st->bucket]))
 950                         return instance_table[st->bucket].first;
 951         }
 952         return NULL;
 953 }
 954
 955 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
 956 {
 957         struct iter_state *st = seq->private;
 958
 959         h = h->next;
 960         while (!h) {
 961                 if (++st->bucket >= INSTANCE_BUCKETS)
 962                         return NULL;
 963
 964                 h = instance_table[st->bucket].first;
 965         }
 966         return h;
 967 }
 968
 969 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 970 {
 971         struct hlist_node *head;
 972         head = get_first(seq);
 973
 974         if (head)
 975                 while (pos && (head = get_next(seq, head)))
 976                         pos--;
 977         return pos ? NULL : head;
 978 }
 979
 980 static void *seq_start(struct seq_file *seq, loff_t *pos)
 981 {
 982         read_lock_bh(&instances_lock);
 983         return get_idx(seq, *pos);
 984 }
 985
 986 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 987 {
 988         (*pos)++;
 989         return get_next(s, v);
 990 }
 991
 992 static void seq_stop(struct seq_file *s, void *v)
 993 {
 994         read_unlock_bh(&instances_lock);
 995 }
 996
 997 static int seq_show(struct seq_file *s, void *v)
 998 {
 999         const struct nfulnl_instance *inst = v;
1000
1001         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
1002                           inst->group_num,
1003                           inst->peer_pid, inst->qlen,
1004                           inst->copy_mode, inst->copy_range,
1005                           inst->flushtimeout, atomic_read(&inst->use));
1006 }
1007
1008 static struct seq_operations nful_seq_ops = {
1009         .start  = seq_start,
1010         .next   = seq_next,
1011         .stop   = seq_stop,
1012         .show   = seq_show,
1013 };
1014
1015 static int nful_open(struct inode *inode, struct file *file)
1016 {
1017         struct seq_file *seq;
1018         struct iter_state *is;
1019         int ret;
1020
1021         is = kzalloc(sizeof(*is), GFP_KERNEL);
1022         if (!is)
1023                 return -ENOMEM;
1024         ret = seq_open(file, &nful_seq_ops);
1025         if (ret < 0)
1026                 goto out_free;
1027         seq = file->private_data;
1028         seq->private = is;
1029         return ret;
1030 out_free:
1031         kfree(is);
1032         return ret;
1033 }
1034
1035 static const struct file_operations nful_file_ops = {
1036         .owner   = THIS_MODULE,
1037         .open    = nful_open,
1038         .read    = seq_read,
1039         .llseek  = seq_lseek,
1040         .release = seq_release_private,
1041 };
1042
1043 #endif /* PROC_FS */
1044
1045 static int __init nfnetlink_log_init(void)
1046 {
1047         int i, status = -ENOMEM;
1048 #ifdef CONFIG_PROC_FS
1049         struct proc_dir_entry *proc_nful;
1050 #endif
1051
1052         for (i = 0; i < INSTANCE_BUCKETS; i++)
1053                 INIT_HLIST_HEAD(&instance_table[i]);
1054
1055         /* it's not really all that important to have a random value, so
1056          * we can do this from the init function, even if there hasn't
1057          * been that much entropy yet */
1058         get_random_bytes(&hash_init, sizeof(hash_init));
1059
1060         netlink_register_notifier(&nfulnl_rtnl_notifier);
1061         status = nfnetlink_subsys_register(&nfulnl_subsys);
1062         if (status < 0) {
1063                 printk(KERN_ERR "log: failed to create netlink socket\n");
1064                 goto cleanup_netlink_notifier;
1065         }
1066
1067 #ifdef CONFIG_PROC_FS
1068         proc_nful = create_proc_entry("nfnetlink_log", 0440,
1069                                       proc_net_netfilter);
1070         if (!proc_nful)
1071                 goto cleanup_subsys;
1072         proc_nful->proc_fops = &nful_file_ops;
1073 #endif
1074         return status;
1075
1076 #ifdef CONFIG_PROC_FS
1077 cleanup_subsys:
1078         nfnetlink_subsys_unregister(&nfulnl_subsys);
1079 #endif
1080 cleanup_netlink_notifier:
1081         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1082         return status;
1083 }
1084
1085 static void __exit nfnetlink_log_fini(void)
1086 {
1087         nf_log_unregister(&nfulnl_logger);
1088 #ifdef CONFIG_PROC_FS
1089         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1090 #endif
1091         nfnetlink_subsys_unregister(&nfulnl_subsys);
1092         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1093 }
1094
1095 MODULE_DESCRIPTION("netfilter userspace logging");
1096 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1097 MODULE_LICENSE("GPL");
1098 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1099
1100 module_init(nfnetlink_log_init);
1101 module_exit(nfnetlink_log_fini);