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 #include <linux/module.h>
  15 #include <linux/skbuff.h>
  16 #include <linux/init.h>
  17 #include <linux/ip.h>
  18 #include <linux/ipv6.h>
  19 #include <linux/netdevice.h>
  20 #include <linux/netfilter.h>
  21 #include <linux/netlink.h>
  22 #include <linux/netfilter/nfnetlink.h>
  23 #include <linux/netfilter/nfnetlink_log.h>
  24 #include <linux/spinlock.h>
  25 #include <linux/sysctl.h>
  26 #include <linux/proc_fs.h>
  27 #include <linux/security.h>
  28 #include <linux/list.h>
  29 #include <linux/jhash.h>
  30 #include <linux/random.h>
  31 #include <net/sock.h>
  32
  33 #include <asm/atomic.h>
  34
  35 #ifdef CONFIG_BRIDGE_NETFILTER
  36 #include "../bridge/br_private.h"
  37 #endif
  38
  39 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
  40 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
  41 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  42
  43 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  44                                      printk(x, ## args); } while (0);
  45
  46 #if 0
  47 #define UDEBUG(x, args ...)     printk(KERN_DEBUG "%s(%d):%s(): " x,       \
  48                                         __FILE__, __LINE__, __FUNCTION__,  \
  49                                         ## args)
  50 #else
  51 #define UDEBUG(x, ...)
  52 #endif
  53
  54 struct nfulnl_instance {
  55         struct hlist_node hlist;        /* global list of instances */
  56         spinlock_t lock;
  57         atomic_t use;                   /* use count */
  58
  59         unsigned int qlen;              /* number of nlmsgs in skb */
  60         struct sk_buff *skb;            /* pre-allocatd skb */
  61         struct nlmsghdr *lastnlh;       /* netlink header of last msg in skb */
  62         struct timer_list timer;
  63         int peer_pid;                   /* PID of the peer process */
  64
  65         /* configurable parameters */
  66         unsigned int flushtimeout;      /* timeout until queue flush */
  67         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  68         unsigned int qthreshold;        /* threshold of the queue */
  69         u_int32_t copy_range;
  70         u_int32_t seq;                  /* instance-local sequential counter */
  71         u_int16_t group_num;            /* number of this queue */
  72         u_int16_t flags;
  73         u_int8_t copy_mode;
  74 };
  75
  76 static DEFINE_RWLOCK(instances_lock);
  77 static atomic_t global_seq;
  78
  79 #define INSTANCE_BUCKETS        16
  80 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  81 static unsigned int hash_init;
  82
  83 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  84 {
  85         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  86 }
  87
  88 static struct nfulnl_instance *
  89 __instance_lookup(u_int16_t group_num)
  90 {
  91         struct hlist_head *head;
  92         struct hlist_node *pos;
  93         struct nfulnl_instance *inst;
  94
  95         UDEBUG("entering (group_num=%u)\n", group_num);
  96
  97         head = &instance_table[instance_hashfn(group_num)];
  98         hlist_for_each_entry(inst, pos, head, hlist) {
  99                 if (inst->group_num == group_num)
 100                         return inst;
 101         }
 102         return NULL;
 103 }
 104
 105 static inline void
 106 instance_get(struct nfulnl_instance *inst)
 107 {
 108         atomic_inc(&inst->use);
 109 }
 110
 111 static struct nfulnl_instance *
 112 instance_lookup_get(u_int16_t group_num)
 113 {
 114         struct nfulnl_instance *inst;
 115
 116         read_lock_bh(&instances_lock);
 117         inst = __instance_lookup(group_num);
 118         if (inst)
 119                 instance_get(inst);
 120         read_unlock_bh(&instances_lock);
 121
 122         return inst;
 123 }
 124
 125 static void
 126 instance_put(struct nfulnl_instance *inst)
 127 {
 128         if (inst && atomic_dec_and_test(&inst->use)) {
 129                 UDEBUG("kfree(inst=%p)\n", inst);
 130                 kfree(inst);
 131                 module_put(THIS_MODULE);
 132         }
 133 }
 134
 135 static void nfulnl_timer(unsigned long data);
 136
 137 static struct nfulnl_instance *
 138 instance_create(u_int16_t group_num, int pid)
 139 {
 140         struct nfulnl_instance *inst;
 141
 142         UDEBUG("entering (group_num=%u, pid=%d)\n", group_num,
 143                 pid);
 144
 145         write_lock_bh(&instances_lock);
 146         if (__instance_lookup(group_num)) {
 147                 inst = NULL;
 148                 UDEBUG("aborting, instance already exists\n");
 149                 goto out_unlock;
 150         }
 151
 152         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 153         if (!inst)
 154                 goto out_unlock;
 155
 156         INIT_HLIST_NODE(&inst->hlist);
 157         spin_lock_init(&inst->lock);
 158         /* needs to be two, since we _put() after creation */
 159         atomic_set(&inst->use, 2);
 160
 161         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
 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_destroy(struct nfulnl_instance *inst)
 196 {
 197         /* first pull it out of the global list */
 198         UDEBUG("removing instance %p (queuenum=%u) from hash\n",
 199                 inst, inst->group_num);
 200
 201         hlist_del(&inst->hlist);
 202
 203         /* then flush all pending packets from skb */
 204
 205         spin_lock_bh(&inst->lock);
 206         if (inst->skb) {
 207                 /* timer "holds" one reference (we have one more) */
 208                 if (del_timer(&inst->timer))
 209                         instance_put(inst);
 210                 if (inst->qlen)
 211                         __nfulnl_send(inst);
 212                 if (inst->skb) {
 213                         kfree_skb(inst->skb);
 214                         inst->skb = NULL;
 215                 }
 216         }
 217         spin_unlock_bh(&inst->lock);
 218
 219         /* and finally put the refcount */
 220         instance_put(inst);
 221 }
 222
 223 static inline void
 224 instance_destroy(struct nfulnl_instance *inst)
 225 {
 226         write_lock_bh(&instances_lock);
 227         __instance_destroy(inst);
 228         write_unlock_bh(&instances_lock);
 229 }
 230
 231 static int
 232 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 233                   unsigned int range)
 234 {
 235         int status = 0;
 236
 237         spin_lock_bh(&inst->lock);
 238
 239         switch (mode) {
 240         case NFULNL_COPY_NONE:
 241         case NFULNL_COPY_META:
 242                 inst->copy_mode = mode;
 243                 inst->copy_range = 0;
 244                 break;
 245
 246         case NFULNL_COPY_PACKET:
 247                 inst->copy_mode = mode;
 248                 /* we're using struct nfattr which has 16bit nfa_len */
 249                 if (range > 0xffff)
 250                         inst->copy_range = 0xffff;
 251                 else
 252                         inst->copy_range = range;
 253                 break;
 254
 255         default:
 256                 status = -EINVAL;
 257                 break;
 258         }
 259
 260         spin_unlock_bh(&inst->lock);
 261
 262         return status;
 263 }
 264
 265 static int
 266 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 267 {
 268         int status;
 269
 270         spin_lock_bh(&inst->lock);
 271         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 272                 status = -ERANGE;
 273         else if (nlbufsiz > 131072)
 274                 status = -ERANGE;
 275         else {
 276                 inst->nlbufsiz = nlbufsiz;
 277                 status = 0;
 278         }
 279         spin_unlock_bh(&inst->lock);
 280
 281         return status;
 282 }
 283
 284 static int
 285 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 286 {
 287         spin_lock_bh(&inst->lock);
 288         inst->flushtimeout = timeout;
 289         spin_unlock_bh(&inst->lock);
 290
 291         return 0;
 292 }
 293
 294 static int
 295 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 296 {
 297         spin_lock_bh(&inst->lock);
 298         inst->qthreshold = qthresh;
 299         spin_unlock_bh(&inst->lock);
 300
 301         return 0;
 302 }
 303
 304 static int
 305 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 306 {
 307         spin_lock_bh(&inst->lock);
 308         inst->flags = flags;
 309         spin_unlock_bh(&inst->lock);
 310
 311         return 0;
 312 }
 313
 314 static struct sk_buff *nfulnl_alloc_skb(unsigned int inst_size,
 315                                         unsigned int pkt_size)
 316 {
 317         struct sk_buff *skb;
 318         unsigned int n;
 319
 320         UDEBUG("entered (%u, %u)\n", inst_size, pkt_size);
 321
 322         /* alloc skb which should be big enough for a whole multipart
 323          * message.  WARNING: has to be <= 128k due to slab restrictions */
 324
 325         n = max(inst_size, pkt_size);
 326         skb = alloc_skb(n, GFP_ATOMIC);
 327         if (!skb) {
 328                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
 329                         inst_size);
 330
 331                 if (n > pkt_size) {
 332                         /* try to allocate only as much as we need for current
 333                          * packet */
 334
 335                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 336                         if (!skb)
 337                                 PRINTR("nfnetlink_log: can't even alloc %u "
 338                                        "bytes\n", pkt_size);
 339                 }
 340         }
 341
 342         return skb;
 343 }
 344
 345 static int
 346 __nfulnl_send(struct nfulnl_instance *inst)
 347 {
 348         int status;
 349
 350         if (inst->qlen > 1)
 351                 inst->lastnlh->nlmsg_type = NLMSG_DONE;
 352
 353         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
 354         if (status < 0) {
 355                 UDEBUG("netlink_unicast() failed\n");
 356                 /* FIXME: statistics */
 357         }
 358
 359         inst->qlen = 0;
 360         inst->skb = NULL;
 361         inst->lastnlh = NULL;
 362
 363         return status;
 364 }
 365
 366 static void nfulnl_timer(unsigned long data)
 367 {
 368         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 369
 370         UDEBUG("timer function called, flushing buffer\n");
 371
 372         spin_lock_bh(&inst->lock);
 373         if (inst->skb)
 374                 __nfulnl_send(inst);
 375         spin_unlock_bh(&inst->lock);
 376         instance_put(inst);
 377 }
 378
 379 /* This is an inline function, we don't really care about a long
 380  * list of arguments */
 381 static inline int
 382 __build_packet_message(struct nfulnl_instance *inst,
 383                         const struct sk_buff *skb,
 384                         unsigned int data_len,
 385                         unsigned int pf,
 386                         unsigned int hooknum,
 387                         const struct net_device *indev,
 388                         const struct net_device *outdev,
 389                         const struct nf_loginfo *li,
 390                         const char *prefix, unsigned int plen)
 391 {
 392         struct nfulnl_msg_packet_hdr pmsg;
 393         struct nlmsghdr *nlh;
 394         struct nfgenmsg *nfmsg;
 395         __be32 tmp_uint;
 396         sk_buff_data_t old_tail = inst->skb->tail;
 397
 398         UDEBUG("entered\n");
 399
 400         nlh = NLMSG_PUT(inst->skb, 0, 0,
 401                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 402                         sizeof(struct nfgenmsg));
 403         nfmsg = NLMSG_DATA(nlh);
 404         nfmsg->nfgen_family = pf;
 405         nfmsg->version = NFNETLINK_V0;
 406         nfmsg->res_id = htons(inst->group_num);
 407
 408         pmsg.hw_protocol        = skb->protocol;
 409         pmsg.hook               = hooknum;
 410
 411         NFA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
 412
 413         if (prefix)
 414                 NFA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
 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 && skb->nf_bridge->physoutdev) {
 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->mark) {
 479                 tmp_uint = htonl(skb->mark);
 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                 int len = skb->dev->hard_header_parse((struct sk_buff *)skb,
 486                                                     phw.hw_addr);
 487                 phw.hw_addrlen = htons(len);
 488                 NFA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
 489         }
 490
 491         if (skb->tstamp.tv64) {
 492                 struct nfulnl_msg_packet_timestamp ts;
 493                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 494                 ts.sec = cpu_to_be64(tv.tv_sec);
 495                 ts.usec = cpu_to_be64(tv.tv_usec);
 496
 497                 NFA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 498         }
 499
 500         /* UID */
 501         if (skb->sk) {
 502                 read_lock_bh(&skb->sk->sk_callback_lock);
 503                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 504                         __be32 uid = htonl(skb->sk->sk_socket->file->f_uid);
 505                         /* need to unlock here since NFA_PUT may goto */
 506                         read_unlock_bh(&skb->sk->sk_callback_lock);
 507                         NFA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
 508                 } else
 509                         read_unlock_bh(&skb->sk->sk_callback_lock);
 510         }
 511
 512         /* local sequence number */
 513         if (inst->flags & NFULNL_CFG_F_SEQ) {
 514                 tmp_uint = htonl(inst->seq++);
 515                 NFA_PUT(inst->skb, NFULA_SEQ, sizeof(tmp_uint), &tmp_uint);
 516         }
 517         /* global sequence number */
 518         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) {
 519                 tmp_uint = htonl(atomic_inc_return(&global_seq));
 520                 NFA_PUT(inst->skb, NFULA_SEQ_GLOBAL, sizeof(tmp_uint), &tmp_uint);
 521         }
 522
 523         if (data_len) {
 524                 struct nfattr *nfa;
 525                 int size = NFA_LENGTH(data_len);
 526
 527                 if (skb_tailroom(inst->skb) < (int)NFA_SPACE(data_len)) {
 528                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 529                         goto nlmsg_failure;
 530                 }
 531
 532                 nfa = (struct nfattr *)skb_put(inst->skb, NFA_ALIGN(size));
 533                 nfa->nfa_type = NFULA_PAYLOAD;
 534                 nfa->nfa_len = size;
 535
 536                 if (skb_copy_bits(skb, 0, NFA_DATA(nfa), data_len))
 537                         BUG();
 538         }
 539
 540         nlh->nlmsg_len = inst->skb->tail - old_tail;
 541         inst->lastnlh = nlh;
 542         return 0;
 543
 544 nlmsg_failure:
 545         UDEBUG("nlmsg_failure\n");
 546 nfattr_failure:
 547         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 548         return -1;
 549 }
 550
 551 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 552
 553 static struct nf_loginfo default_loginfo = {
 554         .type =         NF_LOG_TYPE_ULOG,
 555         .u = {
 556                 .ulog = {
 557                         .copy_len       = 0xffff,
 558                         .group          = 0,
 559                         .qthreshold     = 1,
 560                 },
 561         },
 562 };
 563
 564 /* log handler for internal netfilter logging api */
 565 static void
 566 nfulnl_log_packet(unsigned int pf,
 567                   unsigned int hooknum,
 568                   const struct sk_buff *skb,
 569                   const struct net_device *in,
 570                   const struct net_device *out,
 571                   const struct nf_loginfo *li_user,
 572                   const char *prefix)
 573 {
 574         unsigned int size, data_len;
 575         struct nfulnl_instance *inst;
 576         const struct nf_loginfo *li;
 577         unsigned int qthreshold;
 578         unsigned int plen;
 579
 580         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 581                 li = li_user;
 582         else
 583                 li = &default_loginfo;
 584
 585         inst = instance_lookup_get(li->u.ulog.group);
 586         if (!inst)
 587                 return;
 588
 589         plen = 0;
 590         if (prefix)
 591                 plen = strlen(prefix) + 1;
 592
 593         /* all macros expand to constant values at compile time */
 594         /* FIXME: do we want to make the size calculation conditional based on
 595          * what is actually present?  way more branches and checks, but more
 596          * memory efficient... */
 597         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 598                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hdr))
 599                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 600                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 601 #ifdef CONFIG_BRIDGE_NETFILTER
 602                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 603                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 604 #endif
 605                 + NFA_SPACE(sizeof(u_int32_t))  /* mark */
 606                 + NFA_SPACE(sizeof(u_int32_t))  /* uid */
 607                 + NFA_SPACE(plen)               /* prefix */
 608                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hw))
 609                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_timestamp));
 610
 611         UDEBUG("initial size=%u\n", size);
 612
 613         spin_lock_bh(&inst->lock);
 614
 615         if (inst->flags & NFULNL_CFG_F_SEQ)
 616                 size += NFA_SPACE(sizeof(u_int32_t));
 617         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 618                 size += NFA_SPACE(sizeof(u_int32_t));
 619
 620         qthreshold = inst->qthreshold;
 621         /* per-rule qthreshold overrides per-instance */
 622         if (qthreshold > li->u.ulog.qthreshold)
 623                 qthreshold = li->u.ulog.qthreshold;
 624
 625         switch (inst->copy_mode) {
 626         case NFULNL_COPY_META:
 627         case NFULNL_COPY_NONE:
 628                 data_len = 0;
 629                 break;
 630
 631         case NFULNL_COPY_PACKET:
 632                 if (inst->copy_range == 0
 633                     || inst->copy_range > skb->len)
 634                         data_len = skb->len;
 635                 else
 636                         data_len = inst->copy_range;
 637
 638                 size += NFA_SPACE(data_len);
 639                 UDEBUG("copy_packet, therefore size now %u\n", size);
 640                 break;
 641
 642         default:
 643                 goto unlock_and_release;
 644         }
 645
 646         if (inst->qlen >= qthreshold ||
 647             (inst->skb && size > skb_tailroom(inst->skb))) {
 648                 /* either the queue len is too high or we don't have
 649                  * enough room in the skb left. flush to userspace. */
 650                 UDEBUG("flushing old skb\n");
 651
 652                 /* timer "holds" one reference (we have another one) */
 653                 if (del_timer(&inst->timer))
 654                         instance_put(inst);
 655                 __nfulnl_send(inst);
 656         }
 657
 658         if (!inst->skb) {
 659                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
 660                 if (!inst->skb)
 661                         goto alloc_failure;
 662         }
 663
 664         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
 665         inst->qlen++;
 666
 667         __build_packet_message(inst, skb, data_len, pf,
 668                                 hooknum, in, out, li, prefix, plen);
 669
 670         /* timer_pending always called within inst->lock, so there
 671          * is no chance of a race here */
 672         if (!timer_pending(&inst->timer)) {
 673                 instance_get(inst);
 674                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 675                 add_timer(&inst->timer);
 676         }
 677
 678 unlock_and_release:
 679         spin_unlock_bh(&inst->lock);
 680         instance_put(inst);
 681         return;
 682
 683 alloc_failure:
 684         UDEBUG("error allocating skb\n");
 685         /* FIXME: statistics */
 686         goto unlock_and_release;
 687 }
 688
 689 static int
 690 nfulnl_rcv_nl_event(struct notifier_block *this,
 691                    unsigned long event, void *ptr)
 692 {
 693         struct netlink_notify *n = ptr;
 694
 695         if (event == NETLINK_URELEASE &&
 696             n->protocol == NETLINK_NETFILTER && n->pid) {
 697                 int i;
 698
 699                 /* destroy all instances for this pid */
 700                 write_lock_bh(&instances_lock);
 701                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 702                         struct hlist_node *tmp, *t2;
 703                         struct nfulnl_instance *inst;
 704                         struct hlist_head *head = &instance_table[i];
 705
 706                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 707                                 UDEBUG("node = %p\n", inst);
 708                                 if (n->pid == inst->peer_pid)
 709                                         __instance_destroy(inst);
 710                         }
 711                 }
 712                 write_unlock_bh(&instances_lock);
 713         }
 714         return NOTIFY_DONE;
 715 }
 716
 717 static struct notifier_block nfulnl_rtnl_notifier = {
 718         .notifier_call  = nfulnl_rcv_nl_event,
 719 };
 720
 721 static int
 722 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 723                   struct nlmsghdr *nlh, struct nfattr *nfqa[])
 724 {
 725         return -ENOTSUPP;
 726 }
 727
 728 static struct nf_logger nfulnl_logger = {
 729         .name   = "nfnetlink_log",
 730         .logfn  = &nfulnl_log_packet,
 731         .me     = THIS_MODULE,
 732 };
 733
 734 static const int nfula_min[NFULA_MAX] = {
 735         [NFULA_PACKET_HDR-1]    = sizeof(struct nfulnl_msg_packet_hdr),
 736         [NFULA_MARK-1]          = sizeof(u_int32_t),
 737         [NFULA_TIMESTAMP-1]     = sizeof(struct nfulnl_msg_packet_timestamp),
 738         [NFULA_IFINDEX_INDEV-1] = sizeof(u_int32_t),
 739         [NFULA_IFINDEX_OUTDEV-1]= sizeof(u_int32_t),
 740         [NFULA_IFINDEX_PHYSINDEV-1]     = sizeof(u_int32_t),
 741         [NFULA_IFINDEX_PHYSOUTDEV-1]    = sizeof(u_int32_t),
 742         [NFULA_HWADDR-1]        = sizeof(struct nfulnl_msg_packet_hw),
 743         [NFULA_PAYLOAD-1]       = 0,
 744         [NFULA_PREFIX-1]        = 0,
 745         [NFULA_UID-1]           = sizeof(u_int32_t),
 746         [NFULA_SEQ-1]           = sizeof(u_int32_t),
 747         [NFULA_SEQ_GLOBAL-1]    = sizeof(u_int32_t),
 748 };
 749
 750 static const int nfula_cfg_min[NFULA_CFG_MAX] = {
 751         [NFULA_CFG_CMD-1]       = sizeof(struct nfulnl_msg_config_cmd),
 752         [NFULA_CFG_MODE-1]      = sizeof(struct nfulnl_msg_config_mode),
 753         [NFULA_CFG_TIMEOUT-1]   = sizeof(u_int32_t),
 754         [NFULA_CFG_QTHRESH-1]   = sizeof(u_int32_t),
 755         [NFULA_CFG_NLBUFSIZ-1]  = sizeof(u_int32_t),
 756         [NFULA_CFG_FLAGS-1]     = sizeof(u_int16_t),
 757 };
 758
 759 static int
 760 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 761                    struct nlmsghdr *nlh, struct nfattr *nfula[])
 762 {
 763         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 764         u_int16_t group_num = ntohs(nfmsg->res_id);
 765         struct nfulnl_instance *inst;
 766         int ret = 0;
 767
 768         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
 769
 770         if (nfattr_bad_size(nfula, NFULA_CFG_MAX, nfula_cfg_min)) {
 771                 UDEBUG("bad attribute size\n");
 772                 return -EINVAL;
 773         }
 774
 775         inst = instance_lookup_get(group_num);
 776         if (nfula[NFULA_CFG_CMD-1]) {
 777                 u_int8_t pf = nfmsg->nfgen_family;
 778                 struct nfulnl_msg_config_cmd *cmd;
 779                 cmd = NFA_DATA(nfula[NFULA_CFG_CMD-1]);
 780                 UDEBUG("found CFG_CMD for\n");
 781
 782                 switch (cmd->command) {
 783                 case NFULNL_CFG_CMD_BIND:
 784                         if (inst) {
 785                                 ret = -EBUSY;
 786                                 goto out_put;
 787                         }
 788
 789                         inst = instance_create(group_num,
 790                                                NETLINK_CB(skb).pid);
 791                         if (!inst) {
 792                                 ret = -EINVAL;
 793                                 goto out;
 794                         }
 795                         break;
 796                 case NFULNL_CFG_CMD_UNBIND:
 797                         if (!inst) {
 798                                 ret = -ENODEV;
 799                                 goto out;
 800                         }
 801
 802                         if (inst->peer_pid != NETLINK_CB(skb).pid) {
 803                                 ret = -EPERM;
 804                                 goto out_put;
 805                         }
 806
 807                         instance_destroy(inst);
 808                         goto out;
 809                 case NFULNL_CFG_CMD_PF_BIND:
 810                         UDEBUG("registering log handler for pf=%u\n", pf);
 811                         ret = nf_log_register(pf, &nfulnl_logger);
 812                         break;
 813                 case NFULNL_CFG_CMD_PF_UNBIND:
 814                         UDEBUG("unregistering log handler for pf=%u\n", pf);
 815                         /* This is a bug and a feature.  We cannot unregister
 816                          * other handlers, like nfnetlink_inst can */
 817                         nf_log_unregister_pf(pf);
 818                         break;
 819                 default:
 820                         ret = -EINVAL;
 821                         break;
 822                 }
 823
 824                 if (!inst)
 825                         goto out;
 826         } else {
 827                 if (!inst) {
 828                         UDEBUG("no config command, and no instance for "
 829                                 "group=%u pid=%u =>ENOENT\n",
 830                                 group_num, NETLINK_CB(skb).pid);
 831                         ret = -ENOENT;
 832                         goto out;
 833                 }
 834
 835                 if (inst->peer_pid != NETLINK_CB(skb).pid) {
 836                         UDEBUG("no config command, and wrong pid\n");
 837                         ret = -EPERM;
 838                         goto out_put;
 839                 }
 840         }
 841
 842         if (nfula[NFULA_CFG_MODE-1]) {
 843                 struct nfulnl_msg_config_mode *params;
 844                 params = NFA_DATA(nfula[NFULA_CFG_MODE-1]);
 845
 846                 nfulnl_set_mode(inst, params->copy_mode,
 847                                 ntohl(params->copy_range));
 848         }
 849
 850         if (nfula[NFULA_CFG_TIMEOUT-1]) {
 851                 __be32 timeout =
 852                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_TIMEOUT-1]);
 853
 854                 nfulnl_set_timeout(inst, ntohl(timeout));
 855         }
 856
 857         if (nfula[NFULA_CFG_NLBUFSIZ-1]) {
 858                 __be32 nlbufsiz =
 859                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_NLBUFSIZ-1]);
 860
 861                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 862         }
 863
 864         if (nfula[NFULA_CFG_QTHRESH-1]) {
 865                 __be32 qthresh =
 866                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_QTHRESH-1]);
 867
 868                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 869         }
 870
 871         if (nfula[NFULA_CFG_FLAGS-1]) {
 872                 __be16 flags =
 873                         *(__be16 *)NFA_DATA(nfula[NFULA_CFG_FLAGS-1]);
 874                 nfulnl_set_flags(inst, ntohs(flags));
 875         }
 876
 877 out_put:
 878         instance_put(inst);
 879 out:
 880         return ret;
 881 }
 882
 883 static struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 884         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 885                                     .attr_count = NFULA_MAX, },
 886         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 887                                     .attr_count = NFULA_CFG_MAX, },
 888 };
 889
 890 static struct nfnetlink_subsystem nfulnl_subsys = {
 891         .name           = "log",
 892         .subsys_id      = NFNL_SUBSYS_ULOG,
 893         .cb_count       = NFULNL_MSG_MAX,
 894         .cb             = nfulnl_cb,
 895 };
 896
 897 #ifdef CONFIG_PROC_FS
 898 struct iter_state {
 899         unsigned int bucket;
 900 };
 901
 902 static struct hlist_node *get_first(struct iter_state *st)
 903 {
 904         if (!st)
 905                 return NULL;
 906
 907         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 908                 if (!hlist_empty(&instance_table[st->bucket]))
 909                         return instance_table[st->bucket].first;
 910         }
 911         return NULL;
 912 }
 913
 914 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
 915 {
 916         h = h->next;
 917         while (!h) {
 918                 if (++st->bucket >= INSTANCE_BUCKETS)
 919                         return NULL;
 920
 921                 h = instance_table[st->bucket].first;
 922         }
 923         return h;
 924 }
 925
 926 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
 927 {
 928         struct hlist_node *head;
 929         head = get_first(st);
 930
 931         if (head)
 932                 while (pos && (head = get_next(st, head)))
 933                         pos--;
 934         return pos ? NULL : head;
 935 }
 936
 937 static void *seq_start(struct seq_file *seq, loff_t *pos)
 938 {
 939         read_lock_bh(&instances_lock);
 940         return get_idx(seq->private, *pos);
 941 }
 942
 943 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 944 {
 945         (*pos)++;
 946         return get_next(s->private, v);
 947 }
 948
 949 static void seq_stop(struct seq_file *s, void *v)
 950 {
 951         read_unlock_bh(&instances_lock);
 952 }
 953
 954 static int seq_show(struct seq_file *s, void *v)
 955 {
 956         const struct nfulnl_instance *inst = v;
 957
 958         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 959                           inst->group_num,
 960                           inst->peer_pid, inst->qlen,
 961                           inst->copy_mode, inst->copy_range,
 962                           inst->flushtimeout, atomic_read(&inst->use));
 963 }
 964
 965 static struct seq_operations nful_seq_ops = {
 966         .start  = seq_start,
 967         .next   = seq_next,
 968         .stop   = seq_stop,
 969         .show   = seq_show,
 970 };
 971
 972 static int nful_open(struct inode *inode, struct file *file)
 973 {
 974         struct seq_file *seq;
 975         struct iter_state *is;
 976         int ret;
 977
 978         is = kzalloc(sizeof(*is), GFP_KERNEL);
 979         if (!is)
 980                 return -ENOMEM;
 981         ret = seq_open(file, &nful_seq_ops);
 982         if (ret < 0)
 983                 goto out_free;
 984         seq = file->private_data;
 985         seq->private = is;
 986         return ret;
 987 out_free:
 988         kfree(is);
 989         return ret;
 990 }
 991
 992 static const struct file_operations nful_file_ops = {
 993         .owner   = THIS_MODULE,
 994         .open    = nful_open,
 995         .read    = seq_read,
 996         .llseek  = seq_lseek,
 997         .release = seq_release_private,
 998 };
 999
1000 #endif /* PROC_FS */
1001
1002 static int __init nfnetlink_log_init(void)
1003 {
1004         int i, status = -ENOMEM;
1005 #ifdef CONFIG_PROC_FS
1006         struct proc_dir_entry *proc_nful;
1007 #endif
1008
1009         for (i = 0; i < INSTANCE_BUCKETS; i++)
1010                 INIT_HLIST_HEAD(&instance_table[i]);
1011
1012         /* it's not really all that important to have a random value, so
1013          * we can do this from the init function, even if there hasn't
1014          * been that much entropy yet */
1015         get_random_bytes(&hash_init, sizeof(hash_init));
1016
1017         netlink_register_notifier(&nfulnl_rtnl_notifier);
1018         status = nfnetlink_subsys_register(&nfulnl_subsys);
1019         if (status < 0) {
1020                 printk(KERN_ERR "log: failed to create netlink socket\n");
1021                 goto cleanup_netlink_notifier;
1022         }
1023
1024 #ifdef CONFIG_PROC_FS
1025         proc_nful = create_proc_entry("nfnetlink_log", 0440,
1026                                       proc_net_netfilter);
1027         if (!proc_nful)
1028                 goto cleanup_subsys;
1029         proc_nful->proc_fops = &nful_file_ops;
1030 #endif
1031         return status;
1032
1033 #ifdef CONFIG_PROC_FS
1034 cleanup_subsys:
1035         nfnetlink_subsys_unregister(&nfulnl_subsys);
1036 #endif
1037 cleanup_netlink_notifier:
1038         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1039         return status;
1040 }
1041
1042 static void __exit nfnetlink_log_fini(void)
1043 {
1044         nf_log_unregister(&nfulnl_logger);
1045 #ifdef CONFIG_PROC_FS
1046         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1047 #endif
1048         nfnetlink_subsys_unregister(&nfulnl_subsys);
1049         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1050 }
1051
1052 MODULE_DESCRIPTION("netfilter userspace logging");
1053 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1054 MODULE_LICENSE("GPL");
1055 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1056
1057 module_init(nfnetlink_log_init);
1058 module_exit(nfnetlink_log_fini);