net/netfilter/nfnetlink_log.c

   1 /*
   2  * This is a module which is used for logging packets to userspace via
   3  * nfetlink.
   4  *
   5  * (C) 2005 by Harald Welte <laforge@netfilter.org>
   6  *
   7  * Based on the old ipv4-only ipt_ULOG.c:
   8  * (C) 2000-2004 by Harald Welte <laforge@netfilter.org>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License version 2 as
  12  * published by the Free Software Foundation.
  13  *
  14  */
  15 #include <linux/module.h>
  16 #include <linux/skbuff.h>
  17 #include <linux/init.h>
  18 #include <linux/ip.h>
  19 #include <linux/ipv6.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/netfilter.h>
  22 #include <linux/netlink.h>
  23 #include <linux/netfilter/nfnetlink.h>
  24 #include <linux/netfilter/nfnetlink_log.h>
  25 #include <linux/spinlock.h>
  26 #include <linux/sysctl.h>
  27 #include <linux/proc_fs.h>
  28 #include <linux/security.h>
  29 #include <linux/list.h>
  30 #include <linux/jhash.h>
  31 #include <linux/random.h>
  32 #include <net/sock.h>
  33
  34 #include <asm/atomic.h>
  35
  36 #ifdef CONFIG_BRIDGE_NETFILTER
  37 #include "../bridge/br_private.h"
  38 #endif
  39
  40 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
  41 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
  42 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  43
  44 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  45                                      printk(x, ## args); } while (0);
  46
  47 #if 0
  48 #define UDEBUG(x, args ...)     printk(KERN_DEBUG "%s(%d):%s(): " x,       \
  49                                         __FILE__, __LINE__, __FUNCTION__,  \
  50                                         ## args)
  51 #else
  52 #define UDEBUG(x, ...)
  53 #endif
  54
  55 struct nfulnl_instance {
  56         struct hlist_node hlist;        /* global list of instances */
  57         spinlock_t lock;
  58         atomic_t use;                   /* use count */
  59
  60         unsigned int qlen;              /* number of nlmsgs in skb */
  61         struct sk_buff *skb;            /* pre-allocatd skb */
  62         struct nlmsghdr *lastnlh;       /* netlink header of last msg in skb */
  63         struct timer_list timer;
  64         int peer_pid;                   /* PID of the peer process */
  65
  66         /* configurable parameters */
  67         unsigned int flushtimeout;      /* timeout until queue flush */
  68         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  69         unsigned int qthreshold;        /* threshold of the queue */
  70         u_int32_t copy_range;
  71         u_int16_t group_num;            /* number of this queue */
  72         u_int8_t copy_mode;
  73 };
  74
  75 static DEFINE_RWLOCK(instances_lock);
  76
  77 #define INSTANCE_BUCKETS        16
  78 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  79 static unsigned int hash_init;
  80
  81 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  82 {
  83         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  84 }
  85
  86 static struct nfulnl_instance *
  87 __instance_lookup(u_int16_t group_num)
  88 {
  89         struct hlist_head *head;
  90         struct hlist_node *pos;
  91         struct nfulnl_instance *inst;
  92
  93         UDEBUG("entering (group_num=%u)\n", group_num);
  94
  95         head = &instance_table[instance_hashfn(group_num)];
  96         hlist_for_each_entry(inst, pos, head, hlist) {
  97                 if (inst->group_num == group_num)
  98                         return inst;
  99         }
 100         return NULL;
 101 }
 102
 103 static inline void
 104 instance_get(struct nfulnl_instance *inst)
 105 {
 106         atomic_inc(&inst->use);
 107 }
 108
 109 static struct nfulnl_instance *
 110 instance_lookup_get(u_int16_t group_num)
 111 {
 112         struct nfulnl_instance *inst;
 113
 114         read_lock_bh(&instances_lock);
 115         inst = __instance_lookup(group_num);
 116         if (inst)
 117                 instance_get(inst);
 118         read_unlock_bh(&instances_lock);
 119
 120         return inst;
 121 }
 122
 123 static void
 124 instance_put(struct nfulnl_instance *inst)
 125 {
 126         if (inst && atomic_dec_and_test(&inst->use)) {
 127                 UDEBUG("kfree(inst=%p)\n", inst);
 128                 kfree(inst);
 129         }
 130 }
 131
 132 static void nfulnl_timer(unsigned long data);
 133
 134 static struct nfulnl_instance *
 135 instance_create(u_int16_t group_num, int pid)
 136 {
 137         struct nfulnl_instance *inst;
 138
 139         UDEBUG("entering (group_num=%u, pid=%d)\n", group_num,
 140                 pid);
 141
 142         write_lock_bh(&instances_lock);
 143         if (__instance_lookup(group_num)) {
 144                 inst = NULL;
 145                 UDEBUG("aborting, instance already exists\n");
 146                 goto out_unlock;
 147         }
 148
 149         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 150         if (!inst)
 151                 goto out_unlock;
 152
 153         INIT_HLIST_NODE(&inst->hlist);
 154         spin_lock_init(&inst->lock);
 155         /* needs to be two, since we _put() after creation */
 156         atomic_set(&inst->use, 2);
 157
 158         init_timer(&inst->timer);
 159         inst->timer.function = nfulnl_timer;
 160         inst->timer.data = (unsigned long)inst;
 161         /* don't start timer yet. (re)start it  with every packet */
 162
 163         inst->peer_pid = pid;
 164         inst->group_num = group_num;
 165
 166         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 167         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 168         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 169         inst->copy_mode         = NFULNL_COPY_PACKET;
 170         inst->copy_range        = 0xffff;
 171
 172         if (!try_module_get(THIS_MODULE))
 173                 goto out_free;
 174
 175         hlist_add_head(&inst->hlist,
 176                        &instance_table[instance_hashfn(group_num)]);
 177
 178         UDEBUG("newly added node: %p, next=%p\n", &inst->hlist,
 179                 inst->hlist.next);
 180
 181         write_unlock_bh(&instances_lock);
 182
 183         return inst;
 184
 185 out_free:
 186         instance_put(inst);
 187 out_unlock:
 188         write_unlock_bh(&instances_lock);
 189         return NULL;
 190 }
 191
 192 static int __nfulnl_send(struct nfulnl_instance *inst);
 193
 194 static void
 195 _instance_destroy2(struct nfulnl_instance *inst, int lock)
 196 {
 197         /* first pull it out of the global list */
 198         if (lock)
 199                 write_lock_bh(&instances_lock);
 200
 201         UDEBUG("removing instance %p (queuenum=%u) from hash\n",
 202                 inst, inst->group_num);
 203
 204         hlist_del(&inst->hlist);
 205
 206         if (lock)
 207                 write_unlock_bh(&instances_lock);
 208
 209         /* then flush all pending packets from skb */
 210
 211         spin_lock_bh(&inst->lock);
 212         if (inst->skb) {
 213                 if (inst->qlen)
 214                         __nfulnl_send(inst);
 215                 if (inst->skb) {
 216                         kfree_skb(inst->skb);
 217                         inst->skb = NULL;
 218                 }
 219         }
 220         spin_unlock_bh(&inst->lock);
 221
 222         /* and finally put the refcount */
 223         instance_put(inst);
 224
 225         module_put(THIS_MODULE);
 226 }
 227
 228 static inline void
 229 __instance_destroy(struct nfulnl_instance *inst)
 230 {
 231         _instance_destroy2(inst, 0);
 232 }
 233
 234 static inline void
 235 instance_destroy(struct nfulnl_instance *inst)
 236 {
 237         _instance_destroy2(inst, 1);
 238 }
 239
 240 static int
 241 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 242                   unsigned int range)
 243 {
 244         int status = 0;
 245
 246         spin_lock_bh(&inst->lock);
 247
 248         switch (mode) {
 249         case NFULNL_COPY_NONE:
 250         case NFULNL_COPY_META:
 251                 inst->copy_mode = mode;
 252                 inst->copy_range = 0;
 253                 break;
 254
 255         case NFULNL_COPY_PACKET:
 256                 inst->copy_mode = mode;
 257                 /* we're using struct nfattr which has 16bit nfa_len */
 258                 if (range > 0xffff)
 259                         inst->copy_range = 0xffff;
 260                 else
 261                         inst->copy_range = range;
 262                 break;
 263
 264         default:
 265                 status = -EINVAL;
 266                 break;
 267         }
 268
 269         spin_unlock_bh(&inst->lock);
 270
 271         return status;
 272 }
 273
 274 static int
 275 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 276 {
 277         int status;
 278
 279         spin_lock_bh(&inst->lock);
 280         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 281                 status = -ERANGE;
 282         else if (nlbufsiz > 131072)
 283                 status = -ERANGE;
 284         else {
 285                 inst->nlbufsiz = nlbufsiz;
 286                 status = 0;
 287         }
 288         spin_unlock_bh(&inst->lock);
 289
 290         return status;
 291 }
 292
 293 static int
 294 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 295 {
 296         spin_lock_bh(&inst->lock);
 297         inst->flushtimeout = timeout;
 298         spin_unlock_bh(&inst->lock);
 299
 300         return 0;
 301 }
 302
 303 static int
 304 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 305 {
 306         spin_lock_bh(&inst->lock);
 307         inst->qthreshold = qthresh;
 308         spin_unlock_bh(&inst->lock);
 309
 310         return 0;
 311 }
 312
 313 static struct sk_buff *nfulnl_alloc_skb(unsigned int inst_size,
 314                                         unsigned int pkt_size)
 315 {
 316         struct sk_buff *skb;
 317         unsigned int n;
 318
 319         UDEBUG("entered (%u, %u)\n", inst_size, pkt_size);
 320
 321         /* alloc skb which should be big enough for a whole multipart
 322          * message.  WARNING: has to be <= 128k due to slab restrictions */
 323
 324         n = max(inst_size, pkt_size);
 325         skb = alloc_skb(n, GFP_ATOMIC);
 326         if (!skb) {
 327                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
 328                         inst_size);
 329
 330                 if (n > pkt_size) {
 331                         /* try to allocate only as much as we need for current
 332                          * packet */
 333
 334                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 335                         if (!skb)
 336                                 PRINTR("nfnetlink_log: can't even alloc %u "
 337                                        "bytes\n", pkt_size);
 338                 }
 339         }
 340
 341         return skb;
 342 }
 343
 344 static int
 345 __nfulnl_send(struct nfulnl_instance *inst)
 346 {
 347         int status;
 348
 349         if (timer_pending(&inst->timer))
 350                 del_timer(&inst->timer);
 351
 352         if (!inst->skb)
 353                 return 0;
 354
 355         if (inst->qlen > 1)
 356                 inst->lastnlh->nlmsg_type = NLMSG_DONE;
 357
 358         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
 359         if (status < 0) {
 360                 UDEBUG("netlink_unicast() failed\n");
 361                 /* FIXME: statistics */
 362         }
 363
 364         inst->qlen = 0;
 365         inst->skb = NULL;
 366         inst->lastnlh = NULL;
 367
 368         return status;
 369 }
 370
 371 static void nfulnl_timer(unsigned long data)
 372 {
 373         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 374
 375         UDEBUG("timer function called, flushing buffer\n");
 376
 377         spin_lock_bh(&inst->lock);
 378         __nfulnl_send(inst);
 379         instance_put(inst);
 380         spin_unlock_bh(&inst->lock);
 381 }
 382
 383 static inline int
 384 __build_packet_message(struct nfulnl_instance *inst,
 385                         const struct sk_buff *skb,
 386                         unsigned int data_len,
 387                         unsigned int pf,
 388                         unsigned int hooknum,
 389                         const struct net_device *indev,
 390                         const struct net_device *outdev,
 391                         const struct nf_loginfo *li,
 392                         const char *prefix)
 393 {
 394         unsigned char *old_tail;
 395         struct nfulnl_msg_packet_hdr pmsg;
 396         struct nlmsghdr *nlh;
 397         struct nfgenmsg *nfmsg;
 398         u_int32_t tmp_uint;
 399
 400         UDEBUG("entered\n");
 401
 402         old_tail = inst->skb->tail;
 403         nlh = NLMSG_PUT(inst->skb, 0, 0,
 404                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 405                         sizeof(struct nfgenmsg));
 406         nfmsg = NLMSG_DATA(nlh);
 407         nfmsg->nfgen_family = pf;
 408         nfmsg->version = NFNETLINK_V0;
 409         nfmsg->res_id = htons(inst->group_num);
 410
 411         pmsg.hw_protocol        = htons(skb->protocol);
 412         pmsg.hook               = hooknum;
 413
 414         NFA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
 415
 416         if (prefix) {
 417                 int slen = strlen(prefix);
 418                 if (slen > NFULNL_PREFIXLEN)
 419                         slen = NFULNL_PREFIXLEN;
 420                 NFA_PUT(inst->skb, NFULA_PREFIX, slen, prefix);
 421         }
 422
 423         if (indev) {
 424                 tmp_uint = htonl(indev->ifindex);
 425 #ifndef CONFIG_BRIDGE_NETFILTER
 426                 NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
 427                         &tmp_uint);
 428 #else
 429                 if (pf == PF_BRIDGE) {
 430                         /* Case 1: outdev is physical input device, we need to
 431                          * look for bridge group (when called from
 432                          * netfilter_bridge) */
 433                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 434                                 sizeof(tmp_uint), &tmp_uint);
 435                         /* this is the bridge group "brX" */
 436                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
 437                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 438                                 sizeof(tmp_uint), &tmp_uint);
 439                 } else {
 440                         /* Case 2: indev is bridge group, we need to look for
 441                          * physical device (when called from ipv4) */
 442                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 443                                 sizeof(tmp_uint), &tmp_uint);
 444                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
 445                                 tmp_uint =
 446                                     htonl(skb->nf_bridge->physindev->ifindex);
 447                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 448                                         sizeof(tmp_uint), &tmp_uint);
 449                         }
 450                 }
 451 #endif
 452         }
 453
 454         if (outdev) {
 455                 tmp_uint = htonl(outdev->ifindex);
 456 #ifndef CONFIG_BRIDGE_NETFILTER
 457                 NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 458                         &tmp_uint);
 459 #else
 460                 if (pf == PF_BRIDGE) {
 461                         /* Case 1: outdev is physical output device, we need to
 462                          * look for bridge group (when called from
 463                          * netfilter_bridge) */
 464                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 465                                 sizeof(tmp_uint), &tmp_uint);
 466                         /* this is the bridge group "brX" */
 467                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
 468                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 469                                 sizeof(tmp_uint), &tmp_uint);
 470                 } else {
 471                         /* Case 2: indev is a bridge group, we need to look
 472                          * for physical device (when called from ipv4) */
 473                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 474                                 sizeof(tmp_uint), &tmp_uint);
 475                         if (skb->nf_bridge) {
 476                                 tmp_uint =
 477                                     htonl(skb->nf_bridge->physoutdev->ifindex);
 478                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 479                                         sizeof(tmp_uint), &tmp_uint);
 480                         }
 481                 }
 482 #endif
 483         }
 484
 485         if (skb->nfmark) {
 486                 tmp_uint = htonl(skb->nfmark);
 487                 NFA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
 488         }
 489
 490         if (indev && skb->dev && skb->dev->hard_header_parse) {
 491                 struct nfulnl_msg_packet_hw phw;
 492
 493                 phw.hw_addrlen =
 494                         skb->dev->hard_header_parse((struct sk_buff *)skb,
 495                                                     phw.hw_addr);
 496                 phw.hw_addrlen = htons(phw.hw_addrlen);
 497                 NFA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
 498         }
 499
 500         if (skb->tstamp.off_sec) {
 501                 struct nfulnl_msg_packet_timestamp ts;
 502
 503                 ts.sec = cpu_to_be64(skb->tstamp.off_sec);
 504                 ts.usec = cpu_to_be64(skb->tstamp.off_usec);
 505
 506                 NFA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 507         }
 508
 509         /* UID */
 510         if (skb->sk) {
 511                 read_lock_bh(&skb->sk->sk_callback_lock);
 512                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 513                         u_int32_t uid = htonl(skb->sk->sk_socket->file->f_uid);
 514                         /* need to unlock here since NFA_PUT may goto */
 515                         read_unlock_bh(&skb->sk->sk_callback_lock);
 516                         NFA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
 517                 } else
 518                         read_unlock_bh(&skb->sk->sk_callback_lock);
 519         }
 520
 521         if (data_len) {
 522                 struct nfattr *nfa;
 523                 int size = NFA_LENGTH(data_len);
 524
 525                 if (skb_tailroom(inst->skb) < (int)NFA_SPACE(data_len)) {
 526                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 527                         goto nlmsg_failure;
 528                 }
 529
 530                 nfa = (struct nfattr *)skb_put(inst->skb, NFA_ALIGN(size));
 531                 nfa->nfa_type = NFULA_PAYLOAD;
 532                 nfa->nfa_len = size;
 533
 534                 if (skb_copy_bits(skb, 0, NFA_DATA(nfa), data_len))
 535                         BUG();
 536         }
 537
 538         nlh->nlmsg_len = inst->skb->tail - old_tail;
 539         return 0;
 540
 541 nlmsg_failure:
 542         UDEBUG("nlmsg_failure\n");
 543 nfattr_failure:
 544         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 545         return -1;
 546 }
 547
 548 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 549
 550 static struct nf_loginfo default_loginfo = {
 551         .type =         NF_LOG_TYPE_ULOG,
 552         .u = {
 553                 .ulog = {
 554                         .copy_len       = 0xffff,
 555                         .group          = 0,
 556                         .qthreshold     = 1,
 557                 },
 558         },
 559 };
 560
 561 /* log handler for internal netfilter logging api */
 562 static void
 563 nfulnl_log_packet(unsigned int pf,
 564                   unsigned int hooknum,
 565                   const struct sk_buff *skb,
 566                   const struct net_device *in,
 567                   const struct net_device *out,
 568                   const struct nf_loginfo *li_user,
 569                   const char *prefix)
 570 {
 571         unsigned int size, data_len;
 572         struct nfulnl_instance *inst;
 573         const struct nf_loginfo *li;
 574         unsigned int qthreshold;
 575         unsigned int nlbufsiz;
 576
 577         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 578                 li = li_user;
 579         else
 580                 li = &default_loginfo;
 581
 582         inst = instance_lookup_get(li->u.ulog.group);
 583         if (!inst)
 584                 inst = instance_lookup_get(0);
 585         if (!inst) {
 586                 PRINTR("nfnetlink_log: trying to log packet, "
 587                         "but no instance for group %u\n", li->u.ulog.group);
 588                 return;
 589         }
 590
 591         /* all macros expand to constant values at compile time */
 592         /* FIXME: do we want to make the size calculation conditional based on
 593          * what is actually present?  way more branches and checks, but more
 594          * memory efficient... */
 595         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 596                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hdr))
 597                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 598                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 599 #ifdef CONFIG_BRIDGE_NETFILTER
 600                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 601                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 602 #endif
 603                 + NFA_SPACE(sizeof(u_int32_t))  /* mark */
 604                 + NFA_SPACE(sizeof(u_int32_t))  /* uid */
 605                 + NFA_SPACE(NFULNL_PREFIXLEN)   /* prefix */
 606                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hw))
 607                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_timestamp));
 608
 609         UDEBUG("initial size=%u\n", size);
 610
 611         spin_lock_bh(&inst->lock);
 612
 613         qthreshold = inst->qthreshold;
 614         /* per-rule qthreshold overrides per-instance */
 615         if (qthreshold > li->u.ulog.qthreshold)
 616                 qthreshold = li->u.ulog.qthreshold;
 617
 618         switch (inst->copy_mode) {
 619         case NFULNL_COPY_META:
 620         case NFULNL_COPY_NONE:
 621                 data_len = 0;
 622                 break;
 623
 624         case NFULNL_COPY_PACKET:
 625                 if (inst->copy_range == 0
 626                     || inst->copy_range > skb->len)
 627                         data_len = skb->len;
 628                 else
 629                         data_len = inst->copy_range;
 630
 631                 size += NFA_SPACE(data_len);
 632                 UDEBUG("copy_packet, therefore size now %u\n", size);
 633                 break;
 634
 635         default:
 636                 spin_unlock_bh(&inst->lock);
 637                 instance_put(inst);
 638                 return;
 639         }
 640
 641         if (size > inst->nlbufsiz)
 642                 nlbufsiz = size;
 643         else
 644                 nlbufsiz = inst->nlbufsiz;
 645
 646         if (!inst->skb) {
 647                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
 648                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
 649                                 inst->nlbufsiz, size);
 650                         goto alloc_failure;
 651                 }
 652         } else if (inst->qlen >= qthreshold ||
 653                    size > skb_tailroom(inst->skb)) {
 654                 /* either the queue len is too high or we don't have
 655                  * enough room in the skb left. flush to userspace. */
 656                 UDEBUG("flushing old skb\n");
 657
 658                 __nfulnl_send(inst);
 659
 660                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
 661                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
 662                                 inst->nlbufsiz, size);
 663                         goto alloc_failure;
 664                 }
 665         }
 666
 667         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
 668         inst->qlen++;
 669
 670         __build_packet_message(inst, skb, data_len, pf,
 671                                 hooknum, in, out, li, prefix);
 672
 673         /* timer_pending always called within inst->lock, so there
 674          * is no chance of a race here */
 675         if (!timer_pending(&inst->timer)) {
 676                 instance_get(inst);
 677                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 678                 add_timer(&inst->timer);
 679         }
 680         spin_unlock_bh(&inst->lock);
 681
 682         return;
 683
 684 alloc_failure:
 685         spin_unlock_bh(&inst->lock);
 686         instance_put(inst);
 687         UDEBUG("error allocating skb\n");
 688         /* FIXME: statistics */
 689 }
 690
 691 static int
 692 nfulnl_rcv_nl_event(struct notifier_block *this,
 693                    unsigned long event, void *ptr)
 694 {
 695         struct netlink_notify *n = ptr;
 696
 697         if (event == NETLINK_URELEASE &&
 698             n->protocol == NETLINK_NETFILTER && n->pid) {
 699                 int i;
 700
 701                 /* destroy all instances for this pid */
 702                 write_lock_bh(&instances_lock);
 703                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 704                         struct hlist_node *tmp, *t2;
 705                         struct nfulnl_instance *inst;
 706                         struct hlist_head *head = &instance_table[i];
 707
 708                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 709                                 UDEBUG("node = %p\n", inst);
 710                                 if (n->pid == inst->peer_pid)
 711                                         __instance_destroy(inst);
 712                         }
 713                 }
 714                 write_unlock_bh(&instances_lock);
 715         }
 716         return NOTIFY_DONE;
 717 }
 718
 719 static struct notifier_block nfulnl_rtnl_notifier = {
 720         .notifier_call  = nfulnl_rcv_nl_event,
 721 };
 722
 723 static int
 724 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 725                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
 726 {
 727         return -ENOTSUPP;
 728 }
 729
 730 static struct nf_logger nfulnl_logger = {
 731         .name   = "nfnetlink_log",
 732         .logfn  = &nfulnl_log_packet,
 733         .me     = THIS_MODULE,
 734 };
 735
 736 static const int nfula_min[NFULA_MAX] = {
 737         [NFULA_PACKET_HDR-1]    = sizeof(struct nfulnl_msg_packet_hdr),
 738         [NFULA_MARK-1]          = sizeof(u_int32_t),
 739         [NFULA_TIMESTAMP-1]     = sizeof(struct nfulnl_msg_packet_timestamp),
 740         [NFULA_IFINDEX_INDEV-1] = sizeof(u_int32_t),
 741         [NFULA_IFINDEX_OUTDEV-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 };
 747
 748 static const int nfula_cfg_min[NFULA_CFG_MAX] = {
 749         [NFULA_CFG_CMD-1]       = sizeof(struct nfulnl_msg_config_cmd),
 750         [NFULA_CFG_MODE-1]      = sizeof(struct nfulnl_msg_config_mode),
 751         [NFULA_CFG_TIMEOUT-1]   = sizeof(u_int32_t),
 752         [NFULA_CFG_QTHRESH-1]   = sizeof(u_int32_t),
 753         [NFULA_CFG_NLBUFSIZ-1]  = sizeof(u_int32_t),
 754 };
 755
 756 static int
 757 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 758                    struct nlmsghdr *nlh, struct nfattr *nfula[], int *errp)
 759 {
 760         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 761         u_int16_t group_num = ntohs(nfmsg->res_id);
 762         struct nfulnl_instance *inst;
 763         int ret = 0;
 764
 765         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
 766
 767         if (nfattr_bad_size(nfula, NFULA_CFG_MAX, nfula_cfg_min)) {
 768                 UDEBUG("bad attribute size\n");
 769                 return -EINVAL;
 770         }
 771
 772         inst = instance_lookup_get(group_num);
 773         if (nfula[NFULA_CFG_CMD-1]) {
 774                 u_int8_t pf = nfmsg->nfgen_family;
 775                 struct nfulnl_msg_config_cmd *cmd;
 776                 cmd = NFA_DATA(nfula[NFULA_CFG_CMD-1]);
 777                 UDEBUG("found CFG_CMD for\n");
 778
 779                 switch (cmd->command) {
 780                 case NFULNL_CFG_CMD_BIND:
 781                         if (inst) {
 782                                 ret = -EBUSY;
 783                                 goto out_put;
 784                         }
 785
 786                         inst = instance_create(group_num,
 787                                                NETLINK_CB(skb).pid);
 788                         if (!inst) {
 789                                 ret = -EINVAL;
 790                                 goto out_put;
 791                         }
 792                         break;
 793                 case NFULNL_CFG_CMD_UNBIND:
 794                         if (!inst) {
 795                                 ret = -ENODEV;
 796                                 goto out_put;
 797                         }
 798
 799                         if (inst->peer_pid != NETLINK_CB(skb).pid) {
 800                                 ret = -EPERM;
 801                                 goto out_put;
 802                         }
 803
 804                         instance_destroy(inst);
 805                         break;
 806                 case NFULNL_CFG_CMD_PF_BIND:
 807                         UDEBUG("registering log handler for pf=%u\n", pf);
 808                         ret = nf_log_register(pf, &nfulnl_logger);
 809                         break;
 810                 case NFULNL_CFG_CMD_PF_UNBIND:
 811                         UDEBUG("unregistering log handler for pf=%u\n", pf);
 812                         /* This is a bug and a feature.  We cannot unregister
 813                          * other handlers, like nfnetlink_inst can */
 814                         nf_log_unregister_pf(pf);
 815                         break;
 816                 default:
 817                         ret = -EINVAL;
 818                         break;
 819                 }
 820         } else {
 821                 if (!inst) {
 822                         UDEBUG("no config command, and no instance for "
 823                                 "group=%u pid=%u =>ENOENT\n",
 824                                 group_num, NETLINK_CB(skb).pid);
 825                         ret = -ENOENT;
 826                         goto out_put;
 827                 }
 828
 829                 if (inst->peer_pid != NETLINK_CB(skb).pid) {
 830                         UDEBUG("no config command, and wrong pid\n");
 831                         ret = -EPERM;
 832                         goto out_put;
 833                 }
 834         }
 835
 836         if (nfula[NFULA_CFG_MODE-1]) {
 837                 struct nfulnl_msg_config_mode *params;
 838                 params = NFA_DATA(nfula[NFULA_CFG_MODE-1]);
 839
 840                 nfulnl_set_mode(inst, params->copy_mode,
 841                                 ntohs(params->copy_range));
 842         }
 843
 844         if (nfula[NFULA_CFG_TIMEOUT-1]) {
 845                 u_int32_t timeout =
 846                         *(u_int32_t *)NFA_DATA(nfula[NFULA_CFG_TIMEOUT-1]);
 847
 848                 nfulnl_set_timeout(inst, ntohl(timeout));
 849         }
 850
 851         if (nfula[NFULA_CFG_NLBUFSIZ-1]) {
 852                 u_int32_t nlbufsiz =
 853                         *(u_int32_t *)NFA_DATA(nfula[NFULA_CFG_NLBUFSIZ-1]);
 854
 855                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 856         }
 857
 858         if (nfula[NFULA_CFG_QTHRESH-1]) {
 859                 u_int32_t qthresh =
 860                         *(u_int16_t *)NFA_DATA(nfula[NFULA_CFG_QTHRESH-1]);
 861
 862                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 863         }
 864
 865 out_put:
 866         instance_put(inst);
 867         return ret;
 868 }
 869
 870 static struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 871         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 872                                     .attr_count = NFULA_MAX, },
 873         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 874                                     .attr_count = NFULA_CFG_MAX, },
 875 };
 876
 877 static struct nfnetlink_subsystem nfulnl_subsys = {
 878         .name           = "log",
 879         .subsys_id      = NFNL_SUBSYS_ULOG,
 880         .cb_count       = NFULNL_MSG_MAX,
 881         .cb             = nfulnl_cb,
 882 };
 883
 884 #ifdef CONFIG_PROC_FS
 885 struct iter_state {
 886         unsigned int bucket;
 887 };
 888
 889 static struct hlist_node *get_first(struct seq_file *seq)
 890 {
 891         struct iter_state *st = seq->private;
 892
 893         if (!st)
 894                 return NULL;
 895
 896         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 897                 if (!hlist_empty(&instance_table[st->bucket]))
 898                         return instance_table[st->bucket].first;
 899         }
 900         return NULL;
 901 }
 902
 903 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
 904 {
 905         struct iter_state *st = seq->private;
 906
 907         h = h->next;
 908         while (!h) {
 909                 if (++st->bucket >= INSTANCE_BUCKETS)
 910                         return NULL;
 911
 912                 h = instance_table[st->bucket].first;
 913         }
 914         return h;
 915 }
 916
 917 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 918 {
 919         struct hlist_node *head;
 920         head = get_first(seq);
 921
 922         if (head)
 923                 while (pos && (head = get_next(seq, head)))
 924                         pos--;
 925         return pos ? NULL : head;
 926 }
 927
 928 static void *seq_start(struct seq_file *seq, loff_t *pos)
 929 {
 930         read_lock_bh(&instances_lock);
 931         return get_idx(seq, *pos);
 932 }
 933
 934 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 935 {
 936         (*pos)++;
 937         return get_next(s, v);
 938 }
 939
 940 static void seq_stop(struct seq_file *s, void *v)
 941 {
 942         read_unlock_bh(&instances_lock);
 943 }
 944
 945 static int seq_show(struct seq_file *s, void *v)
 946 {
 947         const struct nfulnl_instance *inst = v;
 948
 949         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 950                           inst->group_num,
 951                           inst->peer_pid, inst->qlen,
 952                           inst->copy_mode, inst->copy_range,
 953                           inst->flushtimeout, atomic_read(&inst->use));
 954 }
 955
 956 static struct seq_operations nful_seq_ops = {
 957         .start  = seq_start,
 958         .next   = seq_next,
 959         .stop   = seq_stop,
 960         .show   = seq_show,
 961 };
 962
 963 static int nful_open(struct inode *inode, struct file *file)
 964 {
 965         struct seq_file *seq;
 966         struct iter_state *is;
 967         int ret;
 968
 969         is = kzalloc(sizeof(*is), GFP_KERNEL);
 970         if (!is)
 971                 return -ENOMEM;
 972         ret = seq_open(file, &nful_seq_ops);
 973         if (ret < 0)
 974                 goto out_free;
 975         seq = file->private_data;
 976         seq->private = is;
 977         return ret;
 978 out_free:
 979         kfree(is);
 980         return ret;
 981 }
 982
 983 static struct file_operations nful_file_ops = {
 984         .owner   = THIS_MODULE,
 985         .open    = nful_open,
 986         .read    = seq_read,
 987         .llseek  = seq_lseek,
 988         .release = seq_release_private,
 989 };
 990
 991 #endif /* PROC_FS */
 992
 993 static int
 994 init_or_cleanup(int init)
 995 {
 996         int i, status = -ENOMEM;
 997 #ifdef CONFIG_PROC_FS
 998         struct proc_dir_entry *proc_nful;
 999 #endif
1000
1001         if (!init)
1002                 goto cleanup;
1003
1004         for (i = 0; i < INSTANCE_BUCKETS; i++)
1005                 INIT_HLIST_HEAD(&instance_table[i]);
1006
1007         /* it's not really all that important to have a random value, so
1008          * we can do this from the init function, even if there hasn't
1009          * been that much entropy yet */
1010         get_random_bytes(&hash_init, sizeof(hash_init));
1011
1012         netlink_register_notifier(&nfulnl_rtnl_notifier);
1013         status = nfnetlink_subsys_register(&nfulnl_subsys);
1014         if (status < 0) {
1015                 printk(KERN_ERR "log: failed to create netlink socket\n");
1016                 goto cleanup_netlink_notifier;
1017         }
1018
1019 #ifdef CONFIG_PROC_FS
1020         proc_nful = create_proc_entry("nfnetlink_log", 0440,
1021                                       proc_net_netfilter);
1022         if (!proc_nful)
1023                 goto cleanup_subsys;
1024         proc_nful->proc_fops = &nful_file_ops;
1025 #endif
1026
1027         return status;
1028
1029 cleanup:
1030         nf_log_unregister_logger(&nfulnl_logger);
1031 #ifdef CONFIG_PROC_FS
1032         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1033 cleanup_subsys:
1034 #endif
1035         nfnetlink_subsys_unregister(&nfulnl_subsys);
1036 cleanup_netlink_notifier:
1037         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1038         return status;
1039 }
1040
1041 static int __init init(void)
1042 {
1043
1044         return init_or_cleanup(1);
1045 }
1046
1047 static void __exit fini(void)
1048 {
1049         init_or_cleanup(0);
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(init);
1058 module_exit(fini);