net/netfilter/nfnetlink_log.c

   1 /*
   2  * This is a module which is used for logging packets to userspace via
   3  * nfetlink.
   4  *
   5  * (C) 2005 by Harald Welte <laforge@netfilter.org>
   6  *
   7  * Based on the old ipv4-only ipt_ULOG.c:
   8  * (C) 2000-2004 by Harald Welte <laforge@netfilter.org>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License version 2 as
  12  * published by the Free Software Foundation.
  13  *
  14  */
  15 #include <linux/module.h>
  16 #include <linux/skbuff.h>
  17 #include <linux/init.h>
  18 #include <linux/ip.h>
  19 #include <linux/ipv6.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/netfilter.h>
  22 #include <linux/netlink.h>
  23 #include <linux/netfilter/nfnetlink.h>
  24 #include <linux/netfilter/nfnetlink_log.h>
  25 #include <linux/spinlock.h>
  26 #include <linux/sysctl.h>
  27 #include <linux/proc_fs.h>
  28 #include <linux/security.h>
  29 #include <linux/list.h>
  30 #include <linux/jhash.h>
  31 #include <linux/random.h>
  32 #include <net/sock.h>
  33
  34 #include <asm/atomic.h>
  35
  36 #ifdef CONFIG_BRIDGE_NETFILTER
  37 #include "../bridge/br_private.h"
  38 #endif
  39
  40 #define NFULNL_NLBUFSIZ_DEFAULT 4096
  41 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
  42 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  43
  44 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  45                                      printk(x, ## args); } while (0);
  46
  47 #if 0
  48 #define UDEBUG(x, args ...)     printk(KERN_DEBUG "%s(%d):%s(): " x,       \
  49                                         __FILE__, __LINE__, __FUNCTION__,  \
  50                                         ## args)
  51 #else
  52 #define UDEBUG(x, ...)
  53 #endif
  54
  55 struct nfulnl_instance {
  56         struct hlist_node hlist;        /* global list of instances */
  57         spinlock_t lock;
  58         atomic_t use;                   /* use count */
  59
  60         unsigned int qlen;              /* number of nlmsgs in skb */
  61         struct sk_buff *skb;            /* pre-allocatd skb */
  62         struct nlmsghdr *lastnlh;       /* netlink header of last msg in skb */
  63         struct timer_list timer;
  64         int peer_pid;                   /* PID of the peer process */
  65
  66         /* configurable parameters */
  67         unsigned int flushtimeout;      /* timeout until queue flush */
  68         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  69         unsigned int qthreshold;        /* threshold of the queue */
  70         u_int32_t copy_range;
  71         u_int16_t group_num;            /* number of this queue */
  72         u_int8_t copy_mode;
  73 };
  74
  75 static DEFINE_RWLOCK(instances_lock);
  76
  77 #define INSTANCE_BUCKETS        16
  78 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  79 static unsigned int hash_init;
  80
  81 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  82 {
  83         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  84 }
  85
  86 static struct nfulnl_instance *
  87 __instance_lookup(u_int16_t group_num)
  88 {
  89         struct hlist_head *head;
  90         struct hlist_node *pos;
  91         struct nfulnl_instance *inst;
  92
  93         UDEBUG("entering (group_num=%u)\n", group_num);
  94
  95         head = &instance_table[instance_hashfn(group_num)];
  96         hlist_for_each_entry(inst, pos, head, hlist) {
  97                 if (inst->group_num == group_num)
  98                         return inst;
  99         }
 100         return NULL;
 101 }
 102
 103 static inline void
 104 instance_get(struct nfulnl_instance *inst)
 105 {
 106         atomic_inc(&inst->use);
 107 }
 108
 109 static struct nfulnl_instance *
 110 instance_lookup_get(u_int16_t group_num)
 111 {
 112         struct nfulnl_instance *inst;
 113
 114         read_lock_bh(&instances_lock);
 115         inst = __instance_lookup(group_num);
 116         if (inst)
 117                 instance_get(inst);
 118         read_unlock_bh(&instances_lock);
 119
 120         return inst;
 121 }
 122
 123 static void
 124 instance_put(struct nfulnl_instance *inst)
 125 {
 126         if (inst && atomic_dec_and_test(&inst->use)) {
 127                 UDEBUG("kfree(inst=%p)\n", inst);
 128                 kfree(inst);
 129         }
 130 }
 131
 132 static void nfulnl_timer(unsigned long data);
 133
 134 static struct nfulnl_instance *
 135 instance_create(u_int16_t group_num, int pid)
 136 {
 137         struct nfulnl_instance *inst;
 138
 139         UDEBUG("entering (group_num=%u, pid=%d)\n", group_num,
 140                 pid);
 141
 142         write_lock_bh(&instances_lock);
 143         if (__instance_lookup(group_num)) {
 144                 inst = NULL;
 145                 UDEBUG("aborting, instance already exists\n");
 146                 goto out_unlock;
 147         }
 148
 149         inst = kmalloc(sizeof(*inst), GFP_ATOMIC);
 150         if (!inst)
 151                 goto out_unlock;
 152
 153         memset(inst, 0, sizeof(*inst));
 154         INIT_HLIST_NODE(&inst->hlist);
 155         inst->lock = SPIN_LOCK_UNLOCKED;
 156         /* needs to be two, since we _put() after creation */
 157         atomic_set(&inst->use, 2);
 158
 159         init_timer(&inst->timer);
 160         inst->timer.function = nfulnl_timer;
 161         inst->timer.data = (unsigned long)inst;
 162         /* don't start timer yet. (re)start it  with every packet */
 163
 164         inst->peer_pid = pid;
 165         inst->group_num = group_num;
 166
 167         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 168         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 169         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 170         inst->copy_mode         = NFULNL_COPY_PACKET;
 171         inst->copy_range        = 0xffff;
 172
 173         if (!try_module_get(THIS_MODULE))
 174                 goto out_free;
 175
 176         hlist_add_head(&inst->hlist,
 177                        &instance_table[instance_hashfn(group_num)]);
 178
 179         UDEBUG("newly added node: %p, next=%p\n", &inst->hlist,
 180                 inst->hlist.next);
 181
 182         write_unlock_bh(&instances_lock);
 183
 184         return inst;
 185
 186 out_free:
 187         instance_put(inst);
 188 out_unlock:
 189         write_unlock_bh(&instances_lock);
 190         return NULL;
 191 }
 192
 193 static int __nfulnl_send(struct nfulnl_instance *inst);
 194
 195 static void
 196 _instance_destroy2(struct nfulnl_instance *inst, int lock)
 197 {
 198         /* first pull it out of the global list */
 199         if (lock)
 200                 write_lock_bh(&instances_lock);
 201
 202         UDEBUG("removing instance %p (queuenum=%u) from hash\n",
 203                 inst, inst->group_num);
 204
 205         hlist_del(&inst->hlist);
 206
 207         if (lock)
 208                 write_unlock_bh(&instances_lock);
 209
 210         /* then flush all pending packets from skb */
 211
 212         spin_lock_bh(&inst->lock);
 213         if (inst->skb) {
 214                 if (inst->qlen)
 215                         __nfulnl_send(inst);
 216                 if (inst->skb) {
 217                         kfree_skb(inst->skb);
 218                         inst->skb = NULL;
 219                 }
 220         }
 221         spin_unlock_bh(&inst->lock);
 222
 223         /* and finally put the refcount */
 224         instance_put(inst);
 225
 226         module_put(THIS_MODULE);
 227 }
 228
 229 static inline void
 230 __instance_destroy(struct nfulnl_instance *inst)
 231 {
 232         _instance_destroy2(inst, 0);
 233 }
 234
 235 static inline void
 236 instance_destroy(struct nfulnl_instance *inst)
 237 {
 238         _instance_destroy2(inst, 1);
 239 }
 240
 241 static int
 242 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 243                   unsigned int range)
 244 {
 245         int status = 0;
 246
 247         spin_lock_bh(&inst->lock);
 248
 249         switch (mode) {
 250         case NFULNL_COPY_NONE:
 251         case NFULNL_COPY_META:
 252                 inst->copy_mode = mode;
 253                 inst->copy_range = 0;
 254                 break;
 255
 256         case NFULNL_COPY_PACKET:
 257                 inst->copy_mode = mode;
 258                 /* we're using struct nfattr which has 16bit nfa_len */
 259                 if (range > 0xffff)
 260                         inst->copy_range = 0xffff;
 261                 else
 262                         inst->copy_range = range;
 263                 break;
 264
 265         default:
 266                 status = -EINVAL;
 267                 break;
 268         }
 269
 270         spin_unlock_bh(&inst->lock);
 271
 272         return status;
 273 }
 274
 275 static int
 276 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 277 {
 278         int status;
 279
 280         spin_lock_bh(&inst->lock);
 281         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 282                 status = -ERANGE;
 283         else if (nlbufsiz > 131072)
 284                 status = -ERANGE;
 285         else {
 286                 inst->nlbufsiz = nlbufsiz;
 287                 status = 0;
 288         }
 289         spin_unlock_bh(&inst->lock);
 290
 291         return status;
 292 }
 293
 294 static int
 295 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 296 {
 297         spin_lock_bh(&inst->lock);
 298         inst->flushtimeout = timeout;
 299         spin_unlock_bh(&inst->lock);
 300
 301         return 0;
 302 }
 303
 304 static int
 305 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 306 {
 307         spin_lock_bh(&inst->lock);
 308         inst->qthreshold = qthresh;
 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
 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         skb = alloc_skb(inst_size, GFP_ATOMIC);
 325         if (!skb) {
 326                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
 327                         inst_size);
 328
 329                 /* try to allocate only as much as we need for current
 330                  * packet */
 331
 332                 skb = alloc_skb(pkt_size, GFP_ATOMIC);
 333                 if (!skb)
 334                         PRINTR("nfnetlink_log: can't even alloc %u bytes\n",
 335                                 pkt_size);
 336         }
 337
 338         return skb;
 339 }
 340
 341 static int
 342 __nfulnl_send(struct nfulnl_instance *inst)
 343 {
 344         int status;
 345
 346         if (timer_pending(&inst->timer))
 347                 del_timer(&inst->timer);
 348
 349         if (inst->qlen > 1)
 350                 inst->lastnlh->nlmsg_type = NLMSG_DONE;
 351
 352         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
 353         if (status < 0) {
 354                 UDEBUG("netlink_unicast() failed\n");
 355                 /* FIXME: statistics */
 356         }
 357
 358         inst->qlen = 0;
 359         inst->skb = NULL;
 360         inst->lastnlh = NULL;
 361
 362         return status;
 363 }
 364
 365 static void nfulnl_timer(unsigned long data)
 366 {
 367         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 368
 369         UDEBUG("timer function called, flushing buffer\n");
 370
 371         spin_lock_bh(&inst->lock);
 372         __nfulnl_send(inst);
 373         instance_put(inst);
 374         spin_unlock_bh(&inst->lock);
 375 }
 376
 377 static inline int
 378 __build_packet_message(struct nfulnl_instance *inst,
 379                         const struct sk_buff *skb,
 380                         unsigned int data_len,
 381                         unsigned int pf,
 382                         unsigned int hooknum,
 383                         const struct net_device *indev,
 384                         const struct net_device *outdev,
 385                         const struct nf_loginfo *li,
 386                         const char *prefix)
 387 {
 388         unsigned char *old_tail;
 389         struct nfulnl_msg_packet_hdr pmsg;
 390         struct nlmsghdr *nlh;
 391         struct nfgenmsg *nfmsg;
 392         u_int32_t tmp_uint;
 393
 394         UDEBUG("entered\n");
 395
 396         old_tail = inst->skb->tail;
 397         nlh = NLMSG_PUT(inst->skb, 0, 0,
 398                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 399                         sizeof(struct nfgenmsg));
 400         nfmsg = NLMSG_DATA(nlh);
 401         nfmsg->nfgen_family = pf;
 402         nfmsg->version = NFNETLINK_V0;
 403         nfmsg->res_id = htons(inst->group_num);
 404
 405         pmsg.hw_protocol        = htons(skb->protocol);
 406         pmsg.hook               = hooknum;
 407
 408         NFA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
 409
 410         if (prefix) {
 411                 int slen = strlen(prefix);
 412                 if (slen > NFULNL_PREFIXLEN)
 413                         slen = NFULNL_PREFIXLEN;
 414                 NFA_PUT(inst->skb, NFULA_PREFIX, slen, prefix);
 415         }
 416
 417         if (indev) {
 418                 tmp_uint = htonl(indev->ifindex);
 419 #ifndef CONFIG_BRIDGE_NETFILTER
 420                 NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
 421                         &tmp_uint);
 422 #else
 423                 if (pf == PF_BRIDGE) {
 424                         /* Case 1: outdev is physical input device, we need to
 425                          * look for bridge group (when called from
 426                          * netfilter_bridge) */
 427                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 428                                 sizeof(tmp_uint), &tmp_uint);
 429                         /* this is the bridge group "brX" */
 430                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
 431                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 432                                 sizeof(tmp_uint), &tmp_uint);
 433                 } else {
 434                         /* Case 2: indev is bridge group, we need to look for
 435                          * physical device (when called from ipv4) */
 436                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 437                                 sizeof(tmp_uint), &tmp_uint);
 438                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
 439                                 tmp_uint =
 440                                     htonl(skb->nf_bridge->physindev->ifindex);
 441                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 442                                         sizeof(tmp_uint), &tmp_uint);
 443                         }
 444                 }
 445 #endif
 446         }
 447
 448         if (outdev) {
 449                 tmp_uint = htonl(outdev->ifindex);
 450 #ifndef CONFIG_BRIDGE_NETFILTER
 451                 NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 452                         &tmp_uint);
 453 #else
 454                 if (pf == PF_BRIDGE) {
 455                         /* Case 1: outdev is physical output device, we need to
 456                          * look for bridge group (when called from
 457                          * netfilter_bridge) */
 458                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 459                                 sizeof(tmp_uint), &tmp_uint);
 460                         /* this is the bridge group "brX" */
 461                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
 462                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 463                                 sizeof(tmp_uint), &tmp_uint);
 464                 } else {
 465                         /* Case 2: indev is a bridge group, we need to look
 466                          * for physical device (when called from ipv4) */
 467                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 468                                 sizeof(tmp_uint), &tmp_uint);
 469                         if (skb->nf_bridge) {
 470                                 tmp_uint =
 471                                     htonl(skb->nf_bridge->physoutdev->ifindex);
 472                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 473                                         sizeof(tmp_uint), &tmp_uint);
 474                         }
 475                 }
 476 #endif
 477         }
 478
 479         if (skb->nfmark) {
 480                 tmp_uint = htonl(skb->nfmark);
 481                 NFA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
 482         }
 483
 484         if (indev && skb->dev && skb->dev->hard_header_parse) {
 485                 struct nfulnl_msg_packet_hw phw;
 486
 487                 phw.hw_addrlen =
 488                         skb->dev->hard_header_parse((struct sk_buff *)skb,
 489                                                     phw.hw_addr);
 490                 phw.hw_addrlen = htons(phw.hw_addrlen);
 491                 NFA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
 492         }
 493
 494         if (skb->tstamp.off_sec) {
 495                 struct nfulnl_msg_packet_timestamp ts;
 496
 497                 ts.sec = cpu_to_be64(skb_tv_base.tv_sec + skb->tstamp.off_sec);
 498                 ts.usec = cpu_to_be64(skb_tv_base.tv_usec + skb->tstamp.off_usec);
 499
 500                 NFA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 501         }
 502
 503         /* UID */
 504         if (skb->sk) {
 505                 read_lock_bh(&skb->sk->sk_callback_lock);
 506                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 507                         u_int32_t uid = htonl(skb->sk->sk_socket->file->f_uid);
 508                         /* need to unlock here since NFA_PUT may goto */
 509                         read_unlock_bh(&skb->sk->sk_callback_lock);
 510                         NFA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
 511                 } else
 512                         read_unlock_bh(&skb->sk->sk_callback_lock);
 513         }
 514
 515         if (data_len) {
 516                 struct nfattr *nfa;
 517                 int size = NFA_LENGTH(data_len);
 518
 519                 if (skb_tailroom(inst->skb) < (int)NFA_SPACE(data_len)) {
 520                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 521                         goto nlmsg_failure;
 522                 }
 523
 524                 nfa = (struct nfattr *)skb_put(inst->skb, NFA_ALIGN(size));
 525                 nfa->nfa_type = NFULA_PAYLOAD;
 526                 nfa->nfa_len = size;
 527
 528                 if (skb_copy_bits(skb, 0, NFA_DATA(nfa), data_len))
 529                         BUG();
 530         }
 531
 532         nlh->nlmsg_len = inst->skb->tail - old_tail;
 533         return 0;
 534
 535 nlmsg_failure:
 536         UDEBUG("nlmsg_failure\n");
 537 nfattr_failure:
 538         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 539         return -1;
 540 }
 541
 542 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 543
 544 static struct nf_loginfo default_loginfo = {
 545         .type =         NF_LOG_TYPE_ULOG,
 546         .u = {
 547                 .ulog = {
 548                         .copy_len       = 0xffff,
 549                         .group          = 0,
 550                         .qthreshold     = 1,
 551                 },
 552         },
 553 };
 554
 555 /* log handler for internal netfilter logging api */
 556 static void
 557 nfulnl_log_packet(unsigned int pf,
 558                   unsigned int hooknum,
 559                   const struct sk_buff *skb,
 560                   const struct net_device *in,
 561                   const struct net_device *out,
 562                   const struct nf_loginfo *li_user,
 563                   const char *prefix)
 564 {
 565         unsigned int size, data_len;
 566         struct nfulnl_instance *inst;
 567         const struct nf_loginfo *li;
 568         unsigned int qthreshold;
 569         unsigned int nlbufsiz;
 570
 571         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 572                 li = li_user;
 573         else
 574                 li = &default_loginfo;
 575
 576         inst = instance_lookup_get(li->u.ulog.group);
 577         if (!inst)
 578                 inst = instance_lookup_get(0);
 579         if (!inst) {
 580                 PRINTR("nfnetlink_log: trying to log packet, "
 581                         "but no instance for group %u\n", li->u.ulog.group);
 582                 return;
 583         }
 584
 585         /* all macros expand to constant values at compile time */
 586         /* FIXME: do we want to make the size calculation conditional based on
 587          * what is actually present?  way more branches and checks, but more
 588          * memory efficient... */
 589         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 590                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hdr))
 591                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 592                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 593 #ifdef CONFIG_BRIDGE_NETFILTER
 594                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 595                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 596 #endif
 597                 + NFA_SPACE(sizeof(u_int32_t))  /* mark */
 598                 + NFA_SPACE(sizeof(u_int32_t))  /* uid */
 599                 + NFA_SPACE(NFULNL_PREFIXLEN)   /* prefix */
 600                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hw))
 601                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_timestamp));
 602
 603         UDEBUG("initial size=%u\n", size);
 604
 605         spin_lock_bh(&inst->lock);
 606
 607         qthreshold = inst->qthreshold;
 608         /* per-rule qthreshold overrides per-instance */
 609         if (qthreshold > li->u.ulog.qthreshold)
 610                 qthreshold = li->u.ulog.qthreshold;
 611
 612         switch (inst->copy_mode) {
 613         case NFULNL_COPY_META:
 614         case NFULNL_COPY_NONE:
 615                 data_len = 0;
 616                 break;
 617
 618         case NFULNL_COPY_PACKET:
 619                 if (inst->copy_range == 0
 620                     || inst->copy_range > skb->len)
 621                         data_len = skb->len;
 622                 else
 623                         data_len = inst->copy_range;
 624
 625                 size += NFA_SPACE(data_len);
 626                 UDEBUG("copy_packet, therefore size now %u\n", size);
 627                 break;
 628
 629         default:
 630                 spin_unlock_bh(&inst->lock);
 631                 instance_put(inst);
 632                 return;
 633         }
 634
 635         if (size > inst->nlbufsiz)
 636                 nlbufsiz = size;
 637         else
 638                 nlbufsiz = inst->nlbufsiz;
 639
 640         if (!inst->skb) {
 641                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
 642                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
 643                                 inst->nlbufsiz, size);
 644                         goto alloc_failure;
 645                 }
 646         } else if (inst->qlen >= qthreshold ||
 647                    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                 __nfulnl_send(inst);
 653
 654                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
 655                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
 656                                 inst->nlbufsiz, size);
 657                         goto alloc_failure;
 658                 }
 659         }
 660
 661         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
 662         inst->qlen++;
 663
 664         __build_packet_message(inst, skb, data_len, pf,
 665                                 hooknum, in, out, li, prefix);
 666
 667         /* timer_pending always called within inst->lock, so there
 668          * is no chance of a race here */
 669         if (!timer_pending(&inst->timer)) {
 670                 instance_get(inst);
 671                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 672                 add_timer(&inst->timer);
 673         }
 674         spin_unlock_bh(&inst->lock);
 675
 676         return;
 677
 678 alloc_failure:
 679         spin_unlock_bh(&inst->lock);
 680         instance_put(inst);
 681         UDEBUG("error allocating skb\n");
 682         /* FIXME: statistics */
 683 }
 684
 685 static int
 686 nfulnl_rcv_nl_event(struct notifier_block *this,
 687                    unsigned long event, void *ptr)
 688 {
 689         struct netlink_notify *n = ptr;
 690
 691         if (event == NETLINK_URELEASE &&
 692             n->protocol == NETLINK_NETFILTER && n->pid) {
 693                 int i;
 694
 695                 /* destroy all instances for this pid */
 696                 write_lock_bh(&instances_lock);
 697                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 698                         struct hlist_node *tmp, *t2;
 699                         struct nfulnl_instance *inst;
 700                         struct hlist_head *head = &instance_table[i];
 701
 702                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 703                                 UDEBUG("node = %p\n", inst);
 704                                 if (n->pid == inst->peer_pid)
 705                                         __instance_destroy(inst);
 706                         }
 707                 }
 708                 write_unlock_bh(&instances_lock);
 709         }
 710         return NOTIFY_DONE;
 711 }
 712
 713 static struct notifier_block nfulnl_rtnl_notifier = {
 714         .notifier_call  = nfulnl_rcv_nl_event,
 715 };
 716
 717 static int
 718 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 719                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
 720 {
 721         return -ENOTSUPP;
 722 }
 723
 724 static struct nf_logger nfulnl_logger = {
 725         .name   = "nfnetlink_log",
 726         .logfn  = &nfulnl_log_packet,
 727         .me     = THIS_MODULE,
 728 };
 729
 730 static const int nfula_min[NFULA_MAX] = {
 731         [NFULA_PACKET_HDR-1]    = sizeof(struct nfulnl_msg_packet_hdr),
 732         [NFULA_MARK-1]          = sizeof(u_int32_t),
 733         [NFULA_TIMESTAMP-1]     = sizeof(struct nfulnl_msg_packet_timestamp),
 734         [NFULA_IFINDEX_INDEV-1] = sizeof(u_int32_t),
 735         [NFULA_IFINDEX_OUTDEV-1]= sizeof(u_int32_t),
 736         [NFULA_HWADDR-1]        = sizeof(struct nfulnl_msg_packet_hw),
 737         [NFULA_PAYLOAD-1]       = 0,
 738         [NFULA_PREFIX-1]        = 0,
 739         [NFULA_UID-1]           = sizeof(u_int32_t),
 740 };
 741
 742 static const int nfula_cfg_min[NFULA_CFG_MAX] = {
 743         [NFULA_CFG_CMD-1]       = sizeof(struct nfulnl_msg_config_cmd),
 744         [NFULA_CFG_MODE-1]      = sizeof(struct nfulnl_msg_config_mode),
 745         [NFULA_CFG_TIMEOUT-1]   = sizeof(u_int32_t),
 746         [NFULA_CFG_QTHRESH-1]   = sizeof(u_int32_t),
 747         [NFULA_CFG_NLBUFSIZ-1]  = sizeof(u_int32_t),
 748 };
 749
 750 static int
 751 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 752                    struct nlmsghdr *nlh, struct nfattr *nfula[], int *errp)
 753 {
 754         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 755         u_int16_t group_num = ntohs(nfmsg->res_id);
 756         struct nfulnl_instance *inst;
 757         int ret = 0;
 758
 759         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
 760
 761         if (nfattr_bad_size(nfula, NFULA_CFG_MAX, nfula_cfg_min)) {
 762                 UDEBUG("bad attribute size\n");
 763                 return -EINVAL;
 764         }
 765
 766         inst = instance_lookup_get(group_num);
 767         if (nfula[NFULA_CFG_CMD-1]) {
 768                 u_int8_t pf = nfmsg->nfgen_family;
 769                 struct nfulnl_msg_config_cmd *cmd;
 770                 cmd = NFA_DATA(nfula[NFULA_CFG_CMD-1]);
 771                 UDEBUG("found CFG_CMD for\n");
 772
 773                 switch (cmd->command) {
 774                 case NFULNL_CFG_CMD_BIND:
 775                         if (inst) {
 776                                 ret = -EBUSY;
 777                                 goto out_put;
 778                         }
 779
 780                         inst = instance_create(group_num,
 781                                                NETLINK_CB(skb).pid);
 782                         if (!inst) {
 783                                 ret = -EINVAL;
 784                                 goto out_put;
 785                         }
 786                         break;
 787                 case NFULNL_CFG_CMD_UNBIND:
 788                         if (!inst) {
 789                                 ret = -ENODEV;
 790                                 goto out_put;
 791                         }
 792
 793                         if (inst->peer_pid != NETLINK_CB(skb).pid) {
 794                                 ret = -EPERM;
 795                                 goto out_put;
 796                         }
 797
 798                         instance_destroy(inst);
 799                         break;
 800                 case NFULNL_CFG_CMD_PF_BIND:
 801                         UDEBUG("registering log handler for pf=%u\n", pf);
 802                         ret = nf_log_register(pf, &nfulnl_logger);
 803                         break;
 804                 case NFULNL_CFG_CMD_PF_UNBIND:
 805                         UDEBUG("unregistering log handler for pf=%u\n", pf);
 806                         /* This is a bug and a feature.  We cannot unregister
 807                          * other handlers, like nfnetlink_inst can */
 808                         nf_log_unregister_pf(pf);
 809                         break;
 810                 default:
 811                         ret = -EINVAL;
 812                         break;
 813                 }
 814         } else {
 815                 if (!inst) {
 816                         UDEBUG("no config command, and no instance for "
 817                                 "group=%u pid=%u =>ENOENT\n",
 818                                 group_num, NETLINK_CB(skb).pid);
 819                         ret = -ENOENT;
 820                         goto out_put;
 821                 }
 822
 823                 if (inst->peer_pid != NETLINK_CB(skb).pid) {
 824                         UDEBUG("no config command, and wrong pid\n");
 825                         ret = -EPERM;
 826                         goto out_put;
 827                 }
 828         }
 829
 830         if (nfula[NFULA_CFG_MODE-1]) {
 831                 struct nfulnl_msg_config_mode *params;
 832                 params = NFA_DATA(nfula[NFULA_CFG_MODE-1]);
 833
 834                 nfulnl_set_mode(inst, params->copy_mode,
 835                                 ntohs(params->copy_range));
 836         }
 837
 838         if (nfula[NFULA_CFG_TIMEOUT-1]) {
 839                 u_int32_t timeout =
 840                         *(u_int32_t *)NFA_DATA(nfula[NFULA_CFG_TIMEOUT-1]);
 841
 842                 nfulnl_set_timeout(inst, ntohl(timeout));
 843         }
 844
 845         if (nfula[NFULA_CFG_NLBUFSIZ-1]) {
 846                 u_int32_t nlbufsiz =
 847                         *(u_int32_t *)NFA_DATA(nfula[NFULA_CFG_NLBUFSIZ-1]);
 848
 849                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 850         }
 851
 852         if (nfula[NFULA_CFG_QTHRESH-1]) {
 853                 u_int32_t qthresh =
 854                         *(u_int16_t *)NFA_DATA(nfula[NFULA_CFG_QTHRESH-1]);
 855
 856                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 857         }
 858
 859 out_put:
 860         instance_put(inst);
 861         return ret;
 862 }
 863
 864 static struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 865         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 866                                     .attr_count = NFULA_MAX,
 867                                     .cap_required = CAP_NET_ADMIN, },
 868         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 869                                     .attr_count = NFULA_CFG_MAX,
 870                                     .cap_required = CAP_NET_ADMIN },
 871 };
 872
 873 static struct nfnetlink_subsystem nfulnl_subsys = {
 874         .name           = "log",
 875         .subsys_id      = NFNL_SUBSYS_ULOG,
 876         .cb_count       = NFULNL_MSG_MAX,
 877         .cb             = nfulnl_cb,
 878 };
 879
 880 #ifdef CONFIG_PROC_FS
 881 struct iter_state {
 882         unsigned int bucket;
 883 };
 884
 885 static struct hlist_node *get_first(struct seq_file *seq)
 886 {
 887         struct iter_state *st = seq->private;
 888
 889         if (!st)
 890                 return NULL;
 891
 892         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 893                 if (!hlist_empty(&instance_table[st->bucket]))
 894                         return instance_table[st->bucket].first;
 895         }
 896         return NULL;
 897 }
 898
 899 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
 900 {
 901         struct iter_state *st = seq->private;
 902
 903         h = h->next;
 904         while (!h) {
 905                 if (++st->bucket >= INSTANCE_BUCKETS)
 906                         return NULL;
 907
 908                 h = instance_table[st->bucket].first;
 909         }
 910         return h;
 911 }
 912
 913 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 914 {
 915         struct hlist_node *head;
 916         head = get_first(seq);
 917
 918         if (head)
 919                 while (pos && (head = get_next(seq, head)))
 920                         pos--;
 921         return pos ? NULL : head;
 922 }
 923
 924 static void *seq_start(struct seq_file *seq, loff_t *pos)
 925 {
 926         read_lock_bh(&instances_lock);
 927         return get_idx(seq, *pos);
 928 }
 929
 930 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 931 {
 932         (*pos)++;
 933         return get_next(s, v);
 934 }
 935
 936 static void seq_stop(struct seq_file *s, void *v)
 937 {
 938         read_unlock_bh(&instances_lock);
 939 }
 940
 941 static int seq_show(struct seq_file *s, void *v)
 942 {
 943         const struct nfulnl_instance *inst = v;
 944
 945         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 946                           inst->group_num,
 947                           inst->peer_pid, inst->qlen,
 948                           inst->copy_mode, inst->copy_range,
 949                           inst->flushtimeout, atomic_read(&inst->use));
 950 }
 951
 952 static struct seq_operations nful_seq_ops = {
 953         .start  = seq_start,
 954         .next   = seq_next,
 955         .stop   = seq_stop,
 956         .show   = seq_show,
 957 };
 958
 959 static int nful_open(struct inode *inode, struct file *file)
 960 {
 961         struct seq_file *seq;
 962         struct iter_state *is;
 963         int ret;
 964
 965         is = kmalloc(sizeof(*is), GFP_KERNEL);
 966         if (!is)
 967                 return -ENOMEM;
 968         memset(is, 0, sizeof(*is));
 969         ret = seq_open(file, &nful_seq_ops);
 970         if (ret < 0)
 971                 goto out_free;
 972         seq = file->private_data;
 973         seq->private = is;
 974         return ret;
 975 out_free:
 976         kfree(is);
 977         return ret;
 978 }
 979
 980 static struct file_operations nful_file_ops = {
 981         .owner   = THIS_MODULE,
 982         .open    = nful_open,
 983         .read    = seq_read,
 984         .llseek  = seq_lseek,
 985         .release = seq_release_private,
 986 };
 987
 988 #endif /* PROC_FS */
 989
 990 static int
 991 init_or_cleanup(int init)
 992 {
 993         int i, status = -ENOMEM;
 994 #ifdef CONFIG_PROC_FS
 995         struct proc_dir_entry *proc_nful;
 996 #endif
 997
 998         if (!init)
 999                 goto cleanup;
1000
1001         for (i = 0; i < INSTANCE_BUCKETS; i++)
1002                 INIT_HLIST_HEAD(&instance_table[i]);
1003
1004         /* it's not really all that important to have a random value, so
1005          * we can do this from the init function, even if there hasn't
1006          * been that much entropy yet */
1007         get_random_bytes(&hash_init, sizeof(hash_init));
1008
1009         netlink_register_notifier(&nfulnl_rtnl_notifier);
1010         status = nfnetlink_subsys_register(&nfulnl_subsys);
1011         if (status < 0) {
1012                 printk(KERN_ERR "log: failed to create netlink socket\n");
1013                 goto cleanup_netlink_notifier;
1014         }
1015
1016 #ifdef CONFIG_PROC_FS
1017         proc_nful = create_proc_entry("nfnetlink_log", 0440,
1018                                       proc_net_netfilter);
1019         if (!proc_nful)
1020                 goto cleanup_subsys;
1021         proc_nful->proc_fops = &nful_file_ops;
1022 #endif
1023
1024         return status;
1025
1026 cleanup:
1027         nf_log_unregister_logger(&nfulnl_logger);
1028 #ifdef CONFIG_PROC_FS
1029         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1030 cleanup_subsys:
1031 #endif
1032         nfnetlink_subsys_unregister(&nfulnl_subsys);
1033 cleanup_netlink_notifier:
1034         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1035         return status;
1036 }
1037
1038 static int __init init(void)
1039 {
1040
1041         return init_or_cleanup(1);
1042 }
1043
1044 static void __exit fini(void)
1045 {
1046         init_or_cleanup(0);
1047 }
1048
1049 MODULE_DESCRIPTION("netfilter userspace logging");
1050 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1051 MODULE_LICENSE("GPL");
1052 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1053
1054 module_init(init);
1055 module_exit(fini);