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

   1 /*
   2  * This is a module which is used for logging packets to userspace via
   3  * nfetlink.
   4  *
   5  * (C) 2005 by Harald Welte <laforge@netfilter.org>
   6  *
   7  * Based on the old ipv4-only ipt_ULOG.c:
   8  * (C) 2000-2004 by Harald Welte <laforge@netfilter.org>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License version 2 as
  12  * published by the Free Software Foundation.
  13  */
  14 #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 #include <net/netfilter/nf_log.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  HZ      /* every second */
  42 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  43 #define NFULNL_COPY_RANGE_MAX   0xFFFF  /* max packet size is limited by 16-bit struct nfattr nfa_len field */
  44
  45 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  46                                      printk(x, ## args); } while (0);
  47
  48 struct nfulnl_instance {
  49         struct hlist_node hlist;        /* global list of instances */
  50         spinlock_t lock;
  51         atomic_t use;                   /* use count */
  52
  53         unsigned int qlen;              /* number of nlmsgs in skb */
  54         struct sk_buff *skb;            /* pre-allocatd skb */
  55         struct timer_list timer;
  56         int peer_pid;                   /* PID of the peer process */
  57
  58         /* configurable parameters */
  59         unsigned int flushtimeout;      /* timeout until queue flush */
  60         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  61         unsigned int qthreshold;        /* threshold of the queue */
  62         u_int32_t copy_range;
  63         u_int32_t seq;                  /* instance-local sequential counter */
  64         u_int16_t group_num;            /* number of this queue */
  65         u_int16_t flags;
  66         u_int8_t copy_mode;
  67 };
  68
  69 static DEFINE_RWLOCK(instances_lock);
  70 static atomic_t global_seq;
  71
  72 #define INSTANCE_BUCKETS        16
  73 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  74 static unsigned int hash_init;
  75
  76 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  77 {
  78         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  79 }
  80
  81 static struct nfulnl_instance *
  82 __instance_lookup(u_int16_t group_num)
  83 {
  84         struct hlist_head *head;
  85         struct hlist_node *pos;
  86         struct nfulnl_instance *inst;
  87
  88         head = &instance_table[instance_hashfn(group_num)];
  89         hlist_for_each_entry(inst, pos, head, hlist) {
  90                 if (inst->group_num == group_num)
  91                         return inst;
  92         }
  93         return NULL;
  94 }
  95
  96 static inline void
  97 instance_get(struct nfulnl_instance *inst)
  98 {
  99         atomic_inc(&inst->use);
 100 }
 101
 102 static struct nfulnl_instance *
 103 instance_lookup_get(u_int16_t group_num)
 104 {
 105         struct nfulnl_instance *inst;
 106
 107         read_lock_bh(&instances_lock);
 108         inst = __instance_lookup(group_num);
 109         if (inst)
 110                 instance_get(inst);
 111         read_unlock_bh(&instances_lock);
 112
 113         return inst;
 114 }
 115
 116 static void
 117 instance_put(struct nfulnl_instance *inst)
 118 {
 119         if (inst && atomic_dec_and_test(&inst->use)) {
 120                 kfree(inst);
 121                 module_put(THIS_MODULE);
 122         }
 123 }
 124
 125 static void nfulnl_timer(unsigned long data);
 126
 127 static struct nfulnl_instance *
 128 instance_create(u_int16_t group_num, int pid)
 129 {
 130         struct nfulnl_instance *inst;
 131         int err;
 132
 133         write_lock_bh(&instances_lock);
 134         if (__instance_lookup(group_num)) {
 135                 err = -EEXIST;
 136                 goto out_unlock;
 137         }
 138
 139         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 140         if (!inst) {
 141                 err = -ENOMEM;
 142                 goto out_unlock;
 143         }
 144
 145         if (!try_module_get(THIS_MODULE)) {
 146                 kfree(inst);
 147                 err = -EAGAIN;
 148                 goto out_unlock;
 149         }
 150
 151         INIT_HLIST_NODE(&inst->hlist);
 152         spin_lock_init(&inst->lock);
 153         /* needs to be two, since we _put() after creation */
 154         atomic_set(&inst->use, 2);
 155
 156         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
 157
 158         inst->peer_pid = pid;
 159         inst->group_num = group_num;
 160
 161         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 162         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 163         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 164         inst->copy_mode         = NFULNL_COPY_PACKET;
 165         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
 166
 167         hlist_add_head(&inst->hlist,
 168                        &instance_table[instance_hashfn(group_num)]);
 169
 170         write_unlock_bh(&instances_lock);
 171
 172         return inst;
 173
 174 out_unlock:
 175         write_unlock_bh(&instances_lock);
 176         return ERR_PTR(err);
 177 }
 178
 179 static void __nfulnl_flush(struct nfulnl_instance *inst);
 180
 181 static void
 182 __instance_destroy(struct nfulnl_instance *inst)
 183 {
 184         /* first pull it out of the global list */
 185         hlist_del(&inst->hlist);
 186
 187         /* then flush all pending packets from skb */
 188
 189         spin_lock_bh(&inst->lock);
 190         if (inst->skb)
 191                 __nfulnl_flush(inst);
 192         spin_unlock_bh(&inst->lock);
 193
 194         /* and finally put the refcount */
 195         instance_put(inst);
 196 }
 197
 198 static inline void
 199 instance_destroy(struct nfulnl_instance *inst)
 200 {
 201         write_lock_bh(&instances_lock);
 202         __instance_destroy(inst);
 203         write_unlock_bh(&instances_lock);
 204 }
 205
 206 static int
 207 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 208                   unsigned int range)
 209 {
 210         int status = 0;
 211
 212         spin_lock_bh(&inst->lock);
 213
 214         switch (mode) {
 215         case NFULNL_COPY_NONE:
 216         case NFULNL_COPY_META:
 217                 inst->copy_mode = mode;
 218                 inst->copy_range = 0;
 219                 break;
 220
 221         case NFULNL_COPY_PACKET:
 222                 inst->copy_mode = mode;
 223                 inst->copy_range = min_t(unsigned int,
 224                                          range, NFULNL_COPY_RANGE_MAX);
 225                 break;
 226
 227         default:
 228                 status = -EINVAL;
 229                 break;
 230         }
 231
 232         spin_unlock_bh(&inst->lock);
 233
 234         return status;
 235 }
 236
 237 static int
 238 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 239 {
 240         int status;
 241
 242         spin_lock_bh(&inst->lock);
 243         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 244                 status = -ERANGE;
 245         else if (nlbufsiz > 131072)
 246                 status = -ERANGE;
 247         else {
 248                 inst->nlbufsiz = nlbufsiz;
 249                 status = 0;
 250         }
 251         spin_unlock_bh(&inst->lock);
 252
 253         return status;
 254 }
 255
 256 static int
 257 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 258 {
 259         spin_lock_bh(&inst->lock);
 260         inst->flushtimeout = timeout;
 261         spin_unlock_bh(&inst->lock);
 262
 263         return 0;
 264 }
 265
 266 static int
 267 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 268 {
 269         spin_lock_bh(&inst->lock);
 270         inst->qthreshold = qthresh;
 271         spin_unlock_bh(&inst->lock);
 272
 273         return 0;
 274 }
 275
 276 static int
 277 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 278 {
 279         spin_lock_bh(&inst->lock);
 280         inst->flags = flags;
 281         spin_unlock_bh(&inst->lock);
 282
 283         return 0;
 284 }
 285
 286 static struct sk_buff *
 287 nfulnl_alloc_skb(unsigned int inst_size, unsigned int pkt_size)
 288 {
 289         struct sk_buff *skb;
 290         unsigned int n;
 291
 292         /* alloc skb which should be big enough for a whole multipart
 293          * message.  WARNING: has to be <= 128k due to slab restrictions */
 294
 295         n = max(inst_size, pkt_size);
 296         skb = alloc_skb(n, GFP_ATOMIC);
 297         if (!skb) {
 298                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
 299                         inst_size);
 300
 301                 if (n > pkt_size) {
 302                         /* try to allocate only as much as we need for current
 303                          * packet */
 304
 305                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 306                         if (!skb)
 307                                 PRINTR("nfnetlink_log: can't even alloc %u "
 308                                        "bytes\n", pkt_size);
 309                 }
 310         }
 311
 312         return skb;
 313 }
 314
 315 static int
 316 __nfulnl_send(struct nfulnl_instance *inst)
 317 {
 318         int status = -1;
 319
 320         if (inst->qlen > 1)
 321                 NLMSG_PUT(inst->skb, 0, 0,
 322                           NLMSG_DONE,
 323                           sizeof(struct nfgenmsg));
 324
 325         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
 326
 327         inst->qlen = 0;
 328         inst->skb = NULL;
 329
 330 nlmsg_failure:
 331         return status;
 332 }
 333
 334 static void
 335 __nfulnl_flush(struct nfulnl_instance *inst)
 336 {
 337         /* timer holds a reference */
 338         if (del_timer(&inst->timer))
 339                 instance_put(inst);
 340         if (inst->skb)
 341                 __nfulnl_send(inst);
 342 }
 343
 344 static void
 345 nfulnl_timer(unsigned long data)
 346 {
 347         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 348
 349         spin_lock_bh(&inst->lock);
 350         if (inst->skb)
 351                 __nfulnl_send(inst);
 352         spin_unlock_bh(&inst->lock);
 353         instance_put(inst);
 354 }
 355
 356 /* This is an inline function, we don't really care about a long
 357  * list of arguments */
 358 static inline int
 359 __build_packet_message(struct nfulnl_instance *inst,
 360                         const struct sk_buff *skb,
 361                         unsigned int data_len,
 362                         unsigned int pf,
 363                         unsigned int hooknum,
 364                         const struct net_device *indev,
 365                         const struct net_device *outdev,
 366                         const struct nf_loginfo *li,
 367                         const char *prefix, unsigned int plen)
 368 {
 369         struct nfulnl_msg_packet_hdr pmsg;
 370         struct nlmsghdr *nlh;
 371         struct nfgenmsg *nfmsg;
 372         __be32 tmp_uint;
 373         sk_buff_data_t old_tail = inst->skb->tail;
 374
 375         nlh = NLMSG_PUT(inst->skb, 0, 0,
 376                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 377                         sizeof(struct nfgenmsg));
 378         nfmsg = NLMSG_DATA(nlh);
 379         nfmsg->nfgen_family = pf;
 380         nfmsg->version = NFNETLINK_V0;
 381         nfmsg->res_id = htons(inst->group_num);
 382
 383         pmsg.hw_protocol        = skb->protocol;
 384         pmsg.hook               = hooknum;
 385
 386         NLA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
 387
 388         if (prefix)
 389                 NLA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
 390
 391         if (indev) {
 392                 tmp_uint = htonl(indev->ifindex);
 393 #ifndef CONFIG_BRIDGE_NETFILTER
 394                 NLA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
 395                         &tmp_uint);
 396 #else
 397                 if (pf == PF_BRIDGE) {
 398                         /* Case 1: outdev is physical input device, we need to
 399                          * look for bridge group (when called from
 400                          * netfilter_bridge) */
 401                         NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 402                                 sizeof(tmp_uint), &tmp_uint);
 403                         /* this is the bridge group "brX" */
 404                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
 405                         NLA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 406                                 sizeof(tmp_uint), &tmp_uint);
 407                 } else {
 408                         /* Case 2: indev is bridge group, we need to look for
 409                          * physical device (when called from ipv4) */
 410                         NLA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 411                                 sizeof(tmp_uint), &tmp_uint);
 412                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
 413                                 tmp_uint =
 414                                     htonl(skb->nf_bridge->physindev->ifindex);
 415                                 NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 416                                         sizeof(tmp_uint), &tmp_uint);
 417                         }
 418                 }
 419 #endif
 420         }
 421
 422         if (outdev) {
 423                 tmp_uint = htonl(outdev->ifindex);
 424 #ifndef CONFIG_BRIDGE_NETFILTER
 425                 NLA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 426                         &tmp_uint);
 427 #else
 428                 if (pf == PF_BRIDGE) {
 429                         /* Case 1: outdev is physical output device, we need to
 430                          * look for bridge group (when called from
 431                          * netfilter_bridge) */
 432                         NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 433                                 sizeof(tmp_uint), &tmp_uint);
 434                         /* this is the bridge group "brX" */
 435                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
 436                         NLA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 437                                 sizeof(tmp_uint), &tmp_uint);
 438                 } else {
 439                         /* Case 2: indev is a bridge group, we need to look
 440                          * for physical device (when called from ipv4) */
 441                         NLA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 442                                 sizeof(tmp_uint), &tmp_uint);
 443                         if (skb->nf_bridge && skb->nf_bridge->physoutdev) {
 444                                 tmp_uint =
 445                                     htonl(skb->nf_bridge->physoutdev->ifindex);
 446                                 NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 447                                         sizeof(tmp_uint), &tmp_uint);
 448                         }
 449                 }
 450 #endif
 451         }
 452
 453         if (skb->mark) {
 454                 tmp_uint = htonl(skb->mark);
 455                 NLA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
 456         }
 457
 458         if (indev && skb->dev) {
 459                 struct nfulnl_msg_packet_hw phw;
 460                 int len = dev_parse_header(skb, phw.hw_addr);
 461                 if (len > 0) {
 462                         phw.hw_addrlen = htons(len);
 463                         NLA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
 464                 }
 465         }
 466
 467         if (skb->tstamp.tv64) {
 468                 struct nfulnl_msg_packet_timestamp ts;
 469                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 470                 ts.sec = cpu_to_be64(tv.tv_sec);
 471                 ts.usec = cpu_to_be64(tv.tv_usec);
 472
 473                 NLA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 474         }
 475
 476         /* UID */
 477         if (skb->sk) {
 478                 read_lock_bh(&skb->sk->sk_callback_lock);
 479                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 480                         __be32 uid = htonl(skb->sk->sk_socket->file->f_uid);
 481                         /* need to unlock here since NLA_PUT may goto */
 482                         read_unlock_bh(&skb->sk->sk_callback_lock);
 483                         NLA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
 484                 } else
 485                         read_unlock_bh(&skb->sk->sk_callback_lock);
 486         }
 487
 488         /* local sequence number */
 489         if (inst->flags & NFULNL_CFG_F_SEQ) {
 490                 tmp_uint = htonl(inst->seq++);
 491                 NLA_PUT(inst->skb, NFULA_SEQ, sizeof(tmp_uint), &tmp_uint);
 492         }
 493         /* global sequence number */
 494         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) {
 495                 tmp_uint = htonl(atomic_inc_return(&global_seq));
 496                 NLA_PUT(inst->skb, NFULA_SEQ_GLOBAL, sizeof(tmp_uint), &tmp_uint);
 497         }
 498
 499         if (data_len) {
 500                 struct nlattr *nla;
 501                 int size = nla_attr_size(data_len);
 502
 503                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
 504                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 505                         goto nlmsg_failure;
 506                 }
 507
 508                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 509                 nla->nla_type = NFULA_PAYLOAD;
 510                 nla->nla_len = size;
 511
 512                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 513                         BUG();
 514         }
 515
 516         nlh->nlmsg_len = inst->skb->tail - old_tail;
 517         return 0;
 518
 519 nlmsg_failure:
 520 nla_put_failure:
 521         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 522         return -1;
 523 }
 524
 525 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 526
 527 static struct nf_loginfo default_loginfo = {
 528         .type =         NF_LOG_TYPE_ULOG,
 529         .u = {
 530                 .ulog = {
 531                         .copy_len       = 0xffff,
 532                         .group          = 0,
 533                         .qthreshold     = 1,
 534                 },
 535         },
 536 };
 537
 538 /* log handler for internal netfilter logging api */
 539 static void
 540 nfulnl_log_packet(unsigned int pf,
 541                   unsigned int hooknum,
 542                   const struct sk_buff *skb,
 543                   const struct net_device *in,
 544                   const struct net_device *out,
 545                   const struct nf_loginfo *li_user,
 546                   const char *prefix)
 547 {
 548         unsigned int size, data_len;
 549         struct nfulnl_instance *inst;
 550         const struct nf_loginfo *li;
 551         unsigned int qthreshold;
 552         unsigned int plen;
 553
 554         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 555                 li = li_user;
 556         else
 557                 li = &default_loginfo;
 558
 559         inst = instance_lookup_get(li->u.ulog.group);
 560         if (!inst)
 561                 return;
 562
 563         plen = 0;
 564         if (prefix)
 565                 plen = strlen(prefix) + 1;
 566
 567         /* FIXME: do we want to make the size calculation conditional based on
 568          * what is actually present?  way more branches and checks, but more
 569          * memory efficient... */
 570         size =    NLMSG_ALIGN(sizeof(struct nfgenmsg))
 571                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 572                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 573                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 574 #ifdef CONFIG_BRIDGE_NETFILTER
 575                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 576                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 577 #endif
 578                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 579                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 580                 + nla_total_size(plen)                  /* prefix */
 581                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 582                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
 583
 584         spin_lock_bh(&inst->lock);
 585
 586         if (inst->flags & NFULNL_CFG_F_SEQ)
 587                 size += nla_total_size(sizeof(u_int32_t));
 588         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 589                 size += nla_total_size(sizeof(u_int32_t));
 590
 591         qthreshold = inst->qthreshold;
 592         /* per-rule qthreshold overrides per-instance */
 593         if (qthreshold > li->u.ulog.qthreshold)
 594                 qthreshold = li->u.ulog.qthreshold;
 595
 596         switch (inst->copy_mode) {
 597         case NFULNL_COPY_META:
 598         case NFULNL_COPY_NONE:
 599                 data_len = 0;
 600                 break;
 601
 602         case NFULNL_COPY_PACKET:
 603                 if (inst->copy_range == 0
 604                     || inst->copy_range > skb->len)
 605                         data_len = skb->len;
 606                 else
 607                         data_len = inst->copy_range;
 608
 609                 size += nla_total_size(data_len);
 610                 break;
 611
 612         default:
 613                 goto unlock_and_release;
 614         }
 615
 616         if (inst->skb &&
 617             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
 618                 /* either the queue len is too high or we don't have
 619                  * enough room in the skb left. flush to userspace. */
 620                 __nfulnl_flush(inst);
 621         }
 622
 623         if (!inst->skb) {
 624                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
 625                 if (!inst->skb)
 626                         goto alloc_failure;
 627         }
 628
 629         inst->qlen++;
 630
 631         __build_packet_message(inst, skb, data_len, pf,
 632                                 hooknum, in, out, li, prefix, plen);
 633
 634         if (inst->qlen >= qthreshold)
 635                 __nfulnl_flush(inst);
 636         /* timer_pending always called within inst->lock, so there
 637          * is no chance of a race here */
 638         else if (!timer_pending(&inst->timer)) {
 639                 instance_get(inst);
 640                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 641                 add_timer(&inst->timer);
 642         }
 643
 644 unlock_and_release:
 645         spin_unlock_bh(&inst->lock);
 646         instance_put(inst);
 647         return;
 648
 649 alloc_failure:
 650         /* FIXME: statistics */
 651         goto unlock_and_release;
 652 }
 653
 654 static int
 655 nfulnl_rcv_nl_event(struct notifier_block *this,
 656                    unsigned long event, void *ptr)
 657 {
 658         struct netlink_notify *n = ptr;
 659
 660         if (event == NETLINK_URELEASE &&
 661             n->protocol == NETLINK_NETFILTER && n->pid) {
 662                 int i;
 663
 664                 /* destroy all instances for this pid */
 665                 write_lock_bh(&instances_lock);
 666                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 667                         struct hlist_node *tmp, *t2;
 668                         struct nfulnl_instance *inst;
 669                         struct hlist_head *head = &instance_table[i];
 670
 671                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 672                                 if ((n->net == &init_net) &&
 673                                     (n->pid == inst->peer_pid))
 674                                         __instance_destroy(inst);
 675                         }
 676                 }
 677                 write_unlock_bh(&instances_lock);
 678         }
 679         return NOTIFY_DONE;
 680 }
 681
 682 static struct notifier_block nfulnl_rtnl_notifier = {
 683         .notifier_call  = nfulnl_rcv_nl_event,
 684 };
 685
 686 static int
 687 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 688                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
 689 {
 690         return -ENOTSUPP;
 691 }
 692
 693 static const struct nf_logger nfulnl_logger = {
 694         .name   = "nfnetlink_log",
 695         .logfn  = &nfulnl_log_packet,
 696         .me     = THIS_MODULE,
 697 };
 698
 699 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 700         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 701         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 702         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 703         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 704         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 705         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 706 };
 707
 708 static int
 709 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 710                    struct nlmsghdr *nlh, struct nlattr *nfula[])
 711 {
 712         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 713         u_int16_t group_num = ntohs(nfmsg->res_id);
 714         struct nfulnl_instance *inst;
 715         int ret = 0;
 716
 717         inst = instance_lookup_get(group_num);
 718         if (inst && inst->peer_pid != NETLINK_CB(skb).pid) {
 719                 ret = -EPERM;
 720                 goto out_put;
 721         }
 722
 723         if (nfula[NFULA_CFG_CMD]) {
 724                 u_int8_t pf = nfmsg->nfgen_family;
 725                 struct nfulnl_msg_config_cmd *cmd;
 726
 727                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 728
 729                 switch (cmd->command) {
 730                 case NFULNL_CFG_CMD_BIND:
 731                         if (inst) {
 732                                 ret = -EBUSY;
 733                                 goto out_put;
 734                         }
 735
 736                         inst = instance_create(group_num,
 737                                                NETLINK_CB(skb).pid);
 738                         if (IS_ERR(inst)) {
 739                                 ret = PTR_ERR(inst);
 740                                 goto out;
 741                         }
 742                         break;
 743                 case NFULNL_CFG_CMD_UNBIND:
 744                         if (!inst) {
 745                                 ret = -ENODEV;
 746                                 goto out;
 747                         }
 748
 749                         instance_destroy(inst);
 750                         goto out;
 751                 case NFULNL_CFG_CMD_PF_BIND:
 752                         ret = nf_log_register(pf, &nfulnl_logger);
 753                         break;
 754                 case NFULNL_CFG_CMD_PF_UNBIND:
 755                         /* This is a bug and a feature.  We cannot unregister
 756                          * other handlers, like nfnetlink_inst can */
 757                         nf_log_unregister_pf(pf);
 758                         break;
 759                 default:
 760                         ret = -ENOTSUPP;
 761                         break;
 762                 }
 763         }
 764
 765         if (nfula[NFULA_CFG_MODE]) {
 766                 struct nfulnl_msg_config_mode *params;
 767                 params = nla_data(nfula[NFULA_CFG_MODE]);
 768
 769                 if (!inst) {
 770                         ret = -ENODEV;
 771                         goto out;
 772                 }
 773                 nfulnl_set_mode(inst, params->copy_mode,
 774                                 ntohl(params->copy_range));
 775         }
 776
 777         if (nfula[NFULA_CFG_TIMEOUT]) {
 778                 __be32 timeout =
 779                         *(__be32 *)nla_data(nfula[NFULA_CFG_TIMEOUT]);
 780
 781                 if (!inst) {
 782                         ret = -ENODEV;
 783                         goto out;
 784                 }
 785                 nfulnl_set_timeout(inst, ntohl(timeout));
 786         }
 787
 788         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 789                 __be32 nlbufsiz =
 790                         *(__be32 *)nla_data(nfula[NFULA_CFG_NLBUFSIZ]);
 791
 792                 if (!inst) {
 793                         ret = -ENODEV;
 794                         goto out;
 795                 }
 796                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 797         }
 798
 799         if (nfula[NFULA_CFG_QTHRESH]) {
 800                 __be32 qthresh =
 801                         *(__be32 *)nla_data(nfula[NFULA_CFG_QTHRESH]);
 802
 803                 if (!inst) {
 804                         ret = -ENODEV;
 805                         goto out;
 806                 }
 807                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 808         }
 809
 810         if (nfula[NFULA_CFG_FLAGS]) {
 811                 __be16 flags =
 812                         *(__be16 *)nla_data(nfula[NFULA_CFG_FLAGS]);
 813
 814                 if (!inst) {
 815                         ret = -ENODEV;
 816                         goto out;
 817                 }
 818                 nfulnl_set_flags(inst, ntohs(flags));
 819         }
 820
 821 out_put:
 822         instance_put(inst);
 823 out:
 824         return ret;
 825 }
 826
 827 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 828         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 829                                     .attr_count = NFULA_MAX, },
 830         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 831                                     .attr_count = NFULA_CFG_MAX,
 832                                     .policy = nfula_cfg_policy },
 833 };
 834
 835 static const struct nfnetlink_subsystem nfulnl_subsys = {
 836         .name           = "log",
 837         .subsys_id      = NFNL_SUBSYS_ULOG,
 838         .cb_count       = NFULNL_MSG_MAX,
 839         .cb             = nfulnl_cb,
 840 };
 841
 842 #ifdef CONFIG_PROC_FS
 843 struct iter_state {
 844         unsigned int bucket;
 845 };
 846
 847 static struct hlist_node *get_first(struct iter_state *st)
 848 {
 849         if (!st)
 850                 return NULL;
 851
 852         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 853                 if (!hlist_empty(&instance_table[st->bucket]))
 854                         return instance_table[st->bucket].first;
 855         }
 856         return NULL;
 857 }
 858
 859 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
 860 {
 861         h = h->next;
 862         while (!h) {
 863                 if (++st->bucket >= INSTANCE_BUCKETS)
 864                         return NULL;
 865
 866                 h = instance_table[st->bucket].first;
 867         }
 868         return h;
 869 }
 870
 871 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
 872 {
 873         struct hlist_node *head;
 874         head = get_first(st);
 875
 876         if (head)
 877                 while (pos && (head = get_next(st, head)))
 878                         pos--;
 879         return pos ? NULL : head;
 880 }
 881
 882 static void *seq_start(struct seq_file *seq, loff_t *pos)
 883 {
 884         read_lock_bh(&instances_lock);
 885         return get_idx(seq->private, *pos);
 886 }
 887
 888 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 889 {
 890         (*pos)++;
 891         return get_next(s->private, v);
 892 }
 893
 894 static void seq_stop(struct seq_file *s, void *v)
 895 {
 896         read_unlock_bh(&instances_lock);
 897 }
 898
 899 static int seq_show(struct seq_file *s, void *v)
 900 {
 901         const struct nfulnl_instance *inst = v;
 902
 903         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 904                           inst->group_num,
 905                           inst->peer_pid, inst->qlen,
 906                           inst->copy_mode, inst->copy_range,
 907                           inst->flushtimeout, atomic_read(&inst->use));
 908 }
 909
 910 static const struct seq_operations nful_seq_ops = {
 911         .start  = seq_start,
 912         .next   = seq_next,
 913         .stop   = seq_stop,
 914         .show   = seq_show,
 915 };
 916
 917 static int nful_open(struct inode *inode, struct file *file)
 918 {
 919         return seq_open_private(file, &nful_seq_ops,
 920                         sizeof(struct iter_state));
 921 }
 922
 923 static const struct file_operations nful_file_ops = {
 924         .owner   = THIS_MODULE,
 925         .open    = nful_open,
 926         .read    = seq_read,
 927         .llseek  = seq_lseek,
 928         .release = seq_release_private,
 929 };
 930
 931 #endif /* PROC_FS */
 932
 933 static int __init nfnetlink_log_init(void)
 934 {
 935         int i, status = -ENOMEM;
 936 #ifdef CONFIG_PROC_FS
 937         struct proc_dir_entry *proc_nful;
 938 #endif
 939
 940         for (i = 0; i < INSTANCE_BUCKETS; i++)
 941                 INIT_HLIST_HEAD(&instance_table[i]);
 942
 943         /* it's not really all that important to have a random value, so
 944          * we can do this from the init function, even if there hasn't
 945          * been that much entropy yet */
 946         get_random_bytes(&hash_init, sizeof(hash_init));
 947
 948         netlink_register_notifier(&nfulnl_rtnl_notifier);
 949         status = nfnetlink_subsys_register(&nfulnl_subsys);
 950         if (status < 0) {
 951                 printk(KERN_ERR "log: failed to create netlink socket\n");
 952                 goto cleanup_netlink_notifier;
 953         }
 954
 955 #ifdef CONFIG_PROC_FS
 956         proc_nful = create_proc_entry("nfnetlink_log", 0440,
 957                                       proc_net_netfilter);
 958         if (!proc_nful)
 959                 goto cleanup_subsys;
 960         proc_nful->proc_fops = &nful_file_ops;
 961 #endif
 962         return status;
 963
 964 #ifdef CONFIG_PROC_FS
 965 cleanup_subsys:
 966         nfnetlink_subsys_unregister(&nfulnl_subsys);
 967 #endif
 968 cleanup_netlink_notifier:
 969         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
 970         return status;
 971 }
 972
 973 static void __exit nfnetlink_log_fini(void)
 974 {
 975         nf_log_unregister(&nfulnl_logger);
 976 #ifdef CONFIG_PROC_FS
 977         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
 978 #endif
 979         nfnetlink_subsys_unregister(&nfulnl_subsys);
 980         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
 981 }
 982
 983 MODULE_DESCRIPTION("netfilter userspace logging");
 984 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
 985 MODULE_LICENSE("GPL");
 986 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
 987
 988 module_init(nfnetlink_log_init);
 989 module_exit(nfnetlink_log_fini);