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 #ifndef CONFIG_BRIDGE_NETFILTER
 393                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
 394                              htonl(indev->ifindex));
 395 #else
 396                 if (pf == PF_BRIDGE) {
 397                         /* Case 1: outdev is physical input device, we need to
 398                          * look for bridge group (when called from
 399                          * netfilter_bridge) */
 400                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 401                                      htonl(indev->ifindex));
 402                         /* this is the bridge group "brX" */
 403                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
 404                                      htonl(indev->br_port->br->dev->ifindex));
 405                 } else {
 406                         /* Case 2: indev is bridge group, we need to look for
 407                          * physical device (when called from ipv4) */
 408                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
 409                                      htonl(indev->ifindex));
 410                         if (skb->nf_bridge && skb->nf_bridge->physindev)
 411                                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 412                                              htonl(skb->nf_bridge->physindev->ifindex));
 413                 }
 414 #endif
 415         }
 416
 417         if (outdev) {
 418                 tmp_uint = htonl(outdev->ifindex);
 419 #ifndef CONFIG_BRIDGE_NETFILTER
 420                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
 421                              htonl(outdev->ifindex));
 422 #else
 423                 if (pf == PF_BRIDGE) {
 424                         /* Case 1: outdev is physical output device, we need to
 425                          * look for bridge group (when called from
 426                          * netfilter_bridge) */
 427                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 428                                      htonl(outdev->ifindex));
 429                         /* this is the bridge group "brX" */
 430                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
 431                                      htonl(outdev->br_port->br->dev->ifindex));
 432                 } else {
 433                         /* Case 2: indev is a bridge group, we need to look
 434                          * for physical device (when called from ipv4) */
 435                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
 436                                      htonl(outdev->ifindex));
 437                         if (skb->nf_bridge && skb->nf_bridge->physoutdev)
 438                                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 439                                              htonl(skb->nf_bridge->physoutdev->ifindex));
 440                 }
 441 #endif
 442         }
 443
 444         if (skb->mark)
 445                 NLA_PUT_BE32(inst->skb, NFULA_MARK, htonl(skb->mark));
 446
 447         if (indev && skb->dev) {
 448                 struct nfulnl_msg_packet_hw phw;
 449                 int len = dev_parse_header(skb, phw.hw_addr);
 450                 if (len > 0) {
 451                         phw.hw_addrlen = htons(len);
 452                         NLA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
 453                 }
 454         }
 455
 456         if (skb->tstamp.tv64) {
 457                 struct nfulnl_msg_packet_timestamp ts;
 458                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 459                 ts.sec = cpu_to_be64(tv.tv_sec);
 460                 ts.usec = cpu_to_be64(tv.tv_usec);
 461
 462                 NLA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 463         }
 464
 465         /* UID */
 466         if (skb->sk) {
 467                 read_lock_bh(&skb->sk->sk_callback_lock);
 468                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 469                         __be32 uid = htonl(skb->sk->sk_socket->file->f_uid);
 470                         /* need to unlock here since NLA_PUT may goto */
 471                         read_unlock_bh(&skb->sk->sk_callback_lock);
 472                         NLA_PUT_BE32(inst->skb, NFULA_UID, uid);
 473                 } else
 474                         read_unlock_bh(&skb->sk->sk_callback_lock);
 475         }
 476
 477         /* local sequence number */
 478         if (inst->flags & NFULNL_CFG_F_SEQ)
 479                 NLA_PUT_BE32(inst->skb, NFULA_SEQ, htonl(inst->seq++));
 480
 481         /* global sequence number */
 482         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 483                 NLA_PUT_BE32(inst->skb, NFULA_SEQ_GLOBAL,
 484                              htonl(atomic_inc_return(&global_seq)));
 485
 486         if (data_len) {
 487                 struct nlattr *nla;
 488                 int size = nla_attr_size(data_len);
 489
 490                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
 491                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 492                         goto nlmsg_failure;
 493                 }
 494
 495                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 496                 nla->nla_type = NFULA_PAYLOAD;
 497                 nla->nla_len = size;
 498
 499                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 500                         BUG();
 501         }
 502
 503         nlh->nlmsg_len = inst->skb->tail - old_tail;
 504         return 0;
 505
 506 nlmsg_failure:
 507 nla_put_failure:
 508         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 509         return -1;
 510 }
 511
 512 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 513
 514 static struct nf_loginfo default_loginfo = {
 515         .type =         NF_LOG_TYPE_ULOG,
 516         .u = {
 517                 .ulog = {
 518                         .copy_len       = 0xffff,
 519                         .group          = 0,
 520                         .qthreshold     = 1,
 521                 },
 522         },
 523 };
 524
 525 /* log handler for internal netfilter logging api */
 526 static void
 527 nfulnl_log_packet(unsigned int pf,
 528                   unsigned int hooknum,
 529                   const struct sk_buff *skb,
 530                   const struct net_device *in,
 531                   const struct net_device *out,
 532                   const struct nf_loginfo *li_user,
 533                   const char *prefix)
 534 {
 535         unsigned int size, data_len;
 536         struct nfulnl_instance *inst;
 537         const struct nf_loginfo *li;
 538         unsigned int qthreshold;
 539         unsigned int plen;
 540
 541         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 542                 li = li_user;
 543         else
 544                 li = &default_loginfo;
 545
 546         inst = instance_lookup_get(li->u.ulog.group);
 547         if (!inst)
 548                 return;
 549
 550         plen = 0;
 551         if (prefix)
 552                 plen = strlen(prefix) + 1;
 553
 554         /* FIXME: do we want to make the size calculation conditional based on
 555          * what is actually present?  way more branches and checks, but more
 556          * memory efficient... */
 557         size =    NLMSG_ALIGN(sizeof(struct nfgenmsg))
 558                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 559                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 560                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 561 #ifdef CONFIG_BRIDGE_NETFILTER
 562                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 563                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 564 #endif
 565                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 566                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 567                 + nla_total_size(plen)                  /* prefix */
 568                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 569                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
 570
 571         spin_lock_bh(&inst->lock);
 572
 573         if (inst->flags & NFULNL_CFG_F_SEQ)
 574                 size += nla_total_size(sizeof(u_int32_t));
 575         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 576                 size += nla_total_size(sizeof(u_int32_t));
 577
 578         qthreshold = inst->qthreshold;
 579         /* per-rule qthreshold overrides per-instance */
 580         if (qthreshold > li->u.ulog.qthreshold)
 581                 qthreshold = li->u.ulog.qthreshold;
 582
 583         switch (inst->copy_mode) {
 584         case NFULNL_COPY_META:
 585         case NFULNL_COPY_NONE:
 586                 data_len = 0;
 587                 break;
 588
 589         case NFULNL_COPY_PACKET:
 590                 if (inst->copy_range == 0
 591                     || inst->copy_range > skb->len)
 592                         data_len = skb->len;
 593                 else
 594                         data_len = inst->copy_range;
 595
 596                 size += nla_total_size(data_len);
 597                 break;
 598
 599         default:
 600                 goto unlock_and_release;
 601         }
 602
 603         if (inst->skb &&
 604             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
 605                 /* either the queue len is too high or we don't have
 606                  * enough room in the skb left. flush to userspace. */
 607                 __nfulnl_flush(inst);
 608         }
 609
 610         if (!inst->skb) {
 611                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
 612                 if (!inst->skb)
 613                         goto alloc_failure;
 614         }
 615
 616         inst->qlen++;
 617
 618         __build_packet_message(inst, skb, data_len, pf,
 619                                 hooknum, in, out, li, prefix, plen);
 620
 621         if (inst->qlen >= qthreshold)
 622                 __nfulnl_flush(inst);
 623         /* timer_pending always called within inst->lock, so there
 624          * is no chance of a race here */
 625         else if (!timer_pending(&inst->timer)) {
 626                 instance_get(inst);
 627                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 628                 add_timer(&inst->timer);
 629         }
 630
 631 unlock_and_release:
 632         spin_unlock_bh(&inst->lock);
 633         instance_put(inst);
 634         return;
 635
 636 alloc_failure:
 637         /* FIXME: statistics */
 638         goto unlock_and_release;
 639 }
 640
 641 static int
 642 nfulnl_rcv_nl_event(struct notifier_block *this,
 643                    unsigned long event, void *ptr)
 644 {
 645         struct netlink_notify *n = ptr;
 646
 647         if (event == NETLINK_URELEASE &&
 648             n->protocol == NETLINK_NETFILTER && n->pid) {
 649                 int i;
 650
 651                 /* destroy all instances for this pid */
 652                 write_lock_bh(&instances_lock);
 653                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 654                         struct hlist_node *tmp, *t2;
 655                         struct nfulnl_instance *inst;
 656                         struct hlist_head *head = &instance_table[i];
 657
 658                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 659                                 if ((n->net == &init_net) &&
 660                                     (n->pid == inst->peer_pid))
 661                                         __instance_destroy(inst);
 662                         }
 663                 }
 664                 write_unlock_bh(&instances_lock);
 665         }
 666         return NOTIFY_DONE;
 667 }
 668
 669 static struct notifier_block nfulnl_rtnl_notifier = {
 670         .notifier_call  = nfulnl_rcv_nl_event,
 671 };
 672
 673 static int
 674 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 675                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
 676 {
 677         return -ENOTSUPP;
 678 }
 679
 680 static const struct nf_logger nfulnl_logger = {
 681         .name   = "nfnetlink_log",
 682         .logfn  = &nfulnl_log_packet,
 683         .me     = THIS_MODULE,
 684 };
 685
 686 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 687         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 688         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 689         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 690         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 691         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 692         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 693 };
 694
 695 static int
 696 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 697                    struct nlmsghdr *nlh, struct nlattr *nfula[])
 698 {
 699         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 700         u_int16_t group_num = ntohs(nfmsg->res_id);
 701         struct nfulnl_instance *inst;
 702         int ret = 0;
 703
 704         inst = instance_lookup_get(group_num);
 705         if (inst && inst->peer_pid != NETLINK_CB(skb).pid) {
 706                 ret = -EPERM;
 707                 goto out_put;
 708         }
 709
 710         if (nfula[NFULA_CFG_CMD]) {
 711                 u_int8_t pf = nfmsg->nfgen_family;
 712                 struct nfulnl_msg_config_cmd *cmd;
 713
 714                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 715
 716                 switch (cmd->command) {
 717                 case NFULNL_CFG_CMD_BIND:
 718                         if (inst) {
 719                                 ret = -EBUSY;
 720                                 goto out_put;
 721                         }
 722
 723                         inst = instance_create(group_num,
 724                                                NETLINK_CB(skb).pid);
 725                         if (IS_ERR(inst)) {
 726                                 ret = PTR_ERR(inst);
 727                                 goto out;
 728                         }
 729                         break;
 730                 case NFULNL_CFG_CMD_UNBIND:
 731                         if (!inst) {
 732                                 ret = -ENODEV;
 733                                 goto out;
 734                         }
 735
 736                         instance_destroy(inst);
 737                         goto out;
 738                 case NFULNL_CFG_CMD_PF_BIND:
 739                         ret = nf_log_register(pf, &nfulnl_logger);
 740                         break;
 741                 case NFULNL_CFG_CMD_PF_UNBIND:
 742                         /* This is a bug and a feature.  We cannot unregister
 743                          * other handlers, like nfnetlink_inst can */
 744                         nf_log_unregister_pf(pf);
 745                         break;
 746                 default:
 747                         ret = -ENOTSUPP;
 748                         break;
 749                 }
 750         }
 751
 752         if (nfula[NFULA_CFG_MODE]) {
 753                 struct nfulnl_msg_config_mode *params;
 754                 params = nla_data(nfula[NFULA_CFG_MODE]);
 755
 756                 if (!inst) {
 757                         ret = -ENODEV;
 758                         goto out;
 759                 }
 760                 nfulnl_set_mode(inst, params->copy_mode,
 761                                 ntohl(params->copy_range));
 762         }
 763
 764         if (nfula[NFULA_CFG_TIMEOUT]) {
 765                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 766
 767                 if (!inst) {
 768                         ret = -ENODEV;
 769                         goto out;
 770                 }
 771                 nfulnl_set_timeout(inst, ntohl(timeout));
 772         }
 773
 774         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 775                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 776
 777                 if (!inst) {
 778                         ret = -ENODEV;
 779                         goto out;
 780                 }
 781                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 782         }
 783
 784         if (nfula[NFULA_CFG_QTHRESH]) {
 785                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 786
 787                 if (!inst) {
 788                         ret = -ENODEV;
 789                         goto out;
 790                 }
 791                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 792         }
 793
 794         if (nfula[NFULA_CFG_FLAGS]) {
 795                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
 796
 797                 if (!inst) {
 798                         ret = -ENODEV;
 799                         goto out;
 800                 }
 801                 nfulnl_set_flags(inst, ntohs(flags));
 802         }
 803
 804 out_put:
 805         instance_put(inst);
 806 out:
 807         return ret;
 808 }
 809
 810 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 811         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 812                                     .attr_count = NFULA_MAX, },
 813         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 814                                     .attr_count = NFULA_CFG_MAX,
 815                                     .policy = nfula_cfg_policy },
 816 };
 817
 818 static const struct nfnetlink_subsystem nfulnl_subsys = {
 819         .name           = "log",
 820         .subsys_id      = NFNL_SUBSYS_ULOG,
 821         .cb_count       = NFULNL_MSG_MAX,
 822         .cb             = nfulnl_cb,
 823 };
 824
 825 #ifdef CONFIG_PROC_FS
 826 struct iter_state {
 827         unsigned int bucket;
 828 };
 829
 830 static struct hlist_node *get_first(struct iter_state *st)
 831 {
 832         if (!st)
 833                 return NULL;
 834
 835         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 836                 if (!hlist_empty(&instance_table[st->bucket]))
 837                         return instance_table[st->bucket].first;
 838         }
 839         return NULL;
 840 }
 841
 842 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
 843 {
 844         h = h->next;
 845         while (!h) {
 846                 if (++st->bucket >= INSTANCE_BUCKETS)
 847                         return NULL;
 848
 849                 h = instance_table[st->bucket].first;
 850         }
 851         return h;
 852 }
 853
 854 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
 855 {
 856         struct hlist_node *head;
 857         head = get_first(st);
 858
 859         if (head)
 860                 while (pos && (head = get_next(st, head)))
 861                         pos--;
 862         return pos ? NULL : head;
 863 }
 864
 865 static void *seq_start(struct seq_file *seq, loff_t *pos)
 866 {
 867         read_lock_bh(&instances_lock);
 868         return get_idx(seq->private, *pos);
 869 }
 870
 871 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 872 {
 873         (*pos)++;
 874         return get_next(s->private, v);
 875 }
 876
 877 static void seq_stop(struct seq_file *s, void *v)
 878 {
 879         read_unlock_bh(&instances_lock);
 880 }
 881
 882 static int seq_show(struct seq_file *s, void *v)
 883 {
 884         const struct nfulnl_instance *inst = v;
 885
 886         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 887                           inst->group_num,
 888                           inst->peer_pid, inst->qlen,
 889                           inst->copy_mode, inst->copy_range,
 890                           inst->flushtimeout, atomic_read(&inst->use));
 891 }
 892
 893 static const struct seq_operations nful_seq_ops = {
 894         .start  = seq_start,
 895         .next   = seq_next,
 896         .stop   = seq_stop,
 897         .show   = seq_show,
 898 };
 899
 900 static int nful_open(struct inode *inode, struct file *file)
 901 {
 902         return seq_open_private(file, &nful_seq_ops,
 903                         sizeof(struct iter_state));
 904 }
 905
 906 static const struct file_operations nful_file_ops = {
 907         .owner   = THIS_MODULE,
 908         .open    = nful_open,
 909         .read    = seq_read,
 910         .llseek  = seq_lseek,
 911         .release = seq_release_private,
 912 };
 913
 914 #endif /* PROC_FS */
 915
 916 static int __init nfnetlink_log_init(void)
 917 {
 918         int i, status = -ENOMEM;
 919 #ifdef CONFIG_PROC_FS
 920         struct proc_dir_entry *proc_nful;
 921 #endif
 922
 923         for (i = 0; i < INSTANCE_BUCKETS; i++)
 924                 INIT_HLIST_HEAD(&instance_table[i]);
 925
 926         /* it's not really all that important to have a random value, so
 927          * we can do this from the init function, even if there hasn't
 928          * been that much entropy yet */
 929         get_random_bytes(&hash_init, sizeof(hash_init));
 930
 931         netlink_register_notifier(&nfulnl_rtnl_notifier);
 932         status = nfnetlink_subsys_register(&nfulnl_subsys);
 933         if (status < 0) {
 934                 printk(KERN_ERR "log: failed to create netlink socket\n");
 935                 goto cleanup_netlink_notifier;
 936         }
 937
 938 #ifdef CONFIG_PROC_FS
 939         proc_nful = create_proc_entry("nfnetlink_log", 0440,
 940                                       proc_net_netfilter);
 941         if (!proc_nful)
 942                 goto cleanup_subsys;
 943         proc_nful->proc_fops = &nful_file_ops;
 944 #endif
 945         return status;
 946
 947 #ifdef CONFIG_PROC_FS
 948 cleanup_subsys:
 949         nfnetlink_subsys_unregister(&nfulnl_subsys);
 950 #endif
 951 cleanup_netlink_notifier:
 952         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
 953         return status;
 954 }
 955
 956 static void __exit nfnetlink_log_fini(void)
 957 {
 958         nf_log_unregister(&nfulnl_logger);
 959 #ifdef CONFIG_PROC_FS
 960         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
 961 #endif
 962         nfnetlink_subsys_unregister(&nfulnl_subsys);
 963         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
 964 }
 965
 966 MODULE_DESCRIPTION("netfilter userspace logging");
 967 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
 968 MODULE_LICENSE("GPL");
 969 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
 970
 971 module_init(nfnetlink_log_init);
 972 module_exit(nfnetlink_log_fini);