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                         u_int8_t 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 (indev && skb_mac_header_was_set(skb)) {
 457                 NLA_PUT_BE16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type));
 458                 NLA_PUT_BE16(inst->skb, NFULA_HWLEN,
 459                              htons(skb->dev->hard_header_len));
 460                 NLA_PUT(inst->skb, NFULA_HWHEADER, skb->dev->hard_header_len,
 461                         skb_mac_header(skb));
 462         }
 463
 464         if (skb->tstamp.tv64) {
 465                 struct nfulnl_msg_packet_timestamp ts;
 466                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 467                 ts.sec = cpu_to_be64(tv.tv_sec);
 468                 ts.usec = cpu_to_be64(tv.tv_usec);
 469
 470                 NLA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 471         }
 472
 473         /* UID */
 474         if (skb->sk) {
 475                 read_lock_bh(&skb->sk->sk_callback_lock);
 476                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 477                         struct file *file = skb->sk->sk_socket->file;
 478                         __be32 uid = htonl(file->f_cred->fsuid);
 479                         __be32 gid = htonl(file->f_cred->fsgid);
 480                         /* need to unlock here since NLA_PUT may goto */
 481                         read_unlock_bh(&skb->sk->sk_callback_lock);
 482                         NLA_PUT_BE32(inst->skb, NFULA_UID, uid);
 483                         NLA_PUT_BE32(inst->skb, NFULA_GID, gid);
 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                 NLA_PUT_BE32(inst->skb, NFULA_SEQ, htonl(inst->seq++));
 491
 492         /* global sequence number */
 493         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 494                 NLA_PUT_BE32(inst->skb, NFULA_SEQ_GLOBAL,
 495                              htonl(atomic_inc_return(&global_seq)));
 496
 497         if (data_len) {
 498                 struct nlattr *nla;
 499                 int size = nla_attr_size(data_len);
 500
 501                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
 502                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 503                         goto nlmsg_failure;
 504                 }
 505
 506                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 507                 nla->nla_type = NFULA_PAYLOAD;
 508                 nla->nla_len = size;
 509
 510                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 511                         BUG();
 512         }
 513
 514         nlh->nlmsg_len = inst->skb->tail - old_tail;
 515         return 0;
 516
 517 nlmsg_failure:
 518 nla_put_failure:
 519         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 520         return -1;
 521 }
 522
 523 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 524
 525 static struct nf_loginfo default_loginfo = {
 526         .type =         NF_LOG_TYPE_ULOG,
 527         .u = {
 528                 .ulog = {
 529                         .copy_len       = 0xffff,
 530                         .group          = 0,
 531                         .qthreshold     = 1,
 532                 },
 533         },
 534 };
 535
 536 /* log handler for internal netfilter logging api */
 537 static void
 538 nfulnl_log_packet(u_int8_t pf,
 539                   unsigned int hooknum,
 540                   const struct sk_buff *skb,
 541                   const struct net_device *in,
 542                   const struct net_device *out,
 543                   const struct nf_loginfo *li_user,
 544                   const char *prefix)
 545 {
 546         unsigned int size, data_len;
 547         struct nfulnl_instance *inst;
 548         const struct nf_loginfo *li;
 549         unsigned int qthreshold;
 550         unsigned int plen;
 551
 552         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 553                 li = li_user;
 554         else
 555                 li = &default_loginfo;
 556
 557         inst = instance_lookup_get(li->u.ulog.group);
 558         if (!inst)
 559                 return;
 560
 561         plen = 0;
 562         if (prefix)
 563                 plen = strlen(prefix) + 1;
 564
 565         /* FIXME: do we want to make the size calculation conditional based on
 566          * what is actually present?  way more branches and checks, but more
 567          * memory efficient... */
 568         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 569                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 570                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 571                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 572 #ifdef CONFIG_BRIDGE_NETFILTER
 573                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 574                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 575 #endif
 576                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 577                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 578                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 579                 + nla_total_size(plen)                  /* prefix */
 580                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 581                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
 582
 583         spin_lock_bh(&inst->lock);
 584
 585         if (inst->flags & NFULNL_CFG_F_SEQ)
 586                 size += nla_total_size(sizeof(u_int32_t));
 587         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 588                 size += nla_total_size(sizeof(u_int32_t));
 589
 590         qthreshold = inst->qthreshold;
 591         /* per-rule qthreshold overrides per-instance */
 592         if (qthreshold > li->u.ulog.qthreshold)
 593                 qthreshold = li->u.ulog.qthreshold;
 594
 595         switch (inst->copy_mode) {
 596         case NFULNL_COPY_META:
 597         case NFULNL_COPY_NONE:
 598                 data_len = 0;
 599                 break;
 600
 601         case NFULNL_COPY_PACKET:
 602                 if (inst->copy_range == 0
 603                     || inst->copy_range > skb->len)
 604                         data_len = skb->len;
 605                 else
 606                         data_len = inst->copy_range;
 607
 608                 size += nla_total_size(data_len);
 609                 break;
 610
 611         default:
 612                 goto unlock_and_release;
 613         }
 614
 615         if (inst->skb &&
 616             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
 617                 /* either the queue len is too high or we don't have
 618                  * enough room in the skb left. flush to userspace. */
 619                 __nfulnl_flush(inst);
 620         }
 621
 622         if (!inst->skb) {
 623                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
 624                 if (!inst->skb)
 625                         goto alloc_failure;
 626         }
 627
 628         inst->qlen++;
 629
 630         __build_packet_message(inst, skb, data_len, pf,
 631                                 hooknum, in, out, li, prefix, plen);
 632
 633         if (inst->qlen >= qthreshold)
 634                 __nfulnl_flush(inst);
 635         /* timer_pending always called within inst->lock, so there
 636          * is no chance of a race here */
 637         else if (!timer_pending(&inst->timer)) {
 638                 instance_get(inst);
 639                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 640                 add_timer(&inst->timer);
 641         }
 642
 643 unlock_and_release:
 644         spin_unlock_bh(&inst->lock);
 645         instance_put(inst);
 646         return;
 647
 648 alloc_failure:
 649         /* FIXME: statistics */
 650         goto unlock_and_release;
 651 }
 652
 653 static int
 654 nfulnl_rcv_nl_event(struct notifier_block *this,
 655                    unsigned long event, void *ptr)
 656 {
 657         struct netlink_notify *n = ptr;
 658
 659         if (event == NETLINK_URELEASE &&
 660             n->protocol == NETLINK_NETFILTER && n->pid) {
 661                 int i;
 662
 663                 /* destroy all instances for this pid */
 664                 write_lock_bh(&instances_lock);
 665                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 666                         struct hlist_node *tmp, *t2;
 667                         struct nfulnl_instance *inst;
 668                         struct hlist_head *head = &instance_table[i];
 669
 670                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 671                                 if ((n->net == &init_net) &&
 672                                     (n->pid == inst->peer_pid))
 673                                         __instance_destroy(inst);
 674                         }
 675                 }
 676                 write_unlock_bh(&instances_lock);
 677         }
 678         return NOTIFY_DONE;
 679 }
 680
 681 static struct notifier_block nfulnl_rtnl_notifier = {
 682         .notifier_call  = nfulnl_rcv_nl_event,
 683 };
 684
 685 static int
 686 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 687                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
 688 {
 689         return -ENOTSUPP;
 690 }
 691
 692 static const struct nf_logger nfulnl_logger = {
 693         .name   = "nfnetlink_log",
 694         .logfn  = &nfulnl_log_packet,
 695         .me     = THIS_MODULE,
 696 };
 697
 698 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 699         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 700         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 701         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 702         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 703         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 704         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 705 };
 706
 707 static int
 708 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 709                    struct nlmsghdr *nlh, struct nlattr *nfula[])
 710 {
 711         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 712         u_int16_t group_num = ntohs(nfmsg->res_id);
 713         struct nfulnl_instance *inst;
 714         struct nfulnl_msg_config_cmd *cmd = NULL;
 715         int ret = 0;
 716
 717         if (nfula[NFULA_CFG_CMD]) {
 718                 u_int8_t pf = nfmsg->nfgen_family;
 719                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 720
 721                 /* Commands without queue context */
 722                 switch (cmd->command) {
 723                 case NFULNL_CFG_CMD_PF_BIND:
 724                         return nf_log_register(pf, &nfulnl_logger);
 725                 case NFULNL_CFG_CMD_PF_UNBIND:
 726                         nf_log_unregister_pf(pf);
 727                         return 0;
 728                 }
 729         }
 730
 731         inst = instance_lookup_get(group_num);
 732         if (inst && inst->peer_pid != NETLINK_CB(skb).pid) {
 733                 ret = -EPERM;
 734                 goto out_put;
 735         }
 736
 737         if (cmd != NULL) {
 738                 switch (cmd->command) {
 739                 case NFULNL_CFG_CMD_BIND:
 740                         if (inst) {
 741                                 ret = -EBUSY;
 742                                 goto out_put;
 743                         }
 744
 745                         inst = instance_create(group_num,
 746                                                NETLINK_CB(skb).pid);
 747                         if (IS_ERR(inst)) {
 748                                 ret = PTR_ERR(inst);
 749                                 goto out;
 750                         }
 751                         break;
 752                 case NFULNL_CFG_CMD_UNBIND:
 753                         if (!inst) {
 754                                 ret = -ENODEV;
 755                                 goto out;
 756                         }
 757
 758                         instance_destroy(inst);
 759                         goto out;
 760                 default:
 761                         ret = -ENOTSUPP;
 762                         break;
 763                 }
 764         }
 765
 766         if (nfula[NFULA_CFG_MODE]) {
 767                 struct nfulnl_msg_config_mode *params;
 768                 params = nla_data(nfula[NFULA_CFG_MODE]);
 769
 770                 if (!inst) {
 771                         ret = -ENODEV;
 772                         goto out;
 773                 }
 774                 nfulnl_set_mode(inst, params->copy_mode,
 775                                 ntohl(params->copy_range));
 776         }
 777
 778         if (nfula[NFULA_CFG_TIMEOUT]) {
 779                 __be32 timeout = nla_get_be32(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 = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 790
 791                 if (!inst) {
 792                         ret = -ENODEV;
 793                         goto out;
 794                 }
 795                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 796         }
 797
 798         if (nfula[NFULA_CFG_QTHRESH]) {
 799                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 800
 801                 if (!inst) {
 802                         ret = -ENODEV;
 803                         goto out;
 804                 }
 805                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 806         }
 807
 808         if (nfula[NFULA_CFG_FLAGS]) {
 809                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
 810
 811                 if (!inst) {
 812                         ret = -ENODEV;
 813                         goto out;
 814                 }
 815                 nfulnl_set_flags(inst, ntohs(flags));
 816         }
 817
 818 out_put:
 819         instance_put(inst);
 820 out:
 821         return ret;
 822 }
 823
 824 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 825         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 826                                     .attr_count = NFULA_MAX, },
 827         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 828                                     .attr_count = NFULA_CFG_MAX,
 829                                     .policy = nfula_cfg_policy },
 830 };
 831
 832 static const struct nfnetlink_subsystem nfulnl_subsys = {
 833         .name           = "log",
 834         .subsys_id      = NFNL_SUBSYS_ULOG,
 835         .cb_count       = NFULNL_MSG_MAX,
 836         .cb             = nfulnl_cb,
 837 };
 838
 839 #ifdef CONFIG_PROC_FS
 840 struct iter_state {
 841         unsigned int bucket;
 842 };
 843
 844 static struct hlist_node *get_first(struct iter_state *st)
 845 {
 846         if (!st)
 847                 return NULL;
 848
 849         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 850                 if (!hlist_empty(&instance_table[st->bucket]))
 851                         return instance_table[st->bucket].first;
 852         }
 853         return NULL;
 854 }
 855
 856 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
 857 {
 858         h = h->next;
 859         while (!h) {
 860                 if (++st->bucket >= INSTANCE_BUCKETS)
 861                         return NULL;
 862
 863                 h = instance_table[st->bucket].first;
 864         }
 865         return h;
 866 }
 867
 868 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
 869 {
 870         struct hlist_node *head;
 871         head = get_first(st);
 872
 873         if (head)
 874                 while (pos && (head = get_next(st, head)))
 875                         pos--;
 876         return pos ? NULL : head;
 877 }
 878
 879 static void *seq_start(struct seq_file *seq, loff_t *pos)
 880         __acquires(instances_lock)
 881 {
 882         read_lock_bh(&instances_lock);
 883         return get_idx(seq->private, *pos);
 884 }
 885
 886 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 887 {
 888         (*pos)++;
 889         return get_next(s->private, v);
 890 }
 891
 892 static void seq_stop(struct seq_file *s, void *v)
 893         __releases(instances_lock)
 894 {
 895         read_unlock_bh(&instances_lock);
 896 }
 897
 898 static int seq_show(struct seq_file *s, void *v)
 899 {
 900         const struct nfulnl_instance *inst = v;
 901
 902         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 903                           inst->group_num,
 904                           inst->peer_pid, inst->qlen,
 905                           inst->copy_mode, inst->copy_range,
 906                           inst->flushtimeout, atomic_read(&inst->use));
 907 }
 908
 909 static const struct seq_operations nful_seq_ops = {
 910         .start  = seq_start,
 911         .next   = seq_next,
 912         .stop   = seq_stop,
 913         .show   = seq_show,
 914 };
 915
 916 static int nful_open(struct inode *inode, struct file *file)
 917 {
 918         return seq_open_private(file, &nful_seq_ops,
 919                         sizeof(struct iter_state));
 920 }
 921
 922 static const struct file_operations nful_file_ops = {
 923         .owner   = THIS_MODULE,
 924         .open    = nful_open,
 925         .read    = seq_read,
 926         .llseek  = seq_lseek,
 927         .release = seq_release_private,
 928 };
 929
 930 #endif /* PROC_FS */
 931
 932 static int __init nfnetlink_log_init(void)
 933 {
 934         int i, status = -ENOMEM;
 935
 936         for (i = 0; i < INSTANCE_BUCKETS; i++)
 937                 INIT_HLIST_HEAD(&instance_table[i]);
 938
 939         /* it's not really all that important to have a random value, so
 940          * we can do this from the init function, even if there hasn't
 941          * been that much entropy yet */
 942         get_random_bytes(&hash_init, sizeof(hash_init));
 943
 944         netlink_register_notifier(&nfulnl_rtnl_notifier);
 945         status = nfnetlink_subsys_register(&nfulnl_subsys);
 946         if (status < 0) {
 947                 printk(KERN_ERR "log: failed to create netlink socket\n");
 948                 goto cleanup_netlink_notifier;
 949         }
 950
 951 #ifdef CONFIG_PROC_FS
 952         if (!proc_create("nfnetlink_log", 0440,
 953                          proc_net_netfilter, &nful_file_ops))
 954                 goto cleanup_subsys;
 955 #endif
 956         return status;
 957
 958 #ifdef CONFIG_PROC_FS
 959 cleanup_subsys:
 960         nfnetlink_subsys_unregister(&nfulnl_subsys);
 961 #endif
 962 cleanup_netlink_notifier:
 963         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
 964         return status;
 965 }
 966
 967 static void __exit nfnetlink_log_fini(void)
 968 {
 969         nf_log_unregister(&nfulnl_logger);
 970 #ifdef CONFIG_PROC_FS
 971         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
 972 #endif
 973         nfnetlink_subsys_unregister(&nfulnl_subsys);
 974         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
 975 }
 976
 977 MODULE_DESCRIPTION("netfilter userspace logging");
 978 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
 979 MODULE_LICENSE("GPL");
 980 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
 981
 982 module_init(nfnetlink_log_init);
 983 module_exit(nfnetlink_log_fini);