git.oblomov.eu Git - 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 #include <net/netfilter/nfnetlink_log.h>
  34
  35 #include <asm/atomic.h>
  36
  37 #ifdef CONFIG_BRIDGE_NETFILTER
  38 #include "../bridge/br_private.h"
  39 #endif
  40
  41 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
  42 #define NFULNL_TIMEOUT_DEFAULT  HZ      /* every second */
  43 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  44 #define NFULNL_COPY_RANGE_MAX   0xFFFF  /* max packet size is limited by 16-bit struct nfattr nfa_len field */
  45
  46 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  47                                      printk(x, ## args); } while (0);
  48
  49 struct nfulnl_instance {
  50         struct hlist_node hlist;        /* global list of instances */
  51         spinlock_t lock;
  52         atomic_t use;                   /* use count */
  53
  54         unsigned int qlen;              /* number of nlmsgs in skb */
  55         struct sk_buff *skb;            /* pre-allocatd skb */
  56         struct timer_list timer;
  57         int peer_pid;                   /* PID of the peer process */
  58
  59         /* configurable parameters */
  60         unsigned int flushtimeout;      /* timeout until queue flush */
  61         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  62         unsigned int qthreshold;        /* threshold of the queue */
  63         u_int32_t copy_range;
  64         u_int32_t seq;                  /* instance-local sequential counter */
  65         u_int16_t group_num;            /* number of this queue */
  66         u_int16_t flags;
  67         u_int8_t copy_mode;
  68 };
  69
  70 static DEFINE_RWLOCK(instances_lock);
  71 static atomic_t global_seq;
  72
  73 #define INSTANCE_BUCKETS        16
  74 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  75 static unsigned int hash_init;
  76
  77 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  78 {
  79         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  80 }
  81
  82 static struct nfulnl_instance *
  83 __instance_lookup(u_int16_t group_num)
  84 {
  85         struct hlist_head *head;
  86         struct hlist_node *pos;
  87         struct nfulnl_instance *inst;
  88
  89         head = &instance_table[instance_hashfn(group_num)];
  90         hlist_for_each_entry(inst, pos, head, hlist) {
  91                 if (inst->group_num == group_num)
  92                         return inst;
  93         }
  94         return NULL;
  95 }
  96
  97 static inline void
  98 instance_get(struct nfulnl_instance *inst)
  99 {
 100         atomic_inc(&inst->use);
 101 }
 102
 103 static struct nfulnl_instance *
 104 instance_lookup_get(u_int16_t group_num)
 105 {
 106         struct nfulnl_instance *inst;
 107
 108         read_lock_bh(&instances_lock);
 109         inst = __instance_lookup(group_num);
 110         if (inst)
 111                 instance_get(inst);
 112         read_unlock_bh(&instances_lock);
 113
 114         return inst;
 115 }
 116
 117 static void
 118 instance_put(struct nfulnl_instance *inst)
 119 {
 120         if (inst && atomic_dec_and_test(&inst->use)) {
 121                 kfree(inst);
 122                 module_put(THIS_MODULE);
 123         }
 124 }
 125
 126 static void nfulnl_timer(unsigned long data);
 127
 128 static struct nfulnl_instance *
 129 instance_create(u_int16_t group_num, int pid)
 130 {
 131         struct nfulnl_instance *inst;
 132         int err;
 133
 134         write_lock_bh(&instances_lock);
 135         if (__instance_lookup(group_num)) {
 136                 err = -EEXIST;
 137                 goto out_unlock;
 138         }
 139
 140         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 141         if (!inst) {
 142                 err = -ENOMEM;
 143                 goto out_unlock;
 144         }
 145
 146         if (!try_module_get(THIS_MODULE)) {
 147                 kfree(inst);
 148                 err = -EAGAIN;
 149                 goto out_unlock;
 150         }
 151
 152         INIT_HLIST_NODE(&inst->hlist);
 153         spin_lock_init(&inst->lock);
 154         /* needs to be two, since we _put() after creation */
 155         atomic_set(&inst->use, 2);
 156
 157         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
 158
 159         inst->peer_pid = pid;
 160         inst->group_num = group_num;
 161
 162         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 163         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 164         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 165         inst->copy_mode         = NFULNL_COPY_PACKET;
 166         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
 167
 168         hlist_add_head(&inst->hlist,
 169                        &instance_table[instance_hashfn(group_num)]);
 170
 171         write_unlock_bh(&instances_lock);
 172
 173         return inst;
 174
 175 out_unlock:
 176         write_unlock_bh(&instances_lock);
 177         return ERR_PTR(err);
 178 }
 179
 180 static void __nfulnl_flush(struct nfulnl_instance *inst);
 181
 182 static void
 183 __instance_destroy(struct nfulnl_instance *inst)
 184 {
 185         /* first pull it out of the global list */
 186         hlist_del(&inst->hlist);
 187
 188         /* then flush all pending packets from skb */
 189
 190         spin_lock_bh(&inst->lock);
 191         if (inst->skb)
 192                 __nfulnl_flush(inst);
 193         spin_unlock_bh(&inst->lock);
 194
 195         /* and finally put the refcount */
 196         instance_put(inst);
 197 }
 198
 199 static inline void
 200 instance_destroy(struct nfulnl_instance *inst)
 201 {
 202         write_lock_bh(&instances_lock);
 203         __instance_destroy(inst);
 204         write_unlock_bh(&instances_lock);
 205 }
 206
 207 static int
 208 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 209                   unsigned int range)
 210 {
 211         int status = 0;
 212
 213         spin_lock_bh(&inst->lock);
 214
 215         switch (mode) {
 216         case NFULNL_COPY_NONE:
 217         case NFULNL_COPY_META:
 218                 inst->copy_mode = mode;
 219                 inst->copy_range = 0;
 220                 break;
 221
 222         case NFULNL_COPY_PACKET:
 223                 inst->copy_mode = mode;
 224                 inst->copy_range = min_t(unsigned int,
 225                                          range, NFULNL_COPY_RANGE_MAX);
 226                 break;
 227
 228         default:
 229                 status = -EINVAL;
 230                 break;
 231         }
 232
 233         spin_unlock_bh(&inst->lock);
 234
 235         return status;
 236 }
 237
 238 static int
 239 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 240 {
 241         int status;
 242
 243         spin_lock_bh(&inst->lock);
 244         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 245                 status = -ERANGE;
 246         else if (nlbufsiz > 131072)
 247                 status = -ERANGE;
 248         else {
 249                 inst->nlbufsiz = nlbufsiz;
 250                 status = 0;
 251         }
 252         spin_unlock_bh(&inst->lock);
 253
 254         return status;
 255 }
 256
 257 static int
 258 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 259 {
 260         spin_lock_bh(&inst->lock);
 261         inst->flushtimeout = timeout;
 262         spin_unlock_bh(&inst->lock);
 263
 264         return 0;
 265 }
 266
 267 static int
 268 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 269 {
 270         spin_lock_bh(&inst->lock);
 271         inst->qthreshold = qthresh;
 272         spin_unlock_bh(&inst->lock);
 273
 274         return 0;
 275 }
 276
 277 static int
 278 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 279 {
 280         spin_lock_bh(&inst->lock);
 281         inst->flags = flags;
 282         spin_unlock_bh(&inst->lock);
 283
 284         return 0;
 285 }
 286
 287 static struct sk_buff *
 288 nfulnl_alloc_skb(unsigned int inst_size, unsigned int pkt_size)
 289 {
 290         struct sk_buff *skb;
 291         unsigned int n;
 292
 293         /* alloc skb which should be big enough for a whole multipart
 294          * message.  WARNING: has to be <= 128k due to slab restrictions */
 295
 296         n = max(inst_size, pkt_size);
 297         skb = alloc_skb(n, GFP_ATOMIC);
 298         if (!skb) {
 299                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
 300                         inst_size);
 301
 302                 if (n > pkt_size) {
 303                         /* try to allocate only as much as we need for current
 304                          * packet */
 305
 306                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 307                         if (!skb)
 308                                 PRINTR("nfnetlink_log: can't even alloc %u "
 309                                        "bytes\n", pkt_size);
 310                 }
 311         }
 312
 313         return skb;
 314 }
 315
 316 static int
 317 __nfulnl_send(struct nfulnl_instance *inst)
 318 {
 319         int status = -1;
 320
 321         if (inst->qlen > 1)
 322                 NLMSG_PUT(inst->skb, 0, 0,
 323                           NLMSG_DONE,
 324                           sizeof(struct nfgenmsg));
 325
 326         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
 327
 328         inst->qlen = 0;
 329         inst->skb = NULL;
 330
 331 nlmsg_failure:
 332         return status;
 333 }
 334
 335 static void
 336 __nfulnl_flush(struct nfulnl_instance *inst)
 337 {
 338         /* timer holds a reference */
 339         if (del_timer(&inst->timer))
 340                 instance_put(inst);
 341         if (inst->skb)
 342                 __nfulnl_send(inst);
 343 }
 344
 345 static void
 346 nfulnl_timer(unsigned long data)
 347 {
 348         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 349
 350         spin_lock_bh(&inst->lock);
 351         if (inst->skb)
 352                 __nfulnl_send(inst);
 353         spin_unlock_bh(&inst->lock);
 354         instance_put(inst);
 355 }
 356
 357 /* This is an inline function, we don't really care about a long
 358  * list of arguments */
 359 static inline int
 360 __build_packet_message(struct nfulnl_instance *inst,
 361                         const struct sk_buff *skb,
 362                         unsigned int data_len,
 363                         u_int8_t pf,
 364                         unsigned int hooknum,
 365                         const struct net_device *indev,
 366                         const struct net_device *outdev,
 367                         const struct nf_loginfo *li,
 368                         const char *prefix, unsigned int plen)
 369 {
 370         struct nfulnl_msg_packet_hdr pmsg;
 371         struct nlmsghdr *nlh;
 372         struct nfgenmsg *nfmsg;
 373         __be32 tmp_uint;
 374         sk_buff_data_t old_tail = inst->skb->tail;
 375
 376         nlh = NLMSG_PUT(inst->skb, 0, 0,
 377                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 378                         sizeof(struct nfgenmsg));
 379         nfmsg = NLMSG_DATA(nlh);
 380         nfmsg->nfgen_family = pf;
 381         nfmsg->version = NFNETLINK_V0;
 382         nfmsg->res_id = htons(inst->group_num);
 383
 384         pmsg.hw_protocol        = skb->protocol;
 385         pmsg.hook               = hooknum;
 386
 387         NLA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
 388
 389         if (prefix)
 390                 NLA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
 391
 392         if (indev) {
 393 #ifndef CONFIG_BRIDGE_NETFILTER
 394                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
 395                              htonl(indev->ifindex));
 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_BE32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 402                                      htonl(indev->ifindex));
 403                         /* this is the bridge group "brX" */
 404                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
 405                                      htonl(indev->br_port->br->dev->ifindex));
 406                 } else {
 407                         /* Case 2: indev is bridge group, we need to look for
 408                          * physical device (when called from ipv4) */
 409                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
 410                                      htonl(indev->ifindex));
 411                         if (skb->nf_bridge && skb->nf_bridge->physindev)
 412                                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 413                                              htonl(skb->nf_bridge->physindev->ifindex));
 414                 }
 415 #endif
 416         }
 417
 418         if (outdev) {
 419                 tmp_uint = htonl(outdev->ifindex);
 420 #ifndef CONFIG_BRIDGE_NETFILTER
 421                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
 422                              htonl(outdev->ifindex));
 423 #else
 424                 if (pf == PF_BRIDGE) {
 425                         /* Case 1: outdev is physical output device, we need to
 426                          * look for bridge group (when called from
 427                          * netfilter_bridge) */
 428                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 429                                      htonl(outdev->ifindex));
 430                         /* this is the bridge group "brX" */
 431                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
 432                                      htonl(outdev->br_port->br->dev->ifindex));
 433                 } else {
 434                         /* Case 2: indev is a bridge group, we need to look
 435                          * for physical device (when called from ipv4) */
 436                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
 437                                      htonl(outdev->ifindex));
 438                         if (skb->nf_bridge && skb->nf_bridge->physoutdev)
 439                                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 440                                              htonl(skb->nf_bridge->physoutdev->ifindex));
 441                 }
 442 #endif
 443         }
 444
 445         if (skb->mark)
 446                 NLA_PUT_BE32(inst->skb, NFULA_MARK, htonl(skb->mark));
 447
 448         if (indev && skb->dev) {
 449                 struct nfulnl_msg_packet_hw phw;
 450                 int len = dev_parse_header(skb, phw.hw_addr);
 451                 if (len > 0) {
 452                         phw.hw_addrlen = htons(len);
 453                         NLA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
 454                 }
 455         }
 456
 457         if (indev && skb_mac_header_was_set(skb)) {
 458                 NLA_PUT_BE16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type));
 459                 NLA_PUT_BE16(inst->skb, NFULA_HWLEN,
 460                              htons(skb->dev->hard_header_len));
 461                 NLA_PUT(inst->skb, NFULA_HWHEADER, skb->dev->hard_header_len,
 462                         skb_mac_header(skb));
 463         }
 464
 465         if (skb->tstamp.tv64) {
 466                 struct nfulnl_msg_packet_timestamp ts;
 467                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 468                 ts.sec = cpu_to_be64(tv.tv_sec);
 469                 ts.usec = cpu_to_be64(tv.tv_usec);
 470
 471                 NLA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 472         }
 473
 474         /* UID */
 475         if (skb->sk) {
 476                 read_lock_bh(&skb->sk->sk_callback_lock);
 477                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 478                         struct file *file = skb->sk->sk_socket->file;
 479                         __be32 uid = htonl(file->f_cred->fsuid);
 480                         __be32 gid = htonl(file->f_cred->fsgid);
 481                         /* need to unlock here since NLA_PUT may goto */
 482                         read_unlock_bh(&skb->sk->sk_callback_lock);
 483                         NLA_PUT_BE32(inst->skb, NFULA_UID, uid);
 484                         NLA_PUT_BE32(inst->skb, NFULA_GID, gid);
 485                 } else
 486                         read_unlock_bh(&skb->sk->sk_callback_lock);
 487         }
 488
 489         /* local sequence number */
 490         if (inst->flags & NFULNL_CFG_F_SEQ)
 491                 NLA_PUT_BE32(inst->skb, NFULA_SEQ, htonl(inst->seq++));
 492
 493         /* global sequence number */
 494         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 495                 NLA_PUT_BE32(inst->skb, NFULA_SEQ_GLOBAL,
 496                              htonl(atomic_inc_return(&global_seq)));
 497
 498         if (data_len) {
 499                 struct nlattr *nla;
 500                 int size = nla_attr_size(data_len);
 501
 502                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
 503                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 504                         goto nlmsg_failure;
 505                 }
 506
 507                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 508                 nla->nla_type = NFULA_PAYLOAD;
 509                 nla->nla_len = size;
 510
 511                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 512                         BUG();
 513         }
 514
 515         nlh->nlmsg_len = inst->skb->tail - old_tail;
 516         return 0;
 517
 518 nlmsg_failure:
 519 nla_put_failure:
 520         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 521         return -1;
 522 }
 523
 524 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 525
 526 static struct nf_loginfo default_loginfo = {
 527         .type =         NF_LOG_TYPE_ULOG,
 528         .u = {
 529                 .ulog = {
 530                         .copy_len       = 0xffff,
 531                         .group          = 0,
 532                         .qthreshold     = 1,
 533                 },
 534         },
 535 };
 536
 537 /* log handler for internal netfilter logging api */
 538 void
 539 nfulnl_log_packet(u_int8_t pf,
 540                   unsigned int hooknum,
 541                   const struct sk_buff *skb,
 542                   const struct net_device *in,
 543                   const struct net_device *out,
 544                   const struct nf_loginfo *li_user,
 545                   const char *prefix)
 546 {
 547         unsigned int size, data_len;
 548         struct nfulnl_instance *inst;
 549         const struct nf_loginfo *li;
 550         unsigned int qthreshold;
 551         unsigned int plen;
 552
 553         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 554                 li = li_user;
 555         else
 556                 li = &default_loginfo;
 557
 558         inst = instance_lookup_get(li->u.ulog.group);
 559         if (!inst)
 560                 return;
 561
 562         plen = 0;
 563         if (prefix)
 564                 plen = strlen(prefix) + 1;
 565
 566         /* FIXME: do we want to make the size calculation conditional based on
 567          * what is actually present?  way more branches and checks, but more
 568          * memory efficient... */
 569         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 570                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 571                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 572                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 573 #ifdef CONFIG_BRIDGE_NETFILTER
 574                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 575                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 576 #endif
 577                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 578                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 579                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 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 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
 654
 655 static int
 656 nfulnl_rcv_nl_event(struct notifier_block *this,
 657                    unsigned long event, void *ptr)
 658 {
 659         struct netlink_notify *n = ptr;
 660
 661         if (event == NETLINK_URELEASE &&
 662             n->protocol == NETLINK_NETFILTER && n->pid) {
 663                 int i;
 664
 665                 /* destroy all instances for this pid */
 666                 write_lock_bh(&instances_lock);
 667                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 668                         struct hlist_node *tmp, *t2;
 669                         struct nfulnl_instance *inst;
 670                         struct hlist_head *head = &instance_table[i];
 671
 672                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 673                                 if ((n->net == &init_net) &&
 674                                     (n->pid == inst->peer_pid))
 675                                         __instance_destroy(inst);
 676                         }
 677                 }
 678                 write_unlock_bh(&instances_lock);
 679         }
 680         return NOTIFY_DONE;
 681 }
 682
 683 static struct notifier_block nfulnl_rtnl_notifier = {
 684         .notifier_call  = nfulnl_rcv_nl_event,
 685 };
 686
 687 static int
 688 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 689                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
 690 {
 691         return -ENOTSUPP;
 692 }
 693
 694 static const struct nf_logger nfulnl_logger = {
 695         .name   = "nfnetlink_log",
 696         .logfn  = &nfulnl_log_packet,
 697         .me     = THIS_MODULE,
 698 };
 699
 700 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 701         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 702         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 703         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 704         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 705         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 706         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 707 };
 708
 709 static int
 710 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 711                    struct nlmsghdr *nlh, struct nlattr *nfula[])
 712 {
 713         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 714         u_int16_t group_num = ntohs(nfmsg->res_id);
 715         struct nfulnl_instance *inst;
 716         struct nfulnl_msg_config_cmd *cmd = NULL;
 717         int ret = 0;
 718
 719         if (nfula[NFULA_CFG_CMD]) {
 720                 u_int8_t pf = nfmsg->nfgen_family;
 721                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 722
 723                 /* Commands without queue context */
 724                 switch (cmd->command) {
 725                 case NFULNL_CFG_CMD_PF_BIND:
 726                         return nf_log_register(pf, &nfulnl_logger);
 727                 case NFULNL_CFG_CMD_PF_UNBIND:
 728                         nf_log_unregister_pf(pf);
 729                         return 0;
 730                 }
 731         }
 732
 733         inst = instance_lookup_get(group_num);
 734         if (inst && inst->peer_pid != NETLINK_CB(skb).pid) {
 735                 ret = -EPERM;
 736                 goto out_put;
 737         }
 738
 739         if (cmd != NULL) {
 740                 switch (cmd->command) {
 741                 case NFULNL_CFG_CMD_BIND:
 742                         if (inst) {
 743                                 ret = -EBUSY;
 744                                 goto out_put;
 745                         }
 746
 747                         inst = instance_create(group_num,
 748                                                NETLINK_CB(skb).pid);
 749                         if (IS_ERR(inst)) {
 750                                 ret = PTR_ERR(inst);
 751                                 goto out;
 752                         }
 753                         break;
 754                 case NFULNL_CFG_CMD_UNBIND:
 755                         if (!inst) {
 756                                 ret = -ENODEV;
 757                                 goto out;
 758                         }
 759
 760                         instance_destroy(inst);
 761                         goto out;
 762                 default:
 763                         ret = -ENOTSUPP;
 764                         break;
 765                 }
 766         }
 767
 768         if (nfula[NFULA_CFG_MODE]) {
 769                 struct nfulnl_msg_config_mode *params;
 770                 params = nla_data(nfula[NFULA_CFG_MODE]);
 771
 772                 if (!inst) {
 773                         ret = -ENODEV;
 774                         goto out;
 775                 }
 776                 nfulnl_set_mode(inst, params->copy_mode,
 777                                 ntohl(params->copy_range));
 778         }
 779
 780         if (nfula[NFULA_CFG_TIMEOUT]) {
 781                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 782
 783                 if (!inst) {
 784                         ret = -ENODEV;
 785                         goto out;
 786                 }
 787                 nfulnl_set_timeout(inst, ntohl(timeout));
 788         }
 789
 790         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 791                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 792
 793                 if (!inst) {
 794                         ret = -ENODEV;
 795                         goto out;
 796                 }
 797                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 798         }
 799
 800         if (nfula[NFULA_CFG_QTHRESH]) {
 801                 __be32 qthresh = nla_get_be32(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 = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
 812
 813                 if (!inst) {
 814                         ret = -ENODEV;
 815                         goto out;
 816                 }
 817                 nfulnl_set_flags(inst, ntohs(flags));
 818         }
 819
 820 out_put:
 821         instance_put(inst);
 822 out:
 823         return ret;
 824 }
 825
 826 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 827         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 828                                     .attr_count = NFULA_MAX, },
 829         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 830                                     .attr_count = NFULA_CFG_MAX,
 831                                     .policy = nfula_cfg_policy },
 832 };
 833
 834 static const struct nfnetlink_subsystem nfulnl_subsys = {
 835         .name           = "log",
 836         .subsys_id      = NFNL_SUBSYS_ULOG,
 837         .cb_count       = NFULNL_MSG_MAX,
 838         .cb             = nfulnl_cb,
 839 };
 840
 841 #ifdef CONFIG_PROC_FS
 842 struct iter_state {
 843         unsigned int bucket;
 844 };
 845
 846 static struct hlist_node *get_first(struct iter_state *st)
 847 {
 848         if (!st)
 849                 return NULL;
 850
 851         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 852                 if (!hlist_empty(&instance_table[st->bucket]))
 853                         return instance_table[st->bucket].first;
 854         }
 855         return NULL;
 856 }
 857
 858 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
 859 {
 860         h = h->next;
 861         while (!h) {
 862                 if (++st->bucket >= INSTANCE_BUCKETS)
 863                         return NULL;
 864
 865                 h = instance_table[st->bucket].first;
 866         }
 867         return h;
 868 }
 869
 870 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
 871 {
 872         struct hlist_node *head;
 873         head = get_first(st);
 874
 875         if (head)
 876                 while (pos && (head = get_next(st, head)))
 877                         pos--;
 878         return pos ? NULL : head;
 879 }
 880
 881 static void *seq_start(struct seq_file *seq, loff_t *pos)
 882         __acquires(instances_lock)
 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         __releases(instances_lock)
 896 {
 897         read_unlock_bh(&instances_lock);
 898 }
 899
 900 static int seq_show(struct seq_file *s, void *v)
 901 {
 902         const struct nfulnl_instance *inst = v;
 903
 904         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 905                           inst->group_num,
 906                           inst->peer_pid, inst->qlen,
 907                           inst->copy_mode, inst->copy_range,
 908                           inst->flushtimeout, atomic_read(&inst->use));
 909 }
 910
 911 static const struct seq_operations nful_seq_ops = {
 912         .start  = seq_start,
 913         .next   = seq_next,
 914         .stop   = seq_stop,
 915         .show   = seq_show,
 916 };
 917
 918 static int nful_open(struct inode *inode, struct file *file)
 919 {
 920         return seq_open_private(file, &nful_seq_ops,
 921                         sizeof(struct iter_state));
 922 }
 923
 924 static const struct file_operations nful_file_ops = {
 925         .owner   = THIS_MODULE,
 926         .open    = nful_open,
 927         .read    = seq_read,
 928         .llseek  = seq_lseek,
 929         .release = seq_release_private,
 930 };
 931
 932 #endif /* PROC_FS */
 933
 934 static int __init nfnetlink_log_init(void)
 935 {
 936         int i, status = -ENOMEM;
 937
 938         for (i = 0; i < INSTANCE_BUCKETS; i++)
 939                 INIT_HLIST_HEAD(&instance_table[i]);
 940
 941         /* it's not really all that important to have a random value, so
 942          * we can do this from the init function, even if there hasn't
 943          * been that much entropy yet */
 944         get_random_bytes(&hash_init, sizeof(hash_init));
 945
 946         netlink_register_notifier(&nfulnl_rtnl_notifier);
 947         status = nfnetlink_subsys_register(&nfulnl_subsys);
 948         if (status < 0) {
 949                 printk(KERN_ERR "log: failed to create netlink socket\n");
 950                 goto cleanup_netlink_notifier;
 951         }
 952
 953 #ifdef CONFIG_PROC_FS
 954         if (!proc_create("nfnetlink_log", 0440,
 955                          proc_net_netfilter, &nful_file_ops))
 956                 goto cleanup_subsys;
 957 #endif
 958         return status;
 959
 960 #ifdef CONFIG_PROC_FS
 961 cleanup_subsys:
 962         nfnetlink_subsys_unregister(&nfulnl_subsys);
 963 #endif
 964 cleanup_netlink_notifier:
 965         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
 966         return status;
 967 }
 968
 969 static void __exit nfnetlink_log_fini(void)
 970 {
 971         nf_log_unregister(&nfulnl_logger);
 972 #ifdef CONFIG_PROC_FS
 973         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
 974 #endif
 975         nfnetlink_subsys_unregister(&nfulnl_subsys);
 976         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
 977 }
 978
 979 MODULE_DESCRIPTION("netfilter userspace logging");
 980 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
 981 MODULE_LICENSE("GPL");
 982 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
 983
 984 module_init(nfnetlink_log_init);
 985 module_exit(nfnetlink_log_fini);