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 <linux/slab.h>
  32 #include <net/sock.h>
  33 #include <net/netfilter/nf_log.h>
  34 #include <net/netfilter/nfnetlink_log.h>
  35
  36 #include <linux/atomic.h>
  37
  38 #ifdef CONFIG_BRIDGE_NETFILTER
  39 #include "../bridge/br_private.h"
  40 #endif
  41
  42 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
  43 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
  44 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  45 #define NFULNL_COPY_RANGE_MAX   0xFFFF  /* max packet size is limited by 16-bit struct nfattr nfa_len field */
  46
  47 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  48                                      printk(x, ## args); } while (0);
  49
  50 struct nfulnl_instance {
  51         struct hlist_node hlist;        /* global list of instances */
  52         spinlock_t lock;
  53         atomic_t use;                   /* use count */
  54
  55         unsigned int qlen;              /* number of nlmsgs in skb */
  56         struct sk_buff *skb;            /* pre-allocatd skb */
  57         struct timer_list timer;
  58         int peer_pid;                   /* PID of the peer process */
  59
  60         /* configurable parameters */
  61         unsigned int flushtimeout;      /* timeout until queue flush */
  62         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  63         unsigned int qthreshold;        /* threshold of the queue */
  64         u_int32_t copy_range;
  65         u_int32_t seq;                  /* instance-local sequential counter */
  66         u_int16_t group_num;            /* number of this queue */
  67         u_int16_t flags;
  68         u_int8_t copy_mode;
  69         struct rcu_head rcu;
  70 };
  71
  72 static DEFINE_SPINLOCK(instances_lock);
  73 static atomic_t global_seq;
  74
  75 #define INSTANCE_BUCKETS        16
  76 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  77 static unsigned int hash_init;
  78
  79 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  80 {
  81         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  82 }
  83
  84 static struct nfulnl_instance *
  85 __instance_lookup(u_int16_t group_num)
  86 {
  87         struct hlist_head *head;
  88         struct hlist_node *pos;
  89         struct nfulnl_instance *inst;
  90
  91         head = &instance_table[instance_hashfn(group_num)];
  92         hlist_for_each_entry_rcu(inst, pos, head, hlist) {
  93                 if (inst->group_num == group_num)
  94                         return inst;
  95         }
  96         return NULL;
  97 }
  98
  99 static inline void
 100 instance_get(struct nfulnl_instance *inst)
 101 {
 102         atomic_inc(&inst->use);
 103 }
 104
 105 static struct nfulnl_instance *
 106 instance_lookup_get(u_int16_t group_num)
 107 {
 108         struct nfulnl_instance *inst;
 109
 110         rcu_read_lock_bh();
 111         inst = __instance_lookup(group_num);
 112         if (inst && !atomic_inc_not_zero(&inst->use))
 113                 inst = NULL;
 114         rcu_read_unlock_bh();
 115
 116         return inst;
 117 }
 118
 119 static void nfulnl_instance_free_rcu(struct rcu_head *head)
 120 {
 121         kfree(container_of(head, struct nfulnl_instance, rcu));
 122         module_put(THIS_MODULE);
 123 }
 124
 125 static void
 126 instance_put(struct nfulnl_instance *inst)
 127 {
 128         if (inst && atomic_dec_and_test(&inst->use))
 129                 call_rcu_bh(&inst->rcu, nfulnl_instance_free_rcu);
 130 }
 131
 132 static void nfulnl_timer(unsigned long data);
 133
 134 static struct nfulnl_instance *
 135 instance_create(u_int16_t group_num, int pid)
 136 {
 137         struct nfulnl_instance *inst;
 138         int err;
 139
 140         spin_lock_bh(&instances_lock);
 141         if (__instance_lookup(group_num)) {
 142                 err = -EEXIST;
 143                 goto out_unlock;
 144         }
 145
 146         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 147         if (!inst) {
 148                 err = -ENOMEM;
 149                 goto out_unlock;
 150         }
 151
 152         if (!try_module_get(THIS_MODULE)) {
 153                 kfree(inst);
 154                 err = -EAGAIN;
 155                 goto out_unlock;
 156         }
 157
 158         INIT_HLIST_NODE(&inst->hlist);
 159         spin_lock_init(&inst->lock);
 160         /* needs to be two, since we _put() after creation */
 161         atomic_set(&inst->use, 2);
 162
 163         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
 164
 165         inst->peer_pid = pid;
 166         inst->group_num = group_num;
 167
 168         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 169         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 170         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 171         inst->copy_mode         = NFULNL_COPY_PACKET;
 172         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
 173
 174         hlist_add_head_rcu(&inst->hlist,
 175                        &instance_table[instance_hashfn(group_num)]);
 176
 177         spin_unlock_bh(&instances_lock);
 178
 179         return inst;
 180
 181 out_unlock:
 182         spin_unlock_bh(&instances_lock);
 183         return ERR_PTR(err);
 184 }
 185
 186 static void __nfulnl_flush(struct nfulnl_instance *inst);
 187
 188 /* called with BH disabled */
 189 static void
 190 __instance_destroy(struct nfulnl_instance *inst)
 191 {
 192         /* first pull it out of the global list */
 193         hlist_del_rcu(&inst->hlist);
 194
 195         /* then flush all pending packets from skb */
 196
 197         spin_lock(&inst->lock);
 198
 199         /* lockless readers wont be able to use us */
 200         inst->copy_mode = NFULNL_COPY_DISABLED;
 201
 202         if (inst->skb)
 203                 __nfulnl_flush(inst);
 204         spin_unlock(&inst->lock);
 205
 206         /* and finally put the refcount */
 207         instance_put(inst);
 208 }
 209
 210 static inline void
 211 instance_destroy(struct nfulnl_instance *inst)
 212 {
 213         spin_lock_bh(&instances_lock);
 214         __instance_destroy(inst);
 215         spin_unlock_bh(&instances_lock);
 216 }
 217
 218 static int
 219 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 220                   unsigned int range)
 221 {
 222         int status = 0;
 223
 224         spin_lock_bh(&inst->lock);
 225
 226         switch (mode) {
 227         case NFULNL_COPY_NONE:
 228         case NFULNL_COPY_META:
 229                 inst->copy_mode = mode;
 230                 inst->copy_range = 0;
 231                 break;
 232
 233         case NFULNL_COPY_PACKET:
 234                 inst->copy_mode = mode;
 235                 inst->copy_range = min_t(unsigned int,
 236                                          range, NFULNL_COPY_RANGE_MAX);
 237                 break;
 238
 239         default:
 240                 status = -EINVAL;
 241                 break;
 242         }
 243
 244         spin_unlock_bh(&inst->lock);
 245
 246         return status;
 247 }
 248
 249 static int
 250 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 251 {
 252         int status;
 253
 254         spin_lock_bh(&inst->lock);
 255         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 256                 status = -ERANGE;
 257         else if (nlbufsiz > 131072)
 258                 status = -ERANGE;
 259         else {
 260                 inst->nlbufsiz = nlbufsiz;
 261                 status = 0;
 262         }
 263         spin_unlock_bh(&inst->lock);
 264
 265         return status;
 266 }
 267
 268 static int
 269 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 270 {
 271         spin_lock_bh(&inst->lock);
 272         inst->flushtimeout = timeout;
 273         spin_unlock_bh(&inst->lock);
 274
 275         return 0;
 276 }
 277
 278 static int
 279 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 280 {
 281         spin_lock_bh(&inst->lock);
 282         inst->qthreshold = qthresh;
 283         spin_unlock_bh(&inst->lock);
 284
 285         return 0;
 286 }
 287
 288 static int
 289 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 290 {
 291         spin_lock_bh(&inst->lock);
 292         inst->flags = flags;
 293         spin_unlock_bh(&inst->lock);
 294
 295         return 0;
 296 }
 297
 298 static struct sk_buff *
 299 nfulnl_alloc_skb(unsigned int inst_size, unsigned int pkt_size)
 300 {
 301         struct sk_buff *skb;
 302         unsigned int n;
 303
 304         /* alloc skb which should be big enough for a whole multipart
 305          * message.  WARNING: has to be <= 128k due to slab restrictions */
 306
 307         n = max(inst_size, pkt_size);
 308         skb = alloc_skb(n, GFP_ATOMIC);
 309         if (!skb) {
 310                 if (n > pkt_size) {
 311                         /* try to allocate only as much as we need for current
 312                          * packet */
 313
 314                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 315                         if (!skb)
 316                                 pr_err("nfnetlink_log: can't even alloc %u bytes\n",
 317                                        pkt_size);
 318                 }
 319         }
 320
 321         return skb;
 322 }
 323
 324 static int
 325 __nfulnl_send(struct nfulnl_instance *inst)
 326 {
 327         int status = -1;
 328
 329         if (inst->qlen > 1) {
 330                 struct nlmsghdr *nlh = nlmsg_put(inst->skb, 0, 0,
 331                                                  NLMSG_DONE,
 332                                                  sizeof(struct nfgenmsg),
 333                                                  0);
 334                 if (!nlh)
 335                         goto out;
 336         }
 337         status = nfnetlink_unicast(inst->skb, &init_net, inst->peer_pid,
 338                                    MSG_DONTWAIT);
 339
 340         inst->qlen = 0;
 341         inst->skb = NULL;
 342 out:
 343         return status;
 344 }
 345
 346 static void
 347 __nfulnl_flush(struct nfulnl_instance *inst)
 348 {
 349         /* timer holds a reference */
 350         if (del_timer(&inst->timer))
 351                 instance_put(inst);
 352         if (inst->skb)
 353                 __nfulnl_send(inst);
 354 }
 355
 356 static void
 357 nfulnl_timer(unsigned long data)
 358 {
 359         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 360
 361         spin_lock_bh(&inst->lock);
 362         if (inst->skb)
 363                 __nfulnl_send(inst);
 364         spin_unlock_bh(&inst->lock);
 365         instance_put(inst);
 366 }
 367
 368 /* This is an inline function, we don't really care about a long
 369  * list of arguments */
 370 static inline int
 371 __build_packet_message(struct nfulnl_instance *inst,
 372                         const struct sk_buff *skb,
 373                         unsigned int data_len,
 374                         u_int8_t pf,
 375                         unsigned int hooknum,
 376                         const struct net_device *indev,
 377                         const struct net_device *outdev,
 378                         const char *prefix, unsigned int plen)
 379 {
 380         struct nfulnl_msg_packet_hdr pmsg;
 381         struct nlmsghdr *nlh;
 382         struct nfgenmsg *nfmsg;
 383         sk_buff_data_t old_tail = inst->skb->tail;
 384         struct sock *sk;
 385
 386         nlh = nlmsg_put(inst->skb, 0, 0,
 387                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 388                         sizeof(struct nfgenmsg), 0);
 389         if (!nlh)
 390                 return -1;
 391         nfmsg = nlmsg_data(nlh);
 392         nfmsg->nfgen_family = pf;
 393         nfmsg->version = NFNETLINK_V0;
 394         nfmsg->res_id = htons(inst->group_num);
 395
 396         pmsg.hw_protocol        = skb->protocol;
 397         pmsg.hook               = hooknum;
 398
 399         if (nla_put(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg))
 400                 goto nla_put_failure;
 401
 402         if (prefix &&
 403             nla_put(inst->skb, NFULA_PREFIX, plen, prefix))
 404                 goto nla_put_failure;
 405
 406         if (indev) {
 407 #ifndef CONFIG_BRIDGE_NETFILTER
 408                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 409                                  htonl(indev->ifindex)))
 410                         goto nla_put_failure;
 411 #else
 412                 if (pf == PF_BRIDGE) {
 413                         /* Case 1: outdev is physical input device, we need to
 414                          * look for bridge group (when called from
 415                          * netfilter_bridge) */
 416                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 417                                          htonl(indev->ifindex)) ||
 418                         /* this is the bridge group "brX" */
 419                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 420                             nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 421                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 422                                 goto nla_put_failure;
 423                 } else {
 424                         /* Case 2: indev is bridge group, we need to look for
 425                          * physical device (when called from ipv4) */
 426                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 427                                          htonl(indev->ifindex)))
 428                                 goto nla_put_failure;
 429                         if (skb->nf_bridge && skb->nf_bridge->physindev &&
 430                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 431                                          htonl(skb->nf_bridge->physindev->ifindex)))
 432                                 goto nla_put_failure;
 433                 }
 434 #endif
 435         }
 436
 437         if (outdev) {
 438 #ifndef CONFIG_BRIDGE_NETFILTER
 439                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 440                                  htonl(outdev->ifindex)))
 441                         goto nla_put_failure;
 442 #else
 443                 if (pf == PF_BRIDGE) {
 444                         /* Case 1: outdev is physical output device, we need to
 445                          * look for bridge group (when called from
 446                          * netfilter_bridge) */
 447                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 448                                          htonl(outdev->ifindex)) ||
 449                         /* this is the bridge group "brX" */
 450                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 451                             nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 452                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 453                                 goto nla_put_failure;
 454                 } else {
 455                         /* Case 2: indev is a bridge group, we need to look
 456                          * for physical device (when called from ipv4) */
 457                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 458                                          htonl(outdev->ifindex)))
 459                                 goto nla_put_failure;
 460                         if (skb->nf_bridge && skb->nf_bridge->physoutdev &&
 461                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 462                                          htonl(skb->nf_bridge->physoutdev->ifindex)))
 463                                 goto nla_put_failure;
 464                 }
 465 #endif
 466         }
 467
 468         if (skb->mark &&
 469             nla_put_be32(inst->skb, NFULA_MARK, htonl(skb->mark)))
 470                 goto nla_put_failure;
 471
 472         if (indev && skb->dev &&
 473             skb->mac_header != skb->network_header) {
 474                 struct nfulnl_msg_packet_hw phw;
 475                 int len = dev_parse_header(skb, phw.hw_addr);
 476                 if (len > 0) {
 477                         phw.hw_addrlen = htons(len);
 478                         if (nla_put(inst->skb, NFULA_HWADDR, sizeof(phw), &phw))
 479                                 goto nla_put_failure;
 480                 }
 481         }
 482
 483         if (indev && skb_mac_header_was_set(skb)) {
 484                 if (nla_put_be16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type)) ||
 485                     nla_put_be16(inst->skb, NFULA_HWLEN,
 486                                  htons(skb->dev->hard_header_len)) ||
 487                     nla_put(inst->skb, NFULA_HWHEADER, skb->dev->hard_header_len,
 488                             skb_mac_header(skb)))
 489                         goto nla_put_failure;
 490         }
 491
 492         if (skb->tstamp.tv64) {
 493                 struct nfulnl_msg_packet_timestamp ts;
 494                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 495                 ts.sec = cpu_to_be64(tv.tv_sec);
 496                 ts.usec = cpu_to_be64(tv.tv_usec);
 497
 498                 if (nla_put(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts))
 499                         goto nla_put_failure;
 500         }
 501
 502         /* UID */
 503         sk = skb->sk;
 504         if (sk && sk->sk_state != TCP_TIME_WAIT) {
 505                 read_lock_bh(&sk->sk_callback_lock);
 506                 if (sk->sk_socket && sk->sk_socket->file) {
 507                         struct file *file = sk->sk_socket->file;
 508                         __be32 uid = htonl(file->f_cred->fsuid);
 509                         __be32 gid = htonl(file->f_cred->fsgid);
 510                         read_unlock_bh(&sk->sk_callback_lock);
 511                         if (nla_put_be32(inst->skb, NFULA_UID, uid) ||
 512                             nla_put_be32(inst->skb, NFULA_GID, gid))
 513                                 goto nla_put_failure;
 514                 } else
 515                         read_unlock_bh(&sk->sk_callback_lock);
 516         }
 517
 518         /* local sequence number */
 519         if ((inst->flags & NFULNL_CFG_F_SEQ) &&
 520             nla_put_be32(inst->skb, NFULA_SEQ, htonl(inst->seq++)))
 521                 goto nla_put_failure;
 522
 523         /* global sequence number */
 524         if ((inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) &&
 525             nla_put_be32(inst->skb, NFULA_SEQ_GLOBAL,
 526                          htonl(atomic_inc_return(&global_seq))))
 527                 goto nla_put_failure;
 528
 529         if (data_len) {
 530                 struct nlattr *nla;
 531                 int size = nla_attr_size(data_len);
 532
 533                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
 534                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 535                         return -1;
 536                 }
 537
 538                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 539                 nla->nla_type = NFULA_PAYLOAD;
 540                 nla->nla_len = size;
 541
 542                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 543                         BUG();
 544         }
 545
 546         nlh->nlmsg_len = inst->skb->tail - old_tail;
 547         return 0;
 548
 549 nla_put_failure:
 550         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 551         return -1;
 552 }
 553
 554 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 555
 556 static struct nf_loginfo default_loginfo = {
 557         .type =         NF_LOG_TYPE_ULOG,
 558         .u = {
 559                 .ulog = {
 560                         .copy_len       = 0xffff,
 561                         .group          = 0,
 562                         .qthreshold     = 1,
 563                 },
 564         },
 565 };
 566
 567 /* log handler for internal netfilter logging api */
 568 void
 569 nfulnl_log_packet(u_int8_t pf,
 570                   unsigned int hooknum,
 571                   const struct sk_buff *skb,
 572                   const struct net_device *in,
 573                   const struct net_device *out,
 574                   const struct nf_loginfo *li_user,
 575                   const char *prefix)
 576 {
 577         unsigned int size, data_len;
 578         struct nfulnl_instance *inst;
 579         const struct nf_loginfo *li;
 580         unsigned int qthreshold;
 581         unsigned int plen;
 582
 583         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 584                 li = li_user;
 585         else
 586                 li = &default_loginfo;
 587
 588         inst = instance_lookup_get(li->u.ulog.group);
 589         if (!inst)
 590                 return;
 591
 592         plen = 0;
 593         if (prefix)
 594                 plen = strlen(prefix) + 1;
 595
 596         /* FIXME: do we want to make the size calculation conditional based on
 597          * what is actually present?  way more branches and checks, but more
 598          * memory efficient... */
 599         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 600                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 601                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 602                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 603 #ifdef CONFIG_BRIDGE_NETFILTER
 604                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 605                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 606 #endif
 607                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 608                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 609                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 610                 + nla_total_size(plen)                  /* prefix */
 611                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 612                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
 613
 614         if (in && skb_mac_header_was_set(skb)) {
 615                 size +=   nla_total_size(skb->dev->hard_header_len)
 616                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 617                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 618         }
 619
 620         spin_lock_bh(&inst->lock);
 621
 622         if (inst->flags & NFULNL_CFG_F_SEQ)
 623                 size += nla_total_size(sizeof(u_int32_t));
 624         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 625                 size += nla_total_size(sizeof(u_int32_t));
 626
 627         qthreshold = inst->qthreshold;
 628         /* per-rule qthreshold overrides per-instance */
 629         if (li->u.ulog.qthreshold)
 630                 if (qthreshold > li->u.ulog.qthreshold)
 631                         qthreshold = li->u.ulog.qthreshold;
 632
 633
 634         switch (inst->copy_mode) {
 635         case NFULNL_COPY_META:
 636         case NFULNL_COPY_NONE:
 637                 data_len = 0;
 638                 break;
 639
 640         case NFULNL_COPY_PACKET:
 641                 if (inst->copy_range == 0
 642                     || inst->copy_range > skb->len)
 643                         data_len = skb->len;
 644                 else
 645                         data_len = inst->copy_range;
 646
 647                 size += nla_total_size(data_len);
 648                 break;
 649
 650         case NFULNL_COPY_DISABLED:
 651         default:
 652                 goto unlock_and_release;
 653         }
 654
 655         if (inst->skb &&
 656             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
 657                 /* either the queue len is too high or we don't have
 658                  * enough room in the skb left. flush to userspace. */
 659                 __nfulnl_flush(inst);
 660         }
 661
 662         if (!inst->skb) {
 663                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
 664                 if (!inst->skb)
 665                         goto alloc_failure;
 666         }
 667
 668         inst->qlen++;
 669
 670         __build_packet_message(inst, skb, data_len, pf,
 671                                 hooknum, in, out, prefix, plen);
 672
 673         if (inst->qlen >= qthreshold)
 674                 __nfulnl_flush(inst);
 675         /* timer_pending always called within inst->lock, so there
 676          * is no chance of a race here */
 677         else if (!timer_pending(&inst->timer)) {
 678                 instance_get(inst);
 679                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 680                 add_timer(&inst->timer);
 681         }
 682
 683 unlock_and_release:
 684         spin_unlock_bh(&inst->lock);
 685         instance_put(inst);
 686         return;
 687
 688 alloc_failure:
 689         /* FIXME: statistics */
 690         goto unlock_and_release;
 691 }
 692 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
 693
 694 static int
 695 nfulnl_rcv_nl_event(struct notifier_block *this,
 696                    unsigned long event, void *ptr)
 697 {
 698         struct netlink_notify *n = ptr;
 699
 700         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 701                 int i;
 702
 703                 /* destroy all instances for this pid */
 704                 spin_lock_bh(&instances_lock);
 705                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 706                         struct hlist_node *tmp, *t2;
 707                         struct nfulnl_instance *inst;
 708                         struct hlist_head *head = &instance_table[i];
 709
 710                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 711                                 if ((net_eq(n->net, &init_net)) &&
 712                                     (n->pid == inst->peer_pid))
 713                                         __instance_destroy(inst);
 714                         }
 715                 }
 716                 spin_unlock_bh(&instances_lock);
 717         }
 718         return NOTIFY_DONE;
 719 }
 720
 721 static struct notifier_block nfulnl_rtnl_notifier = {
 722         .notifier_call  = nfulnl_rcv_nl_event,
 723 };
 724
 725 static int
 726 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 727                    const struct nlmsghdr *nlh,
 728                    const struct nlattr * const nfqa[])
 729 {
 730         return -ENOTSUPP;
 731 }
 732
 733 static struct nf_logger nfulnl_logger __read_mostly = {
 734         .name   = "nfnetlink_log",
 735         .logfn  = &nfulnl_log_packet,
 736         .me     = THIS_MODULE,
 737 };
 738
 739 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 740         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 741         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 742         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 743         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 744         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 745         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 746 };
 747
 748 static int
 749 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 750                    const struct nlmsghdr *nlh,
 751                    const struct nlattr * const nfula[])
 752 {
 753         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 754         u_int16_t group_num = ntohs(nfmsg->res_id);
 755         struct nfulnl_instance *inst;
 756         struct nfulnl_msg_config_cmd *cmd = NULL;
 757         int ret = 0;
 758
 759         if (nfula[NFULA_CFG_CMD]) {
 760                 u_int8_t pf = nfmsg->nfgen_family;
 761                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 762
 763                 /* Commands without queue context */
 764                 switch (cmd->command) {
 765                 case NFULNL_CFG_CMD_PF_BIND:
 766                         return nf_log_bind_pf(pf, &nfulnl_logger);
 767                 case NFULNL_CFG_CMD_PF_UNBIND:
 768                         nf_log_unbind_pf(pf);
 769                         return 0;
 770                 }
 771         }
 772
 773         inst = instance_lookup_get(group_num);
 774         if (inst && inst->peer_pid != NETLINK_CB(skb).pid) {
 775                 ret = -EPERM;
 776                 goto out_put;
 777         }
 778
 779         if (cmd != NULL) {
 780                 switch (cmd->command) {
 781                 case NFULNL_CFG_CMD_BIND:
 782                         if (inst) {
 783                                 ret = -EBUSY;
 784                                 goto out_put;
 785                         }
 786
 787                         inst = instance_create(group_num,
 788                                                NETLINK_CB(skb).pid);
 789                         if (IS_ERR(inst)) {
 790                                 ret = PTR_ERR(inst);
 791                                 goto out;
 792                         }
 793                         break;
 794                 case NFULNL_CFG_CMD_UNBIND:
 795                         if (!inst) {
 796                                 ret = -ENODEV;
 797                                 goto out;
 798                         }
 799
 800                         instance_destroy(inst);
 801                         goto out_put;
 802                 default:
 803                         ret = -ENOTSUPP;
 804                         break;
 805                 }
 806         }
 807
 808         if (nfula[NFULA_CFG_MODE]) {
 809                 struct nfulnl_msg_config_mode *params;
 810                 params = nla_data(nfula[NFULA_CFG_MODE]);
 811
 812                 if (!inst) {
 813                         ret = -ENODEV;
 814                         goto out;
 815                 }
 816                 nfulnl_set_mode(inst, params->copy_mode,
 817                                 ntohl(params->copy_range));
 818         }
 819
 820         if (nfula[NFULA_CFG_TIMEOUT]) {
 821                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 822
 823                 if (!inst) {
 824                         ret = -ENODEV;
 825                         goto out;
 826                 }
 827                 nfulnl_set_timeout(inst, ntohl(timeout));
 828         }
 829
 830         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 831                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 832
 833                 if (!inst) {
 834                         ret = -ENODEV;
 835                         goto out;
 836                 }
 837                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 838         }
 839
 840         if (nfula[NFULA_CFG_QTHRESH]) {
 841                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 842
 843                 if (!inst) {
 844                         ret = -ENODEV;
 845                         goto out;
 846                 }
 847                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 848         }
 849
 850         if (nfula[NFULA_CFG_FLAGS]) {
 851                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
 852
 853                 if (!inst) {
 854                         ret = -ENODEV;
 855                         goto out;
 856                 }
 857                 nfulnl_set_flags(inst, ntohs(flags));
 858         }
 859
 860 out_put:
 861         instance_put(inst);
 862 out:
 863         return ret;
 864 }
 865
 866 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 867         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 868                                     .attr_count = NFULA_MAX, },
 869         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 870                                     .attr_count = NFULA_CFG_MAX,
 871                                     .policy = nfula_cfg_policy },
 872 };
 873
 874 static const struct nfnetlink_subsystem nfulnl_subsys = {
 875         .name           = "log",
 876         .subsys_id      = NFNL_SUBSYS_ULOG,
 877         .cb_count       = NFULNL_MSG_MAX,
 878         .cb             = nfulnl_cb,
 879 };
 880
 881 #ifdef CONFIG_PROC_FS
 882 struct iter_state {
 883         unsigned int bucket;
 884 };
 885
 886 static struct hlist_node *get_first(struct iter_state *st)
 887 {
 888         if (!st)
 889                 return NULL;
 890
 891         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 892                 if (!hlist_empty(&instance_table[st->bucket]))
 893                         return rcu_dereference_bh(hlist_first_rcu(&instance_table[st->bucket]));
 894         }
 895         return NULL;
 896 }
 897
 898 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
 899 {
 900         h = rcu_dereference_bh(hlist_next_rcu(h));
 901         while (!h) {
 902                 if (++st->bucket >= INSTANCE_BUCKETS)
 903                         return NULL;
 904
 905                 h = rcu_dereference_bh(hlist_first_rcu(&instance_table[st->bucket]));
 906         }
 907         return h;
 908 }
 909
 910 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
 911 {
 912         struct hlist_node *head;
 913         head = get_first(st);
 914
 915         if (head)
 916                 while (pos && (head = get_next(st, head)))
 917                         pos--;
 918         return pos ? NULL : head;
 919 }
 920
 921 static void *seq_start(struct seq_file *seq, loff_t *pos)
 922         __acquires(rcu_bh)
 923 {
 924         rcu_read_lock_bh();
 925         return get_idx(seq->private, *pos);
 926 }
 927
 928 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 929 {
 930         (*pos)++;
 931         return get_next(s->private, v);
 932 }
 933
 934 static void seq_stop(struct seq_file *s, void *v)
 935         __releases(rcu_bh)
 936 {
 937         rcu_read_unlock_bh();
 938 }
 939
 940 static int seq_show(struct seq_file *s, void *v)
 941 {
 942         const struct nfulnl_instance *inst = v;
 943
 944         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 945                           inst->group_num,
 946                           inst->peer_pid, inst->qlen,
 947                           inst->copy_mode, inst->copy_range,
 948                           inst->flushtimeout, atomic_read(&inst->use));
 949 }
 950
 951 static const struct seq_operations nful_seq_ops = {
 952         .start  = seq_start,
 953         .next   = seq_next,
 954         .stop   = seq_stop,
 955         .show   = seq_show,
 956 };
 957
 958 static int nful_open(struct inode *inode, struct file *file)
 959 {
 960         return seq_open_private(file, &nful_seq_ops,
 961                         sizeof(struct iter_state));
 962 }
 963
 964 static const struct file_operations nful_file_ops = {
 965         .owner   = THIS_MODULE,
 966         .open    = nful_open,
 967         .read    = seq_read,
 968         .llseek  = seq_lseek,
 969         .release = seq_release_private,
 970 };
 971
 972 #endif /* PROC_FS */
 973
 974 static int __init nfnetlink_log_init(void)
 975 {
 976         int i, status = -ENOMEM;
 977
 978         for (i = 0; i < INSTANCE_BUCKETS; i++)
 979                 INIT_HLIST_HEAD(&instance_table[i]);
 980
 981         /* it's not really all that important to have a random value, so
 982          * we can do this from the init function, even if there hasn't
 983          * been that much entropy yet */
 984         get_random_bytes(&hash_init, sizeof(hash_init));
 985
 986         netlink_register_notifier(&nfulnl_rtnl_notifier);
 987         status = nfnetlink_subsys_register(&nfulnl_subsys);
 988         if (status < 0) {
 989                 printk(KERN_ERR "log: failed to create netlink socket\n");
 990                 goto cleanup_netlink_notifier;
 991         }
 992
 993         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
 994         if (status < 0) {
 995                 printk(KERN_ERR "log: failed to register logger\n");
 996                 goto cleanup_subsys;
 997         }
 998
 999 #ifdef CONFIG_PROC_FS
1000         if (!proc_create("nfnetlink_log", 0440,
1001                          proc_net_netfilter, &nful_file_ops)) {
1002                 status = -ENOMEM;
1003                 goto cleanup_logger;
1004         }
1005 #endif
1006         return status;
1007
1008 #ifdef CONFIG_PROC_FS
1009 cleanup_logger:
1010         nf_log_unregister(&nfulnl_logger);
1011 #endif
1012 cleanup_subsys:
1013         nfnetlink_subsys_unregister(&nfulnl_subsys);
1014 cleanup_netlink_notifier:
1015         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1016         return status;
1017 }
1018
1019 static void __exit nfnetlink_log_fini(void)
1020 {
1021         nf_log_unregister(&nfulnl_logger);
1022 #ifdef CONFIG_PROC_FS
1023         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1024 #endif
1025         nfnetlink_subsys_unregister(&nfulnl_subsys);
1026         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1027 }
1028
1029 MODULE_DESCRIPTION("netfilter userspace logging");
1030 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1031 MODULE_LICENSE("GPL");
1032 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1033
1034 module_init(nfnetlink_log_init);
1035 module_exit(nfnetlink_log_fini);