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
 385         nlh = nlmsg_put(inst->skb, 0, 0,
 386                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 387                         sizeof(struct nfgenmsg), 0);
 388         if (!nlh)
 389                 return -1;
 390         nfmsg = nlmsg_data(nlh);
 391         nfmsg->nfgen_family = pf;
 392         nfmsg->version = NFNETLINK_V0;
 393         nfmsg->res_id = htons(inst->group_num);
 394
 395         pmsg.hw_protocol        = skb->protocol;
 396         pmsg.hook               = hooknum;
 397
 398         if (nla_put(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg))
 399                 goto nla_put_failure;
 400
 401         if (prefix &&
 402             nla_put(inst->skb, NFULA_PREFIX, plen, prefix))
 403                 goto nla_put_failure;
 404
 405         if (indev) {
 406 #ifndef CONFIG_BRIDGE_NETFILTER
 407                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 408                                  htonl(indev->ifindex)))
 409                         goto nla_put_failure;
 410 #else
 411                 if (pf == PF_BRIDGE) {
 412                         /* Case 1: outdev is physical input device, we need to
 413                          * look for bridge group (when called from
 414                          * netfilter_bridge) */
 415                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 416                                          htonl(indev->ifindex)) ||
 417                         /* this is the bridge group "brX" */
 418                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 419                             nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 420                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 421                                 goto nla_put_failure;
 422                 } else {
 423                         /* Case 2: indev is bridge group, we need to look for
 424                          * physical device (when called from ipv4) */
 425                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 426                                          htonl(indev->ifindex)))
 427                                 goto nla_put_failure;
 428                         if (skb->nf_bridge && skb->nf_bridge->physindev &&
 429                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 430                                          htonl(skb->nf_bridge->physindev->ifindex)))
 431                                 goto nla_put_failure;
 432                 }
 433 #endif
 434         }
 435
 436         if (outdev) {
 437 #ifndef CONFIG_BRIDGE_NETFILTER
 438                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 439                                  htonl(outdev->ifindex)))
 440                         goto nla_put_failure;
 441 #else
 442                 if (pf == PF_BRIDGE) {
 443                         /* Case 1: outdev is physical output device, we need to
 444                          * look for bridge group (when called from
 445                          * netfilter_bridge) */
 446                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 447                                          htonl(outdev->ifindex)) ||
 448                         /* this is the bridge group "brX" */
 449                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 450                             nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 451                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 452                                 goto nla_put_failure;
 453                 } else {
 454                         /* Case 2: indev is a bridge group, we need to look
 455                          * for physical device (when called from ipv4) */
 456                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 457                                          htonl(outdev->ifindex)))
 458                                 goto nla_put_failure;
 459                         if (skb->nf_bridge && skb->nf_bridge->physoutdev &&
 460                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 461                                          htonl(skb->nf_bridge->physoutdev->ifindex)))
 462                                 goto nla_put_failure;
 463                 }
 464 #endif
 465         }
 466
 467         if (skb->mark &&
 468             nla_put_be32(inst->skb, NFULA_MARK, htonl(skb->mark)))
 469                 goto nla_put_failure;
 470
 471         if (indev && skb->dev &&
 472             skb->mac_header != skb->network_header) {
 473                 struct nfulnl_msg_packet_hw phw;
 474                 int len = dev_parse_header(skb, phw.hw_addr);
 475                 if (len > 0) {
 476                         phw.hw_addrlen = htons(len);
 477                         if (nla_put(inst->skb, NFULA_HWADDR, sizeof(phw), &phw))
 478                                 goto nla_put_failure;
 479                 }
 480         }
 481
 482         if (indev && skb_mac_header_was_set(skb)) {
 483                 if (nla_put_be32(inst->skb, NFULA_HWTYPE, htons(skb->dev->type)) ||
 484                     nla_put_be16(inst->skb, NFULA_HWLEN,
 485                                  htons(skb->dev->hard_header_len)) ||
 486                     nla_put(inst->skb, NFULA_HWHEADER, skb->dev->hard_header_len,
 487                             skb_mac_header(skb)))
 488                         goto nla_put_failure;
 489         }
 490
 491         if (skb->tstamp.tv64) {
 492                 struct nfulnl_msg_packet_timestamp ts;
 493                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 494                 ts.sec = cpu_to_be64(tv.tv_sec);
 495                 ts.usec = cpu_to_be64(tv.tv_usec);
 496
 497                 if (nla_put(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts))
 498                         goto nla_put_failure;
 499         }
 500
 501         /* UID */
 502         if (skb->sk) {
 503                 read_lock_bh(&skb->sk->sk_callback_lock);
 504                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 505                         struct file *file = skb->sk->sk_socket->file;
 506                         __be32 uid = htonl(file->f_cred->fsuid);
 507                         __be32 gid = htonl(file->f_cred->fsgid);
 508                         read_unlock_bh(&skb->sk->sk_callback_lock);
 509                         if (nla_put_be32(inst->skb, NFULA_UID, uid) ||
 510                             nla_put_be32(inst->skb, NFULA_GID, gid))
 511                                 goto nla_put_failure;
 512                 } else
 513                         read_unlock_bh(&skb->sk->sk_callback_lock);
 514         }
 515
 516         /* local sequence number */
 517         if ((inst->flags & NFULNL_CFG_F_SEQ) &&
 518             nla_put_be32(inst->skb, NFULA_SEQ, htonl(inst->seq++)))
 519                 goto nla_put_failure;
 520
 521         /* global sequence number */
 522         if ((inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) &&
 523             nla_put_be32(inst->skb, NFULA_SEQ_GLOBAL,
 524                          htonl(atomic_inc_return(&global_seq))))
 525                 goto nla_put_failure;
 526
 527         if (data_len) {
 528                 struct nlattr *nla;
 529                 int size = nla_attr_size(data_len);
 530
 531                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
 532                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 533                         return -1;
 534                 }
 535
 536                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 537                 nla->nla_type = NFULA_PAYLOAD;
 538                 nla->nla_len = size;
 539
 540                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 541                         BUG();
 542         }
 543
 544         nlh->nlmsg_len = inst->skb->tail - old_tail;
 545         return 0;
 546
 547 nla_put_failure:
 548         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 549         return -1;
 550 }
 551
 552 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 553
 554 static struct nf_loginfo default_loginfo = {
 555         .type =         NF_LOG_TYPE_ULOG,
 556         .u = {
 557                 .ulog = {
 558                         .copy_len       = 0xffff,
 559                         .group          = 0,
 560                         .qthreshold     = 1,
 561                 },
 562         },
 563 };
 564
 565 /* log handler for internal netfilter logging api */
 566 void
 567 nfulnl_log_packet(u_int8_t pf,
 568                   unsigned int hooknum,
 569                   const struct sk_buff *skb,
 570                   const struct net_device *in,
 571                   const struct net_device *out,
 572                   const struct nf_loginfo *li_user,
 573                   const char *prefix)
 574 {
 575         unsigned int size, data_len;
 576         struct nfulnl_instance *inst;
 577         const struct nf_loginfo *li;
 578         unsigned int qthreshold;
 579         unsigned int plen;
 580
 581         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 582                 li = li_user;
 583         else
 584                 li = &default_loginfo;
 585
 586         inst = instance_lookup_get(li->u.ulog.group);
 587         if (!inst)
 588                 return;
 589
 590         plen = 0;
 591         if (prefix)
 592                 plen = strlen(prefix) + 1;
 593
 594         /* FIXME: do we want to make the size calculation conditional based on
 595          * what is actually present?  way more branches and checks, but more
 596          * memory efficient... */
 597         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 598                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 599                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 600                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 601 #ifdef CONFIG_BRIDGE_NETFILTER
 602                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 603                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 604 #endif
 605                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 606                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 607                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 608                 + nla_total_size(plen)                  /* prefix */
 609                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 610                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
 611
 612         if (in && skb_mac_header_was_set(skb)) {
 613                 size +=   nla_total_size(skb->dev->hard_header_len)
 614                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 615                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 616         }
 617
 618         spin_lock_bh(&inst->lock);
 619
 620         if (inst->flags & NFULNL_CFG_F_SEQ)
 621                 size += nla_total_size(sizeof(u_int32_t));
 622         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 623                 size += nla_total_size(sizeof(u_int32_t));
 624
 625         qthreshold = inst->qthreshold;
 626         /* per-rule qthreshold overrides per-instance */
 627         if (li->u.ulog.qthreshold)
 628                 if (qthreshold > li->u.ulog.qthreshold)
 629                         qthreshold = li->u.ulog.qthreshold;
 630
 631
 632         switch (inst->copy_mode) {
 633         case NFULNL_COPY_META:
 634         case NFULNL_COPY_NONE:
 635                 data_len = 0;
 636                 break;
 637
 638         case NFULNL_COPY_PACKET:
 639                 if (inst->copy_range == 0
 640                     || inst->copy_range > skb->len)
 641                         data_len = skb->len;
 642                 else
 643                         data_len = inst->copy_range;
 644
 645                 size += nla_total_size(data_len);
 646                 break;
 647
 648         case NFULNL_COPY_DISABLED:
 649         default:
 650                 goto unlock_and_release;
 651         }
 652
 653         if (inst->skb &&
 654             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
 655                 /* either the queue len is too high or we don't have
 656                  * enough room in the skb left. flush to userspace. */
 657                 __nfulnl_flush(inst);
 658         }
 659
 660         if (!inst->skb) {
 661                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
 662                 if (!inst->skb)
 663                         goto alloc_failure;
 664         }
 665
 666         inst->qlen++;
 667
 668         __build_packet_message(inst, skb, data_len, pf,
 669                                 hooknum, in, out, prefix, plen);
 670
 671         if (inst->qlen >= qthreshold)
 672                 __nfulnl_flush(inst);
 673         /* timer_pending always called within inst->lock, so there
 674          * is no chance of a race here */
 675         else if (!timer_pending(&inst->timer)) {
 676                 instance_get(inst);
 677                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 678                 add_timer(&inst->timer);
 679         }
 680
 681 unlock_and_release:
 682         spin_unlock_bh(&inst->lock);
 683         instance_put(inst);
 684         return;
 685
 686 alloc_failure:
 687         /* FIXME: statistics */
 688         goto unlock_and_release;
 689 }
 690 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
 691
 692 static int
 693 nfulnl_rcv_nl_event(struct notifier_block *this,
 694                    unsigned long event, void *ptr)
 695 {
 696         struct netlink_notify *n = ptr;
 697
 698         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 699                 int i;
 700
 701                 /* destroy all instances for this pid */
 702                 spin_lock_bh(&instances_lock);
 703                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 704                         struct hlist_node *tmp, *t2;
 705                         struct nfulnl_instance *inst;
 706                         struct hlist_head *head = &instance_table[i];
 707
 708                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 709                                 if ((net_eq(n->net, &init_net)) &&
 710                                     (n->pid == inst->peer_pid))
 711                                         __instance_destroy(inst);
 712                         }
 713                 }
 714                 spin_unlock_bh(&instances_lock);
 715         }
 716         return NOTIFY_DONE;
 717 }
 718
 719 static struct notifier_block nfulnl_rtnl_notifier = {
 720         .notifier_call  = nfulnl_rcv_nl_event,
 721 };
 722
 723 static int
 724 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 725                    const struct nlmsghdr *nlh,
 726                    const struct nlattr * const nfqa[])
 727 {
 728         return -ENOTSUPP;
 729 }
 730
 731 static struct nf_logger nfulnl_logger __read_mostly = {
 732         .name   = "nfnetlink_log",
 733         .logfn  = &nfulnl_log_packet,
 734         .me     = THIS_MODULE,
 735 };
 736
 737 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 738         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 739         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 740         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 741         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 742         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 743         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 744 };
 745
 746 static int
 747 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 748                    const struct nlmsghdr *nlh,
 749                    const struct nlattr * const nfula[])
 750 {
 751         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 752         u_int16_t group_num = ntohs(nfmsg->res_id);
 753         struct nfulnl_instance *inst;
 754         struct nfulnl_msg_config_cmd *cmd = NULL;
 755         int ret = 0;
 756
 757         if (nfula[NFULA_CFG_CMD]) {
 758                 u_int8_t pf = nfmsg->nfgen_family;
 759                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 760
 761                 /* Commands without queue context */
 762                 switch (cmd->command) {
 763                 case NFULNL_CFG_CMD_PF_BIND:
 764                         return nf_log_bind_pf(pf, &nfulnl_logger);
 765                 case NFULNL_CFG_CMD_PF_UNBIND:
 766                         nf_log_unbind_pf(pf);
 767                         return 0;
 768                 }
 769         }
 770
 771         inst = instance_lookup_get(group_num);
 772         if (inst && inst->peer_pid != NETLINK_CB(skb).pid) {
 773                 ret = -EPERM;
 774                 goto out_put;
 775         }
 776
 777         if (cmd != NULL) {
 778                 switch (cmd->command) {
 779                 case NFULNL_CFG_CMD_BIND:
 780                         if (inst) {
 781                                 ret = -EBUSY;
 782                                 goto out_put;
 783                         }
 784
 785                         inst = instance_create(group_num,
 786                                                NETLINK_CB(skb).pid);
 787                         if (IS_ERR(inst)) {
 788                                 ret = PTR_ERR(inst);
 789                                 goto out;
 790                         }
 791                         break;
 792                 case NFULNL_CFG_CMD_UNBIND:
 793                         if (!inst) {
 794                                 ret = -ENODEV;
 795                                 goto out;
 796                         }
 797
 798                         instance_destroy(inst);
 799                         goto out_put;
 800                 default:
 801                         ret = -ENOTSUPP;
 802                         break;
 803                 }
 804         }
 805
 806         if (nfula[NFULA_CFG_MODE]) {
 807                 struct nfulnl_msg_config_mode *params;
 808                 params = nla_data(nfula[NFULA_CFG_MODE]);
 809
 810                 if (!inst) {
 811                         ret = -ENODEV;
 812                         goto out;
 813                 }
 814                 nfulnl_set_mode(inst, params->copy_mode,
 815                                 ntohl(params->copy_range));
 816         }
 817
 818         if (nfula[NFULA_CFG_TIMEOUT]) {
 819                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 820
 821                 if (!inst) {
 822                         ret = -ENODEV;
 823                         goto out;
 824                 }
 825                 nfulnl_set_timeout(inst, ntohl(timeout));
 826         }
 827
 828         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 829                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 830
 831                 if (!inst) {
 832                         ret = -ENODEV;
 833                         goto out;
 834                 }
 835                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 836         }
 837
 838         if (nfula[NFULA_CFG_QTHRESH]) {
 839                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 840
 841                 if (!inst) {
 842                         ret = -ENODEV;
 843                         goto out;
 844                 }
 845                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 846         }
 847
 848         if (nfula[NFULA_CFG_FLAGS]) {
 849                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
 850
 851                 if (!inst) {
 852                         ret = -ENODEV;
 853                         goto out;
 854                 }
 855                 nfulnl_set_flags(inst, ntohs(flags));
 856         }
 857
 858 out_put:
 859         instance_put(inst);
 860 out:
 861         return ret;
 862 }
 863
 864 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 865         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 866                                     .attr_count = NFULA_MAX, },
 867         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 868                                     .attr_count = NFULA_CFG_MAX,
 869                                     .policy = nfula_cfg_policy },
 870 };
 871
 872 static const struct nfnetlink_subsystem nfulnl_subsys = {
 873         .name           = "log",
 874         .subsys_id      = NFNL_SUBSYS_ULOG,
 875         .cb_count       = NFULNL_MSG_MAX,
 876         .cb             = nfulnl_cb,
 877 };
 878
 879 #ifdef CONFIG_PROC_FS
 880 struct iter_state {
 881         unsigned int bucket;
 882 };
 883
 884 static struct hlist_node *get_first(struct iter_state *st)
 885 {
 886         if (!st)
 887                 return NULL;
 888
 889         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 890                 if (!hlist_empty(&instance_table[st->bucket]))
 891                         return rcu_dereference_bh(hlist_first_rcu(&instance_table[st->bucket]));
 892         }
 893         return NULL;
 894 }
 895
 896 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
 897 {
 898         h = rcu_dereference_bh(hlist_next_rcu(h));
 899         while (!h) {
 900                 if (++st->bucket >= INSTANCE_BUCKETS)
 901                         return NULL;
 902
 903                 h = rcu_dereference_bh(hlist_first_rcu(&instance_table[st->bucket]));
 904         }
 905         return h;
 906 }
 907
 908 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
 909 {
 910         struct hlist_node *head;
 911         head = get_first(st);
 912
 913         if (head)
 914                 while (pos && (head = get_next(st, head)))
 915                         pos--;
 916         return pos ? NULL : head;
 917 }
 918
 919 static void *seq_start(struct seq_file *seq, loff_t *pos)
 920         __acquires(rcu_bh)
 921 {
 922         rcu_read_lock_bh();
 923         return get_idx(seq->private, *pos);
 924 }
 925
 926 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 927 {
 928         (*pos)++;
 929         return get_next(s->private, v);
 930 }
 931
 932 static void seq_stop(struct seq_file *s, void *v)
 933         __releases(rcu_bh)
 934 {
 935         rcu_read_unlock_bh();
 936 }
 937
 938 static int seq_show(struct seq_file *s, void *v)
 939 {
 940         const struct nfulnl_instance *inst = v;
 941
 942         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 943                           inst->group_num,
 944                           inst->peer_pid, inst->qlen,
 945                           inst->copy_mode, inst->copy_range,
 946                           inst->flushtimeout, atomic_read(&inst->use));
 947 }
 948
 949 static const struct seq_operations nful_seq_ops = {
 950         .start  = seq_start,
 951         .next   = seq_next,
 952         .stop   = seq_stop,
 953         .show   = seq_show,
 954 };
 955
 956 static int nful_open(struct inode *inode, struct file *file)
 957 {
 958         return seq_open_private(file, &nful_seq_ops,
 959                         sizeof(struct iter_state));
 960 }
 961
 962 static const struct file_operations nful_file_ops = {
 963         .owner   = THIS_MODULE,
 964         .open    = nful_open,
 965         .read    = seq_read,
 966         .llseek  = seq_lseek,
 967         .release = seq_release_private,
 968 };
 969
 970 #endif /* PROC_FS */
 971
 972 static int __init nfnetlink_log_init(void)
 973 {
 974         int i, status = -ENOMEM;
 975
 976         for (i = 0; i < INSTANCE_BUCKETS; i++)
 977                 INIT_HLIST_HEAD(&instance_table[i]);
 978
 979         /* it's not really all that important to have a random value, so
 980          * we can do this from the init function, even if there hasn't
 981          * been that much entropy yet */
 982         get_random_bytes(&hash_init, sizeof(hash_init));
 983
 984         netlink_register_notifier(&nfulnl_rtnl_notifier);
 985         status = nfnetlink_subsys_register(&nfulnl_subsys);
 986         if (status < 0) {
 987                 printk(KERN_ERR "log: failed to create netlink socket\n");
 988                 goto cleanup_netlink_notifier;
 989         }
 990
 991         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
 992         if (status < 0) {
 993                 printk(KERN_ERR "log: failed to register logger\n");
 994                 goto cleanup_subsys;
 995         }
 996
 997 #ifdef CONFIG_PROC_FS
 998         if (!proc_create("nfnetlink_log", 0440,
 999                          proc_net_netfilter, &nful_file_ops))
1000                 goto cleanup_logger;
1001 #endif
1002         return status;
1003
1004 #ifdef CONFIG_PROC_FS
1005 cleanup_logger:
1006         nf_log_unregister(&nfulnl_logger);
1007 #endif
1008 cleanup_subsys:
1009         nfnetlink_subsys_unregister(&nfulnl_subsys);
1010 cleanup_netlink_notifier:
1011         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1012         return status;
1013 }
1014
1015 static void __exit nfnetlink_log_fini(void)
1016 {
1017         nf_log_unregister(&nfulnl_logger);
1018 #ifdef CONFIG_PROC_FS
1019         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1020 #endif
1021         nfnetlink_subsys_unregister(&nfulnl_subsys);
1022         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1023 }
1024
1025 MODULE_DESCRIPTION("netfilter userspace logging");
1026 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1027 MODULE_LICENSE("GPL");
1028 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1029
1030 module_init(nfnetlink_log_init);
1031 module_exit(nfnetlink_log_fini);