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  * 2006-01-26 Harald Welte <laforge@netfilter.org>
  15  *      - Add optional local and global sequence number to detect lost
  16  *        events from userspace
  17  *
  18  */
  19 #include <linux/module.h>
  20 #include <linux/skbuff.h>
  21 #include <linux/init.h>
  22 #include <linux/ip.h>
  23 #include <linux/ipv6.h>
  24 #include <linux/netdevice.h>
  25 #include <linux/netfilter.h>
  26 #include <linux/netlink.h>
  27 #include <linux/netfilter/nfnetlink.h>
  28 #include <linux/netfilter/nfnetlink_log.h>
  29 #include <linux/spinlock.h>
  30 #include <linux/sysctl.h>
  31 #include <linux/proc_fs.h>
  32 #include <linux/security.h>
  33 #include <linux/list.h>
  34 #include <linux/jhash.h>
  35 #include <linux/random.h>
  36 #include <net/sock.h>
  37
  38 #include <asm/atomic.h>
  39
  40 #ifdef CONFIG_BRIDGE_NETFILTER
  41 #include "../bridge/br_private.h"
  42 #endif
  43
  44 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
  45 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
  46 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  47
  48 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  49                                      printk(x, ## args); } while (0);
  50
  51 #if 0
  52 #define UDEBUG(x, args ...)     printk(KERN_DEBUG "%s(%d):%s(): " x,       \
  53                                         __FILE__, __LINE__, __FUNCTION__,  \
  54                                         ## args)
  55 #else
  56 #define UDEBUG(x, ...)
  57 #endif
  58
  59 struct nfulnl_instance {
  60         struct hlist_node hlist;        /* global list of instances */
  61         spinlock_t lock;
  62         atomic_t use;                   /* use count */
  63
  64         unsigned int qlen;              /* number of nlmsgs in skb */
  65         struct sk_buff *skb;            /* pre-allocatd skb */
  66         struct nlmsghdr *lastnlh;       /* netlink header of last msg in skb */
  67         struct timer_list timer;
  68         int peer_pid;                   /* PID of the peer process */
  69
  70         /* configurable parameters */
  71         unsigned int flushtimeout;      /* timeout until queue flush */
  72         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  73         unsigned int qthreshold;        /* threshold of the queue */
  74         u_int32_t copy_range;
  75         u_int32_t seq;                  /* instance-local sequential counter */
  76         u_int16_t group_num;            /* number of this queue */
  77         u_int16_t flags;
  78         u_int8_t copy_mode;
  79 };
  80
  81 static DEFINE_RWLOCK(instances_lock);
  82 static atomic_t global_seq;
  83
  84 #define INSTANCE_BUCKETS        16
  85 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  86 static unsigned int hash_init;
  87
  88 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  89 {
  90         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  91 }
  92
  93 static struct nfulnl_instance *
  94 __instance_lookup(u_int16_t group_num)
  95 {
  96         struct hlist_head *head;
  97         struct hlist_node *pos;
  98         struct nfulnl_instance *inst;
  99
 100         UDEBUG("entering (group_num=%u)\n", group_num);
 101
 102         head = &instance_table[instance_hashfn(group_num)];
 103         hlist_for_each_entry(inst, pos, head, hlist) {
 104                 if (inst->group_num == group_num)
 105                         return inst;
 106         }
 107         return NULL;
 108 }
 109
 110 static inline void
 111 instance_get(struct nfulnl_instance *inst)
 112 {
 113         atomic_inc(&inst->use);
 114 }
 115
 116 static struct nfulnl_instance *
 117 instance_lookup_get(u_int16_t group_num)
 118 {
 119         struct nfulnl_instance *inst;
 120
 121         read_lock_bh(&instances_lock);
 122         inst = __instance_lookup(group_num);
 123         if (inst)
 124                 instance_get(inst);
 125         read_unlock_bh(&instances_lock);
 126
 127         return inst;
 128 }
 129
 130 static void
 131 instance_put(struct nfulnl_instance *inst)
 132 {
 133         if (inst && atomic_dec_and_test(&inst->use)) {
 134                 UDEBUG("kfree(inst=%p)\n", inst);
 135                 kfree(inst);
 136         }
 137 }
 138
 139 static void nfulnl_timer(unsigned long data);
 140
 141 static struct nfulnl_instance *
 142 instance_create(u_int16_t group_num, int pid)
 143 {
 144         struct nfulnl_instance *inst;
 145
 146         UDEBUG("entering (group_num=%u, pid=%d)\n", group_num,
 147                 pid);
 148
 149         write_lock_bh(&instances_lock);
 150         if (__instance_lookup(group_num)) {
 151                 inst = NULL;
 152                 UDEBUG("aborting, instance already exists\n");
 153                 goto out_unlock;
 154         }
 155
 156         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 157         if (!inst)
 158                 goto out_unlock;
 159
 160         INIT_HLIST_NODE(&inst->hlist);
 161         spin_lock_init(&inst->lock);
 162         /* needs to be two, since we _put() after creation */
 163         atomic_set(&inst->use, 2);
 164
 165         init_timer(&inst->timer);
 166         inst->timer.function = nfulnl_timer;
 167         inst->timer.data = (unsigned long)inst;
 168         /* don't start timer yet. (re)start it  with every packet */
 169
 170         inst->peer_pid = pid;
 171         inst->group_num = group_num;
 172
 173         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 174         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 175         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 176         inst->copy_mode         = NFULNL_COPY_PACKET;
 177         inst->copy_range        = 0xffff;
 178
 179         if (!try_module_get(THIS_MODULE))
 180                 goto out_free;
 181
 182         hlist_add_head(&inst->hlist,
 183                        &instance_table[instance_hashfn(group_num)]);
 184
 185         UDEBUG("newly added node: %p, next=%p\n", &inst->hlist,
 186                 inst->hlist.next);
 187
 188         write_unlock_bh(&instances_lock);
 189
 190         return inst;
 191
 192 out_free:
 193         instance_put(inst);
 194 out_unlock:
 195         write_unlock_bh(&instances_lock);
 196         return NULL;
 197 }
 198
 199 static int __nfulnl_send(struct nfulnl_instance *inst);
 200
 201 static void
 202 _instance_destroy2(struct nfulnl_instance *inst, int lock)
 203 {
 204         /* first pull it out of the global list */
 205         if (lock)
 206                 write_lock_bh(&instances_lock);
 207
 208         UDEBUG("removing instance %p (queuenum=%u) from hash\n",
 209                 inst, inst->group_num);
 210
 211         hlist_del(&inst->hlist);
 212
 213         if (lock)
 214                 write_unlock_bh(&instances_lock);
 215
 216         /* then flush all pending packets from skb */
 217
 218         spin_lock_bh(&inst->lock);
 219         if (inst->skb) {
 220                 /* timer "holds" one reference (we have one more) */
 221                 if (del_timer(&inst->timer))
 222                         instance_put(inst);
 223                 if (inst->qlen)
 224                         __nfulnl_send(inst);
 225                 if (inst->skb) {
 226                         kfree_skb(inst->skb);
 227                         inst->skb = NULL;
 228                 }
 229         }
 230         spin_unlock_bh(&inst->lock);
 231
 232         /* and finally put the refcount */
 233         instance_put(inst);
 234
 235         module_put(THIS_MODULE);
 236 }
 237
 238 static inline void
 239 __instance_destroy(struct nfulnl_instance *inst)
 240 {
 241         _instance_destroy2(inst, 0);
 242 }
 243
 244 static inline void
 245 instance_destroy(struct nfulnl_instance *inst)
 246 {
 247         _instance_destroy2(inst, 1);
 248 }
 249
 250 static int
 251 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 252                   unsigned int range)
 253 {
 254         int status = 0;
 255
 256         spin_lock_bh(&inst->lock);
 257
 258         switch (mode) {
 259         case NFULNL_COPY_NONE:
 260         case NFULNL_COPY_META:
 261                 inst->copy_mode = mode;
 262                 inst->copy_range = 0;
 263                 break;
 264
 265         case NFULNL_COPY_PACKET:
 266                 inst->copy_mode = mode;
 267                 /* we're using struct nfattr which has 16bit nfa_len */
 268                 if (range > 0xffff)
 269                         inst->copy_range = 0xffff;
 270                 else
 271                         inst->copy_range = range;
 272                 break;
 273
 274         default:
 275                 status = -EINVAL;
 276                 break;
 277         }
 278
 279         spin_unlock_bh(&inst->lock);
 280
 281         return status;
 282 }
 283
 284 static int
 285 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 286 {
 287         int status;
 288
 289         spin_lock_bh(&inst->lock);
 290         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 291                 status = -ERANGE;
 292         else if (nlbufsiz > 131072)
 293                 status = -ERANGE;
 294         else {
 295                 inst->nlbufsiz = nlbufsiz;
 296                 status = 0;
 297         }
 298         spin_unlock_bh(&inst->lock);
 299
 300         return status;
 301 }
 302
 303 static int
 304 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 305 {
 306         spin_lock_bh(&inst->lock);
 307         inst->flushtimeout = timeout;
 308         spin_unlock_bh(&inst->lock);
 309
 310         return 0;
 311 }
 312
 313 static int
 314 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 315 {
 316         spin_lock_bh(&inst->lock);
 317         inst->qthreshold = qthresh;
 318         spin_unlock_bh(&inst->lock);
 319
 320         return 0;
 321 }
 322
 323 static int
 324 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 325 {
 326         spin_lock_bh(&inst->lock);
 327         inst->flags = flags;
 328         spin_unlock_bh(&inst->lock);
 329
 330         return 0;
 331 }
 332
 333 static struct sk_buff *nfulnl_alloc_skb(unsigned int inst_size,
 334                                         unsigned int pkt_size)
 335 {
 336         struct sk_buff *skb;
 337         unsigned int n;
 338
 339         UDEBUG("entered (%u, %u)\n", inst_size, pkt_size);
 340
 341         /* alloc skb which should be big enough for a whole multipart
 342          * message.  WARNING: has to be <= 128k due to slab restrictions */
 343
 344         n = max(inst_size, pkt_size);
 345         skb = alloc_skb(n, GFP_ATOMIC);
 346         if (!skb) {
 347                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
 348                         inst_size);
 349
 350                 if (n > pkt_size) {
 351                         /* try to allocate only as much as we need for current
 352                          * packet */
 353
 354                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 355                         if (!skb)
 356                                 PRINTR("nfnetlink_log: can't even alloc %u "
 357                                        "bytes\n", pkt_size);
 358                 }
 359         }
 360
 361         return skb;
 362 }
 363
 364 static int
 365 __nfulnl_send(struct nfulnl_instance *inst)
 366 {
 367         int status;
 368
 369         if (!inst->skb)
 370                 return 0;
 371
 372         if (inst->qlen > 1)
 373                 inst->lastnlh->nlmsg_type = NLMSG_DONE;
 374
 375         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
 376         if (status < 0) {
 377                 UDEBUG("netlink_unicast() failed\n");
 378                 /* FIXME: statistics */
 379         }
 380
 381         inst->qlen = 0;
 382         inst->skb = NULL;
 383         inst->lastnlh = NULL;
 384
 385         return status;
 386 }
 387
 388 static void nfulnl_timer(unsigned long data)
 389 {
 390         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 391
 392         UDEBUG("timer function called, flushing buffer\n");
 393
 394         spin_lock_bh(&inst->lock);
 395         if (timer_pending(&inst->timer))        /* is it always true or false here? */
 396                 del_timer(&inst->timer);
 397         __nfulnl_send(inst);
 398         spin_unlock_bh(&inst->lock);
 399         instance_put(inst);
 400 }
 401
 402 /* This is an inline function, we don't really care about a long
 403  * list of arguments */
 404 static inline int
 405 __build_packet_message(struct nfulnl_instance *inst,
 406                         const struct sk_buff *skb,
 407                         unsigned int data_len,
 408                         unsigned int pf,
 409                         unsigned int hooknum,
 410                         const struct net_device *indev,
 411                         const struct net_device *outdev,
 412                         const struct nf_loginfo *li,
 413                         const char *prefix, unsigned int plen)
 414 {
 415         unsigned char *old_tail;
 416         struct nfulnl_msg_packet_hdr pmsg;
 417         struct nlmsghdr *nlh;
 418         struct nfgenmsg *nfmsg;
 419         __be32 tmp_uint;
 420
 421         UDEBUG("entered\n");
 422
 423         old_tail = inst->skb->tail;
 424         nlh = NLMSG_PUT(inst->skb, 0, 0,
 425                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 426                         sizeof(struct nfgenmsg));
 427         nfmsg = NLMSG_DATA(nlh);
 428         nfmsg->nfgen_family = pf;
 429         nfmsg->version = NFNETLINK_V0;
 430         nfmsg->res_id = htons(inst->group_num);
 431
 432         pmsg.hw_protocol        = skb->protocol;
 433         pmsg.hook               = hooknum;
 434
 435         NFA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
 436
 437         if (prefix)
 438                 NFA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
 439
 440         if (indev) {
 441                 tmp_uint = htonl(indev->ifindex);
 442 #ifndef CONFIG_BRIDGE_NETFILTER
 443                 NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
 444                         &tmp_uint);
 445 #else
 446                 if (pf == PF_BRIDGE) {
 447                         /* Case 1: outdev is physical input device, we need to
 448                          * look for bridge group (when called from
 449                          * netfilter_bridge) */
 450                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 451                                 sizeof(tmp_uint), &tmp_uint);
 452                         /* this is the bridge group "brX" */
 453                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
 454                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 455                                 sizeof(tmp_uint), &tmp_uint);
 456                 } else {
 457                         /* Case 2: indev is bridge group, we need to look for
 458                          * physical device (when called from ipv4) */
 459                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 460                                 sizeof(tmp_uint), &tmp_uint);
 461                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
 462                                 tmp_uint =
 463                                     htonl(skb->nf_bridge->physindev->ifindex);
 464                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 465                                         sizeof(tmp_uint), &tmp_uint);
 466                         }
 467                 }
 468 #endif
 469         }
 470
 471         if (outdev) {
 472                 tmp_uint = htonl(outdev->ifindex);
 473 #ifndef CONFIG_BRIDGE_NETFILTER
 474                 NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 475                         &tmp_uint);
 476 #else
 477                 if (pf == PF_BRIDGE) {
 478                         /* Case 1: outdev is physical output device, we need to
 479                          * look for bridge group (when called from
 480                          * netfilter_bridge) */
 481                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 482                                 sizeof(tmp_uint), &tmp_uint);
 483                         /* this is the bridge group "brX" */
 484                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
 485                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 486                                 sizeof(tmp_uint), &tmp_uint);
 487                 } else {
 488                         /* Case 2: indev is a bridge group, we need to look
 489                          * for physical device (when called from ipv4) */
 490                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 491                                 sizeof(tmp_uint), &tmp_uint);
 492                         if (skb->nf_bridge && skb->nf_bridge->physoutdev) {
 493                                 tmp_uint =
 494                                     htonl(skb->nf_bridge->physoutdev->ifindex);
 495                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 496                                         sizeof(tmp_uint), &tmp_uint);
 497                         }
 498                 }
 499 #endif
 500         }
 501
 502         if (skb->mark) {
 503                 tmp_uint = htonl(skb->mark);
 504                 NFA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
 505         }
 506
 507         if (indev && skb->dev && skb->dev->hard_header_parse) {
 508                 struct nfulnl_msg_packet_hw phw;
 509                 int len = skb->dev->hard_header_parse((struct sk_buff *)skb,
 510                                                     phw.hw_addr);
 511                 phw.hw_addrlen = htons(len);
 512                 NFA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
 513         }
 514
 515         if (skb->tstamp.off_sec) {
 516                 struct nfulnl_msg_packet_timestamp ts;
 517
 518                 ts.sec = cpu_to_be64(skb->tstamp.off_sec);
 519                 ts.usec = cpu_to_be64(skb->tstamp.off_usec);
 520
 521                 NFA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 522         }
 523
 524         /* UID */
 525         if (skb->sk) {
 526                 read_lock_bh(&skb->sk->sk_callback_lock);
 527                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 528                         __be32 uid = htonl(skb->sk->sk_socket->file->f_uid);
 529                         /* need to unlock here since NFA_PUT may goto */
 530                         read_unlock_bh(&skb->sk->sk_callback_lock);
 531                         NFA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
 532                 } else
 533                         read_unlock_bh(&skb->sk->sk_callback_lock);
 534         }
 535
 536         /* local sequence number */
 537         if (inst->flags & NFULNL_CFG_F_SEQ) {
 538                 tmp_uint = htonl(inst->seq++);
 539                 NFA_PUT(inst->skb, NFULA_SEQ, sizeof(tmp_uint), &tmp_uint);
 540         }
 541         /* global sequence number */
 542         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) {
 543                 tmp_uint = htonl(atomic_inc_return(&global_seq));
 544                 NFA_PUT(inst->skb, NFULA_SEQ_GLOBAL, sizeof(tmp_uint), &tmp_uint);
 545         }
 546
 547         if (data_len) {
 548                 struct nfattr *nfa;
 549                 int size = NFA_LENGTH(data_len);
 550
 551                 if (skb_tailroom(inst->skb) < (int)NFA_SPACE(data_len)) {
 552                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 553                         goto nlmsg_failure;
 554                 }
 555
 556                 nfa = (struct nfattr *)skb_put(inst->skb, NFA_ALIGN(size));
 557                 nfa->nfa_type = NFULA_PAYLOAD;
 558                 nfa->nfa_len = size;
 559
 560                 if (skb_copy_bits(skb, 0, NFA_DATA(nfa), data_len))
 561                         BUG();
 562         }
 563
 564         nlh->nlmsg_len = inst->skb->tail - old_tail;
 565         inst->lastnlh = nlh;
 566         return 0;
 567
 568 nlmsg_failure:
 569         UDEBUG("nlmsg_failure\n");
 570 nfattr_failure:
 571         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 572         return -1;
 573 }
 574
 575 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 576
 577 static struct nf_loginfo default_loginfo = {
 578         .type =         NF_LOG_TYPE_ULOG,
 579         .u = {
 580                 .ulog = {
 581                         .copy_len       = 0xffff,
 582                         .group          = 0,
 583                         .qthreshold     = 1,
 584                 },
 585         },
 586 };
 587
 588 /* log handler for internal netfilter logging api */
 589 static void
 590 nfulnl_log_packet(unsigned int pf,
 591                   unsigned int hooknum,
 592                   const struct sk_buff *skb,
 593                   const struct net_device *in,
 594                   const struct net_device *out,
 595                   const struct nf_loginfo *li_user,
 596                   const char *prefix)
 597 {
 598         unsigned int size, data_len;
 599         struct nfulnl_instance *inst;
 600         const struct nf_loginfo *li;
 601         unsigned int qthreshold;
 602         unsigned int nlbufsiz;
 603         unsigned int plen;
 604
 605         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 606                 li = li_user;
 607         else
 608                 li = &default_loginfo;
 609
 610         inst = instance_lookup_get(li->u.ulog.group);
 611         if (!inst)
 612                 inst = instance_lookup_get(0);
 613         if (!inst) {
 614                 PRINTR("nfnetlink_log: trying to log packet, "
 615                         "but no instance for group %u\n", li->u.ulog.group);
 616                 return;
 617         }
 618
 619         plen = 0;
 620         if (prefix)
 621                 plen = strlen(prefix) + 1;
 622
 623         /* all macros expand to constant values at compile time */
 624         /* FIXME: do we want to make the size calculation conditional based on
 625          * what is actually present?  way more branches and checks, but more
 626          * memory efficient... */
 627         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 628                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hdr))
 629                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 630                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 631 #ifdef CONFIG_BRIDGE_NETFILTER
 632                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 633                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 634 #endif
 635                 + NFA_SPACE(sizeof(u_int32_t))  /* mark */
 636                 + NFA_SPACE(sizeof(u_int32_t))  /* uid */
 637                 + NFA_SPACE(plen)               /* prefix */
 638                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hw))
 639                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_timestamp));
 640
 641         UDEBUG("initial size=%u\n", size);
 642
 643         spin_lock_bh(&inst->lock);
 644
 645         if (inst->flags & NFULNL_CFG_F_SEQ)
 646                 size += NFA_SPACE(sizeof(u_int32_t));
 647         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 648                 size += NFA_SPACE(sizeof(u_int32_t));
 649
 650         qthreshold = inst->qthreshold;
 651         /* per-rule qthreshold overrides per-instance */
 652         if (qthreshold > li->u.ulog.qthreshold)
 653                 qthreshold = li->u.ulog.qthreshold;
 654
 655         switch (inst->copy_mode) {
 656         case NFULNL_COPY_META:
 657         case NFULNL_COPY_NONE:
 658                 data_len = 0;
 659                 break;
 660
 661         case NFULNL_COPY_PACKET:
 662                 if (inst->copy_range == 0
 663                     || inst->copy_range > skb->len)
 664                         data_len = skb->len;
 665                 else
 666                         data_len = inst->copy_range;
 667
 668                 size += NFA_SPACE(data_len);
 669                 UDEBUG("copy_packet, therefore size now %u\n", size);
 670                 break;
 671
 672         default:
 673                 spin_unlock_bh(&inst->lock);
 674                 instance_put(inst);
 675                 return;
 676         }
 677
 678         if (size > inst->nlbufsiz)
 679                 nlbufsiz = size;
 680         else
 681                 nlbufsiz = inst->nlbufsiz;
 682
 683         if (!inst->skb) {
 684                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
 685                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
 686                                 inst->nlbufsiz, size);
 687                         goto alloc_failure;
 688                 }
 689         } else if (inst->qlen >= qthreshold ||
 690                    size > skb_tailroom(inst->skb)) {
 691                 /* either the queue len is too high or we don't have
 692                  * enough room in the skb left. flush to userspace. */
 693                 UDEBUG("flushing old skb\n");
 694
 695                 /* timer "holds" one reference (we have another one) */
 696                 if (del_timer(&inst->timer))
 697                         instance_put(inst);
 698                 __nfulnl_send(inst);
 699
 700                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
 701                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
 702                                 inst->nlbufsiz, size);
 703                         goto alloc_failure;
 704                 }
 705         }
 706
 707         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
 708         inst->qlen++;
 709
 710         __build_packet_message(inst, skb, data_len, pf,
 711                                 hooknum, in, out, li, prefix, plen);
 712
 713         /* timer_pending always called within inst->lock, so there
 714          * is no chance of a race here */
 715         if (!timer_pending(&inst->timer)) {
 716                 instance_get(inst);
 717                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 718                 add_timer(&inst->timer);
 719         }
 720
 721 unlock_and_release:
 722         spin_unlock_bh(&inst->lock);
 723         instance_put(inst);
 724         return;
 725
 726 alloc_failure:
 727         UDEBUG("error allocating skb\n");
 728         /* FIXME: statistics */
 729         goto unlock_and_release;
 730 }
 731
 732 static int
 733 nfulnl_rcv_nl_event(struct notifier_block *this,
 734                    unsigned long event, void *ptr)
 735 {
 736         struct netlink_notify *n = ptr;
 737
 738         if (event == NETLINK_URELEASE &&
 739             n->protocol == NETLINK_NETFILTER && n->pid) {
 740                 int i;
 741
 742                 /* destroy all instances for this pid */
 743                 write_lock_bh(&instances_lock);
 744                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 745                         struct hlist_node *tmp, *t2;
 746                         struct nfulnl_instance *inst;
 747                         struct hlist_head *head = &instance_table[i];
 748
 749                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 750                                 UDEBUG("node = %p\n", inst);
 751                                 if (n->pid == inst->peer_pid)
 752                                         __instance_destroy(inst);
 753                         }
 754                 }
 755                 write_unlock_bh(&instances_lock);
 756         }
 757         return NOTIFY_DONE;
 758 }
 759
 760 static struct notifier_block nfulnl_rtnl_notifier = {
 761         .notifier_call  = nfulnl_rcv_nl_event,
 762 };
 763
 764 static int
 765 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 766                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
 767 {
 768         return -ENOTSUPP;
 769 }
 770
 771 static struct nf_logger nfulnl_logger = {
 772         .name   = "nfnetlink_log",
 773         .logfn  = &nfulnl_log_packet,
 774         .me     = THIS_MODULE,
 775 };
 776
 777 static const int nfula_min[NFULA_MAX] = {
 778         [NFULA_PACKET_HDR-1]    = sizeof(struct nfulnl_msg_packet_hdr),
 779         [NFULA_MARK-1]          = sizeof(u_int32_t),
 780         [NFULA_TIMESTAMP-1]     = sizeof(struct nfulnl_msg_packet_timestamp),
 781         [NFULA_IFINDEX_INDEV-1] = sizeof(u_int32_t),
 782         [NFULA_IFINDEX_OUTDEV-1]= sizeof(u_int32_t),
 783         [NFULA_IFINDEX_PHYSINDEV-1]     = sizeof(u_int32_t),
 784         [NFULA_IFINDEX_PHYSOUTDEV-1]    = sizeof(u_int32_t),
 785         [NFULA_HWADDR-1]        = sizeof(struct nfulnl_msg_packet_hw),
 786         [NFULA_PAYLOAD-1]       = 0,
 787         [NFULA_PREFIX-1]        = 0,
 788         [NFULA_UID-1]           = sizeof(u_int32_t),
 789         [NFULA_SEQ-1]           = sizeof(u_int32_t),
 790         [NFULA_SEQ_GLOBAL-1]    = sizeof(u_int32_t),
 791 };
 792
 793 static const int nfula_cfg_min[NFULA_CFG_MAX] = {
 794         [NFULA_CFG_CMD-1]       = sizeof(struct nfulnl_msg_config_cmd),
 795         [NFULA_CFG_MODE-1]      = sizeof(struct nfulnl_msg_config_mode),
 796         [NFULA_CFG_TIMEOUT-1]   = sizeof(u_int32_t),
 797         [NFULA_CFG_QTHRESH-1]   = sizeof(u_int32_t),
 798         [NFULA_CFG_NLBUFSIZ-1]  = sizeof(u_int32_t),
 799         [NFULA_CFG_FLAGS-1]     = sizeof(u_int16_t),
 800 };
 801
 802 static int
 803 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 804                    struct nlmsghdr *nlh, struct nfattr *nfula[], int *errp)
 805 {
 806         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 807         u_int16_t group_num = ntohs(nfmsg->res_id);
 808         struct nfulnl_instance *inst;
 809         int ret = 0;
 810
 811         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
 812
 813         if (nfattr_bad_size(nfula, NFULA_CFG_MAX, nfula_cfg_min)) {
 814                 UDEBUG("bad attribute size\n");
 815                 return -EINVAL;
 816         }
 817
 818         inst = instance_lookup_get(group_num);
 819         if (nfula[NFULA_CFG_CMD-1]) {
 820                 u_int8_t pf = nfmsg->nfgen_family;
 821                 struct nfulnl_msg_config_cmd *cmd;
 822                 cmd = NFA_DATA(nfula[NFULA_CFG_CMD-1]);
 823                 UDEBUG("found CFG_CMD for\n");
 824
 825                 switch (cmd->command) {
 826                 case NFULNL_CFG_CMD_BIND:
 827                         if (inst) {
 828                                 ret = -EBUSY;
 829                                 goto out_put;
 830                         }
 831
 832                         inst = instance_create(group_num,
 833                                                NETLINK_CB(skb).pid);
 834                         if (!inst) {
 835                                 ret = -EINVAL;
 836                                 goto out_put;
 837                         }
 838                         break;
 839                 case NFULNL_CFG_CMD_UNBIND:
 840                         if (!inst) {
 841                                 ret = -ENODEV;
 842                                 goto out_put;
 843                         }
 844
 845                         if (inst->peer_pid != NETLINK_CB(skb).pid) {
 846                                 ret = -EPERM;
 847                                 goto out_put;
 848                         }
 849
 850                         instance_destroy(inst);
 851                         break;
 852                 case NFULNL_CFG_CMD_PF_BIND:
 853                         UDEBUG("registering log handler for pf=%u\n", pf);
 854                         ret = nf_log_register(pf, &nfulnl_logger);
 855                         break;
 856                 case NFULNL_CFG_CMD_PF_UNBIND:
 857                         UDEBUG("unregistering log handler for pf=%u\n", pf);
 858                         /* This is a bug and a feature.  We cannot unregister
 859                          * other handlers, like nfnetlink_inst can */
 860                         nf_log_unregister_pf(pf);
 861                         break;
 862                 default:
 863                         ret = -EINVAL;
 864                         break;
 865                 }
 866
 867                 if (!inst)
 868                         goto out;
 869         } else {
 870                 if (!inst) {
 871                         UDEBUG("no config command, and no instance for "
 872                                 "group=%u pid=%u =>ENOENT\n",
 873                                 group_num, NETLINK_CB(skb).pid);
 874                         ret = -ENOENT;
 875                         goto out_put;
 876                 }
 877
 878                 if (inst->peer_pid != NETLINK_CB(skb).pid) {
 879                         UDEBUG("no config command, and wrong pid\n");
 880                         ret = -EPERM;
 881                         goto out_put;
 882                 }
 883         }
 884
 885         if (nfula[NFULA_CFG_MODE-1]) {
 886                 struct nfulnl_msg_config_mode *params;
 887                 params = NFA_DATA(nfula[NFULA_CFG_MODE-1]);
 888
 889                 nfulnl_set_mode(inst, params->copy_mode,
 890                                 ntohl(params->copy_range));
 891         }
 892
 893         if (nfula[NFULA_CFG_TIMEOUT-1]) {
 894                 __be32 timeout =
 895                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_TIMEOUT-1]);
 896
 897                 nfulnl_set_timeout(inst, ntohl(timeout));
 898         }
 899
 900         if (nfula[NFULA_CFG_NLBUFSIZ-1]) {
 901                 __be32 nlbufsiz =
 902                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_NLBUFSIZ-1]);
 903
 904                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 905         }
 906
 907         if (nfula[NFULA_CFG_QTHRESH-1]) {
 908                 __be32 qthresh =
 909                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_QTHRESH-1]);
 910
 911                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 912         }
 913
 914         if (nfula[NFULA_CFG_FLAGS-1]) {
 915                 __be16 flags =
 916                         *(__be16 *)NFA_DATA(nfula[NFULA_CFG_FLAGS-1]);
 917                 nfulnl_set_flags(inst, ntohs(flags));
 918         }
 919
 920 out_put:
 921         instance_put(inst);
 922 out:
 923         return ret;
 924 }
 925
 926 static struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 927         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 928                                     .attr_count = NFULA_MAX, },
 929         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 930                                     .attr_count = NFULA_CFG_MAX, },
 931 };
 932
 933 static struct nfnetlink_subsystem nfulnl_subsys = {
 934         .name           = "log",
 935         .subsys_id      = NFNL_SUBSYS_ULOG,
 936         .cb_count       = NFULNL_MSG_MAX,
 937         .cb             = nfulnl_cb,
 938 };
 939
 940 #ifdef CONFIG_PROC_FS
 941 struct iter_state {
 942         unsigned int bucket;
 943 };
 944
 945 static struct hlist_node *get_first(struct seq_file *seq)
 946 {
 947         struct iter_state *st = seq->private;
 948
 949         if (!st)
 950                 return NULL;
 951
 952         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 953                 if (!hlist_empty(&instance_table[st->bucket]))
 954                         return instance_table[st->bucket].first;
 955         }
 956         return NULL;
 957 }
 958
 959 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
 960 {
 961         struct iter_state *st = seq->private;
 962
 963         h = h->next;
 964         while (!h) {
 965                 if (++st->bucket >= INSTANCE_BUCKETS)
 966                         return NULL;
 967
 968                 h = instance_table[st->bucket].first;
 969         }
 970         return h;
 971 }
 972
 973 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 974 {
 975         struct hlist_node *head;
 976         head = get_first(seq);
 977
 978         if (head)
 979                 while (pos && (head = get_next(seq, head)))
 980                         pos--;
 981         return pos ? NULL : head;
 982 }
 983
 984 static void *seq_start(struct seq_file *seq, loff_t *pos)
 985 {
 986         read_lock_bh(&instances_lock);
 987         return get_idx(seq, *pos);
 988 }
 989
 990 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 991 {
 992         (*pos)++;
 993         return get_next(s, v);
 994 }
 995
 996 static void seq_stop(struct seq_file *s, void *v)
 997 {
 998         read_unlock_bh(&instances_lock);
 999 }
1000
1001 static int seq_show(struct seq_file *s, void *v)
1002 {
1003         const struct nfulnl_instance *inst = v;
1004
1005         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
1006                           inst->group_num,
1007                           inst->peer_pid, inst->qlen,
1008                           inst->copy_mode, inst->copy_range,
1009                           inst->flushtimeout, atomic_read(&inst->use));
1010 }
1011
1012 static struct seq_operations nful_seq_ops = {
1013         .start  = seq_start,
1014         .next   = seq_next,
1015         .stop   = seq_stop,
1016         .show   = seq_show,
1017 };
1018
1019 static int nful_open(struct inode *inode, struct file *file)
1020 {
1021         struct seq_file *seq;
1022         struct iter_state *is;
1023         int ret;
1024
1025         is = kzalloc(sizeof(*is), GFP_KERNEL);
1026         if (!is)
1027                 return -ENOMEM;
1028         ret = seq_open(file, &nful_seq_ops);
1029         if (ret < 0)
1030                 goto out_free;
1031         seq = file->private_data;
1032         seq->private = is;
1033         return ret;
1034 out_free:
1035         kfree(is);
1036         return ret;
1037 }
1038
1039 static struct file_operations nful_file_ops = {
1040         .owner   = THIS_MODULE,
1041         .open    = nful_open,
1042         .read    = seq_read,
1043         .llseek  = seq_lseek,
1044         .release = seq_release_private,
1045 };
1046
1047 #endif /* PROC_FS */
1048
1049 static int __init nfnetlink_log_init(void)
1050 {
1051         int i, status = -ENOMEM;
1052 #ifdef CONFIG_PROC_FS
1053         struct proc_dir_entry *proc_nful;
1054 #endif
1055
1056         for (i = 0; i < INSTANCE_BUCKETS; i++)
1057                 INIT_HLIST_HEAD(&instance_table[i]);
1058
1059         /* it's not really all that important to have a random value, so
1060          * we can do this from the init function, even if there hasn't
1061          * been that much entropy yet */
1062         get_random_bytes(&hash_init, sizeof(hash_init));
1063
1064         netlink_register_notifier(&nfulnl_rtnl_notifier);
1065         status = nfnetlink_subsys_register(&nfulnl_subsys);
1066         if (status < 0) {
1067                 printk(KERN_ERR "log: failed to create netlink socket\n");
1068                 goto cleanup_netlink_notifier;
1069         }
1070
1071 #ifdef CONFIG_PROC_FS
1072         proc_nful = create_proc_entry("nfnetlink_log", 0440,
1073                                       proc_net_netfilter);
1074         if (!proc_nful)
1075                 goto cleanup_subsys;
1076         proc_nful->proc_fops = &nful_file_ops;
1077 #endif
1078         return status;
1079
1080 #ifdef CONFIG_PROC_FS
1081 cleanup_subsys:
1082         nfnetlink_subsys_unregister(&nfulnl_subsys);
1083 #endif
1084 cleanup_netlink_notifier:
1085         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1086         return status;
1087 }
1088
1089 static void __exit nfnetlink_log_fini(void)
1090 {
1091         nf_log_unregister_logger(&nfulnl_logger);
1092 #ifdef CONFIG_PROC_FS
1093         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1094 #endif
1095         nfnetlink_subsys_unregister(&nfulnl_subsys);
1096         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1097 }
1098
1099 MODULE_DESCRIPTION("netfilter userspace logging");
1100 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1101 MODULE_LICENSE("GPL");
1102 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1103
1104 module_init(nfnetlink_log_init);
1105 module_exit(nfnetlink_log_fini);