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 <asm/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                 pr_notice("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
 311                         inst_size);
 312
 313                 if (n > pkt_size) {
 314                         /* try to allocate only as much as we need for current
 315                          * packet */
 316
 317                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 318                         if (!skb)
 319                                 pr_err("nfnetlink_log: can't even alloc %u "
 320                                        "bytes\n", pkt_size);
 321                 }
 322         }
 323
 324         return skb;
 325 }
 326
 327 static int
 328 __nfulnl_send(struct nfulnl_instance *inst)
 329 {
 330         int status = -1;
 331
 332         if (inst->qlen > 1)
 333                 NLMSG_PUT(inst->skb, 0, 0,
 334                           NLMSG_DONE,
 335                           sizeof(struct nfgenmsg));
 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
 343 nlmsg_failure:
 344         return status;
 345 }
 346
 347 static void
 348 __nfulnl_flush(struct nfulnl_instance *inst)
 349 {
 350         /* timer holds a reference */
 351         if (del_timer(&inst->timer))
 352                 instance_put(inst);
 353         if (inst->skb)
 354                 __nfulnl_send(inst);
 355 }
 356
 357 static void
 358 nfulnl_timer(unsigned long data)
 359 {
 360         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 361
 362         spin_lock_bh(&inst->lock);
 363         if (inst->skb)
 364                 __nfulnl_send(inst);
 365         spin_unlock_bh(&inst->lock);
 366         instance_put(inst);
 367 }
 368
 369 /* This is an inline function, we don't really care about a long
 370  * list of arguments */
 371 static inline int
 372 __build_packet_message(struct nfulnl_instance *inst,
 373                         const struct sk_buff *skb,
 374                         unsigned int data_len,
 375                         u_int8_t pf,
 376                         unsigned int hooknum,
 377                         const struct net_device *indev,
 378                         const struct net_device *outdev,
 379                         const char *prefix, unsigned int plen)
 380 {
 381         struct nfulnl_msg_packet_hdr pmsg;
 382         struct nlmsghdr *nlh;
 383         struct nfgenmsg *nfmsg;
 384         sk_buff_data_t old_tail = inst->skb->tail;
 385
 386         nlh = NLMSG_PUT(inst->skb, 0, 0,
 387                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 388                         sizeof(struct nfgenmsg));
 389         nfmsg = NLMSG_DATA(nlh);
 390         nfmsg->nfgen_family = pf;
 391         nfmsg->version = NFNETLINK_V0;
 392         nfmsg->res_id = htons(inst->group_num);
 393
 394         pmsg.hw_protocol        = skb->protocol;
 395         pmsg.hook               = hooknum;
 396
 397         NLA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
 398
 399         if (prefix)
 400                 NLA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
 401
 402         if (indev) {
 403 #ifndef CONFIG_BRIDGE_NETFILTER
 404                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
 405                              htonl(indev->ifindex));
 406 #else
 407                 if (pf == PF_BRIDGE) {
 408                         /* Case 1: outdev is physical input device, we need to
 409                          * look for bridge group (when called from
 410                          * netfilter_bridge) */
 411                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 412                                      htonl(indev->ifindex));
 413                         /* this is the bridge group "brX" */
 414                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 415                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
 416                                      htonl(br_port_get_rcu(indev)->br->dev->ifindex));
 417                 } else {
 418                         /* Case 2: indev is bridge group, we need to look for
 419                          * physical device (when called from ipv4) */
 420                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
 421                                      htonl(indev->ifindex));
 422                         if (skb->nf_bridge && skb->nf_bridge->physindev)
 423                                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 424                                              htonl(skb->nf_bridge->physindev->ifindex));
 425                 }
 426 #endif
 427         }
 428
 429         if (outdev) {
 430 #ifndef CONFIG_BRIDGE_NETFILTER
 431                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
 432                              htonl(outdev->ifindex));
 433 #else
 434                 if (pf == PF_BRIDGE) {
 435                         /* Case 1: outdev is physical output device, we need to
 436                          * look for bridge group (when called from
 437                          * netfilter_bridge) */
 438                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 439                                      htonl(outdev->ifindex));
 440                         /* this is the bridge group "brX" */
 441                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 442                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
 443                                      htonl(br_port_get_rcu(outdev)->br->dev->ifindex));
 444                 } else {
 445                         /* Case 2: indev is a bridge group, we need to look
 446                          * for physical device (when called from ipv4) */
 447                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
 448                                      htonl(outdev->ifindex));
 449                         if (skb->nf_bridge && skb->nf_bridge->physoutdev)
 450                                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 451                                              htonl(skb->nf_bridge->physoutdev->ifindex));
 452                 }
 453 #endif
 454         }
 455
 456         if (skb->mark)
 457                 NLA_PUT_BE32(inst->skb, NFULA_MARK, htonl(skb->mark));
 458
 459         if (indev && skb->dev) {
 460                 struct nfulnl_msg_packet_hw phw;
 461                 int len = dev_parse_header(skb, phw.hw_addr);
 462                 if (len > 0) {
 463                         phw.hw_addrlen = htons(len);
 464                         NLA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
 465                 }
 466         }
 467
 468         if (indev && skb_mac_header_was_set(skb)) {
 469                 NLA_PUT_BE16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type));
 470                 NLA_PUT_BE16(inst->skb, NFULA_HWLEN,
 471                              htons(skb->dev->hard_header_len));
 472                 NLA_PUT(inst->skb, NFULA_HWHEADER, skb->dev->hard_header_len,
 473                         skb_mac_header(skb));
 474         }
 475
 476         if (skb->tstamp.tv64) {
 477                 struct nfulnl_msg_packet_timestamp ts;
 478                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 479                 ts.sec = cpu_to_be64(tv.tv_sec);
 480                 ts.usec = cpu_to_be64(tv.tv_usec);
 481
 482                 NLA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 483         }
 484
 485         /* UID */
 486         if (skb->sk) {
 487                 read_lock_bh(&skb->sk->sk_callback_lock);
 488                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 489                         struct file *file = skb->sk->sk_socket->file;
 490                         __be32 uid = htonl(file->f_cred->fsuid);
 491                         __be32 gid = htonl(file->f_cred->fsgid);
 492                         /* need to unlock here since NLA_PUT may goto */
 493                         read_unlock_bh(&skb->sk->sk_callback_lock);
 494                         NLA_PUT_BE32(inst->skb, NFULA_UID, uid);
 495                         NLA_PUT_BE32(inst->skb, NFULA_GID, gid);
 496                 } else
 497                         read_unlock_bh(&skb->sk->sk_callback_lock);
 498         }
 499
 500         /* local sequence number */
 501         if (inst->flags & NFULNL_CFG_F_SEQ)
 502                 NLA_PUT_BE32(inst->skb, NFULA_SEQ, htonl(inst->seq++));
 503
 504         /* global sequence number */
 505         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 506                 NLA_PUT_BE32(inst->skb, NFULA_SEQ_GLOBAL,
 507                              htonl(atomic_inc_return(&global_seq)));
 508
 509         if (data_len) {
 510                 struct nlattr *nla;
 511                 int size = nla_attr_size(data_len);
 512
 513                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
 514                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 515                         goto nlmsg_failure;
 516                 }
 517
 518                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 519                 nla->nla_type = NFULA_PAYLOAD;
 520                 nla->nla_len = size;
 521
 522                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 523                         BUG();
 524         }
 525
 526         nlh->nlmsg_len = inst->skb->tail - old_tail;
 527         return 0;
 528
 529 nlmsg_failure:
 530 nla_put_failure:
 531         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 532         return -1;
 533 }
 534
 535 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 536
 537 static struct nf_loginfo default_loginfo = {
 538         .type =         NF_LOG_TYPE_ULOG,
 539         .u = {
 540                 .ulog = {
 541                         .copy_len       = 0xffff,
 542                         .group          = 0,
 543                         .qthreshold     = 1,
 544                 },
 545         },
 546 };
 547
 548 /* log handler for internal netfilter logging api */
 549 void
 550 nfulnl_log_packet(u_int8_t pf,
 551                   unsigned int hooknum,
 552                   const struct sk_buff *skb,
 553                   const struct net_device *in,
 554                   const struct net_device *out,
 555                   const struct nf_loginfo *li_user,
 556                   const char *prefix)
 557 {
 558         unsigned int size, data_len;
 559         struct nfulnl_instance *inst;
 560         const struct nf_loginfo *li;
 561         unsigned int qthreshold;
 562         unsigned int plen;
 563
 564         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 565                 li = li_user;
 566         else
 567                 li = &default_loginfo;
 568
 569         inst = instance_lookup_get(li->u.ulog.group);
 570         if (!inst)
 571                 return;
 572
 573         plen = 0;
 574         if (prefix)
 575                 plen = strlen(prefix) + 1;
 576
 577         /* FIXME: do we want to make the size calculation conditional based on
 578          * what is actually present?  way more branches and checks, but more
 579          * memory efficient... */
 580         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 581                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 582                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 583                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 584 #ifdef CONFIG_BRIDGE_NETFILTER
 585                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 586                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 587 #endif
 588                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 589                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 590                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 591                 + nla_total_size(plen)                  /* prefix */
 592                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 593                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
 594
 595         if (in && skb_mac_header_was_set(skb)) {
 596                 size +=   nla_total_size(skb->dev->hard_header_len)
 597                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 598                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 599         }
 600
 601         spin_lock_bh(&inst->lock);
 602
 603         if (inst->flags & NFULNL_CFG_F_SEQ)
 604                 size += nla_total_size(sizeof(u_int32_t));
 605         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 606                 size += nla_total_size(sizeof(u_int32_t));
 607
 608         qthreshold = inst->qthreshold;
 609         /* per-rule qthreshold overrides per-instance */
 610         if (li->u.ulog.qthreshold)
 611                 if (qthreshold > li->u.ulog.qthreshold)
 612                         qthreshold = li->u.ulog.qthreshold;
 613
 614
 615         switch (inst->copy_mode) {
 616         case NFULNL_COPY_META:
 617         case NFULNL_COPY_NONE:
 618                 data_len = 0;
 619                 break;
 620
 621         case NFULNL_COPY_PACKET:
 622                 if (inst->copy_range == 0
 623                     || inst->copy_range > skb->len)
 624                         data_len = skb->len;
 625                 else
 626                         data_len = inst->copy_range;
 627
 628                 size += nla_total_size(data_len);
 629                 break;
 630
 631         case NFULNL_COPY_DISABLED:
 632         default:
 633                 goto unlock_and_release;
 634         }
 635
 636         if (inst->skb &&
 637             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
 638                 /* either the queue len is too high or we don't have
 639                  * enough room in the skb left. flush to userspace. */
 640                 __nfulnl_flush(inst);
 641         }
 642
 643         if (!inst->skb) {
 644                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
 645                 if (!inst->skb)
 646                         goto alloc_failure;
 647         }
 648
 649         inst->qlen++;
 650
 651         __build_packet_message(inst, skb, data_len, pf,
 652                                 hooknum, in, out, prefix, plen);
 653
 654         if (inst->qlen >= qthreshold)
 655                 __nfulnl_flush(inst);
 656         /* timer_pending always called within inst->lock, so there
 657          * is no chance of a race here */
 658         else if (!timer_pending(&inst->timer)) {
 659                 instance_get(inst);
 660                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 661                 add_timer(&inst->timer);
 662         }
 663
 664 unlock_and_release:
 665         spin_unlock_bh(&inst->lock);
 666         instance_put(inst);
 667         return;
 668
 669 alloc_failure:
 670         /* FIXME: statistics */
 671         goto unlock_and_release;
 672 }
 673 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
 674
 675 static int
 676 nfulnl_rcv_nl_event(struct notifier_block *this,
 677                    unsigned long event, void *ptr)
 678 {
 679         struct netlink_notify *n = ptr;
 680
 681         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 682                 int i;
 683
 684                 /* destroy all instances for this pid */
 685                 spin_lock_bh(&instances_lock);
 686                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 687                         struct hlist_node *tmp, *t2;
 688                         struct nfulnl_instance *inst;
 689                         struct hlist_head *head = &instance_table[i];
 690
 691                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 692                                 if ((net_eq(n->net, &init_net)) &&
 693                                     (n->pid == inst->peer_pid))
 694                                         __instance_destroy(inst);
 695                         }
 696                 }
 697                 spin_unlock_bh(&instances_lock);
 698         }
 699         return NOTIFY_DONE;
 700 }
 701
 702 static struct notifier_block nfulnl_rtnl_notifier = {
 703         .notifier_call  = nfulnl_rcv_nl_event,
 704 };
 705
 706 static int
 707 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 708                    const struct nlmsghdr *nlh,
 709                    const struct nlattr * const nfqa[])
 710 {
 711         return -ENOTSUPP;
 712 }
 713
 714 static struct nf_logger nfulnl_logger __read_mostly = {
 715         .name   = "nfnetlink_log",
 716         .logfn  = &nfulnl_log_packet,
 717         .me     = THIS_MODULE,
 718 };
 719
 720 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 721         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 722         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 723         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 724         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 725         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 726         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 727 };
 728
 729 static int
 730 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 731                    const struct nlmsghdr *nlh,
 732                    const struct nlattr * const nfula[])
 733 {
 734         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 735         u_int16_t group_num = ntohs(nfmsg->res_id);
 736         struct nfulnl_instance *inst;
 737         struct nfulnl_msg_config_cmd *cmd = NULL;
 738         int ret = 0;
 739
 740         if (nfula[NFULA_CFG_CMD]) {
 741                 u_int8_t pf = nfmsg->nfgen_family;
 742                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 743
 744                 /* Commands without queue context */
 745                 switch (cmd->command) {
 746                 case NFULNL_CFG_CMD_PF_BIND:
 747                         return nf_log_bind_pf(pf, &nfulnl_logger);
 748                 case NFULNL_CFG_CMD_PF_UNBIND:
 749                         nf_log_unbind_pf(pf);
 750                         return 0;
 751                 }
 752         }
 753
 754         inst = instance_lookup_get(group_num);
 755         if (inst && inst->peer_pid != NETLINK_CB(skb).pid) {
 756                 ret = -EPERM;
 757                 goto out_put;
 758         }
 759
 760         if (cmd != NULL) {
 761                 switch (cmd->command) {
 762                 case NFULNL_CFG_CMD_BIND:
 763                         if (inst) {
 764                                 ret = -EBUSY;
 765                                 goto out_put;
 766                         }
 767
 768                         inst = instance_create(group_num,
 769                                                NETLINK_CB(skb).pid);
 770                         if (IS_ERR(inst)) {
 771                                 ret = PTR_ERR(inst);
 772                                 goto out;
 773                         }
 774                         break;
 775                 case NFULNL_CFG_CMD_UNBIND:
 776                         if (!inst) {
 777                                 ret = -ENODEV;
 778                                 goto out;
 779                         }
 780
 781                         instance_destroy(inst);
 782                         goto out_put;
 783                 default:
 784                         ret = -ENOTSUPP;
 785                         break;
 786                 }
 787         }
 788
 789         if (nfula[NFULA_CFG_MODE]) {
 790                 struct nfulnl_msg_config_mode *params;
 791                 params = nla_data(nfula[NFULA_CFG_MODE]);
 792
 793                 if (!inst) {
 794                         ret = -ENODEV;
 795                         goto out;
 796                 }
 797                 nfulnl_set_mode(inst, params->copy_mode,
 798                                 ntohl(params->copy_range));
 799         }
 800
 801         if (nfula[NFULA_CFG_TIMEOUT]) {
 802                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 803
 804                 if (!inst) {
 805                         ret = -ENODEV;
 806                         goto out;
 807                 }
 808                 nfulnl_set_timeout(inst, ntohl(timeout));
 809         }
 810
 811         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 812                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 813
 814                 if (!inst) {
 815                         ret = -ENODEV;
 816                         goto out;
 817                 }
 818                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 819         }
 820
 821         if (nfula[NFULA_CFG_QTHRESH]) {
 822                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 823
 824                 if (!inst) {
 825                         ret = -ENODEV;
 826                         goto out;
 827                 }
 828                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 829         }
 830
 831         if (nfula[NFULA_CFG_FLAGS]) {
 832                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
 833
 834                 if (!inst) {
 835                         ret = -ENODEV;
 836                         goto out;
 837                 }
 838                 nfulnl_set_flags(inst, ntohs(flags));
 839         }
 840
 841 out_put:
 842         instance_put(inst);
 843 out:
 844         return ret;
 845 }
 846
 847 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 848         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 849                                     .attr_count = NFULA_MAX, },
 850         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 851                                     .attr_count = NFULA_CFG_MAX,
 852                                     .policy = nfula_cfg_policy },
 853 };
 854
 855 static const struct nfnetlink_subsystem nfulnl_subsys = {
 856         .name           = "log",
 857         .subsys_id      = NFNL_SUBSYS_ULOG,
 858         .cb_count       = NFULNL_MSG_MAX,
 859         .cb             = nfulnl_cb,
 860 };
 861
 862 #ifdef CONFIG_PROC_FS
 863 struct iter_state {
 864         unsigned int bucket;
 865 };
 866
 867 static struct hlist_node *get_first(struct iter_state *st)
 868 {
 869         if (!st)
 870                 return NULL;
 871
 872         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 873                 if (!hlist_empty(&instance_table[st->bucket]))
 874                         return rcu_dereference_bh(hlist_first_rcu(&instance_table[st->bucket]));
 875         }
 876         return NULL;
 877 }
 878
 879 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
 880 {
 881         h = rcu_dereference_bh(hlist_next_rcu(h));
 882         while (!h) {
 883                 if (++st->bucket >= INSTANCE_BUCKETS)
 884                         return NULL;
 885
 886                 h = rcu_dereference_bh(hlist_first_rcu(&instance_table[st->bucket]));
 887         }
 888         return h;
 889 }
 890
 891 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
 892 {
 893         struct hlist_node *head;
 894         head = get_first(st);
 895
 896         if (head)
 897                 while (pos && (head = get_next(st, head)))
 898                         pos--;
 899         return pos ? NULL : head;
 900 }
 901
 902 static void *seq_start(struct seq_file *seq, loff_t *pos)
 903         __acquires(rcu_bh)
 904 {
 905         rcu_read_lock_bh();
 906         return get_idx(seq->private, *pos);
 907 }
 908
 909 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 910 {
 911         (*pos)++;
 912         return get_next(s->private, v);
 913 }
 914
 915 static void seq_stop(struct seq_file *s, void *v)
 916         __releases(rcu_bh)
 917 {
 918         rcu_read_unlock_bh();
 919 }
 920
 921 static int seq_show(struct seq_file *s, void *v)
 922 {
 923         const struct nfulnl_instance *inst = v;
 924
 925         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 926                           inst->group_num,
 927                           inst->peer_pid, inst->qlen,
 928                           inst->copy_mode, inst->copy_range,
 929                           inst->flushtimeout, atomic_read(&inst->use));
 930 }
 931
 932 static const struct seq_operations nful_seq_ops = {
 933         .start  = seq_start,
 934         .next   = seq_next,
 935         .stop   = seq_stop,
 936         .show   = seq_show,
 937 };
 938
 939 static int nful_open(struct inode *inode, struct file *file)
 940 {
 941         return seq_open_private(file, &nful_seq_ops,
 942                         sizeof(struct iter_state));
 943 }
 944
 945 static const struct file_operations nful_file_ops = {
 946         .owner   = THIS_MODULE,
 947         .open    = nful_open,
 948         .read    = seq_read,
 949         .llseek  = seq_lseek,
 950         .release = seq_release_private,
 951 };
 952
 953 #endif /* PROC_FS */
 954
 955 static int __init nfnetlink_log_init(void)
 956 {
 957         int i, status = -ENOMEM;
 958
 959         for (i = 0; i < INSTANCE_BUCKETS; i++)
 960                 INIT_HLIST_HEAD(&instance_table[i]);
 961
 962         /* it's not really all that important to have a random value, so
 963          * we can do this from the init function, even if there hasn't
 964          * been that much entropy yet */
 965         get_random_bytes(&hash_init, sizeof(hash_init));
 966
 967         netlink_register_notifier(&nfulnl_rtnl_notifier);
 968         status = nfnetlink_subsys_register(&nfulnl_subsys);
 969         if (status < 0) {
 970                 printk(KERN_ERR "log: failed to create netlink socket\n");
 971                 goto cleanup_netlink_notifier;
 972         }
 973
 974         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
 975         if (status < 0) {
 976                 printk(KERN_ERR "log: failed to register logger\n");
 977                 goto cleanup_subsys;
 978         }
 979
 980 #ifdef CONFIG_PROC_FS
 981         if (!proc_create("nfnetlink_log", 0440,
 982                          proc_net_netfilter, &nful_file_ops))
 983                 goto cleanup_logger;
 984 #endif
 985         return status;
 986
 987 #ifdef CONFIG_PROC_FS
 988 cleanup_logger:
 989         nf_log_unregister(&nfulnl_logger);
 990 #endif
 991 cleanup_subsys:
 992         nfnetlink_subsys_unregister(&nfulnl_subsys);
 993 cleanup_netlink_notifier:
 994         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
 995         return status;
 996 }
 997
 998 static void __exit nfnetlink_log_fini(void)
 999 {
1000         nf_log_unregister(&nfulnl_logger);
1001 #ifdef CONFIG_PROC_FS
1002         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1003 #endif
1004         nfnetlink_subsys_unregister(&nfulnl_subsys);
1005         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1006 }
1007
1008 MODULE_DESCRIPTION("netfilter userspace logging");
1009 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1010 MODULE_LICENSE("GPL");
1011 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1012
1013 module_init(nfnetlink_log_init);
1014 module_exit(nfnetlink_log_fini);