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                 NLMSG_PUT(inst->skb, 0, 0,
 331                           NLMSG_DONE,
 332                           sizeof(struct nfgenmsg));
 333
 334         status = nfnetlink_unicast(inst->skb, &init_net, inst->peer_pid,
 335                                    MSG_DONTWAIT);
 336
 337         inst->qlen = 0;
 338         inst->skb = NULL;
 339
 340 nlmsg_failure:
 341         return status;
 342 }
 343
 344 static void
 345 __nfulnl_flush(struct nfulnl_instance *inst)
 346 {
 347         /* timer holds a reference */
 348         if (del_timer(&inst->timer))
 349                 instance_put(inst);
 350         if (inst->skb)
 351                 __nfulnl_send(inst);
 352 }
 353
 354 static void
 355 nfulnl_timer(unsigned long data)
 356 {
 357         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 358
 359         spin_lock_bh(&inst->lock);
 360         if (inst->skb)
 361                 __nfulnl_send(inst);
 362         spin_unlock_bh(&inst->lock);
 363         instance_put(inst);
 364 }
 365
 366 /* This is an inline function, we don't really care about a long
 367  * list of arguments */
 368 static inline int
 369 __build_packet_message(struct nfulnl_instance *inst,
 370                         const struct sk_buff *skb,
 371                         unsigned int data_len,
 372                         u_int8_t pf,
 373                         unsigned int hooknum,
 374                         const struct net_device *indev,
 375                         const struct net_device *outdev,
 376                         const char *prefix, unsigned int plen)
 377 {
 378         struct nfulnl_msg_packet_hdr pmsg;
 379         struct nlmsghdr *nlh;
 380         struct nfgenmsg *nfmsg;
 381         sk_buff_data_t old_tail = inst->skb->tail;
 382
 383         nlh = NLMSG_PUT(inst->skb, 0, 0,
 384                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 385                         sizeof(struct nfgenmsg));
 386         nfmsg = NLMSG_DATA(nlh);
 387         nfmsg->nfgen_family = pf;
 388         nfmsg->version = NFNETLINK_V0;
 389         nfmsg->res_id = htons(inst->group_num);
 390
 391         pmsg.hw_protocol        = skb->protocol;
 392         pmsg.hook               = hooknum;
 393
 394         NLA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
 395
 396         if (prefix)
 397                 NLA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
 398
 399         if (indev) {
 400 #ifndef CONFIG_BRIDGE_NETFILTER
 401                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
 402                              htonl(indev->ifindex));
 403 #else
 404                 if (pf == PF_BRIDGE) {
 405                         /* Case 1: outdev is physical input device, we need to
 406                          * look for bridge group (when called from
 407                          * netfilter_bridge) */
 408                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 409                                      htonl(indev->ifindex));
 410                         /* this is the bridge group "brX" */
 411                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 412                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
 413                                      htonl(br_port_get_rcu(indev)->br->dev->ifindex));
 414                 } else {
 415                         /* Case 2: indev is bridge group, we need to look for
 416                          * physical device (when called from ipv4) */
 417                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
 418                                      htonl(indev->ifindex));
 419                         if (skb->nf_bridge && skb->nf_bridge->physindev)
 420                                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 421                                              htonl(skb->nf_bridge->physindev->ifindex));
 422                 }
 423 #endif
 424         }
 425
 426         if (outdev) {
 427 #ifndef CONFIG_BRIDGE_NETFILTER
 428                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
 429                              htonl(outdev->ifindex));
 430 #else
 431                 if (pf == PF_BRIDGE) {
 432                         /* Case 1: outdev is physical output device, we need to
 433                          * look for bridge group (when called from
 434                          * netfilter_bridge) */
 435                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 436                                      htonl(outdev->ifindex));
 437                         /* this is the bridge group "brX" */
 438                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 439                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
 440                                      htonl(br_port_get_rcu(outdev)->br->dev->ifindex));
 441                 } else {
 442                         /* Case 2: indev is a bridge group, we need to look
 443                          * for physical device (when called from ipv4) */
 444                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
 445                                      htonl(outdev->ifindex));
 446                         if (skb->nf_bridge && skb->nf_bridge->physoutdev)
 447                                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 448                                              htonl(skb->nf_bridge->physoutdev->ifindex));
 449                 }
 450 #endif
 451         }
 452
 453         if (skb->mark)
 454                 NLA_PUT_BE32(inst->skb, NFULA_MARK, htonl(skb->mark));
 455
 456         if (indev && skb->dev &&
 457             skb->mac_header != skb->network_header) {
 458                 struct nfulnl_msg_packet_hw phw;
 459                 int len = dev_parse_header(skb, phw.hw_addr);
 460                 if (len > 0) {
 461                         phw.hw_addrlen = htons(len);
 462                         NLA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
 463                 }
 464         }
 465
 466         if (indev && skb_mac_header_was_set(skb)) {
 467                 NLA_PUT_BE16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type));
 468                 NLA_PUT_BE16(inst->skb, NFULA_HWLEN,
 469                              htons(skb->dev->hard_header_len));
 470                 NLA_PUT(inst->skb, NFULA_HWHEADER, skb->dev->hard_header_len,
 471                         skb_mac_header(skb));
 472         }
 473
 474         if (skb->tstamp.tv64) {
 475                 struct nfulnl_msg_packet_timestamp ts;
 476                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 477                 ts.sec = cpu_to_be64(tv.tv_sec);
 478                 ts.usec = cpu_to_be64(tv.tv_usec);
 479
 480                 NLA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 481         }
 482
 483         /* UID */
 484         if (skb->sk) {
 485                 read_lock_bh(&skb->sk->sk_callback_lock);
 486                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 487                         struct file *file = skb->sk->sk_socket->file;
 488                         __be32 uid = htonl(file->f_cred->fsuid);
 489                         __be32 gid = htonl(file->f_cred->fsgid);
 490                         /* need to unlock here since NLA_PUT may goto */
 491                         read_unlock_bh(&skb->sk->sk_callback_lock);
 492                         NLA_PUT_BE32(inst->skb, NFULA_UID, uid);
 493                         NLA_PUT_BE32(inst->skb, NFULA_GID, gid);
 494                 } else
 495                         read_unlock_bh(&skb->sk->sk_callback_lock);
 496         }
 497
 498         /* local sequence number */
 499         if (inst->flags & NFULNL_CFG_F_SEQ)
 500                 NLA_PUT_BE32(inst->skb, NFULA_SEQ, htonl(inst->seq++));
 501
 502         /* global sequence number */
 503         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 504                 NLA_PUT_BE32(inst->skb, NFULA_SEQ_GLOBAL,
 505                              htonl(atomic_inc_return(&global_seq)));
 506
 507         if (data_len) {
 508                 struct nlattr *nla;
 509                 int size = nla_attr_size(data_len);
 510
 511                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
 512                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 513                         goto nlmsg_failure;
 514                 }
 515
 516                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 517                 nla->nla_type = NFULA_PAYLOAD;
 518                 nla->nla_len = size;
 519
 520                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 521                         BUG();
 522         }
 523
 524         nlh->nlmsg_len = inst->skb->tail - old_tail;
 525         return 0;
 526
 527 nlmsg_failure:
 528 nla_put_failure:
 529         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 530         return -1;
 531 }
 532
 533 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 534
 535 static struct nf_loginfo default_loginfo = {
 536         .type =         NF_LOG_TYPE_ULOG,
 537         .u = {
 538                 .ulog = {
 539                         .copy_len       = 0xffff,
 540                         .group          = 0,
 541                         .qthreshold     = 1,
 542                 },
 543         },
 544 };
 545
 546 /* log handler for internal netfilter logging api */
 547 void
 548 nfulnl_log_packet(u_int8_t pf,
 549                   unsigned int hooknum,
 550                   const struct sk_buff *skb,
 551                   const struct net_device *in,
 552                   const struct net_device *out,
 553                   const struct nf_loginfo *li_user,
 554                   const char *prefix)
 555 {
 556         unsigned int size, data_len;
 557         struct nfulnl_instance *inst;
 558         const struct nf_loginfo *li;
 559         unsigned int qthreshold;
 560         unsigned int plen;
 561
 562         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 563                 li = li_user;
 564         else
 565                 li = &default_loginfo;
 566
 567         inst = instance_lookup_get(li->u.ulog.group);
 568         if (!inst)
 569                 return;
 570
 571         plen = 0;
 572         if (prefix)
 573                 plen = strlen(prefix) + 1;
 574
 575         /* FIXME: do we want to make the size calculation conditional based on
 576          * what is actually present?  way more branches and checks, but more
 577          * memory efficient... */
 578         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 579                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 580                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 581                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 582 #ifdef CONFIG_BRIDGE_NETFILTER
 583                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 584                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 585 #endif
 586                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 587                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 588                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 589                 + nla_total_size(plen)                  /* prefix */
 590                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 591                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
 592
 593         if (in && skb_mac_header_was_set(skb)) {
 594                 size +=   nla_total_size(skb->dev->hard_header_len)
 595                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 596                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 597         }
 598
 599         spin_lock_bh(&inst->lock);
 600
 601         if (inst->flags & NFULNL_CFG_F_SEQ)
 602                 size += nla_total_size(sizeof(u_int32_t));
 603         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 604                 size += nla_total_size(sizeof(u_int32_t));
 605
 606         qthreshold = inst->qthreshold;
 607         /* per-rule qthreshold overrides per-instance */
 608         if (li->u.ulog.qthreshold)
 609                 if (qthreshold > li->u.ulog.qthreshold)
 610                         qthreshold = li->u.ulog.qthreshold;
 611
 612
 613         switch (inst->copy_mode) {
 614         case NFULNL_COPY_META:
 615         case NFULNL_COPY_NONE:
 616                 data_len = 0;
 617                 break;
 618
 619         case NFULNL_COPY_PACKET:
 620                 if (inst->copy_range == 0
 621                     || inst->copy_range > skb->len)
 622                         data_len = skb->len;
 623                 else
 624                         data_len = inst->copy_range;
 625
 626                 size += nla_total_size(data_len);
 627                 break;
 628
 629         case NFULNL_COPY_DISABLED:
 630         default:
 631                 goto unlock_and_release;
 632         }
 633
 634         if (inst->skb &&
 635             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
 636                 /* either the queue len is too high or we don't have
 637                  * enough room in the skb left. flush to userspace. */
 638                 __nfulnl_flush(inst);
 639         }
 640
 641         if (!inst->skb) {
 642                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
 643                 if (!inst->skb)
 644                         goto alloc_failure;
 645         }
 646
 647         inst->qlen++;
 648
 649         __build_packet_message(inst, skb, data_len, pf,
 650                                 hooknum, in, out, prefix, plen);
 651
 652         if (inst->qlen >= qthreshold)
 653                 __nfulnl_flush(inst);
 654         /* timer_pending always called within inst->lock, so there
 655          * is no chance of a race here */
 656         else if (!timer_pending(&inst->timer)) {
 657                 instance_get(inst);
 658                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 659                 add_timer(&inst->timer);
 660         }
 661
 662 unlock_and_release:
 663         spin_unlock_bh(&inst->lock);
 664         instance_put(inst);
 665         return;
 666
 667 alloc_failure:
 668         /* FIXME: statistics */
 669         goto unlock_and_release;
 670 }
 671 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
 672
 673 static int
 674 nfulnl_rcv_nl_event(struct notifier_block *this,
 675                    unsigned long event, void *ptr)
 676 {
 677         struct netlink_notify *n = ptr;
 678
 679         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 680                 int i;
 681
 682                 /* destroy all instances for this pid */
 683                 spin_lock_bh(&instances_lock);
 684                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 685                         struct hlist_node *tmp, *t2;
 686                         struct nfulnl_instance *inst;
 687                         struct hlist_head *head = &instance_table[i];
 688
 689                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 690                                 if ((net_eq(n->net, &init_net)) &&
 691                                     (n->pid == inst->peer_pid))
 692                                         __instance_destroy(inst);
 693                         }
 694                 }
 695                 spin_unlock_bh(&instances_lock);
 696         }
 697         return NOTIFY_DONE;
 698 }
 699
 700 static struct notifier_block nfulnl_rtnl_notifier = {
 701         .notifier_call  = nfulnl_rcv_nl_event,
 702 };
 703
 704 static int
 705 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 706                    const struct nlmsghdr *nlh,
 707                    const struct nlattr * const nfqa[])
 708 {
 709         return -ENOTSUPP;
 710 }
 711
 712 static struct nf_logger nfulnl_logger __read_mostly = {
 713         .name   = "nfnetlink_log",
 714         .logfn  = &nfulnl_log_packet,
 715         .me     = THIS_MODULE,
 716 };
 717
 718 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 719         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 720         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 721         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 722         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 723         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 724         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 725 };
 726
 727 static int
 728 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 729                    const struct nlmsghdr *nlh,
 730                    const struct nlattr * const nfula[])
 731 {
 732         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 733         u_int16_t group_num = ntohs(nfmsg->res_id);
 734         struct nfulnl_instance *inst;
 735         struct nfulnl_msg_config_cmd *cmd = NULL;
 736         int ret = 0;
 737
 738         if (nfula[NFULA_CFG_CMD]) {
 739                 u_int8_t pf = nfmsg->nfgen_family;
 740                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 741
 742                 /* Commands without queue context */
 743                 switch (cmd->command) {
 744                 case NFULNL_CFG_CMD_PF_BIND:
 745                         return nf_log_bind_pf(pf, &nfulnl_logger);
 746                 case NFULNL_CFG_CMD_PF_UNBIND:
 747                         nf_log_unbind_pf(pf);
 748                         return 0;
 749                 }
 750         }
 751
 752         inst = instance_lookup_get(group_num);
 753         if (inst && inst->peer_pid != NETLINK_CB(skb).pid) {
 754                 ret = -EPERM;
 755                 goto out_put;
 756         }
 757
 758         if (cmd != NULL) {
 759                 switch (cmd->command) {
 760                 case NFULNL_CFG_CMD_BIND:
 761                         if (inst) {
 762                                 ret = -EBUSY;
 763                                 goto out_put;
 764                         }
 765
 766                         inst = instance_create(group_num,
 767                                                NETLINK_CB(skb).pid);
 768                         if (IS_ERR(inst)) {
 769                                 ret = PTR_ERR(inst);
 770                                 goto out;
 771                         }
 772                         break;
 773                 case NFULNL_CFG_CMD_UNBIND:
 774                         if (!inst) {
 775                                 ret = -ENODEV;
 776                                 goto out;
 777                         }
 778
 779                         instance_destroy(inst);
 780                         goto out_put;
 781                 default:
 782                         ret = -ENOTSUPP;
 783                         break;
 784                 }
 785         }
 786
 787         if (nfula[NFULA_CFG_MODE]) {
 788                 struct nfulnl_msg_config_mode *params;
 789                 params = nla_data(nfula[NFULA_CFG_MODE]);
 790
 791                 if (!inst) {
 792                         ret = -ENODEV;
 793                         goto out;
 794                 }
 795                 nfulnl_set_mode(inst, params->copy_mode,
 796                                 ntohl(params->copy_range));
 797         }
 798
 799         if (nfula[NFULA_CFG_TIMEOUT]) {
 800                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 801
 802                 if (!inst) {
 803                         ret = -ENODEV;
 804                         goto out;
 805                 }
 806                 nfulnl_set_timeout(inst, ntohl(timeout));
 807         }
 808
 809         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 810                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 811
 812                 if (!inst) {
 813                         ret = -ENODEV;
 814                         goto out;
 815                 }
 816                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 817         }
 818
 819         if (nfula[NFULA_CFG_QTHRESH]) {
 820                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 821
 822                 if (!inst) {
 823                         ret = -ENODEV;
 824                         goto out;
 825                 }
 826                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 827         }
 828
 829         if (nfula[NFULA_CFG_FLAGS]) {
 830                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
 831
 832                 if (!inst) {
 833                         ret = -ENODEV;
 834                         goto out;
 835                 }
 836                 nfulnl_set_flags(inst, ntohs(flags));
 837         }
 838
 839 out_put:
 840         instance_put(inst);
 841 out:
 842         return ret;
 843 }
 844
 845 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 846         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 847                                     .attr_count = NFULA_MAX, },
 848         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 849                                     .attr_count = NFULA_CFG_MAX,
 850                                     .policy = nfula_cfg_policy },
 851 };
 852
 853 static const struct nfnetlink_subsystem nfulnl_subsys = {
 854         .name           = "log",
 855         .subsys_id      = NFNL_SUBSYS_ULOG,
 856         .cb_count       = NFULNL_MSG_MAX,
 857         .cb             = nfulnl_cb,
 858 };
 859
 860 #ifdef CONFIG_PROC_FS
 861 struct iter_state {
 862         unsigned int bucket;
 863 };
 864
 865 static struct hlist_node *get_first(struct iter_state *st)
 866 {
 867         if (!st)
 868                 return NULL;
 869
 870         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 871                 if (!hlist_empty(&instance_table[st->bucket]))
 872                         return rcu_dereference_bh(hlist_first_rcu(&instance_table[st->bucket]));
 873         }
 874         return NULL;
 875 }
 876
 877 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
 878 {
 879         h = rcu_dereference_bh(hlist_next_rcu(h));
 880         while (!h) {
 881                 if (++st->bucket >= INSTANCE_BUCKETS)
 882                         return NULL;
 883
 884                 h = rcu_dereference_bh(hlist_first_rcu(&instance_table[st->bucket]));
 885         }
 886         return h;
 887 }
 888
 889 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
 890 {
 891         struct hlist_node *head;
 892         head = get_first(st);
 893
 894         if (head)
 895                 while (pos && (head = get_next(st, head)))
 896                         pos--;
 897         return pos ? NULL : head;
 898 }
 899
 900 static void *seq_start(struct seq_file *seq, loff_t *pos)
 901         __acquires(rcu_bh)
 902 {
 903         rcu_read_lock_bh();
 904         return get_idx(seq->private, *pos);
 905 }
 906
 907 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 908 {
 909         (*pos)++;
 910         return get_next(s->private, v);
 911 }
 912
 913 static void seq_stop(struct seq_file *s, void *v)
 914         __releases(rcu_bh)
 915 {
 916         rcu_read_unlock_bh();
 917 }
 918
 919 static int seq_show(struct seq_file *s, void *v)
 920 {
 921         const struct nfulnl_instance *inst = v;
 922
 923         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 924                           inst->group_num,
 925                           inst->peer_pid, inst->qlen,
 926                           inst->copy_mode, inst->copy_range,
 927                           inst->flushtimeout, atomic_read(&inst->use));
 928 }
 929
 930 static const struct seq_operations nful_seq_ops = {
 931         .start  = seq_start,
 932         .next   = seq_next,
 933         .stop   = seq_stop,
 934         .show   = seq_show,
 935 };
 936
 937 static int nful_open(struct inode *inode, struct file *file)
 938 {
 939         return seq_open_private(file, &nful_seq_ops,
 940                         sizeof(struct iter_state));
 941 }
 942
 943 static const struct file_operations nful_file_ops = {
 944         .owner   = THIS_MODULE,
 945         .open    = nful_open,
 946         .read    = seq_read,
 947         .llseek  = seq_lseek,
 948         .release = seq_release_private,
 949 };
 950
 951 #endif /* PROC_FS */
 952
 953 static int __init nfnetlink_log_init(void)
 954 {
 955         int i, status = -ENOMEM;
 956
 957         for (i = 0; i < INSTANCE_BUCKETS; i++)
 958                 INIT_HLIST_HEAD(&instance_table[i]);
 959
 960         /* it's not really all that important to have a random value, so
 961          * we can do this from the init function, even if there hasn't
 962          * been that much entropy yet */
 963         get_random_bytes(&hash_init, sizeof(hash_init));
 964
 965         netlink_register_notifier(&nfulnl_rtnl_notifier);
 966         status = nfnetlink_subsys_register(&nfulnl_subsys);
 967         if (status < 0) {
 968                 printk(KERN_ERR "log: failed to create netlink socket\n");
 969                 goto cleanup_netlink_notifier;
 970         }
 971
 972         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
 973         if (status < 0) {
 974                 printk(KERN_ERR "log: failed to register logger\n");
 975                 goto cleanup_subsys;
 976         }
 977
 978 #ifdef CONFIG_PROC_FS
 979         if (!proc_create("nfnetlink_log", 0440,
 980                          proc_net_netfilter, &nful_file_ops))
 981                 goto cleanup_logger;
 982 #endif
 983         return status;
 984
 985 #ifdef CONFIG_PROC_FS
 986 cleanup_logger:
 987         nf_log_unregister(&nfulnl_logger);
 988 #endif
 989 cleanup_subsys:
 990         nfnetlink_subsys_unregister(&nfulnl_subsys);
 991 cleanup_netlink_notifier:
 992         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
 993         return status;
 994 }
 995
 996 static void __exit nfnetlink_log_fini(void)
 997 {
 998         nf_log_unregister(&nfulnl_logger);
 999 #ifdef CONFIG_PROC_FS
1000         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1001 #endif
1002         nfnetlink_subsys_unregister(&nfulnl_subsys);
1003         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1004 }
1005
1006 MODULE_DESCRIPTION("netfilter userspace logging");
1007 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1008 MODULE_LICENSE("GPL");
1009 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1010
1011 module_init(nfnetlink_log_init);
1012 module_exit(nfnetlink_log_fini);