Pileus Git - ~andy/linux/blob - 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         struct user_namespace *peer_user_ns;    /* User namespace of the peer process */
  59         int peer_portid;                        /* PORTID of the peer process */
  60
  61         /* configurable parameters */
  62         unsigned int flushtimeout;      /* timeout until queue flush */
  63         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  64         unsigned int qthreshold;        /* threshold of the queue */
  65         u_int32_t copy_range;
  66         u_int32_t seq;                  /* instance-local sequential counter */
  67         u_int16_t group_num;            /* number of this queue */
  68         u_int16_t flags;
  69         u_int8_t copy_mode;
  70         struct rcu_head rcu;
  71 };
  72
  73 static DEFINE_SPINLOCK(instances_lock);
  74 static atomic_t global_seq;
  75
  76 #define INSTANCE_BUCKETS        16
  77 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  78 static unsigned int hash_init;
  79
  80 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  81 {
  82         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  83 }
  84
  85 static struct nfulnl_instance *
  86 __instance_lookup(u_int16_t group_num)
  87 {
  88         struct hlist_head *head;
  89         struct hlist_node *pos;
  90         struct nfulnl_instance *inst;
  91
  92         head = &instance_table[instance_hashfn(group_num)];
  93         hlist_for_each_entry_rcu(inst, pos, head, hlist) {
  94                 if (inst->group_num == group_num)
  95                         return inst;
  96         }
  97         return NULL;
  98 }
  99
 100 static inline void
 101 instance_get(struct nfulnl_instance *inst)
 102 {
 103         atomic_inc(&inst->use);
 104 }
 105
 106 static struct nfulnl_instance *
 107 instance_lookup_get(u_int16_t group_num)
 108 {
 109         struct nfulnl_instance *inst;
 110
 111         rcu_read_lock_bh();
 112         inst = __instance_lookup(group_num);
 113         if (inst && !atomic_inc_not_zero(&inst->use))
 114                 inst = NULL;
 115         rcu_read_unlock_bh();
 116
 117         return inst;
 118 }
 119
 120 static void nfulnl_instance_free_rcu(struct rcu_head *head)
 121 {
 122         kfree(container_of(head, struct nfulnl_instance, rcu));
 123         module_put(THIS_MODULE);
 124 }
 125
 126 static void
 127 instance_put(struct nfulnl_instance *inst)
 128 {
 129         if (inst && atomic_dec_and_test(&inst->use))
 130                 call_rcu_bh(&inst->rcu, nfulnl_instance_free_rcu);
 131 }
 132
 133 static void nfulnl_timer(unsigned long data);
 134
 135 static struct nfulnl_instance *
 136 instance_create(u_int16_t group_num, int portid, struct user_namespace *user_ns)
 137 {
 138         struct nfulnl_instance *inst;
 139         int err;
 140
 141         spin_lock_bh(&instances_lock);
 142         if (__instance_lookup(group_num)) {
 143                 err = -EEXIST;
 144                 goto out_unlock;
 145         }
 146
 147         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 148         if (!inst) {
 149                 err = -ENOMEM;
 150                 goto out_unlock;
 151         }
 152
 153         if (!try_module_get(THIS_MODULE)) {
 154                 kfree(inst);
 155                 err = -EAGAIN;
 156                 goto out_unlock;
 157         }
 158
 159         INIT_HLIST_NODE(&inst->hlist);
 160         spin_lock_init(&inst->lock);
 161         /* needs to be two, since we _put() after creation */
 162         atomic_set(&inst->use, 2);
 163
 164         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
 165
 166         inst->peer_user_ns = user_ns;
 167         inst->peer_portid = portid;
 168         inst->group_num = group_num;
 169
 170         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 171         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 172         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 173         inst->copy_mode         = NFULNL_COPY_PACKET;
 174         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
 175
 176         hlist_add_head_rcu(&inst->hlist,
 177                        &instance_table[instance_hashfn(group_num)]);
 178
 179         spin_unlock_bh(&instances_lock);
 180
 181         return inst;
 182
 183 out_unlock:
 184         spin_unlock_bh(&instances_lock);
 185         return ERR_PTR(err);
 186 }
 187
 188 static void __nfulnl_flush(struct nfulnl_instance *inst);
 189
 190 /* called with BH disabled */
 191 static void
 192 __instance_destroy(struct nfulnl_instance *inst)
 193 {
 194         /* first pull it out of the global list */
 195         hlist_del_rcu(&inst->hlist);
 196
 197         /* then flush all pending packets from skb */
 198
 199         spin_lock(&inst->lock);
 200
 201         /* lockless readers wont be able to use us */
 202         inst->copy_mode = NFULNL_COPY_DISABLED;
 203
 204         if (inst->skb)
 205                 __nfulnl_flush(inst);
 206         spin_unlock(&inst->lock);
 207
 208         /* and finally put the refcount */
 209         instance_put(inst);
 210 }
 211
 212 static inline void
 213 instance_destroy(struct nfulnl_instance *inst)
 214 {
 215         spin_lock_bh(&instances_lock);
 216         __instance_destroy(inst);
 217         spin_unlock_bh(&instances_lock);
 218 }
 219
 220 static int
 221 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 222                   unsigned int range)
 223 {
 224         int status = 0;
 225
 226         spin_lock_bh(&inst->lock);
 227
 228         switch (mode) {
 229         case NFULNL_COPY_NONE:
 230         case NFULNL_COPY_META:
 231                 inst->copy_mode = mode;
 232                 inst->copy_range = 0;
 233                 break;
 234
 235         case NFULNL_COPY_PACKET:
 236                 inst->copy_mode = mode;
 237                 inst->copy_range = min_t(unsigned int,
 238                                          range, NFULNL_COPY_RANGE_MAX);
 239                 break;
 240
 241         default:
 242                 status = -EINVAL;
 243                 break;
 244         }
 245
 246         spin_unlock_bh(&inst->lock);
 247
 248         return status;
 249 }
 250
 251 static int
 252 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 253 {
 254         int status;
 255
 256         spin_lock_bh(&inst->lock);
 257         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 258                 status = -ERANGE;
 259         else if (nlbufsiz > 131072)
 260                 status = -ERANGE;
 261         else {
 262                 inst->nlbufsiz = nlbufsiz;
 263                 status = 0;
 264         }
 265         spin_unlock_bh(&inst->lock);
 266
 267         return status;
 268 }
 269
 270 static int
 271 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 272 {
 273         spin_lock_bh(&inst->lock);
 274         inst->flushtimeout = timeout;
 275         spin_unlock_bh(&inst->lock);
 276
 277         return 0;
 278 }
 279
 280 static int
 281 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 282 {
 283         spin_lock_bh(&inst->lock);
 284         inst->qthreshold = qthresh;
 285         spin_unlock_bh(&inst->lock);
 286
 287         return 0;
 288 }
 289
 290 static int
 291 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 292 {
 293         spin_lock_bh(&inst->lock);
 294         inst->flags = flags;
 295         spin_unlock_bh(&inst->lock);
 296
 297         return 0;
 298 }
 299
 300 static struct sk_buff *
 301 nfulnl_alloc_skb(unsigned int inst_size, unsigned int pkt_size)
 302 {
 303         struct sk_buff *skb;
 304         unsigned int n;
 305
 306         /* alloc skb which should be big enough for a whole multipart
 307          * message.  WARNING: has to be <= 128k due to slab restrictions */
 308
 309         n = max(inst_size, pkt_size);
 310         skb = alloc_skb(n, GFP_ATOMIC);
 311         if (!skb) {
 312                 if (n > pkt_size) {
 313                         /* try to allocate only as much as we need for current
 314                          * packet */
 315
 316                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 317                         if (!skb)
 318                                 pr_err("nfnetlink_log: can't even alloc %u bytes\n",
 319                                        pkt_size);
 320                 }
 321         }
 322
 323         return skb;
 324 }
 325
 326 static int
 327 __nfulnl_send(struct nfulnl_instance *inst)
 328 {
 329         int status = -1;
 330
 331         if (inst->qlen > 1) {
 332                 struct nlmsghdr *nlh = nlmsg_put(inst->skb, 0, 0,
 333                                                  NLMSG_DONE,
 334                                                  sizeof(struct nfgenmsg),
 335                                                  0);
 336                 if (!nlh)
 337                         goto out;
 338         }
 339         status = nfnetlink_unicast(inst->skb, &init_net, inst->peer_portid,
 340                                    MSG_DONTWAIT);
 341
 342         inst->qlen = 0;
 343         inst->skb = NULL;
 344 out:
 345         return status;
 346 }
 347
 348 static void
 349 __nfulnl_flush(struct nfulnl_instance *inst)
 350 {
 351         /* timer holds a reference */
 352         if (del_timer(&inst->timer))
 353                 instance_put(inst);
 354         if (inst->skb)
 355                 __nfulnl_send(inst);
 356 }
 357
 358 static void
 359 nfulnl_timer(unsigned long data)
 360 {
 361         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 362
 363         spin_lock_bh(&inst->lock);
 364         if (inst->skb)
 365                 __nfulnl_send(inst);
 366         spin_unlock_bh(&inst->lock);
 367         instance_put(inst);
 368 }
 369
 370 /* This is an inline function, we don't really care about a long
 371  * list of arguments */
 372 static inline int
 373 __build_packet_message(struct nfulnl_instance *inst,
 374                         const struct sk_buff *skb,
 375                         unsigned int data_len,
 376                         u_int8_t pf,
 377                         unsigned int hooknum,
 378                         const struct net_device *indev,
 379                         const struct net_device *outdev,
 380                         const char *prefix, unsigned int plen)
 381 {
 382         struct nfulnl_msg_packet_hdr pmsg;
 383         struct nlmsghdr *nlh;
 384         struct nfgenmsg *nfmsg;
 385         sk_buff_data_t old_tail = inst->skb->tail;
 386         struct sock *sk;
 387
 388         nlh = nlmsg_put(inst->skb, 0, 0,
 389                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 390                         sizeof(struct nfgenmsg), 0);
 391         if (!nlh)
 392                 return -1;
 393         nfmsg = nlmsg_data(nlh);
 394         nfmsg->nfgen_family = pf;
 395         nfmsg->version = NFNETLINK_V0;
 396         nfmsg->res_id = htons(inst->group_num);
 397
 398         pmsg.hw_protocol        = skb->protocol;
 399         pmsg.hook               = hooknum;
 400
 401         if (nla_put(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg))
 402                 goto nla_put_failure;
 403
 404         if (prefix &&
 405             nla_put(inst->skb, NFULA_PREFIX, plen, prefix))
 406                 goto nla_put_failure;
 407
 408         if (indev) {
 409 #ifndef CONFIG_BRIDGE_NETFILTER
 410                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 411                                  htonl(indev->ifindex)))
 412                         goto nla_put_failure;
 413 #else
 414                 if (pf == PF_BRIDGE) {
 415                         /* Case 1: outdev is physical input device, we need to
 416                          * look for bridge group (when called from
 417                          * netfilter_bridge) */
 418                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 419                                          htonl(indev->ifindex)) ||
 420                         /* this is the bridge group "brX" */
 421                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 422                             nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 423                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 424                                 goto nla_put_failure;
 425                 } else {
 426                         /* Case 2: indev is bridge group, we need to look for
 427                          * physical device (when called from ipv4) */
 428                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 429                                          htonl(indev->ifindex)))
 430                                 goto nla_put_failure;
 431                         if (skb->nf_bridge && skb->nf_bridge->physindev &&
 432                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 433                                          htonl(skb->nf_bridge->physindev->ifindex)))
 434                                 goto nla_put_failure;
 435                 }
 436 #endif
 437         }
 438
 439         if (outdev) {
 440 #ifndef CONFIG_BRIDGE_NETFILTER
 441                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 442                                  htonl(outdev->ifindex)))
 443                         goto nla_put_failure;
 444 #else
 445                 if (pf == PF_BRIDGE) {
 446                         /* Case 1: outdev is physical output device, we need to
 447                          * look for bridge group (when called from
 448                          * netfilter_bridge) */
 449                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 450                                          htonl(outdev->ifindex)) ||
 451                         /* this is the bridge group "brX" */
 452                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 453                             nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 454                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 455                                 goto nla_put_failure;
 456                 } else {
 457                         /* Case 2: indev is a bridge group, we need to look
 458                          * for physical device (when called from ipv4) */
 459                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 460                                          htonl(outdev->ifindex)))
 461                                 goto nla_put_failure;
 462                         if (skb->nf_bridge && skb->nf_bridge->physoutdev &&
 463                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 464                                          htonl(skb->nf_bridge->physoutdev->ifindex)))
 465                                 goto nla_put_failure;
 466                 }
 467 #endif
 468         }
 469
 470         if (skb->mark &&
 471             nla_put_be32(inst->skb, NFULA_MARK, htonl(skb->mark)))
 472                 goto nla_put_failure;
 473
 474         if (indev && skb->dev &&
 475             skb->mac_header != skb->network_header) {
 476                 struct nfulnl_msg_packet_hw phw;
 477                 int len = dev_parse_header(skb, phw.hw_addr);
 478                 if (len > 0) {
 479                         phw.hw_addrlen = htons(len);
 480                         if (nla_put(inst->skb, NFULA_HWADDR, sizeof(phw), &phw))
 481                                 goto nla_put_failure;
 482                 }
 483         }
 484
 485         if (indev && skb_mac_header_was_set(skb)) {
 486                 if (nla_put_be16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type)) ||
 487                     nla_put_be16(inst->skb, NFULA_HWLEN,
 488                                  htons(skb->dev->hard_header_len)) ||
 489                     nla_put(inst->skb, NFULA_HWHEADER, skb->dev->hard_header_len,
 490                             skb_mac_header(skb)))
 491                         goto nla_put_failure;
 492         }
 493
 494         if (skb->tstamp.tv64) {
 495                 struct nfulnl_msg_packet_timestamp ts;
 496                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 497                 ts.sec = cpu_to_be64(tv.tv_sec);
 498                 ts.usec = cpu_to_be64(tv.tv_usec);
 499
 500                 if (nla_put(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts))
 501                         goto nla_put_failure;
 502         }
 503
 504         /* UID */
 505         sk = skb->sk;
 506         if (sk && sk->sk_state != TCP_TIME_WAIT) {
 507                 read_lock_bh(&sk->sk_callback_lock);
 508                 if (sk->sk_socket && sk->sk_socket->file) {
 509                         struct file *file = sk->sk_socket->file;
 510                         const struct cred *cred = file->f_cred;
 511                         struct user_namespace *user_ns = inst->peer_user_ns;
 512                         __be32 uid = htonl(from_kuid_munged(user_ns, cred->fsuid));
 513                         __be32 gid = htonl(from_kgid_munged(user_ns, cred->fsgid));
 514                         read_unlock_bh(&sk->sk_callback_lock);
 515                         if (nla_put_be32(inst->skb, NFULA_UID, uid) ||
 516                             nla_put_be32(inst->skb, NFULA_GID, gid))
 517                                 goto nla_put_failure;
 518                 } else
 519                         read_unlock_bh(&sk->sk_callback_lock);
 520         }
 521
 522         /* local sequence number */
 523         if ((inst->flags & NFULNL_CFG_F_SEQ) &&
 524             nla_put_be32(inst->skb, NFULA_SEQ, htonl(inst->seq++)))
 525                 goto nla_put_failure;
 526
 527         /* global sequence number */
 528         if ((inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) &&
 529             nla_put_be32(inst->skb, NFULA_SEQ_GLOBAL,
 530                          htonl(atomic_inc_return(&global_seq))))
 531                 goto nla_put_failure;
 532
 533         if (data_len) {
 534                 struct nlattr *nla;
 535                 int size = nla_attr_size(data_len);
 536
 537                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
 538                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 539                         return -1;
 540                 }
 541
 542                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 543                 nla->nla_type = NFULA_PAYLOAD;
 544                 nla->nla_len = size;
 545
 546                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 547                         BUG();
 548         }
 549
 550         nlh->nlmsg_len = inst->skb->tail - old_tail;
 551         return 0;
 552
 553 nla_put_failure:
 554         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 555         return -1;
 556 }
 557
 558 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 559
 560 static struct nf_loginfo default_loginfo = {
 561         .type =         NF_LOG_TYPE_ULOG,
 562         .u = {
 563                 .ulog = {
 564                         .copy_len       = 0xffff,
 565                         .group          = 0,
 566                         .qthreshold     = 1,
 567                 },
 568         },
 569 };
 570
 571 /* log handler for internal netfilter logging api */
 572 void
 573 nfulnl_log_packet(u_int8_t pf,
 574                   unsigned int hooknum,
 575                   const struct sk_buff *skb,
 576                   const struct net_device *in,
 577                   const struct net_device *out,
 578                   const struct nf_loginfo *li_user,
 579                   const char *prefix)
 580 {
 581         unsigned int size, data_len;
 582         struct nfulnl_instance *inst;
 583         const struct nf_loginfo *li;
 584         unsigned int qthreshold;
 585         unsigned int plen;
 586
 587         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 588                 li = li_user;
 589         else
 590                 li = &default_loginfo;
 591
 592         inst = instance_lookup_get(li->u.ulog.group);
 593         if (!inst)
 594                 return;
 595
 596         plen = 0;
 597         if (prefix)
 598                 plen = strlen(prefix) + 1;
 599
 600         /* FIXME: do we want to make the size calculation conditional based on
 601          * what is actually present?  way more branches and checks, but more
 602          * memory efficient... */
 603         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 604                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 605                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 606                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 607 #ifdef CONFIG_BRIDGE_NETFILTER
 608                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 609                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 610 #endif
 611                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 612                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 613                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 614                 + nla_total_size(plen)                  /* prefix */
 615                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 616                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
 617
 618         if (in && skb_mac_header_was_set(skb)) {
 619                 size +=   nla_total_size(skb->dev->hard_header_len)
 620                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 621                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 622         }
 623
 624         spin_lock_bh(&inst->lock);
 625
 626         if (inst->flags & NFULNL_CFG_F_SEQ)
 627                 size += nla_total_size(sizeof(u_int32_t));
 628         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 629                 size += nla_total_size(sizeof(u_int32_t));
 630
 631         qthreshold = inst->qthreshold;
 632         /* per-rule qthreshold overrides per-instance */
 633         if (li->u.ulog.qthreshold)
 634                 if (qthreshold > li->u.ulog.qthreshold)
 635                         qthreshold = li->u.ulog.qthreshold;
 636
 637
 638         switch (inst->copy_mode) {
 639         case NFULNL_COPY_META:
 640         case NFULNL_COPY_NONE:
 641                 data_len = 0;
 642                 break;
 643
 644         case NFULNL_COPY_PACKET:
 645                 if (inst->copy_range == 0
 646                     || inst->copy_range > skb->len)
 647                         data_len = skb->len;
 648                 else
 649                         data_len = inst->copy_range;
 650
 651                 size += nla_total_size(data_len);
 652                 break;
 653
 654         case NFULNL_COPY_DISABLED:
 655         default:
 656                 goto unlock_and_release;
 657         }
 658
 659         if (inst->skb &&
 660             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
 661                 /* either the queue len is too high or we don't have
 662                  * enough room in the skb left. flush to userspace. */
 663                 __nfulnl_flush(inst);
 664         }
 665
 666         if (!inst->skb) {
 667                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
 668                 if (!inst->skb)
 669                         goto alloc_failure;
 670         }
 671
 672         inst->qlen++;
 673
 674         __build_packet_message(inst, skb, data_len, pf,
 675                                 hooknum, in, out, prefix, plen);
 676
 677         if (inst->qlen >= qthreshold)
 678                 __nfulnl_flush(inst);
 679         /* timer_pending always called within inst->lock, so there
 680          * is no chance of a race here */
 681         else if (!timer_pending(&inst->timer)) {
 682                 instance_get(inst);
 683                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 684                 add_timer(&inst->timer);
 685         }
 686
 687 unlock_and_release:
 688         spin_unlock_bh(&inst->lock);
 689         instance_put(inst);
 690         return;
 691
 692 alloc_failure:
 693         /* FIXME: statistics */
 694         goto unlock_and_release;
 695 }
 696 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
 697
 698 static int
 699 nfulnl_rcv_nl_event(struct notifier_block *this,
 700                    unsigned long event, void *ptr)
 701 {
 702         struct netlink_notify *n = ptr;
 703
 704         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 705                 int i;
 706
 707                 /* destroy all instances for this portid */
 708                 spin_lock_bh(&instances_lock);
 709                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 710                         struct hlist_node *tmp, *t2;
 711                         struct nfulnl_instance *inst;
 712                         struct hlist_head *head = &instance_table[i];
 713
 714                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 715                                 if ((net_eq(n->net, &init_net)) &&
 716                                     (n->portid == inst->peer_portid))
 717                                         __instance_destroy(inst);
 718                         }
 719                 }
 720                 spin_unlock_bh(&instances_lock);
 721         }
 722         return NOTIFY_DONE;
 723 }
 724
 725 static struct notifier_block nfulnl_rtnl_notifier = {
 726         .notifier_call  = nfulnl_rcv_nl_event,
 727 };
 728
 729 static int
 730 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 731                    const struct nlmsghdr *nlh,
 732                    const struct nlattr * const nfqa[])
 733 {
 734         return -ENOTSUPP;
 735 }
 736
 737 static struct nf_logger nfulnl_logger __read_mostly = {
 738         .name   = "nfnetlink_log",
 739         .logfn  = &nfulnl_log_packet,
 740         .me     = THIS_MODULE,
 741 };
 742
 743 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 744         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 745         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 746         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 747         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 748         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 749         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 750 };
 751
 752 static int
 753 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 754                    const struct nlmsghdr *nlh,
 755                    const struct nlattr * const nfula[])
 756 {
 757         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 758         u_int16_t group_num = ntohs(nfmsg->res_id);
 759         struct nfulnl_instance *inst;
 760         struct nfulnl_msg_config_cmd *cmd = NULL;
 761         int ret = 0;
 762
 763         if (nfula[NFULA_CFG_CMD]) {
 764                 u_int8_t pf = nfmsg->nfgen_family;
 765                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 766
 767                 /* Commands without queue context */
 768                 switch (cmd->command) {
 769                 case NFULNL_CFG_CMD_PF_BIND:
 770                         return nf_log_bind_pf(pf, &nfulnl_logger);
 771                 case NFULNL_CFG_CMD_PF_UNBIND:
 772                         nf_log_unbind_pf(pf);
 773                         return 0;
 774                 }
 775         }
 776
 777         inst = instance_lookup_get(group_num);
 778         if (inst && inst->peer_portid != NETLINK_CB(skb).portid) {
 779                 ret = -EPERM;
 780                 goto out_put;
 781         }
 782
 783         if (cmd != NULL) {
 784                 switch (cmd->command) {
 785                 case NFULNL_CFG_CMD_BIND:
 786                         if (inst) {
 787                                 ret = -EBUSY;
 788                                 goto out_put;
 789                         }
 790
 791                         inst = instance_create(group_num,
 792                                                NETLINK_CB(skb).portid,
 793                                                sk_user_ns(NETLINK_CB(skb).ssk));
 794                         if (IS_ERR(inst)) {
 795                                 ret = PTR_ERR(inst);
 796                                 goto out;
 797                         }
 798                         break;
 799                 case NFULNL_CFG_CMD_UNBIND:
 800                         if (!inst) {
 801                                 ret = -ENODEV;
 802                                 goto out;
 803                         }
 804
 805                         instance_destroy(inst);
 806                         goto out_put;
 807                 default:
 808                         ret = -ENOTSUPP;
 809                         break;
 810                 }
 811         }
 812
 813         if (nfula[NFULA_CFG_MODE]) {
 814                 struct nfulnl_msg_config_mode *params;
 815                 params = nla_data(nfula[NFULA_CFG_MODE]);
 816
 817                 if (!inst) {
 818                         ret = -ENODEV;
 819                         goto out;
 820                 }
 821                 nfulnl_set_mode(inst, params->copy_mode,
 822                                 ntohl(params->copy_range));
 823         }
 824
 825         if (nfula[NFULA_CFG_TIMEOUT]) {
 826                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 827
 828                 if (!inst) {
 829                         ret = -ENODEV;
 830                         goto out;
 831                 }
 832                 nfulnl_set_timeout(inst, ntohl(timeout));
 833         }
 834
 835         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 836                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 837
 838                 if (!inst) {
 839                         ret = -ENODEV;
 840                         goto out;
 841                 }
 842                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 843         }
 844
 845         if (nfula[NFULA_CFG_QTHRESH]) {
 846                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 847
 848                 if (!inst) {
 849                         ret = -ENODEV;
 850                         goto out;
 851                 }
 852                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 853         }
 854
 855         if (nfula[NFULA_CFG_FLAGS]) {
 856                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
 857
 858                 if (!inst) {
 859                         ret = -ENODEV;
 860                         goto out;
 861                 }
 862                 nfulnl_set_flags(inst, ntohs(flags));
 863         }
 864
 865 out_put:
 866         instance_put(inst);
 867 out:
 868         return ret;
 869 }
 870
 871 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 872         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 873                                     .attr_count = NFULA_MAX, },
 874         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 875                                     .attr_count = NFULA_CFG_MAX,
 876                                     .policy = nfula_cfg_policy },
 877 };
 878
 879 static const struct nfnetlink_subsystem nfulnl_subsys = {
 880         .name           = "log",
 881         .subsys_id      = NFNL_SUBSYS_ULOG,
 882         .cb_count       = NFULNL_MSG_MAX,
 883         .cb             = nfulnl_cb,
 884 };
 885
 886 #ifdef CONFIG_PROC_FS
 887 struct iter_state {
 888         unsigned int bucket;
 889 };
 890
 891 static struct hlist_node *get_first(struct iter_state *st)
 892 {
 893         if (!st)
 894                 return NULL;
 895
 896         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 897                 if (!hlist_empty(&instance_table[st->bucket]))
 898                         return rcu_dereference_bh(hlist_first_rcu(&instance_table[st->bucket]));
 899         }
 900         return NULL;
 901 }
 902
 903 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
 904 {
 905         h = rcu_dereference_bh(hlist_next_rcu(h));
 906         while (!h) {
 907                 if (++st->bucket >= INSTANCE_BUCKETS)
 908                         return NULL;
 909
 910                 h = rcu_dereference_bh(hlist_first_rcu(&instance_table[st->bucket]));
 911         }
 912         return h;
 913 }
 914
 915 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
 916 {
 917         struct hlist_node *head;
 918         head = get_first(st);
 919
 920         if (head)
 921                 while (pos && (head = get_next(st, head)))
 922                         pos--;
 923         return pos ? NULL : head;
 924 }
 925
 926 static void *seq_start(struct seq_file *seq, loff_t *pos)
 927         __acquires(rcu_bh)
 928 {
 929         rcu_read_lock_bh();
 930         return get_idx(seq->private, *pos);
 931 }
 932
 933 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 934 {
 935         (*pos)++;
 936         return get_next(s->private, v);
 937 }
 938
 939 static void seq_stop(struct seq_file *s, void *v)
 940         __releases(rcu_bh)
 941 {
 942         rcu_read_unlock_bh();
 943 }
 944
 945 static int seq_show(struct seq_file *s, void *v)
 946 {
 947         const struct nfulnl_instance *inst = v;
 948
 949         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 950                           inst->group_num,
 951                           inst->peer_portid, inst->qlen,
 952                           inst->copy_mode, inst->copy_range,
 953                           inst->flushtimeout, atomic_read(&inst->use));
 954 }
 955
 956 static const struct seq_operations nful_seq_ops = {
 957         .start  = seq_start,
 958         .next   = seq_next,
 959         .stop   = seq_stop,
 960         .show   = seq_show,
 961 };
 962
 963 static int nful_open(struct inode *inode, struct file *file)
 964 {
 965         return seq_open_private(file, &nful_seq_ops,
 966                         sizeof(struct iter_state));
 967 }
 968
 969 static const struct file_operations nful_file_ops = {
 970         .owner   = THIS_MODULE,
 971         .open    = nful_open,
 972         .read    = seq_read,
 973         .llseek  = seq_lseek,
 974         .release = seq_release_private,
 975 };
 976
 977 #endif /* PROC_FS */
 978
 979 static int __init nfnetlink_log_init(void)
 980 {
 981         int i, status = -ENOMEM;
 982
 983         for (i = 0; i < INSTANCE_BUCKETS; i++)
 984                 INIT_HLIST_HEAD(&instance_table[i]);
 985
 986         /* it's not really all that important to have a random value, so
 987          * we can do this from the init function, even if there hasn't
 988          * been that much entropy yet */
 989         get_random_bytes(&hash_init, sizeof(hash_init));
 990
 991         netlink_register_notifier(&nfulnl_rtnl_notifier);
 992         status = nfnetlink_subsys_register(&nfulnl_subsys);
 993         if (status < 0) {
 994                 printk(KERN_ERR "log: failed to create netlink socket\n");
 995                 goto cleanup_netlink_notifier;
 996         }
 997
 998         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
 999         if (status < 0) {
1000                 printk(KERN_ERR "log: failed to register logger\n");
1001                 goto cleanup_subsys;
1002         }
1003
1004 #ifdef CONFIG_PROC_FS
1005         if (!proc_create("nfnetlink_log", 0440,
1006                          proc_net_netfilter, &nful_file_ops)) {
1007                 status = -ENOMEM;
1008                 goto cleanup_logger;
1009         }
1010 #endif
1011         return status;
1012
1013 #ifdef CONFIG_PROC_FS
1014 cleanup_logger:
1015         nf_log_unregister(&nfulnl_logger);
1016 #endif
1017 cleanup_subsys:
1018         nfnetlink_subsys_unregister(&nfulnl_subsys);
1019 cleanup_netlink_notifier:
1020         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1021         return status;
1022 }
1023
1024 static void __exit nfnetlink_log_fini(void)
1025 {
1026         nf_log_unregister(&nfulnl_logger);
1027 #ifdef CONFIG_PROC_FS
1028         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1029 #endif
1030         nfnetlink_subsys_unregister(&nfulnl_subsys);
1031         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1032 }
1033
1034 MODULE_DESCRIPTION("netfilter userspace logging");
1035 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1036 MODULE_LICENSE("GPL");
1037 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1038
1039 module_init(nfnetlink_log_init);
1040 module_exit(nfnetlink_log_fini);