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/if_arp.h>
  17 #include <linux/init.h>
  18 #include <linux/ip.h>
  19 #include <linux/ipv6.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/netfilter.h>
  22 #include <net/netlink.h>
  23 #include <linux/netfilter/nfnetlink.h>
  24 #include <linux/netfilter/nfnetlink_log.h>
  25 #include <linux/spinlock.h>
  26 #include <linux/sysctl.h>
  27 #include <linux/proc_fs.h>
  28 #include <linux/security.h>
  29 #include <linux/list.h>
  30 #include <linux/jhash.h>
  31 #include <linux/random.h>
  32 #include <linux/slab.h>
  33 #include <net/sock.h>
  34 #include <net/netfilter/nf_log.h>
  35 #include <net/netns/generic.h>
  36 #include <net/netfilter/nfnetlink_log.h>
  37
  38 #include <linux/atomic.h>
  39
  40 #ifdef CONFIG_BRIDGE_NETFILTER
  41 #include "../bridge/br_private.h"
  42 #endif
  43
  44 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
  45 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
  46 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  47 #define NFULNL_COPY_RANGE_MAX   0xFFFF  /* max packet size is limited by 16-bit struct nfattr nfa_len field */
  48
  49 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  50                                      printk(x, ## args); } while (0);
  51
  52 struct nfulnl_instance {
  53         struct hlist_node hlist;        /* global list of instances */
  54         spinlock_t lock;
  55         atomic_t use;                   /* use count */
  56
  57         unsigned int qlen;              /* number of nlmsgs in skb */
  58         struct sk_buff *skb;            /* pre-allocatd skb */
  59         struct timer_list timer;
  60         struct net *net;
  61         struct user_namespace *peer_user_ns;    /* User namespace of the peer process */
  62         int peer_portid;                        /* PORTID of the peer process */
  63
  64         /* configurable parameters */
  65         unsigned int flushtimeout;      /* timeout until queue flush */
  66         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  67         unsigned int qthreshold;        /* threshold of the queue */
  68         u_int32_t copy_range;
  69         u_int32_t seq;                  /* instance-local sequential counter */
  70         u_int16_t group_num;            /* number of this queue */
  71         u_int16_t flags;
  72         u_int8_t copy_mode;
  73         struct rcu_head rcu;
  74 };
  75
  76 #define INSTANCE_BUCKETS        16
  77 static unsigned int hash_init;
  78
  79 static int nfnl_log_net_id __read_mostly;
  80
  81 struct nfnl_log_net {
  82         spinlock_t instances_lock;
  83         struct hlist_head instance_table[INSTANCE_BUCKETS];
  84         atomic_t global_seq;
  85 };
  86
  87 static struct nfnl_log_net *nfnl_log_pernet(struct net *net)
  88 {
  89         return net_generic(net, nfnl_log_net_id);
  90 }
  91
  92 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  93 {
  94         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  95 }
  96
  97 static struct nfulnl_instance *
  98 __instance_lookup(struct nfnl_log_net *log, u_int16_t group_num)
  99 {
 100         struct hlist_head *head;
 101         struct nfulnl_instance *inst;
 102
 103         head = &log->instance_table[instance_hashfn(group_num)];
 104         hlist_for_each_entry_rcu(inst, head, hlist) {
 105                 if (inst->group_num == group_num)
 106                         return inst;
 107         }
 108         return NULL;
 109 }
 110
 111 static inline void
 112 instance_get(struct nfulnl_instance *inst)
 113 {
 114         atomic_inc(&inst->use);
 115 }
 116
 117 static struct nfulnl_instance *
 118 instance_lookup_get(struct nfnl_log_net *log, u_int16_t group_num)
 119 {
 120         struct nfulnl_instance *inst;
 121
 122         rcu_read_lock_bh();
 123         inst = __instance_lookup(log, group_num);
 124         if (inst && !atomic_inc_not_zero(&inst->use))
 125                 inst = NULL;
 126         rcu_read_unlock_bh();
 127
 128         return inst;
 129 }
 130
 131 static void nfulnl_instance_free_rcu(struct rcu_head *head)
 132 {
 133         struct nfulnl_instance *inst =
 134                 container_of(head, struct nfulnl_instance, rcu);
 135
 136         put_net(inst->net);
 137         kfree(inst);
 138         module_put(THIS_MODULE);
 139 }
 140
 141 static void
 142 instance_put(struct nfulnl_instance *inst)
 143 {
 144         if (inst && atomic_dec_and_test(&inst->use))
 145                 call_rcu_bh(&inst->rcu, nfulnl_instance_free_rcu);
 146 }
 147
 148 static void nfulnl_timer(unsigned long data);
 149
 150 static struct nfulnl_instance *
 151 instance_create(struct net *net, u_int16_t group_num,
 152                 int portid, struct user_namespace *user_ns)
 153 {
 154         struct nfulnl_instance *inst;
 155         struct nfnl_log_net *log = nfnl_log_pernet(net);
 156         int err;
 157
 158         spin_lock_bh(&log->instances_lock);
 159         if (__instance_lookup(log, group_num)) {
 160                 err = -EEXIST;
 161                 goto out_unlock;
 162         }
 163
 164         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 165         if (!inst) {
 166                 err = -ENOMEM;
 167                 goto out_unlock;
 168         }
 169
 170         if (!try_module_get(THIS_MODULE)) {
 171                 kfree(inst);
 172                 err = -EAGAIN;
 173                 goto out_unlock;
 174         }
 175
 176         INIT_HLIST_NODE(&inst->hlist);
 177         spin_lock_init(&inst->lock);
 178         /* needs to be two, since we _put() after creation */
 179         atomic_set(&inst->use, 2);
 180
 181         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
 182
 183         inst->net = get_net(net);
 184         inst->peer_user_ns = user_ns;
 185         inst->peer_portid = portid;
 186         inst->group_num = group_num;
 187
 188         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 189         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 190         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 191         inst->copy_mode         = NFULNL_COPY_PACKET;
 192         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
 193
 194         hlist_add_head_rcu(&inst->hlist,
 195                        &log->instance_table[instance_hashfn(group_num)]);
 196
 197
 198         spin_unlock_bh(&log->instances_lock);
 199
 200         return inst;
 201
 202 out_unlock:
 203         spin_unlock_bh(&log->instances_lock);
 204         return ERR_PTR(err);
 205 }
 206
 207 static void __nfulnl_flush(struct nfulnl_instance *inst);
 208
 209 /* called with BH disabled */
 210 static void
 211 __instance_destroy(struct nfulnl_instance *inst)
 212 {
 213         /* first pull it out of the global list */
 214         hlist_del_rcu(&inst->hlist);
 215
 216         /* then flush all pending packets from skb */
 217
 218         spin_lock(&inst->lock);
 219
 220         /* lockless readers wont be able to use us */
 221         inst->copy_mode = NFULNL_COPY_DISABLED;
 222
 223         if (inst->skb)
 224                 __nfulnl_flush(inst);
 225         spin_unlock(&inst->lock);
 226
 227         /* and finally put the refcount */
 228         instance_put(inst);
 229 }
 230
 231 static inline void
 232 instance_destroy(struct nfnl_log_net *log,
 233                  struct nfulnl_instance *inst)
 234 {
 235         spin_lock_bh(&log->instances_lock);
 236         __instance_destroy(inst);
 237         spin_unlock_bh(&log->instances_lock);
 238 }
 239
 240 static int
 241 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 242                   unsigned int range)
 243 {
 244         int status = 0;
 245
 246         spin_lock_bh(&inst->lock);
 247
 248         switch (mode) {
 249         case NFULNL_COPY_NONE:
 250         case NFULNL_COPY_META:
 251                 inst->copy_mode = mode;
 252                 inst->copy_range = 0;
 253                 break;
 254
 255         case NFULNL_COPY_PACKET:
 256                 inst->copy_mode = mode;
 257                 inst->copy_range = min_t(unsigned int,
 258                                          range, NFULNL_COPY_RANGE_MAX);
 259                 break;
 260
 261         default:
 262                 status = -EINVAL;
 263                 break;
 264         }
 265
 266         spin_unlock_bh(&inst->lock);
 267
 268         return status;
 269 }
 270
 271 static int
 272 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 273 {
 274         int status;
 275
 276         spin_lock_bh(&inst->lock);
 277         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 278                 status = -ERANGE;
 279         else if (nlbufsiz > 131072)
 280                 status = -ERANGE;
 281         else {
 282                 inst->nlbufsiz = nlbufsiz;
 283                 status = 0;
 284         }
 285         spin_unlock_bh(&inst->lock);
 286
 287         return status;
 288 }
 289
 290 static int
 291 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 292 {
 293         spin_lock_bh(&inst->lock);
 294         inst->flushtimeout = timeout;
 295         spin_unlock_bh(&inst->lock);
 296
 297         return 0;
 298 }
 299
 300 static int
 301 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 302 {
 303         spin_lock_bh(&inst->lock);
 304         inst->qthreshold = qthresh;
 305         spin_unlock_bh(&inst->lock);
 306
 307         return 0;
 308 }
 309
 310 static int
 311 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 312 {
 313         spin_lock_bh(&inst->lock);
 314         inst->flags = flags;
 315         spin_unlock_bh(&inst->lock);
 316
 317         return 0;
 318 }
 319
 320 static struct sk_buff *
 321 nfulnl_alloc_skb(unsigned int inst_size, unsigned int pkt_size)
 322 {
 323         struct sk_buff *skb;
 324         unsigned int n;
 325
 326         /* alloc skb which should be big enough for a whole multipart
 327          * message.  WARNING: has to be <= 128k due to slab restrictions */
 328
 329         n = max(inst_size, pkt_size);
 330         skb = alloc_skb(n, GFP_ATOMIC);
 331         if (!skb) {
 332                 if (n > pkt_size) {
 333                         /* try to allocate only as much as we need for current
 334                          * packet */
 335
 336                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 337                         if (!skb)
 338                                 pr_err("nfnetlink_log: can't even alloc %u bytes\n",
 339                                        pkt_size);
 340                 }
 341         }
 342
 343         return skb;
 344 }
 345
 346 static int
 347 __nfulnl_send(struct nfulnl_instance *inst)
 348 {
 349         int status = -1;
 350
 351         if (inst->qlen > 1) {
 352                 struct nlmsghdr *nlh = nlmsg_put(inst->skb, 0, 0,
 353                                                  NLMSG_DONE,
 354                                                  sizeof(struct nfgenmsg),
 355                                                  0);
 356                 if (!nlh)
 357                         goto out;
 358         }
 359         status = nfnetlink_unicast(inst->skb, inst->net, inst->peer_portid,
 360                                    MSG_DONTWAIT);
 361
 362         inst->qlen = 0;
 363         inst->skb = NULL;
 364 out:
 365         return status;
 366 }
 367
 368 static void
 369 __nfulnl_flush(struct nfulnl_instance *inst)
 370 {
 371         /* timer holds a reference */
 372         if (del_timer(&inst->timer))
 373                 instance_put(inst);
 374         if (inst->skb)
 375                 __nfulnl_send(inst);
 376 }
 377
 378 static void
 379 nfulnl_timer(unsigned long data)
 380 {
 381         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 382
 383         spin_lock_bh(&inst->lock);
 384         if (inst->skb)
 385                 __nfulnl_send(inst);
 386         spin_unlock_bh(&inst->lock);
 387         instance_put(inst);
 388 }
 389
 390 /* This is an inline function, we don't really care about a long
 391  * list of arguments */
 392 static inline int
 393 __build_packet_message(struct nfnl_log_net *log,
 394                         struct nfulnl_instance *inst,
 395                         const struct sk_buff *skb,
 396                         unsigned int data_len,
 397                         u_int8_t pf,
 398                         unsigned int hooknum,
 399                         const struct net_device *indev,
 400                         const struct net_device *outdev,
 401                         const char *prefix, unsigned int plen)
 402 {
 403         struct nfulnl_msg_packet_hdr pmsg;
 404         struct nlmsghdr *nlh;
 405         struct nfgenmsg *nfmsg;
 406         sk_buff_data_t old_tail = inst->skb->tail;
 407         struct sock *sk;
 408         const unsigned char *hwhdrp;
 409
 410         nlh = nlmsg_put(inst->skb, 0, 0,
 411                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 412                         sizeof(struct nfgenmsg), 0);
 413         if (!nlh)
 414                 return -1;
 415         nfmsg = nlmsg_data(nlh);
 416         nfmsg->nfgen_family = pf;
 417         nfmsg->version = NFNETLINK_V0;
 418         nfmsg->res_id = htons(inst->group_num);
 419
 420         pmsg.hw_protocol        = skb->protocol;
 421         pmsg.hook               = hooknum;
 422
 423         if (nla_put(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg))
 424                 goto nla_put_failure;
 425
 426         if (prefix &&
 427             nla_put(inst->skb, NFULA_PREFIX, plen, prefix))
 428                 goto nla_put_failure;
 429
 430         if (indev) {
 431 #ifndef CONFIG_BRIDGE_NETFILTER
 432                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 433                                  htonl(indev->ifindex)))
 434                         goto nla_put_failure;
 435 #else
 436                 if (pf == PF_BRIDGE) {
 437                         /* Case 1: outdev is physical input device, we need to
 438                          * look for bridge group (when called from
 439                          * netfilter_bridge) */
 440                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 441                                          htonl(indev->ifindex)) ||
 442                         /* this is the bridge group "brX" */
 443                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 444                             nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 445                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 446                                 goto nla_put_failure;
 447                 } else {
 448                         /* Case 2: indev is bridge group, we need to look for
 449                          * physical device (when called from ipv4) */
 450                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 451                                          htonl(indev->ifindex)))
 452                                 goto nla_put_failure;
 453                         if (skb->nf_bridge && skb->nf_bridge->physindev &&
 454                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 455                                          htonl(skb->nf_bridge->physindev->ifindex)))
 456                                 goto nla_put_failure;
 457                 }
 458 #endif
 459         }
 460
 461         if (outdev) {
 462 #ifndef CONFIG_BRIDGE_NETFILTER
 463                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 464                                  htonl(outdev->ifindex)))
 465                         goto nla_put_failure;
 466 #else
 467                 if (pf == PF_BRIDGE) {
 468                         /* Case 1: outdev is physical output device, we need to
 469                          * look for bridge group (when called from
 470                          * netfilter_bridge) */
 471                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 472                                          htonl(outdev->ifindex)) ||
 473                         /* this is the bridge group "brX" */
 474                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 475                             nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 476                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 477                                 goto nla_put_failure;
 478                 } else {
 479                         /* Case 2: indev is a bridge group, we need to look
 480                          * for physical device (when called from ipv4) */
 481                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 482                                          htonl(outdev->ifindex)))
 483                                 goto nla_put_failure;
 484                         if (skb->nf_bridge && skb->nf_bridge->physoutdev &&
 485                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 486                                          htonl(skb->nf_bridge->physoutdev->ifindex)))
 487                                 goto nla_put_failure;
 488                 }
 489 #endif
 490         }
 491
 492         if (skb->mark &&
 493             nla_put_be32(inst->skb, NFULA_MARK, htonl(skb->mark)))
 494                 goto nla_put_failure;
 495
 496         if (indev && skb->dev &&
 497             skb->mac_header != skb->network_header) {
 498                 struct nfulnl_msg_packet_hw phw;
 499                 int len = dev_parse_header(skb, phw.hw_addr);
 500                 if (len > 0) {
 501                         phw.hw_addrlen = htons(len);
 502                         if (nla_put(inst->skb, NFULA_HWADDR, sizeof(phw), &phw))
 503                                 goto nla_put_failure;
 504                 }
 505         }
 506
 507         if (indev && skb_mac_header_was_set(skb)) {
 508                 if (nla_put_be16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type)) ||
 509                     nla_put_be16(inst->skb, NFULA_HWLEN,
 510                                  htons(skb->dev->hard_header_len)))
 511                         goto nla_put_failure;
 512
 513                 hwhdrp = skb_mac_header(skb);
 514
 515                 if (skb->dev->type == ARPHRD_SIT)
 516                         hwhdrp -= ETH_HLEN;
 517
 518                 if (hwhdrp >= skb->head &&
 519                     nla_put(inst->skb, NFULA_HWHEADER,
 520                             skb->dev->hard_header_len, hwhdrp))
 521                         goto nla_put_failure;
 522         }
 523
 524         if (skb->tstamp.tv64) {
 525                 struct nfulnl_msg_packet_timestamp ts;
 526                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 527                 ts.sec = cpu_to_be64(tv.tv_sec);
 528                 ts.usec = cpu_to_be64(tv.tv_usec);
 529
 530                 if (nla_put(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts))
 531                         goto nla_put_failure;
 532         }
 533
 534         /* UID */
 535         sk = skb->sk;
 536         if (sk && sk->sk_state != TCP_TIME_WAIT) {
 537                 read_lock_bh(&sk->sk_callback_lock);
 538                 if (sk->sk_socket && sk->sk_socket->file) {
 539                         struct file *file = sk->sk_socket->file;
 540                         const struct cred *cred = file->f_cred;
 541                         struct user_namespace *user_ns = inst->peer_user_ns;
 542                         __be32 uid = htonl(from_kuid_munged(user_ns, cred->fsuid));
 543                         __be32 gid = htonl(from_kgid_munged(user_ns, cred->fsgid));
 544                         read_unlock_bh(&sk->sk_callback_lock);
 545                         if (nla_put_be32(inst->skb, NFULA_UID, uid) ||
 546                             nla_put_be32(inst->skb, NFULA_GID, gid))
 547                                 goto nla_put_failure;
 548                 } else
 549                         read_unlock_bh(&sk->sk_callback_lock);
 550         }
 551
 552         /* local sequence number */
 553         if ((inst->flags & NFULNL_CFG_F_SEQ) &&
 554             nla_put_be32(inst->skb, NFULA_SEQ, htonl(inst->seq++)))
 555                 goto nla_put_failure;
 556
 557         /* global sequence number */
 558         if ((inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) &&
 559             nla_put_be32(inst->skb, NFULA_SEQ_GLOBAL,
 560                          htonl(atomic_inc_return(&log->global_seq))))
 561                 goto nla_put_failure;
 562
 563         if (data_len) {
 564                 struct nlattr *nla;
 565                 int size = nla_attr_size(data_len);
 566
 567                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
 568                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 569                         return -1;
 570                 }
 571
 572                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 573                 nla->nla_type = NFULA_PAYLOAD;
 574                 nla->nla_len = size;
 575
 576                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 577                         BUG();
 578         }
 579
 580         nlh->nlmsg_len = inst->skb->tail - old_tail;
 581         return 0;
 582
 583 nla_put_failure:
 584         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 585         return -1;
 586 }
 587
 588 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 589
 590 static struct nf_loginfo default_loginfo = {
 591         .type =         NF_LOG_TYPE_ULOG,
 592         .u = {
 593                 .ulog = {
 594                         .copy_len       = 0xffff,
 595                         .group          = 0,
 596                         .qthreshold     = 1,
 597                 },
 598         },
 599 };
 600
 601 /* log handler for internal netfilter logging api */
 602 void
 603 nfulnl_log_packet(u_int8_t pf,
 604                   unsigned int hooknum,
 605                   const struct sk_buff *skb,
 606                   const struct net_device *in,
 607                   const struct net_device *out,
 608                   const struct nf_loginfo *li_user,
 609                   const char *prefix)
 610 {
 611         unsigned int size, data_len;
 612         struct nfulnl_instance *inst;
 613         const struct nf_loginfo *li;
 614         unsigned int qthreshold;
 615         unsigned int plen;
 616         struct net *net = dev_net(in ? in : out);
 617         struct nfnl_log_net *log = nfnl_log_pernet(net);
 618
 619         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 620                 li = li_user;
 621         else
 622                 li = &default_loginfo;
 623
 624         inst = instance_lookup_get(log, li->u.ulog.group);
 625         if (!inst)
 626                 return;
 627
 628         plen = 0;
 629         if (prefix)
 630                 plen = strlen(prefix) + 1;
 631
 632         /* FIXME: do we want to make the size calculation conditional based on
 633          * what is actually present?  way more branches and checks, but more
 634          * memory efficient... */
 635         size =    nlmsg_total_size(sizeof(struct nfgenmsg))
 636                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 637                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 638                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 639 #ifdef CONFIG_BRIDGE_NETFILTER
 640                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 641                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 642 #endif
 643                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 644                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 645                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 646                 + nla_total_size(plen)                  /* prefix */
 647                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 648                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
 649
 650         if (in && skb_mac_header_was_set(skb)) {
 651                 size +=   nla_total_size(skb->dev->hard_header_len)
 652                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 653                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 654         }
 655
 656         spin_lock_bh(&inst->lock);
 657
 658         if (inst->flags & NFULNL_CFG_F_SEQ)
 659                 size += nla_total_size(sizeof(u_int32_t));
 660         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 661                 size += nla_total_size(sizeof(u_int32_t));
 662
 663         qthreshold = inst->qthreshold;
 664         /* per-rule qthreshold overrides per-instance */
 665         if (li->u.ulog.qthreshold)
 666                 if (qthreshold > li->u.ulog.qthreshold)
 667                         qthreshold = li->u.ulog.qthreshold;
 668
 669
 670         switch (inst->copy_mode) {
 671         case NFULNL_COPY_META:
 672         case NFULNL_COPY_NONE:
 673                 data_len = 0;
 674                 break;
 675
 676         case NFULNL_COPY_PACKET:
 677                 if (inst->copy_range == 0
 678                     || inst->copy_range > skb->len)
 679                         data_len = skb->len;
 680                 else
 681                         data_len = inst->copy_range;
 682
 683                 size += nla_total_size(data_len);
 684                 break;
 685
 686         case NFULNL_COPY_DISABLED:
 687         default:
 688                 goto unlock_and_release;
 689         }
 690
 691         if (inst->skb &&
 692             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
 693                 /* either the queue len is too high or we don't have
 694                  * enough room in the skb left. flush to userspace. */
 695                 __nfulnl_flush(inst);
 696         }
 697
 698         if (!inst->skb) {
 699                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
 700                 if (!inst->skb)
 701                         goto alloc_failure;
 702         }
 703
 704         inst->qlen++;
 705
 706         __build_packet_message(log, inst, skb, data_len, pf,
 707                                 hooknum, in, out, prefix, plen);
 708
 709         if (inst->qlen >= qthreshold)
 710                 __nfulnl_flush(inst);
 711         /* timer_pending always called within inst->lock, so there
 712          * is no chance of a race here */
 713         else if (!timer_pending(&inst->timer)) {
 714                 instance_get(inst);
 715                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 716                 add_timer(&inst->timer);
 717         }
 718
 719 unlock_and_release:
 720         spin_unlock_bh(&inst->lock);
 721         instance_put(inst);
 722         return;
 723
 724 alloc_failure:
 725         /* FIXME: statistics */
 726         goto unlock_and_release;
 727 }
 728 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
 729
 730 static int
 731 nfulnl_rcv_nl_event(struct notifier_block *this,
 732                    unsigned long event, void *ptr)
 733 {
 734         struct netlink_notify *n = ptr;
 735         struct nfnl_log_net *log = nfnl_log_pernet(n->net);
 736
 737         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 738                 int i;
 739
 740                 /* destroy all instances for this portid */
 741                 spin_lock_bh(&log->instances_lock);
 742                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 743                         struct hlist_node *t2;
 744                         struct nfulnl_instance *inst;
 745                         struct hlist_head *head = &log->instance_table[i];
 746
 747                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 748                                 if (n->portid == inst->peer_portid)
 749                                         __instance_destroy(inst);
 750                         }
 751                 }
 752                 spin_unlock_bh(&log->instances_lock);
 753         }
 754         return NOTIFY_DONE;
 755 }
 756
 757 static struct notifier_block nfulnl_rtnl_notifier = {
 758         .notifier_call  = nfulnl_rcv_nl_event,
 759 };
 760
 761 static int
 762 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 763                    const struct nlmsghdr *nlh,
 764                    const struct nlattr * const nfqa[])
 765 {
 766         return -ENOTSUPP;
 767 }
 768
 769 static struct nf_logger nfulnl_logger __read_mostly = {
 770         .name   = "nfnetlink_log",
 771         .logfn  = &nfulnl_log_packet,
 772         .me     = THIS_MODULE,
 773 };
 774
 775 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 776         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 777         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 778         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 779         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 780         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 781         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 782 };
 783
 784 static int
 785 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 786                    const struct nlmsghdr *nlh,
 787                    const struct nlattr * const nfula[])
 788 {
 789         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 790         u_int16_t group_num = ntohs(nfmsg->res_id);
 791         struct nfulnl_instance *inst;
 792         struct nfulnl_msg_config_cmd *cmd = NULL;
 793         struct net *net = sock_net(ctnl);
 794         struct nfnl_log_net *log = nfnl_log_pernet(net);
 795         int ret = 0;
 796
 797         if (nfula[NFULA_CFG_CMD]) {
 798                 u_int8_t pf = nfmsg->nfgen_family;
 799                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 800
 801                 /* Commands without queue context */
 802                 switch (cmd->command) {
 803                 case NFULNL_CFG_CMD_PF_BIND:
 804                         return nf_log_bind_pf(net, pf, &nfulnl_logger);
 805                 case NFULNL_CFG_CMD_PF_UNBIND:
 806                         nf_log_unbind_pf(net, pf);
 807                         return 0;
 808                 }
 809         }
 810
 811         inst = instance_lookup_get(log, group_num);
 812         if (inst && inst->peer_portid != NETLINK_CB(skb).portid) {
 813                 ret = -EPERM;
 814                 goto out_put;
 815         }
 816
 817         if (cmd != NULL) {
 818                 switch (cmd->command) {
 819                 case NFULNL_CFG_CMD_BIND:
 820                         if (inst) {
 821                                 ret = -EBUSY;
 822                                 goto out_put;
 823                         }
 824
 825                         inst = instance_create(net, group_num,
 826                                                NETLINK_CB(skb).portid,
 827                                                sk_user_ns(NETLINK_CB(skb).sk));
 828                         if (IS_ERR(inst)) {
 829                                 ret = PTR_ERR(inst);
 830                                 goto out;
 831                         }
 832                         break;
 833                 case NFULNL_CFG_CMD_UNBIND:
 834                         if (!inst) {
 835                                 ret = -ENODEV;
 836                                 goto out;
 837                         }
 838
 839                         instance_destroy(log, inst);
 840                         goto out_put;
 841                 default:
 842                         ret = -ENOTSUPP;
 843                         break;
 844                 }
 845         }
 846
 847         if (nfula[NFULA_CFG_MODE]) {
 848                 struct nfulnl_msg_config_mode *params;
 849                 params = nla_data(nfula[NFULA_CFG_MODE]);
 850
 851                 if (!inst) {
 852                         ret = -ENODEV;
 853                         goto out;
 854                 }
 855                 nfulnl_set_mode(inst, params->copy_mode,
 856                                 ntohl(params->copy_range));
 857         }
 858
 859         if (nfula[NFULA_CFG_TIMEOUT]) {
 860                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 861
 862                 if (!inst) {
 863                         ret = -ENODEV;
 864                         goto out;
 865                 }
 866                 nfulnl_set_timeout(inst, ntohl(timeout));
 867         }
 868
 869         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 870                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 871
 872                 if (!inst) {
 873                         ret = -ENODEV;
 874                         goto out;
 875                 }
 876                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 877         }
 878
 879         if (nfula[NFULA_CFG_QTHRESH]) {
 880                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 881
 882                 if (!inst) {
 883                         ret = -ENODEV;
 884                         goto out;
 885                 }
 886                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 887         }
 888
 889         if (nfula[NFULA_CFG_FLAGS]) {
 890                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
 891
 892                 if (!inst) {
 893                         ret = -ENODEV;
 894                         goto out;
 895                 }
 896                 nfulnl_set_flags(inst, ntohs(flags));
 897         }
 898
 899 out_put:
 900         instance_put(inst);
 901 out:
 902         return ret;
 903 }
 904
 905 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 906         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 907                                     .attr_count = NFULA_MAX, },
 908         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 909                                     .attr_count = NFULA_CFG_MAX,
 910                                     .policy = nfula_cfg_policy },
 911 };
 912
 913 static const struct nfnetlink_subsystem nfulnl_subsys = {
 914         .name           = "log",
 915         .subsys_id      = NFNL_SUBSYS_ULOG,
 916         .cb_count       = NFULNL_MSG_MAX,
 917         .cb             = nfulnl_cb,
 918 };
 919
 920 #ifdef CONFIG_PROC_FS
 921 struct iter_state {
 922         struct seq_net_private p;
 923         unsigned int bucket;
 924 };
 925
 926 static struct hlist_node *get_first(struct net *net, struct iter_state *st)
 927 {
 928         struct nfnl_log_net *log;
 929         if (!st)
 930                 return NULL;
 931
 932         log = nfnl_log_pernet(net);
 933
 934         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 935                 struct hlist_head *head = &log->instance_table[st->bucket];
 936
 937                 if (!hlist_empty(head))
 938                         return rcu_dereference_bh(hlist_first_rcu(head));
 939         }
 940         return NULL;
 941 }
 942
 943 static struct hlist_node *get_next(struct net *net, struct iter_state *st,
 944                                    struct hlist_node *h)
 945 {
 946         h = rcu_dereference_bh(hlist_next_rcu(h));
 947         while (!h) {
 948                 struct nfnl_log_net *log;
 949                 struct hlist_head *head;
 950
 951                 if (++st->bucket >= INSTANCE_BUCKETS)
 952                         return NULL;
 953
 954                 log = nfnl_log_pernet(net);
 955                 head = &log->instance_table[st->bucket];
 956                 h = rcu_dereference_bh(hlist_first_rcu(head));
 957         }
 958         return h;
 959 }
 960
 961 static struct hlist_node *get_idx(struct net *net, struct iter_state *st,
 962                                   loff_t pos)
 963 {
 964         struct hlist_node *head;
 965         head = get_first(net, st);
 966
 967         if (head)
 968                 while (pos && (head = get_next(net, st, head)))
 969                         pos--;
 970         return pos ? NULL : head;
 971 }
 972
 973 static void *seq_start(struct seq_file *s, loff_t *pos)
 974         __acquires(rcu_bh)
 975 {
 976         rcu_read_lock_bh();
 977         return get_idx(seq_file_net(s), s->private, *pos);
 978 }
 979
 980 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 981 {
 982         (*pos)++;
 983         return get_next(seq_file_net(s), s->private, v);
 984 }
 985
 986 static void seq_stop(struct seq_file *s, void *v)
 987         __releases(rcu_bh)
 988 {
 989         rcu_read_unlock_bh();
 990 }
 991
 992 static int seq_show(struct seq_file *s, void *v)
 993 {
 994         const struct nfulnl_instance *inst = v;
 995
 996         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 997                           inst->group_num,
 998                           inst->peer_portid, inst->qlen,
 999                           inst->copy_mode, inst->copy_range,
1000                           inst->flushtimeout, atomic_read(&inst->use));
1001 }
1002
1003 static const struct seq_operations nful_seq_ops = {
1004         .start  = seq_start,
1005         .next   = seq_next,
1006         .stop   = seq_stop,
1007         .show   = seq_show,
1008 };
1009
1010 static int nful_open(struct inode *inode, struct file *file)
1011 {
1012         return seq_open_net(inode, file, &nful_seq_ops,
1013                             sizeof(struct iter_state));
1014 }
1015
1016 static const struct file_operations nful_file_ops = {
1017         .owner   = THIS_MODULE,
1018         .open    = nful_open,
1019         .read    = seq_read,
1020         .llseek  = seq_lseek,
1021         .release = seq_release_net,
1022 };
1023
1024 #endif /* PROC_FS */
1025
1026 static int __net_init nfnl_log_net_init(struct net *net)
1027 {
1028         unsigned int i;
1029         struct nfnl_log_net *log = nfnl_log_pernet(net);
1030
1031         for (i = 0; i < INSTANCE_BUCKETS; i++)
1032                 INIT_HLIST_HEAD(&log->instance_table[i]);
1033         spin_lock_init(&log->instances_lock);
1034
1035 #ifdef CONFIG_PROC_FS
1036         if (!proc_create("nfnetlink_log", 0440,
1037                          net->nf.proc_netfilter, &nful_file_ops))
1038                 return -ENOMEM;
1039 #endif
1040         return 0;
1041 }
1042
1043 static void __net_exit nfnl_log_net_exit(struct net *net)
1044 {
1045         remove_proc_entry("nfnetlink_log", net->nf.proc_netfilter);
1046 }
1047
1048 static struct pernet_operations nfnl_log_net_ops = {
1049         .init   = nfnl_log_net_init,
1050         .exit   = nfnl_log_net_exit,
1051         .id     = &nfnl_log_net_id,
1052         .size   = sizeof(struct nfnl_log_net),
1053 };
1054
1055 static int __init nfnetlink_log_init(void)
1056 {
1057         int status = -ENOMEM;
1058
1059         /* it's not really all that important to have a random value, so
1060          * we can do this from the init function, even if there hasn't
1061          * been that much entropy yet */
1062         get_random_bytes(&hash_init, sizeof(hash_init));
1063
1064         netlink_register_notifier(&nfulnl_rtnl_notifier);
1065         status = nfnetlink_subsys_register(&nfulnl_subsys);
1066         if (status < 0) {
1067                 pr_err("log: failed to create netlink socket\n");
1068                 goto cleanup_netlink_notifier;
1069         }
1070
1071         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
1072         if (status < 0) {
1073                 pr_err("log: failed to register logger\n");
1074                 goto cleanup_subsys;
1075         }
1076
1077         status = register_pernet_subsys(&nfnl_log_net_ops);
1078         if (status < 0) {
1079                 pr_err("log: failed to register pernet ops\n");
1080                 goto cleanup_logger;
1081         }
1082         return status;
1083
1084 cleanup_logger:
1085         nf_log_unregister(&nfulnl_logger);
1086 cleanup_subsys:
1087         nfnetlink_subsys_unregister(&nfulnl_subsys);
1088 cleanup_netlink_notifier:
1089         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1090         return status;
1091 }
1092
1093 static void __exit nfnetlink_log_fini(void)
1094 {
1095         unregister_pernet_subsys(&nfnl_log_net_ops);
1096         nf_log_unregister(&nfulnl_logger);
1097         nfnetlink_subsys_unregister(&nfulnl_subsys);
1098         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1099 }
1100
1101 MODULE_DESCRIPTION("netfilter userspace logging");
1102 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1103 MODULE_LICENSE("GPL");
1104 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1105
1106 module_init(nfnetlink_log_init);
1107 module_exit(nfnetlink_log_fini);