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  * (C) 2006-2012 Patrick McHardy <kaber@trash.net>
   7  *
   8  * Based on the old ipv4-only ipt_ULOG.c:
   9  * (C) 2000-2004 by Harald Welte <laforge@netfilter.org>
  10  *
  11  * This program is free software; you can redistribute it and/or modify
  12  * it under the terms of the GNU General Public License version 2 as
  13  * published by the Free Software Foundation.
  14  */
  15 #include <linux/module.h>
  16 #include <linux/skbuff.h>
  17 #include <linux/if_arp.h>
  18 #include <linux/init.h>
  19 #include <linux/ip.h>
  20 #include <linux/ipv6.h>
  21 #include <linux/netdevice.h>
  22 #include <linux/netfilter.h>
  23 #include <net/netlink.h>
  24 #include <linux/netfilter/nfnetlink.h>
  25 #include <linux/netfilter/nfnetlink_log.h>
  26 #include <linux/spinlock.h>
  27 #include <linux/sysctl.h>
  28 #include <linux/proc_fs.h>
  29 #include <linux/security.h>
  30 #include <linux/list.h>
  31 #include <linux/jhash.h>
  32 #include <linux/random.h>
  33 #include <linux/slab.h>
  34 #include <net/sock.h>
  35 #include <net/netfilter/nf_log.h>
  36 #include <net/netns/generic.h>
  37 #include <net/netfilter/nfnetlink_log.h>
  38
  39 #include <linux/atomic.h>
  40
  41 #ifdef CONFIG_BRIDGE_NETFILTER
  42 #include "../bridge/br_private.h"
  43 #endif
  44
  45 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
  46 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
  47 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  48 #define NFULNL_COPY_RANGE_MAX   0xFFFF  /* max packet size is limited by 16-bit struct nfattr nfa_len field */
  49
  50 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  51                                      printk(x, ## args); } while (0);
  52
  53 struct nfulnl_instance {
  54         struct hlist_node hlist;        /* global list of instances */
  55         spinlock_t lock;
  56         atomic_t use;                   /* use count */
  57
  58         unsigned int qlen;              /* number of nlmsgs in skb */
  59         struct sk_buff *skb;            /* pre-allocatd skb */
  60         struct timer_list timer;
  61         struct net *net;
  62         struct user_namespace *peer_user_ns;    /* User namespace of the peer process */
  63         int peer_portid;                        /* PORTID of the peer process */
  64
  65         /* configurable parameters */
  66         unsigned int flushtimeout;      /* timeout until queue flush */
  67         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  68         unsigned int qthreshold;        /* threshold of the queue */
  69         u_int32_t copy_range;
  70         u_int32_t seq;                  /* instance-local sequential counter */
  71         u_int16_t group_num;            /* number of this queue */
  72         u_int16_t flags;
  73         u_int8_t copy_mode;
  74         struct rcu_head rcu;
  75 };
  76
  77 #define INSTANCE_BUCKETS        16
  78 static unsigned int hash_init;
  79
  80 static int nfnl_log_net_id __read_mostly;
  81
  82 struct nfnl_log_net {
  83         spinlock_t instances_lock;
  84         struct hlist_head instance_table[INSTANCE_BUCKETS];
  85         atomic_t global_seq;
  86 };
  87
  88 static struct nfnl_log_net *nfnl_log_pernet(struct net *net)
  89 {
  90         return net_generic(net, nfnl_log_net_id);
  91 }
  92
  93 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  94 {
  95         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  96 }
  97
  98 static struct nfulnl_instance *
  99 __instance_lookup(struct nfnl_log_net *log, u_int16_t group_num)
 100 {
 101         struct hlist_head *head;
 102         struct nfulnl_instance *inst;
 103
 104         head = &log->instance_table[instance_hashfn(group_num)];
 105         hlist_for_each_entry_rcu(inst, head, hlist) {
 106                 if (inst->group_num == group_num)
 107                         return inst;
 108         }
 109         return NULL;
 110 }
 111
 112 static inline void
 113 instance_get(struct nfulnl_instance *inst)
 114 {
 115         atomic_inc(&inst->use);
 116 }
 117
 118 static struct nfulnl_instance *
 119 instance_lookup_get(struct nfnl_log_net *log, u_int16_t group_num)
 120 {
 121         struct nfulnl_instance *inst;
 122
 123         rcu_read_lock_bh();
 124         inst = __instance_lookup(log, group_num);
 125         if (inst && !atomic_inc_not_zero(&inst->use))
 126                 inst = NULL;
 127         rcu_read_unlock_bh();
 128
 129         return inst;
 130 }
 131
 132 static void nfulnl_instance_free_rcu(struct rcu_head *head)
 133 {
 134         struct nfulnl_instance *inst =
 135                 container_of(head, struct nfulnl_instance, rcu);
 136
 137         put_net(inst->net);
 138         kfree(inst);
 139         module_put(THIS_MODULE);
 140 }
 141
 142 static void
 143 instance_put(struct nfulnl_instance *inst)
 144 {
 145         if (inst && atomic_dec_and_test(&inst->use))
 146                 call_rcu_bh(&inst->rcu, nfulnl_instance_free_rcu);
 147 }
 148
 149 static void nfulnl_timer(unsigned long data);
 150
 151 static struct nfulnl_instance *
 152 instance_create(struct net *net, u_int16_t group_num,
 153                 int portid, struct user_namespace *user_ns)
 154 {
 155         struct nfulnl_instance *inst;
 156         struct nfnl_log_net *log = nfnl_log_pernet(net);
 157         int err;
 158
 159         spin_lock_bh(&log->instances_lock);
 160         if (__instance_lookup(log, group_num)) {
 161                 err = -EEXIST;
 162                 goto out_unlock;
 163         }
 164
 165         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 166         if (!inst) {
 167                 err = -ENOMEM;
 168                 goto out_unlock;
 169         }
 170
 171         if (!try_module_get(THIS_MODULE)) {
 172                 kfree(inst);
 173                 err = -EAGAIN;
 174                 goto out_unlock;
 175         }
 176
 177         INIT_HLIST_NODE(&inst->hlist);
 178         spin_lock_init(&inst->lock);
 179         /* needs to be two, since we _put() after creation */
 180         atomic_set(&inst->use, 2);
 181
 182         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
 183
 184         inst->net = get_net(net);
 185         inst->peer_user_ns = user_ns;
 186         inst->peer_portid = portid;
 187         inst->group_num = group_num;
 188
 189         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 190         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 191         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 192         inst->copy_mode         = NFULNL_COPY_PACKET;
 193         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
 194
 195         hlist_add_head_rcu(&inst->hlist,
 196                        &log->instance_table[instance_hashfn(group_num)]);
 197
 198
 199         spin_unlock_bh(&log->instances_lock);
 200
 201         return inst;
 202
 203 out_unlock:
 204         spin_unlock_bh(&log->instances_lock);
 205         return ERR_PTR(err);
 206 }
 207
 208 static void __nfulnl_flush(struct nfulnl_instance *inst);
 209
 210 /* called with BH disabled */
 211 static void
 212 __instance_destroy(struct nfulnl_instance *inst)
 213 {
 214         /* first pull it out of the global list */
 215         hlist_del_rcu(&inst->hlist);
 216
 217         /* then flush all pending packets from skb */
 218
 219         spin_lock(&inst->lock);
 220
 221         /* lockless readers wont be able to use us */
 222         inst->copy_mode = NFULNL_COPY_DISABLED;
 223
 224         if (inst->skb)
 225                 __nfulnl_flush(inst);
 226         spin_unlock(&inst->lock);
 227
 228         /* and finally put the refcount */
 229         instance_put(inst);
 230 }
 231
 232 static inline void
 233 instance_destroy(struct nfnl_log_net *log,
 234                  struct nfulnl_instance *inst)
 235 {
 236         spin_lock_bh(&log->instances_lock);
 237         __instance_destroy(inst);
 238         spin_unlock_bh(&log->instances_lock);
 239 }
 240
 241 static int
 242 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 243                   unsigned int range)
 244 {
 245         int status = 0;
 246
 247         spin_lock_bh(&inst->lock);
 248
 249         switch (mode) {
 250         case NFULNL_COPY_NONE:
 251         case NFULNL_COPY_META:
 252                 inst->copy_mode = mode;
 253                 inst->copy_range = 0;
 254                 break;
 255
 256         case NFULNL_COPY_PACKET:
 257                 inst->copy_mode = mode;
 258                 inst->copy_range = min_t(unsigned int,
 259                                          range, NFULNL_COPY_RANGE_MAX);
 260                 break;
 261
 262         default:
 263                 status = -EINVAL;
 264                 break;
 265         }
 266
 267         spin_unlock_bh(&inst->lock);
 268
 269         return status;
 270 }
 271
 272 static int
 273 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 274 {
 275         int status;
 276
 277         spin_lock_bh(&inst->lock);
 278         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 279                 status = -ERANGE;
 280         else if (nlbufsiz > 131072)
 281                 status = -ERANGE;
 282         else {
 283                 inst->nlbufsiz = nlbufsiz;
 284                 status = 0;
 285         }
 286         spin_unlock_bh(&inst->lock);
 287
 288         return status;
 289 }
 290
 291 static int
 292 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 293 {
 294         spin_lock_bh(&inst->lock);
 295         inst->flushtimeout = timeout;
 296         spin_unlock_bh(&inst->lock);
 297
 298         return 0;
 299 }
 300
 301 static int
 302 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 303 {
 304         spin_lock_bh(&inst->lock);
 305         inst->qthreshold = qthresh;
 306         spin_unlock_bh(&inst->lock);
 307
 308         return 0;
 309 }
 310
 311 static int
 312 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 313 {
 314         spin_lock_bh(&inst->lock);
 315         inst->flags = flags;
 316         spin_unlock_bh(&inst->lock);
 317
 318         return 0;
 319 }
 320
 321 static struct sk_buff *
 322 nfulnl_alloc_skb(u32 peer_portid, unsigned int inst_size, unsigned int pkt_size)
 323 {
 324         struct sk_buff *skb;
 325         unsigned int n;
 326
 327         /* alloc skb which should be big enough for a whole multipart
 328          * message.  WARNING: has to be <= 128k due to slab restrictions */
 329
 330         n = max(inst_size, pkt_size);
 331         skb = nfnetlink_alloc_skb(&init_net, n, peer_portid, GFP_ATOMIC);
 332         if (!skb) {
 333                 if (n > pkt_size) {
 334                         /* try to allocate only as much as we need for current
 335                          * packet */
 336
 337                         skb = nfnetlink_alloc_skb(&init_net, pkt_size,
 338                                                   peer_portid, GFP_ATOMIC);
 339                         if (!skb)
 340                                 pr_err("nfnetlink_log: can't even alloc %u bytes\n",
 341                                        pkt_size);
 342                 }
 343         }
 344
 345         return skb;
 346 }
 347
 348 static int
 349 __nfulnl_send(struct nfulnl_instance *inst)
 350 {
 351         int status = -1;
 352
 353         if (inst->qlen > 1) {
 354                 struct nlmsghdr *nlh = nlmsg_put(inst->skb, 0, 0,
 355                                                  NLMSG_DONE,
 356                                                  sizeof(struct nfgenmsg),
 357                                                  0);
 358                 if (!nlh)
 359                         goto out;
 360         }
 361         status = nfnetlink_unicast(inst->skb, inst->net, inst->peer_portid,
 362                                    MSG_DONTWAIT);
 363
 364         inst->qlen = 0;
 365         inst->skb = NULL;
 366 out:
 367         return status;
 368 }
 369
 370 static void
 371 __nfulnl_flush(struct nfulnl_instance *inst)
 372 {
 373         /* timer holds a reference */
 374         if (del_timer(&inst->timer))
 375                 instance_put(inst);
 376         if (inst->skb)
 377                 __nfulnl_send(inst);
 378 }
 379
 380 static void
 381 nfulnl_timer(unsigned long data)
 382 {
 383         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 384
 385         spin_lock_bh(&inst->lock);
 386         if (inst->skb)
 387                 __nfulnl_send(inst);
 388         spin_unlock_bh(&inst->lock);
 389         instance_put(inst);
 390 }
 391
 392 /* This is an inline function, we don't really care about a long
 393  * list of arguments */
 394 static inline int
 395 __build_packet_message(struct nfnl_log_net *log,
 396                         struct nfulnl_instance *inst,
 397                         const struct sk_buff *skb,
 398                         unsigned int data_len,
 399                         u_int8_t pf,
 400                         unsigned int hooknum,
 401                         const struct net_device *indev,
 402                         const struct net_device *outdev,
 403                         const char *prefix, unsigned int plen)
 404 {
 405         struct nfulnl_msg_packet_hdr pmsg;
 406         struct nlmsghdr *nlh;
 407         struct nfgenmsg *nfmsg;
 408         sk_buff_data_t old_tail = inst->skb->tail;
 409         struct sock *sk;
 410         const unsigned char *hwhdrp;
 411
 412         nlh = nlmsg_put(inst->skb, 0, 0,
 413                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 414                         sizeof(struct nfgenmsg), 0);
 415         if (!nlh)
 416                 return -1;
 417         nfmsg = nlmsg_data(nlh);
 418         nfmsg->nfgen_family = pf;
 419         nfmsg->version = NFNETLINK_V0;
 420         nfmsg->res_id = htons(inst->group_num);
 421
 422         pmsg.hw_protocol        = skb->protocol;
 423         pmsg.hook               = hooknum;
 424
 425         if (nla_put(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg))
 426                 goto nla_put_failure;
 427
 428         if (prefix &&
 429             nla_put(inst->skb, NFULA_PREFIX, plen, prefix))
 430                 goto nla_put_failure;
 431
 432         if (indev) {
 433 #ifndef CONFIG_BRIDGE_NETFILTER
 434                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 435                                  htonl(indev->ifindex)))
 436                         goto nla_put_failure;
 437 #else
 438                 if (pf == PF_BRIDGE) {
 439                         /* Case 1: outdev is physical input device, we need to
 440                          * look for bridge group (when called from
 441                          * netfilter_bridge) */
 442                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 443                                          htonl(indev->ifindex)) ||
 444                         /* this is the bridge group "brX" */
 445                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 446                             nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 447                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 448                                 goto nla_put_failure;
 449                 } else {
 450                         /* Case 2: indev is bridge group, we need to look for
 451                          * physical device (when called from ipv4) */
 452                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 453                                          htonl(indev->ifindex)))
 454                                 goto nla_put_failure;
 455                         if (skb->nf_bridge && skb->nf_bridge->physindev &&
 456                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 457                                          htonl(skb->nf_bridge->physindev->ifindex)))
 458                                 goto nla_put_failure;
 459                 }
 460 #endif
 461         }
 462
 463         if (outdev) {
 464 #ifndef CONFIG_BRIDGE_NETFILTER
 465                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 466                                  htonl(outdev->ifindex)))
 467                         goto nla_put_failure;
 468 #else
 469                 if (pf == PF_BRIDGE) {
 470                         /* Case 1: outdev is physical output device, we need to
 471                          * look for bridge group (when called from
 472                          * netfilter_bridge) */
 473                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 474                                          htonl(outdev->ifindex)) ||
 475                         /* this is the bridge group "brX" */
 476                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 477                             nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 478                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 479                                 goto nla_put_failure;
 480                 } else {
 481                         /* Case 2: indev is a bridge group, we need to look
 482                          * for physical device (when called from ipv4) */
 483                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 484                                          htonl(outdev->ifindex)))
 485                                 goto nla_put_failure;
 486                         if (skb->nf_bridge && skb->nf_bridge->physoutdev &&
 487                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 488                                          htonl(skb->nf_bridge->physoutdev->ifindex)))
 489                                 goto nla_put_failure;
 490                 }
 491 #endif
 492         }
 493
 494         if (skb->mark &&
 495             nla_put_be32(inst->skb, NFULA_MARK, htonl(skb->mark)))
 496                 goto nla_put_failure;
 497
 498         if (indev && skb->dev &&
 499             skb->mac_header != skb->network_header) {
 500                 struct nfulnl_msg_packet_hw phw;
 501                 int len = dev_parse_header(skb, phw.hw_addr);
 502                 if (len > 0) {
 503                         phw.hw_addrlen = htons(len);
 504                         if (nla_put(inst->skb, NFULA_HWADDR, sizeof(phw), &phw))
 505                                 goto nla_put_failure;
 506                 }
 507         }
 508
 509         if (indev && skb_mac_header_was_set(skb)) {
 510                 if (nla_put_be16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type)) ||
 511                     nla_put_be16(inst->skb, NFULA_HWLEN,
 512                                  htons(skb->dev->hard_header_len)))
 513                         goto nla_put_failure;
 514
 515                 hwhdrp = skb_mac_header(skb);
 516
 517                 if (skb->dev->type == ARPHRD_SIT)
 518                         hwhdrp -= ETH_HLEN;
 519
 520                 if (hwhdrp >= skb->head &&
 521                     nla_put(inst->skb, NFULA_HWHEADER,
 522                             skb->dev->hard_header_len, hwhdrp))
 523                         goto nla_put_failure;
 524         }
 525
 526         if (skb->tstamp.tv64) {
 527                 struct nfulnl_msg_packet_timestamp ts;
 528                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 529                 ts.sec = cpu_to_be64(tv.tv_sec);
 530                 ts.usec = cpu_to_be64(tv.tv_usec);
 531
 532                 if (nla_put(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts))
 533                         goto nla_put_failure;
 534         }
 535
 536         /* UID */
 537         sk = skb->sk;
 538         if (sk && sk->sk_state != TCP_TIME_WAIT) {
 539                 read_lock_bh(&sk->sk_callback_lock);
 540                 if (sk->sk_socket && sk->sk_socket->file) {
 541                         struct file *file = sk->sk_socket->file;
 542                         const struct cred *cred = file->f_cred;
 543                         struct user_namespace *user_ns = inst->peer_user_ns;
 544                         __be32 uid = htonl(from_kuid_munged(user_ns, cred->fsuid));
 545                         __be32 gid = htonl(from_kgid_munged(user_ns, cred->fsgid));
 546                         read_unlock_bh(&sk->sk_callback_lock);
 547                         if (nla_put_be32(inst->skb, NFULA_UID, uid) ||
 548                             nla_put_be32(inst->skb, NFULA_GID, gid))
 549                                 goto nla_put_failure;
 550                 } else
 551                         read_unlock_bh(&sk->sk_callback_lock);
 552         }
 553
 554         /* local sequence number */
 555         if ((inst->flags & NFULNL_CFG_F_SEQ) &&
 556             nla_put_be32(inst->skb, NFULA_SEQ, htonl(inst->seq++)))
 557                 goto nla_put_failure;
 558
 559         /* global sequence number */
 560         if ((inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) &&
 561             nla_put_be32(inst->skb, NFULA_SEQ_GLOBAL,
 562                          htonl(atomic_inc_return(&log->global_seq))))
 563                 goto nla_put_failure;
 564
 565         if (data_len) {
 566                 struct nlattr *nla;
 567                 int size = nla_attr_size(data_len);
 568
 569                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
 570                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 571                         return -1;
 572                 }
 573
 574                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 575                 nla->nla_type = NFULA_PAYLOAD;
 576                 nla->nla_len = size;
 577
 578                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 579                         BUG();
 580         }
 581
 582         nlh->nlmsg_len = inst->skb->tail - old_tail;
 583         return 0;
 584
 585 nla_put_failure:
 586         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 587         return -1;
 588 }
 589
 590 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 591
 592 static struct nf_loginfo default_loginfo = {
 593         .type =         NF_LOG_TYPE_ULOG,
 594         .u = {
 595                 .ulog = {
 596                         .copy_len       = 0xffff,
 597                         .group          = 0,
 598                         .qthreshold     = 1,
 599                 },
 600         },
 601 };
 602
 603 /* log handler for internal netfilter logging api */
 604 void
 605 nfulnl_log_packet(struct net *net,
 606                   u_int8_t pf,
 607                   unsigned int hooknum,
 608                   const struct sk_buff *skb,
 609                   const struct net_device *in,
 610                   const struct net_device *out,
 611                   const struct nf_loginfo *li_user,
 612                   const char *prefix)
 613 {
 614         unsigned int size, data_len;
 615         struct nfulnl_instance *inst;
 616         const struct nf_loginfo *li;
 617         unsigned int qthreshold;
 618         unsigned int plen;
 619         struct nfnl_log_net *log = nfnl_log_pernet(net);
 620
 621         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 622                 li = li_user;
 623         else
 624                 li = &default_loginfo;
 625
 626         inst = instance_lookup_get(log, li->u.ulog.group);
 627         if (!inst)
 628                 return;
 629
 630         plen = 0;
 631         if (prefix)
 632                 plen = strlen(prefix) + 1;
 633
 634         /* FIXME: do we want to make the size calculation conditional based on
 635          * what is actually present?  way more branches and checks, but more
 636          * memory efficient... */
 637         size =    nlmsg_total_size(sizeof(struct nfgenmsg))
 638                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 639                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 640                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 641 #ifdef CONFIG_BRIDGE_NETFILTER
 642                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 643                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 644 #endif
 645                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 646                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 647                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 648                 + nla_total_size(plen)                  /* prefix */
 649                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 650                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
 651
 652         if (in && skb_mac_header_was_set(skb)) {
 653                 size +=   nla_total_size(skb->dev->hard_header_len)
 654                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 655                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 656         }
 657
 658         spin_lock_bh(&inst->lock);
 659
 660         if (inst->flags & NFULNL_CFG_F_SEQ)
 661                 size += nla_total_size(sizeof(u_int32_t));
 662         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 663                 size += nla_total_size(sizeof(u_int32_t));
 664
 665         qthreshold = inst->qthreshold;
 666         /* per-rule qthreshold overrides per-instance */
 667         if (li->u.ulog.qthreshold)
 668                 if (qthreshold > li->u.ulog.qthreshold)
 669                         qthreshold = li->u.ulog.qthreshold;
 670
 671
 672         switch (inst->copy_mode) {
 673         case NFULNL_COPY_META:
 674         case NFULNL_COPY_NONE:
 675                 data_len = 0;
 676                 break;
 677
 678         case NFULNL_COPY_PACKET:
 679                 if (inst->copy_range == 0
 680                     || inst->copy_range > skb->len)
 681                         data_len = skb->len;
 682                 else
 683                         data_len = inst->copy_range;
 684
 685                 size += nla_total_size(data_len);
 686                 break;
 687
 688         case NFULNL_COPY_DISABLED:
 689         default:
 690                 goto unlock_and_release;
 691         }
 692
 693         if (inst->skb &&
 694             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
 695                 /* either the queue len is too high or we don't have
 696                  * enough room in the skb left. flush to userspace. */
 697                 __nfulnl_flush(inst);
 698         }
 699
 700         if (!inst->skb) {
 701                 inst->skb = nfulnl_alloc_skb(inst->peer_portid, inst->nlbufsiz,
 702                                              size);
 703                 if (!inst->skb)
 704                         goto alloc_failure;
 705         }
 706
 707         inst->qlen++;
 708
 709         __build_packet_message(log, inst, skb, data_len, pf,
 710                                 hooknum, in, out, prefix, plen);
 711
 712         if (inst->qlen >= qthreshold)
 713                 __nfulnl_flush(inst);
 714         /* timer_pending always called within inst->lock, so there
 715          * is no chance of a race here */
 716         else if (!timer_pending(&inst->timer)) {
 717                 instance_get(inst);
 718                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 719                 add_timer(&inst->timer);
 720         }
 721
 722 unlock_and_release:
 723         spin_unlock_bh(&inst->lock);
 724         instance_put(inst);
 725         return;
 726
 727 alloc_failure:
 728         /* FIXME: statistics */
 729         goto unlock_and_release;
 730 }
 731 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
 732
 733 static int
 734 nfulnl_rcv_nl_event(struct notifier_block *this,
 735                    unsigned long event, void *ptr)
 736 {
 737         struct netlink_notify *n = ptr;
 738         struct nfnl_log_net *log = nfnl_log_pernet(n->net);
 739
 740         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 741                 int i;
 742
 743                 /* destroy all instances for this portid */
 744                 spin_lock_bh(&log->instances_lock);
 745                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 746                         struct hlist_node *t2;
 747                         struct nfulnl_instance *inst;
 748                         struct hlist_head *head = &log->instance_table[i];
 749
 750                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 751                                 if (n->portid == inst->peer_portid)
 752                                         __instance_destroy(inst);
 753                         }
 754                 }
 755                 spin_unlock_bh(&log->instances_lock);
 756         }
 757         return NOTIFY_DONE;
 758 }
 759
 760 static struct notifier_block nfulnl_rtnl_notifier = {
 761         .notifier_call  = nfulnl_rcv_nl_event,
 762 };
 763
 764 static int
 765 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 766                    const struct nlmsghdr *nlh,
 767                    const struct nlattr * const nfqa[])
 768 {
 769         return -ENOTSUPP;
 770 }
 771
 772 static struct nf_logger nfulnl_logger __read_mostly = {
 773         .name   = "nfnetlink_log",
 774         .logfn  = &nfulnl_log_packet,
 775         .me     = THIS_MODULE,
 776 };
 777
 778 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 779         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 780         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 781         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 782         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 783         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 784         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 785 };
 786
 787 static int
 788 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 789                    const struct nlmsghdr *nlh,
 790                    const struct nlattr * const nfula[])
 791 {
 792         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 793         u_int16_t group_num = ntohs(nfmsg->res_id);
 794         struct nfulnl_instance *inst;
 795         struct nfulnl_msg_config_cmd *cmd = NULL;
 796         struct net *net = sock_net(ctnl);
 797         struct nfnl_log_net *log = nfnl_log_pernet(net);
 798         int ret = 0;
 799
 800         if (nfula[NFULA_CFG_CMD]) {
 801                 u_int8_t pf = nfmsg->nfgen_family;
 802                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 803
 804                 /* Commands without queue context */
 805                 switch (cmd->command) {
 806                 case NFULNL_CFG_CMD_PF_BIND:
 807                         return nf_log_bind_pf(net, pf, &nfulnl_logger);
 808                 case NFULNL_CFG_CMD_PF_UNBIND:
 809                         nf_log_unbind_pf(net, pf);
 810                         return 0;
 811                 }
 812         }
 813
 814         inst = instance_lookup_get(log, group_num);
 815         if (inst && inst->peer_portid != NETLINK_CB(skb).portid) {
 816                 ret = -EPERM;
 817                 goto out_put;
 818         }
 819
 820         if (cmd != NULL) {
 821                 switch (cmd->command) {
 822                 case NFULNL_CFG_CMD_BIND:
 823                         if (inst) {
 824                                 ret = -EBUSY;
 825                                 goto out_put;
 826                         }
 827
 828                         inst = instance_create(net, group_num,
 829                                                NETLINK_CB(skb).portid,
 830                                                sk_user_ns(NETLINK_CB(skb).sk));
 831                         if (IS_ERR(inst)) {
 832                                 ret = PTR_ERR(inst);
 833                                 goto out;
 834                         }
 835                         break;
 836                 case NFULNL_CFG_CMD_UNBIND:
 837                         if (!inst) {
 838                                 ret = -ENODEV;
 839                                 goto out;
 840                         }
 841
 842                         instance_destroy(log, inst);
 843                         goto out_put;
 844                 default:
 845                         ret = -ENOTSUPP;
 846                         break;
 847                 }
 848         }
 849
 850         if (nfula[NFULA_CFG_MODE]) {
 851                 struct nfulnl_msg_config_mode *params;
 852                 params = nla_data(nfula[NFULA_CFG_MODE]);
 853
 854                 if (!inst) {
 855                         ret = -ENODEV;
 856                         goto out;
 857                 }
 858                 nfulnl_set_mode(inst, params->copy_mode,
 859                                 ntohl(params->copy_range));
 860         }
 861
 862         if (nfula[NFULA_CFG_TIMEOUT]) {
 863                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 864
 865                 if (!inst) {
 866                         ret = -ENODEV;
 867                         goto out;
 868                 }
 869                 nfulnl_set_timeout(inst, ntohl(timeout));
 870         }
 871
 872         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 873                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 874
 875                 if (!inst) {
 876                         ret = -ENODEV;
 877                         goto out;
 878                 }
 879                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 880         }
 881
 882         if (nfula[NFULA_CFG_QTHRESH]) {
 883                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 884
 885                 if (!inst) {
 886                         ret = -ENODEV;
 887                         goto out;
 888                 }
 889                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 890         }
 891
 892         if (nfula[NFULA_CFG_FLAGS]) {
 893                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
 894
 895                 if (!inst) {
 896                         ret = -ENODEV;
 897                         goto out;
 898                 }
 899                 nfulnl_set_flags(inst, ntohs(flags));
 900         }
 901
 902 out_put:
 903         instance_put(inst);
 904 out:
 905         return ret;
 906 }
 907
 908 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 909         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 910                                     .attr_count = NFULA_MAX, },
 911         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 912                                     .attr_count = NFULA_CFG_MAX,
 913                                     .policy = nfula_cfg_policy },
 914 };
 915
 916 static const struct nfnetlink_subsystem nfulnl_subsys = {
 917         .name           = "log",
 918         .subsys_id      = NFNL_SUBSYS_ULOG,
 919         .cb_count       = NFULNL_MSG_MAX,
 920         .cb             = nfulnl_cb,
 921 };
 922
 923 #ifdef CONFIG_PROC_FS
 924 struct iter_state {
 925         struct seq_net_private p;
 926         unsigned int bucket;
 927 };
 928
 929 static struct hlist_node *get_first(struct net *net, struct iter_state *st)
 930 {
 931         struct nfnl_log_net *log;
 932         if (!st)
 933                 return NULL;
 934
 935         log = nfnl_log_pernet(net);
 936
 937         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 938                 struct hlist_head *head = &log->instance_table[st->bucket];
 939
 940                 if (!hlist_empty(head))
 941                         return rcu_dereference_bh(hlist_first_rcu(head));
 942         }
 943         return NULL;
 944 }
 945
 946 static struct hlist_node *get_next(struct net *net, struct iter_state *st,
 947                                    struct hlist_node *h)
 948 {
 949         h = rcu_dereference_bh(hlist_next_rcu(h));
 950         while (!h) {
 951                 struct nfnl_log_net *log;
 952                 struct hlist_head *head;
 953
 954                 if (++st->bucket >= INSTANCE_BUCKETS)
 955                         return NULL;
 956
 957                 log = nfnl_log_pernet(net);
 958                 head = &log->instance_table[st->bucket];
 959                 h = rcu_dereference_bh(hlist_first_rcu(head));
 960         }
 961         return h;
 962 }
 963
 964 static struct hlist_node *get_idx(struct net *net, struct iter_state *st,
 965                                   loff_t pos)
 966 {
 967         struct hlist_node *head;
 968         head = get_first(net, st);
 969
 970         if (head)
 971                 while (pos && (head = get_next(net, st, head)))
 972                         pos--;
 973         return pos ? NULL : head;
 974 }
 975
 976 static void *seq_start(struct seq_file *s, loff_t *pos)
 977         __acquires(rcu_bh)
 978 {
 979         rcu_read_lock_bh();
 980         return get_idx(seq_file_net(s), s->private, *pos);
 981 }
 982
 983 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 984 {
 985         (*pos)++;
 986         return get_next(seq_file_net(s), s->private, v);
 987 }
 988
 989 static void seq_stop(struct seq_file *s, void *v)
 990         __releases(rcu_bh)
 991 {
 992         rcu_read_unlock_bh();
 993 }
 994
 995 static int seq_show(struct seq_file *s, void *v)
 996 {
 997         const struct nfulnl_instance *inst = v;
 998
 999         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
1000                           inst->group_num,
1001                           inst->peer_portid, inst->qlen,
1002                           inst->copy_mode, inst->copy_range,
1003                           inst->flushtimeout, atomic_read(&inst->use));
1004 }
1005
1006 static const struct seq_operations nful_seq_ops = {
1007         .start  = seq_start,
1008         .next   = seq_next,
1009         .stop   = seq_stop,
1010         .show   = seq_show,
1011 };
1012
1013 static int nful_open(struct inode *inode, struct file *file)
1014 {
1015         return seq_open_net(inode, file, &nful_seq_ops,
1016                             sizeof(struct iter_state));
1017 }
1018
1019 static const struct file_operations nful_file_ops = {
1020         .owner   = THIS_MODULE,
1021         .open    = nful_open,
1022         .read    = seq_read,
1023         .llseek  = seq_lseek,
1024         .release = seq_release_net,
1025 };
1026
1027 #endif /* PROC_FS */
1028
1029 static int __net_init nfnl_log_net_init(struct net *net)
1030 {
1031         unsigned int i;
1032         struct nfnl_log_net *log = nfnl_log_pernet(net);
1033
1034         for (i = 0; i < INSTANCE_BUCKETS; i++)
1035                 INIT_HLIST_HEAD(&log->instance_table[i]);
1036         spin_lock_init(&log->instances_lock);
1037
1038 #ifdef CONFIG_PROC_FS
1039         if (!proc_create("nfnetlink_log", 0440,
1040                          net->nf.proc_netfilter, &nful_file_ops))
1041                 return -ENOMEM;
1042 #endif
1043         return 0;
1044 }
1045
1046 static void __net_exit nfnl_log_net_exit(struct net *net)
1047 {
1048 #ifdef CONFIG_PROC_FS
1049         remove_proc_entry("nfnetlink_log", net->nf.proc_netfilter);
1050 #endif
1051 }
1052
1053 static struct pernet_operations nfnl_log_net_ops = {
1054         .init   = nfnl_log_net_init,
1055         .exit   = nfnl_log_net_exit,
1056         .id     = &nfnl_log_net_id,
1057         .size   = sizeof(struct nfnl_log_net),
1058 };
1059
1060 static int __init nfnetlink_log_init(void)
1061 {
1062         int status = -ENOMEM;
1063
1064         /* it's not really all that important to have a random value, so
1065          * we can do this from the init function, even if there hasn't
1066          * been that much entropy yet */
1067         get_random_bytes(&hash_init, sizeof(hash_init));
1068
1069         netlink_register_notifier(&nfulnl_rtnl_notifier);
1070         status = nfnetlink_subsys_register(&nfulnl_subsys);
1071         if (status < 0) {
1072                 pr_err("log: failed to create netlink socket\n");
1073                 goto cleanup_netlink_notifier;
1074         }
1075
1076         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
1077         if (status < 0) {
1078                 pr_err("log: failed to register logger\n");
1079                 goto cleanup_subsys;
1080         }
1081
1082         status = register_pernet_subsys(&nfnl_log_net_ops);
1083         if (status < 0) {
1084                 pr_err("log: failed to register pernet ops\n");
1085                 goto cleanup_logger;
1086         }
1087         return status;
1088
1089 cleanup_logger:
1090         nf_log_unregister(&nfulnl_logger);
1091 cleanup_subsys:
1092         nfnetlink_subsys_unregister(&nfulnl_subsys);
1093 cleanup_netlink_notifier:
1094         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1095         return status;
1096 }
1097
1098 static void __exit nfnetlink_log_fini(void)
1099 {
1100         unregister_pernet_subsys(&nfnl_log_net_ops);
1101         nf_log_unregister(&nfulnl_logger);
1102         nfnetlink_subsys_unregister(&nfulnl_subsys);
1103         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1104 }
1105
1106 MODULE_DESCRIPTION("netfilter userspace logging");
1107 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1108 MODULE_LICENSE("GPL");
1109 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1110
1111 module_init(nfnetlink_log_init);
1112 module_exit(nfnetlink_log_fini);