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 /* max packet size is limited by 16-bit struct nfattr nfa_len field */
  49 #define NFULNL_COPY_RANGE_MAX   (0xFFFF - NLA_HDRLEN)
  50
  51 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  52                                      printk(x, ## args); } while (0);
  53
  54 struct nfulnl_instance {
  55         struct hlist_node hlist;        /* global list of instances */
  56         spinlock_t lock;
  57         atomic_t use;                   /* use count */
  58
  59         unsigned int qlen;              /* number of nlmsgs in skb */
  60         struct sk_buff *skb;            /* pre-allocatd skb */
  61         struct timer_list timer;
  62         struct net *net;
  63         struct user_namespace *peer_user_ns;    /* User namespace of the peer process */
  64         int peer_portid;                        /* PORTID of the peer process */
  65
  66         /* configurable parameters */
  67         unsigned int flushtimeout;      /* timeout until queue flush */
  68         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  69         unsigned int qthreshold;        /* threshold of the queue */
  70         u_int32_t copy_range;
  71         u_int32_t seq;                  /* instance-local sequential counter */
  72         u_int16_t group_num;            /* number of this queue */
  73         u_int16_t flags;
  74         u_int8_t copy_mode;
  75         struct rcu_head rcu;
  76 };
  77
  78 #define INSTANCE_BUCKETS        16
  79 static unsigned int hash_init;
  80
  81 static int nfnl_log_net_id __read_mostly;
  82
  83 struct nfnl_log_net {
  84         spinlock_t instances_lock;
  85         struct hlist_head instance_table[INSTANCE_BUCKETS];
  86         atomic_t global_seq;
  87 };
  88
  89 static struct nfnl_log_net *nfnl_log_pernet(struct net *net)
  90 {
  91         return net_generic(net, nfnl_log_net_id);
  92 }
  93
  94 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  95 {
  96         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  97 }
  98
  99 static struct nfulnl_instance *
 100 __instance_lookup(struct nfnl_log_net *log, u_int16_t group_num)
 101 {
 102         struct hlist_head *head;
 103         struct nfulnl_instance *inst;
 104
 105         head = &log->instance_table[instance_hashfn(group_num)];
 106         hlist_for_each_entry_rcu(inst, head, hlist) {
 107                 if (inst->group_num == group_num)
 108                         return inst;
 109         }
 110         return NULL;
 111 }
 112
 113 static inline void
 114 instance_get(struct nfulnl_instance *inst)
 115 {
 116         atomic_inc(&inst->use);
 117 }
 118
 119 static struct nfulnl_instance *
 120 instance_lookup_get(struct nfnl_log_net *log, u_int16_t group_num)
 121 {
 122         struct nfulnl_instance *inst;
 123
 124         rcu_read_lock_bh();
 125         inst = __instance_lookup(log, group_num);
 126         if (inst && !atomic_inc_not_zero(&inst->use))
 127                 inst = NULL;
 128         rcu_read_unlock_bh();
 129
 130         return inst;
 131 }
 132
 133 static void nfulnl_instance_free_rcu(struct rcu_head *head)
 134 {
 135         struct nfulnl_instance *inst =
 136                 container_of(head, struct nfulnl_instance, rcu);
 137
 138         put_net(inst->net);
 139         kfree(inst);
 140         module_put(THIS_MODULE);
 141 }
 142
 143 static void
 144 instance_put(struct nfulnl_instance *inst)
 145 {
 146         if (inst && atomic_dec_and_test(&inst->use))
 147                 call_rcu_bh(&inst->rcu, nfulnl_instance_free_rcu);
 148 }
 149
 150 static void nfulnl_timer(unsigned long data);
 151
 152 static struct nfulnl_instance *
 153 instance_create(struct net *net, u_int16_t group_num,
 154                 int portid, struct user_namespace *user_ns)
 155 {
 156         struct nfulnl_instance *inst;
 157         struct nfnl_log_net *log = nfnl_log_pernet(net);
 158         int err;
 159
 160         spin_lock_bh(&log->instances_lock);
 161         if (__instance_lookup(log, group_num)) {
 162                 err = -EEXIST;
 163                 goto out_unlock;
 164         }
 165
 166         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 167         if (!inst) {
 168                 err = -ENOMEM;
 169                 goto out_unlock;
 170         }
 171
 172         if (!try_module_get(THIS_MODULE)) {
 173                 kfree(inst);
 174                 err = -EAGAIN;
 175                 goto out_unlock;
 176         }
 177
 178         INIT_HLIST_NODE(&inst->hlist);
 179         spin_lock_init(&inst->lock);
 180         /* needs to be two, since we _put() after creation */
 181         atomic_set(&inst->use, 2);
 182
 183         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
 184
 185         inst->net = get_net(net);
 186         inst->peer_user_ns = user_ns;
 187         inst->peer_portid = portid;
 188         inst->group_num = group_num;
 189
 190         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 191         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 192         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 193         inst->copy_mode         = NFULNL_COPY_PACKET;
 194         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
 195
 196         hlist_add_head_rcu(&inst->hlist,
 197                        &log->instance_table[instance_hashfn(group_num)]);
 198
 199
 200         spin_unlock_bh(&log->instances_lock);
 201
 202         return inst;
 203
 204 out_unlock:
 205         spin_unlock_bh(&log->instances_lock);
 206         return ERR_PTR(err);
 207 }
 208
 209 static void __nfulnl_flush(struct nfulnl_instance *inst);
 210
 211 /* called with BH disabled */
 212 static void
 213 __instance_destroy(struct nfulnl_instance *inst)
 214 {
 215         /* first pull it out of the global list */
 216         hlist_del_rcu(&inst->hlist);
 217
 218         /* then flush all pending packets from skb */
 219
 220         spin_lock(&inst->lock);
 221
 222         /* lockless readers wont be able to use us */
 223         inst->copy_mode = NFULNL_COPY_DISABLED;
 224
 225         if (inst->skb)
 226                 __nfulnl_flush(inst);
 227         spin_unlock(&inst->lock);
 228
 229         /* and finally put the refcount */
 230         instance_put(inst);
 231 }
 232
 233 static inline void
 234 instance_destroy(struct nfnl_log_net *log,
 235                  struct nfulnl_instance *inst)
 236 {
 237         spin_lock_bh(&log->instances_lock);
 238         __instance_destroy(inst);
 239         spin_unlock_bh(&log->instances_lock);
 240 }
 241
 242 static int
 243 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 244                   unsigned int range)
 245 {
 246         int status = 0;
 247
 248         spin_lock_bh(&inst->lock);
 249
 250         switch (mode) {
 251         case NFULNL_COPY_NONE:
 252         case NFULNL_COPY_META:
 253                 inst->copy_mode = mode;
 254                 inst->copy_range = 0;
 255                 break;
 256
 257         case NFULNL_COPY_PACKET:
 258                 inst->copy_mode = mode;
 259                 if (range == 0)
 260                         range = NFULNL_COPY_RANGE_MAX;
 261                 inst->copy_range = min_t(unsigned int,
 262                                          range, NFULNL_COPY_RANGE_MAX);
 263                 break;
 264
 265         default:
 266                 status = -EINVAL;
 267                 break;
 268         }
 269
 270         spin_unlock_bh(&inst->lock);
 271
 272         return status;
 273 }
 274
 275 static int
 276 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 277 {
 278         int status;
 279
 280         spin_lock_bh(&inst->lock);
 281         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 282                 status = -ERANGE;
 283         else if (nlbufsiz > 131072)
 284                 status = -ERANGE;
 285         else {
 286                 inst->nlbufsiz = nlbufsiz;
 287                 status = 0;
 288         }
 289         spin_unlock_bh(&inst->lock);
 290
 291         return status;
 292 }
 293
 294 static int
 295 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 296 {
 297         spin_lock_bh(&inst->lock);
 298         inst->flushtimeout = timeout;
 299         spin_unlock_bh(&inst->lock);
 300
 301         return 0;
 302 }
 303
 304 static int
 305 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 306 {
 307         spin_lock_bh(&inst->lock);
 308         inst->qthreshold = qthresh;
 309         spin_unlock_bh(&inst->lock);
 310
 311         return 0;
 312 }
 313
 314 static int
 315 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 316 {
 317         spin_lock_bh(&inst->lock);
 318         inst->flags = flags;
 319         spin_unlock_bh(&inst->lock);
 320
 321         return 0;
 322 }
 323
 324 static struct sk_buff *
 325 nfulnl_alloc_skb(struct net *net, u32 peer_portid, unsigned int inst_size,
 326                  unsigned int pkt_size)
 327 {
 328         struct sk_buff *skb;
 329         unsigned int n;
 330
 331         /* alloc skb which should be big enough for a whole multipart
 332          * message.  WARNING: has to be <= 128k due to slab restrictions */
 333
 334         n = max(inst_size, pkt_size);
 335         skb = nfnetlink_alloc_skb(net, n, peer_portid, GFP_ATOMIC);
 336         if (!skb) {
 337                 if (n > pkt_size) {
 338                         /* try to allocate only as much as we need for current
 339                          * packet */
 340
 341                         skb = nfnetlink_alloc_skb(net, pkt_size,
 342                                                   peer_portid, GFP_ATOMIC);
 343                         if (!skb)
 344                                 pr_err("nfnetlink_log: can't even alloc %u bytes\n",
 345                                        pkt_size);
 346                 }
 347         }
 348
 349         return skb;
 350 }
 351
 352 static void
 353 __nfulnl_send(struct nfulnl_instance *inst)
 354 {
 355         if (inst->qlen > 1) {
 356                 struct nlmsghdr *nlh = nlmsg_put(inst->skb, 0, 0,
 357                                                  NLMSG_DONE,
 358                                                  sizeof(struct nfgenmsg),
 359                                                  0);
 360                 if (WARN_ONCE(!nlh, "bad nlskb size: %u, tailroom %d\n",
 361                               inst->skb->len, skb_tailroom(inst->skb))) {
 362                         kfree_skb(inst->skb);
 363                         goto out;
 364                 }
 365         }
 366         nfnetlink_unicast(inst->skb, inst->net, inst->peer_portid,
 367                           MSG_DONTWAIT);
 368 out:
 369         inst->qlen = 0;
 370         inst->skb = NULL;
 371 }
 372
 373 static void
 374 __nfulnl_flush(struct nfulnl_instance *inst)
 375 {
 376         /* timer holds a reference */
 377         if (del_timer(&inst->timer))
 378                 instance_put(inst);
 379         if (inst->skb)
 380                 __nfulnl_send(inst);
 381 }
 382
 383 static void
 384 nfulnl_timer(unsigned long data)
 385 {
 386         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 387
 388         spin_lock_bh(&inst->lock);
 389         if (inst->skb)
 390                 __nfulnl_send(inst);
 391         spin_unlock_bh(&inst->lock);
 392         instance_put(inst);
 393 }
 394
 395 /* This is an inline function, we don't really care about a long
 396  * list of arguments */
 397 static inline int
 398 __build_packet_message(struct nfnl_log_net *log,
 399                         struct nfulnl_instance *inst,
 400                         const struct sk_buff *skb,
 401                         unsigned int data_len,
 402                         u_int8_t pf,
 403                         unsigned int hooknum,
 404                         const struct net_device *indev,
 405                         const struct net_device *outdev,
 406                         const char *prefix, unsigned int plen)
 407 {
 408         struct nfulnl_msg_packet_hdr pmsg;
 409         struct nlmsghdr *nlh;
 410         struct nfgenmsg *nfmsg;
 411         sk_buff_data_t old_tail = inst->skb->tail;
 412         struct sock *sk;
 413         const unsigned char *hwhdrp;
 414
 415         nlh = nlmsg_put(inst->skb, 0, 0,
 416                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 417                         sizeof(struct nfgenmsg), 0);
 418         if (!nlh)
 419                 return -1;
 420         nfmsg = nlmsg_data(nlh);
 421         nfmsg->nfgen_family = pf;
 422         nfmsg->version = NFNETLINK_V0;
 423         nfmsg->res_id = htons(inst->group_num);
 424
 425         memset(&pmsg, 0, sizeof(pmsg));
 426         pmsg.hw_protocol        = skb->protocol;
 427         pmsg.hook               = hooknum;
 428
 429         if (nla_put(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg))
 430                 goto nla_put_failure;
 431
 432         if (prefix &&
 433             nla_put(inst->skb, NFULA_PREFIX, plen, prefix))
 434                 goto nla_put_failure;
 435
 436         if (indev) {
 437 #ifndef CONFIG_BRIDGE_NETFILTER
 438                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 439                                  htonl(indev->ifindex)))
 440                         goto nla_put_failure;
 441 #else
 442                 if (pf == PF_BRIDGE) {
 443                         /* Case 1: outdev is physical input device, we need to
 444                          * look for bridge group (when called from
 445                          * netfilter_bridge) */
 446                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 447                                          htonl(indev->ifindex)) ||
 448                         /* this is the bridge group "brX" */
 449                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 450                             nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 451                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 452                                 goto nla_put_failure;
 453                 } else {
 454                         /* Case 2: indev is bridge group, we need to look for
 455                          * physical device (when called from ipv4) */
 456                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 457                                          htonl(indev->ifindex)))
 458                                 goto nla_put_failure;
 459                         if (skb->nf_bridge && skb->nf_bridge->physindev &&
 460                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 461                                          htonl(skb->nf_bridge->physindev->ifindex)))
 462                                 goto nla_put_failure;
 463                 }
 464 #endif
 465         }
 466
 467         if (outdev) {
 468 #ifndef CONFIG_BRIDGE_NETFILTER
 469                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 470                                  htonl(outdev->ifindex)))
 471                         goto nla_put_failure;
 472 #else
 473                 if (pf == PF_BRIDGE) {
 474                         /* Case 1: outdev is physical output device, we need to
 475                          * look for bridge group (when called from
 476                          * netfilter_bridge) */
 477                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 478                                          htonl(outdev->ifindex)) ||
 479                         /* this is the bridge group "brX" */
 480                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 481                             nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 482                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 483                                 goto nla_put_failure;
 484                 } else {
 485                         /* Case 2: indev is a bridge group, we need to look
 486                          * for physical device (when called from ipv4) */
 487                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 488                                          htonl(outdev->ifindex)))
 489                                 goto nla_put_failure;
 490                         if (skb->nf_bridge && skb->nf_bridge->physoutdev &&
 491                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 492                                          htonl(skb->nf_bridge->physoutdev->ifindex)))
 493                                 goto nla_put_failure;
 494                 }
 495 #endif
 496         }
 497
 498         if (skb->mark &&
 499             nla_put_be32(inst->skb, NFULA_MARK, htonl(skb->mark)))
 500                 goto nla_put_failure;
 501
 502         if (indev && skb->dev &&
 503             skb->mac_header != skb->network_header) {
 504                 struct nfulnl_msg_packet_hw phw;
 505                 int len;
 506
 507                 memset(&phw, 0, sizeof(phw));
 508                 len = dev_parse_header(skb, phw.hw_addr);
 509                 if (len > 0) {
 510                         phw.hw_addrlen = htons(len);
 511                         if (nla_put(inst->skb, NFULA_HWADDR, sizeof(phw), &phw))
 512                                 goto nla_put_failure;
 513                 }
 514         }
 515
 516         if (indev && skb_mac_header_was_set(skb)) {
 517                 if (nla_put_be16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type)) ||
 518                     nla_put_be16(inst->skb, NFULA_HWLEN,
 519                                  htons(skb->dev->hard_header_len)))
 520                         goto nla_put_failure;
 521
 522                 hwhdrp = skb_mac_header(skb);
 523
 524                 if (skb->dev->type == ARPHRD_SIT)
 525                         hwhdrp -= ETH_HLEN;
 526
 527                 if (hwhdrp >= skb->head &&
 528                     nla_put(inst->skb, NFULA_HWHEADER,
 529                             skb->dev->hard_header_len, hwhdrp))
 530                         goto nla_put_failure;
 531         }
 532
 533         if (skb->tstamp.tv64) {
 534                 struct nfulnl_msg_packet_timestamp ts;
 535                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 536                 ts.sec = cpu_to_be64(tv.tv_sec);
 537                 ts.usec = cpu_to_be64(tv.tv_usec);
 538
 539                 if (nla_put(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts))
 540                         goto nla_put_failure;
 541         }
 542
 543         /* UID */
 544         sk = skb->sk;
 545         if (sk && sk->sk_state != TCP_TIME_WAIT) {
 546                 read_lock_bh(&sk->sk_callback_lock);
 547                 if (sk->sk_socket && sk->sk_socket->file) {
 548                         struct file *file = sk->sk_socket->file;
 549                         const struct cred *cred = file->f_cred;
 550                         struct user_namespace *user_ns = inst->peer_user_ns;
 551                         __be32 uid = htonl(from_kuid_munged(user_ns, cred->fsuid));
 552                         __be32 gid = htonl(from_kgid_munged(user_ns, cred->fsgid));
 553                         read_unlock_bh(&sk->sk_callback_lock);
 554                         if (nla_put_be32(inst->skb, NFULA_UID, uid) ||
 555                             nla_put_be32(inst->skb, NFULA_GID, gid))
 556                                 goto nla_put_failure;
 557                 } else
 558                         read_unlock_bh(&sk->sk_callback_lock);
 559         }
 560
 561         /* local sequence number */
 562         if ((inst->flags & NFULNL_CFG_F_SEQ) &&
 563             nla_put_be32(inst->skb, NFULA_SEQ, htonl(inst->seq++)))
 564                 goto nla_put_failure;
 565
 566         /* global sequence number */
 567         if ((inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) &&
 568             nla_put_be32(inst->skb, NFULA_SEQ_GLOBAL,
 569                          htonl(atomic_inc_return(&log->global_seq))))
 570                 goto nla_put_failure;
 571
 572         if (data_len) {
 573                 struct nlattr *nla;
 574                 int size = nla_attr_size(data_len);
 575
 576                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
 577                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 578                         return -1;
 579                 }
 580
 581                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 582                 nla->nla_type = NFULA_PAYLOAD;
 583                 nla->nla_len = size;
 584
 585                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 586                         BUG();
 587         }
 588
 589         nlh->nlmsg_len = inst->skb->tail - old_tail;
 590         return 0;
 591
 592 nla_put_failure:
 593         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 594         return -1;
 595 }
 596
 597 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 598
 599 static struct nf_loginfo default_loginfo = {
 600         .type =         NF_LOG_TYPE_ULOG,
 601         .u = {
 602                 .ulog = {
 603                         .copy_len       = 0xffff,
 604                         .group          = 0,
 605                         .qthreshold     = 1,
 606                 },
 607         },
 608 };
 609
 610 /* log handler for internal netfilter logging api */
 611 void
 612 nfulnl_log_packet(struct net *net,
 613                   u_int8_t pf,
 614                   unsigned int hooknum,
 615                   const struct sk_buff *skb,
 616                   const struct net_device *in,
 617                   const struct net_device *out,
 618                   const struct nf_loginfo *li_user,
 619                   const char *prefix)
 620 {
 621         unsigned int size, data_len;
 622         struct nfulnl_instance *inst;
 623         const struct nf_loginfo *li;
 624         unsigned int qthreshold;
 625         unsigned int plen;
 626         struct nfnl_log_net *log = nfnl_log_pernet(net);
 627
 628         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 629                 li = li_user;
 630         else
 631                 li = &default_loginfo;
 632
 633         inst = instance_lookup_get(log, li->u.ulog.group);
 634         if (!inst)
 635                 return;
 636
 637         plen = 0;
 638         if (prefix)
 639                 plen = strlen(prefix) + 1;
 640
 641         /* FIXME: do we want to make the size calculation conditional based on
 642          * what is actually present?  way more branches and checks, but more
 643          * memory efficient... */
 644         size =    nlmsg_total_size(sizeof(struct nfgenmsg))
 645                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 646                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 647                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 648 #ifdef CONFIG_BRIDGE_NETFILTER
 649                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 650                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 651 #endif
 652                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 653                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 654                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 655                 + nla_total_size(plen)                  /* prefix */
 656                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 657                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp))
 658                 + nla_total_size(sizeof(struct nfgenmsg));      /* NLMSG_DONE */
 659
 660         if (in && skb_mac_header_was_set(skb)) {
 661                 size +=   nla_total_size(skb->dev->hard_header_len)
 662                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 663                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 664         }
 665
 666         spin_lock_bh(&inst->lock);
 667
 668         if (inst->flags & NFULNL_CFG_F_SEQ)
 669                 size += nla_total_size(sizeof(u_int32_t));
 670         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 671                 size += nla_total_size(sizeof(u_int32_t));
 672
 673         qthreshold = inst->qthreshold;
 674         /* per-rule qthreshold overrides per-instance */
 675         if (li->u.ulog.qthreshold)
 676                 if (qthreshold > li->u.ulog.qthreshold)
 677                         qthreshold = li->u.ulog.qthreshold;
 678
 679
 680         switch (inst->copy_mode) {
 681         case NFULNL_COPY_META:
 682         case NFULNL_COPY_NONE:
 683                 data_len = 0;
 684                 break;
 685
 686         case NFULNL_COPY_PACKET:
 687                 if (inst->copy_range > skb->len)
 688                         data_len = skb->len;
 689                 else
 690                         data_len = inst->copy_range;
 691
 692                 size += nla_total_size(data_len);
 693                 break;
 694
 695         case NFULNL_COPY_DISABLED:
 696         default:
 697                 goto unlock_and_release;
 698         }
 699
 700         if (inst->skb && size > skb_tailroom(inst->skb)) {
 701                 /* either the queue len is too high or we don't have
 702                  * enough room in the skb left. flush to userspace. */
 703                 __nfulnl_flush(inst);
 704         }
 705
 706         if (!inst->skb) {
 707                 inst->skb = nfulnl_alloc_skb(net, inst->peer_portid,
 708                                              inst->nlbufsiz, size);
 709                 if (!inst->skb)
 710                         goto alloc_failure;
 711         }
 712
 713         inst->qlen++;
 714
 715         __build_packet_message(log, inst, skb, data_len, pf,
 716                                 hooknum, in, out, prefix, plen);
 717
 718         if (inst->qlen >= qthreshold)
 719                 __nfulnl_flush(inst);
 720         /* timer_pending always called within inst->lock, so there
 721          * is no chance of a race here */
 722         else if (!timer_pending(&inst->timer)) {
 723                 instance_get(inst);
 724                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 725                 add_timer(&inst->timer);
 726         }
 727
 728 unlock_and_release:
 729         spin_unlock_bh(&inst->lock);
 730         instance_put(inst);
 731         return;
 732
 733 alloc_failure:
 734         /* FIXME: statistics */
 735         goto unlock_and_release;
 736 }
 737 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
 738
 739 static int
 740 nfulnl_rcv_nl_event(struct notifier_block *this,
 741                    unsigned long event, void *ptr)
 742 {
 743         struct netlink_notify *n = ptr;
 744         struct nfnl_log_net *log = nfnl_log_pernet(n->net);
 745
 746         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 747                 int i;
 748
 749                 /* destroy all instances for this portid */
 750                 spin_lock_bh(&log->instances_lock);
 751                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 752                         struct hlist_node *t2;
 753                         struct nfulnl_instance *inst;
 754                         struct hlist_head *head = &log->instance_table[i];
 755
 756                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 757                                 if (n->portid == inst->peer_portid)
 758                                         __instance_destroy(inst);
 759                         }
 760                 }
 761                 spin_unlock_bh(&log->instances_lock);
 762         }
 763         return NOTIFY_DONE;
 764 }
 765
 766 static struct notifier_block nfulnl_rtnl_notifier = {
 767         .notifier_call  = nfulnl_rcv_nl_event,
 768 };
 769
 770 static int
 771 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 772                    const struct nlmsghdr *nlh,
 773                    const struct nlattr * const nfqa[])
 774 {
 775         return -ENOTSUPP;
 776 }
 777
 778 static struct nf_logger nfulnl_logger __read_mostly = {
 779         .name   = "nfnetlink_log",
 780         .logfn  = &nfulnl_log_packet,
 781         .me     = THIS_MODULE,
 782 };
 783
 784 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 785         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 786         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 787         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 788         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 789         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 790         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 791 };
 792
 793 static int
 794 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 795                    const struct nlmsghdr *nlh,
 796                    const struct nlattr * const nfula[])
 797 {
 798         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 799         u_int16_t group_num = ntohs(nfmsg->res_id);
 800         struct nfulnl_instance *inst;
 801         struct nfulnl_msg_config_cmd *cmd = NULL;
 802         struct net *net = sock_net(ctnl);
 803         struct nfnl_log_net *log = nfnl_log_pernet(net);
 804         int ret = 0;
 805
 806         if (nfula[NFULA_CFG_CMD]) {
 807                 u_int8_t pf = nfmsg->nfgen_family;
 808                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 809
 810                 /* Commands without queue context */
 811                 switch (cmd->command) {
 812                 case NFULNL_CFG_CMD_PF_BIND:
 813                         return nf_log_bind_pf(net, pf, &nfulnl_logger);
 814                 case NFULNL_CFG_CMD_PF_UNBIND:
 815                         nf_log_unbind_pf(net, pf);
 816                         return 0;
 817                 }
 818         }
 819
 820         inst = instance_lookup_get(log, group_num);
 821         if (inst && inst->peer_portid != NETLINK_CB(skb).portid) {
 822                 ret = -EPERM;
 823                 goto out_put;
 824         }
 825
 826         if (cmd != NULL) {
 827                 switch (cmd->command) {
 828                 case NFULNL_CFG_CMD_BIND:
 829                         if (inst) {
 830                                 ret = -EBUSY;
 831                                 goto out_put;
 832                         }
 833
 834                         inst = instance_create(net, group_num,
 835                                                NETLINK_CB(skb).portid,
 836                                                sk_user_ns(NETLINK_CB(skb).sk));
 837                         if (IS_ERR(inst)) {
 838                                 ret = PTR_ERR(inst);
 839                                 goto out;
 840                         }
 841                         break;
 842                 case NFULNL_CFG_CMD_UNBIND:
 843                         if (!inst) {
 844                                 ret = -ENODEV;
 845                                 goto out;
 846                         }
 847
 848                         instance_destroy(log, inst);
 849                         goto out_put;
 850                 default:
 851                         ret = -ENOTSUPP;
 852                         break;
 853                 }
 854         }
 855
 856         if (nfula[NFULA_CFG_MODE]) {
 857                 struct nfulnl_msg_config_mode *params;
 858                 params = nla_data(nfula[NFULA_CFG_MODE]);
 859
 860                 if (!inst) {
 861                         ret = -ENODEV;
 862                         goto out;
 863                 }
 864                 nfulnl_set_mode(inst, params->copy_mode,
 865                                 ntohl(params->copy_range));
 866         }
 867
 868         if (nfula[NFULA_CFG_TIMEOUT]) {
 869                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 870
 871                 if (!inst) {
 872                         ret = -ENODEV;
 873                         goto out;
 874                 }
 875                 nfulnl_set_timeout(inst, ntohl(timeout));
 876         }
 877
 878         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 879                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 880
 881                 if (!inst) {
 882                         ret = -ENODEV;
 883                         goto out;
 884                 }
 885                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 886         }
 887
 888         if (nfula[NFULA_CFG_QTHRESH]) {
 889                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 890
 891                 if (!inst) {
 892                         ret = -ENODEV;
 893                         goto out;
 894                 }
 895                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 896         }
 897
 898         if (nfula[NFULA_CFG_FLAGS]) {
 899                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
 900
 901                 if (!inst) {
 902                         ret = -ENODEV;
 903                         goto out;
 904                 }
 905                 nfulnl_set_flags(inst, ntohs(flags));
 906         }
 907
 908 out_put:
 909         instance_put(inst);
 910 out:
 911         return ret;
 912 }
 913
 914 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 915         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 916                                     .attr_count = NFULA_MAX, },
 917         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 918                                     .attr_count = NFULA_CFG_MAX,
 919                                     .policy = nfula_cfg_policy },
 920 };
 921
 922 static const struct nfnetlink_subsystem nfulnl_subsys = {
 923         .name           = "log",
 924         .subsys_id      = NFNL_SUBSYS_ULOG,
 925         .cb_count       = NFULNL_MSG_MAX,
 926         .cb             = nfulnl_cb,
 927 };
 928
 929 #ifdef CONFIG_PROC_FS
 930 struct iter_state {
 931         struct seq_net_private p;
 932         unsigned int bucket;
 933 };
 934
 935 static struct hlist_node *get_first(struct net *net, struct iter_state *st)
 936 {
 937         struct nfnl_log_net *log;
 938         if (!st)
 939                 return NULL;
 940
 941         log = nfnl_log_pernet(net);
 942
 943         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 944                 struct hlist_head *head = &log->instance_table[st->bucket];
 945
 946                 if (!hlist_empty(head))
 947                         return rcu_dereference_bh(hlist_first_rcu(head));
 948         }
 949         return NULL;
 950 }
 951
 952 static struct hlist_node *get_next(struct net *net, struct iter_state *st,
 953                                    struct hlist_node *h)
 954 {
 955         h = rcu_dereference_bh(hlist_next_rcu(h));
 956         while (!h) {
 957                 struct nfnl_log_net *log;
 958                 struct hlist_head *head;
 959
 960                 if (++st->bucket >= INSTANCE_BUCKETS)
 961                         return NULL;
 962
 963                 log = nfnl_log_pernet(net);
 964                 head = &log->instance_table[st->bucket];
 965                 h = rcu_dereference_bh(hlist_first_rcu(head));
 966         }
 967         return h;
 968 }
 969
 970 static struct hlist_node *get_idx(struct net *net, struct iter_state *st,
 971                                   loff_t pos)
 972 {
 973         struct hlist_node *head;
 974         head = get_first(net, st);
 975
 976         if (head)
 977                 while (pos && (head = get_next(net, st, head)))
 978                         pos--;
 979         return pos ? NULL : head;
 980 }
 981
 982 static void *seq_start(struct seq_file *s, loff_t *pos)
 983         __acquires(rcu_bh)
 984 {
 985         rcu_read_lock_bh();
 986         return get_idx(seq_file_net(s), s->private, *pos);
 987 }
 988
 989 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 990 {
 991         (*pos)++;
 992         return get_next(seq_file_net(s), s->private, v);
 993 }
 994
 995 static void seq_stop(struct seq_file *s, void *v)
 996         __releases(rcu_bh)
 997 {
 998         rcu_read_unlock_bh();
 999 }
1000
1001 static int seq_show(struct seq_file *s, void *v)
1002 {
1003         const struct nfulnl_instance *inst = v;
1004
1005         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
1006                           inst->group_num,
1007                           inst->peer_portid, inst->qlen,
1008                           inst->copy_mode, inst->copy_range,
1009                           inst->flushtimeout, atomic_read(&inst->use));
1010 }
1011
1012 static const struct seq_operations nful_seq_ops = {
1013         .start  = seq_start,
1014         .next   = seq_next,
1015         .stop   = seq_stop,
1016         .show   = seq_show,
1017 };
1018
1019 static int nful_open(struct inode *inode, struct file *file)
1020 {
1021         return seq_open_net(inode, file, &nful_seq_ops,
1022                             sizeof(struct iter_state));
1023 }
1024
1025 static const struct file_operations nful_file_ops = {
1026         .owner   = THIS_MODULE,
1027         .open    = nful_open,
1028         .read    = seq_read,
1029         .llseek  = seq_lseek,
1030         .release = seq_release_net,
1031 };
1032
1033 #endif /* PROC_FS */
1034
1035 static int __net_init nfnl_log_net_init(struct net *net)
1036 {
1037         unsigned int i;
1038         struct nfnl_log_net *log = nfnl_log_pernet(net);
1039
1040         for (i = 0; i < INSTANCE_BUCKETS; i++)
1041                 INIT_HLIST_HEAD(&log->instance_table[i]);
1042         spin_lock_init(&log->instances_lock);
1043
1044 #ifdef CONFIG_PROC_FS
1045         if (!proc_create("nfnetlink_log", 0440,
1046                          net->nf.proc_netfilter, &nful_file_ops))
1047                 return -ENOMEM;
1048 #endif
1049         return 0;
1050 }
1051
1052 static void __net_exit nfnl_log_net_exit(struct net *net)
1053 {
1054 #ifdef CONFIG_PROC_FS
1055         remove_proc_entry("nfnetlink_log", net->nf.proc_netfilter);
1056 #endif
1057         nf_log_unset(net, &nfulnl_logger);
1058 }
1059
1060 static struct pernet_operations nfnl_log_net_ops = {
1061         .init   = nfnl_log_net_init,
1062         .exit   = nfnl_log_net_exit,
1063         .id     = &nfnl_log_net_id,
1064         .size   = sizeof(struct nfnl_log_net),
1065 };
1066
1067 static int __init nfnetlink_log_init(void)
1068 {
1069         int status = -ENOMEM;
1070
1071         /* it's not really all that important to have a random value, so
1072          * we can do this from the init function, even if there hasn't
1073          * been that much entropy yet */
1074         get_random_bytes(&hash_init, sizeof(hash_init));
1075
1076         netlink_register_notifier(&nfulnl_rtnl_notifier);
1077         status = nfnetlink_subsys_register(&nfulnl_subsys);
1078         if (status < 0) {
1079                 pr_err("log: failed to create netlink socket\n");
1080                 goto cleanup_netlink_notifier;
1081         }
1082
1083         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
1084         if (status < 0) {
1085                 pr_err("log: failed to register logger\n");
1086                 goto cleanup_subsys;
1087         }
1088
1089         status = register_pernet_subsys(&nfnl_log_net_ops);
1090         if (status < 0) {
1091                 pr_err("log: failed to register pernet ops\n");
1092                 goto cleanup_logger;
1093         }
1094         return status;
1095
1096 cleanup_logger:
1097         nf_log_unregister(&nfulnl_logger);
1098 cleanup_subsys:
1099         nfnetlink_subsys_unregister(&nfulnl_subsys);
1100 cleanup_netlink_notifier:
1101         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1102         return status;
1103 }
1104
1105 static void __exit nfnetlink_log_fini(void)
1106 {
1107         unregister_pernet_subsys(&nfnl_log_net_ops);
1108         nf_log_unregister(&nfulnl_logger);
1109         nfnetlink_subsys_unregister(&nfulnl_subsys);
1110         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1111 }
1112
1113 MODULE_DESCRIPTION("netfilter userspace logging");
1114 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1115 MODULE_LICENSE("GPL");
1116 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1117
1118 module_init(nfnetlink_log_init);
1119 module_exit(nfnetlink_log_fini);