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  * (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(struct net *net, u32 peer_portid, unsigned int inst_size,
 323                  unsigned int pkt_size)
 324 {
 325         struct sk_buff *skb;
 326         unsigned int n;
 327
 328         /* alloc skb which should be big enough for a whole multipart
 329          * message.  WARNING: has to be <= 128k due to slab restrictions */
 330
 331         n = max(inst_size, pkt_size);
 332         skb = nfnetlink_alloc_skb(net, n, peer_portid, GFP_ATOMIC);
 333         if (!skb) {
 334                 if (n > pkt_size) {
 335                         /* try to allocate only as much as we need for current
 336                          * packet */
 337
 338                         skb = nfnetlink_alloc_skb(net, pkt_size,
 339                                                   peer_portid, GFP_ATOMIC);
 340                         if (!skb)
 341                                 pr_err("nfnetlink_log: can't even alloc %u bytes\n",
 342                                        pkt_size);
 343                 }
 344         }
 345
 346         return skb;
 347 }
 348
 349 static int
 350 __nfulnl_send(struct nfulnl_instance *inst)
 351 {
 352         int status = -1;
 353
 354         if (inst->qlen > 1) {
 355                 struct nlmsghdr *nlh = nlmsg_put(inst->skb, 0, 0,
 356                                                  NLMSG_DONE,
 357                                                  sizeof(struct nfgenmsg),
 358                                                  0);
 359                 if (!nlh)
 360                         goto out;
 361         }
 362         status = nfnetlink_unicast(inst->skb, inst->net, inst->peer_portid,
 363                                    MSG_DONTWAIT);
 364
 365         inst->qlen = 0;
 366         inst->skb = NULL;
 367 out:
 368         return status;
 369 }
 370
 371 static void
 372 __nfulnl_flush(struct nfulnl_instance *inst)
 373 {
 374         /* timer holds a reference */
 375         if (del_timer(&inst->timer))
 376                 instance_put(inst);
 377         if (inst->skb)
 378                 __nfulnl_send(inst);
 379 }
 380
 381 static void
 382 nfulnl_timer(unsigned long data)
 383 {
 384         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 385
 386         spin_lock_bh(&inst->lock);
 387         if (inst->skb)
 388                 __nfulnl_send(inst);
 389         spin_unlock_bh(&inst->lock);
 390         instance_put(inst);
 391 }
 392
 393 /* This is an inline function, we don't really care about a long
 394  * list of arguments */
 395 static inline int
 396 __build_packet_message(struct nfnl_log_net *log,
 397                         struct nfulnl_instance *inst,
 398                         const struct sk_buff *skb,
 399                         unsigned int data_len,
 400                         u_int8_t pf,
 401                         unsigned int hooknum,
 402                         const struct net_device *indev,
 403                         const struct net_device *outdev,
 404                         const char *prefix, unsigned int plen)
 405 {
 406         struct nfulnl_msg_packet_hdr pmsg;
 407         struct nlmsghdr *nlh;
 408         struct nfgenmsg *nfmsg;
 409         sk_buff_data_t old_tail = inst->skb->tail;
 410         struct sock *sk;
 411         const unsigned char *hwhdrp;
 412
 413         nlh = nlmsg_put(inst->skb, 0, 0,
 414                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 415                         sizeof(struct nfgenmsg), 0);
 416         if (!nlh)
 417                 return -1;
 418         nfmsg = nlmsg_data(nlh);
 419         nfmsg->nfgen_family = pf;
 420         nfmsg->version = NFNETLINK_V0;
 421         nfmsg->res_id = htons(inst->group_num);
 422
 423         memset(&pmsg, 0, sizeof(pmsg));
 424         pmsg.hw_protocol        = skb->protocol;
 425         pmsg.hook               = hooknum;
 426
 427         if (nla_put(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg))
 428                 goto nla_put_failure;
 429
 430         if (prefix &&
 431             nla_put(inst->skb, NFULA_PREFIX, plen, prefix))
 432                 goto nla_put_failure;
 433
 434         if (indev) {
 435 #ifndef CONFIG_BRIDGE_NETFILTER
 436                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 437                                  htonl(indev->ifindex)))
 438                         goto nla_put_failure;
 439 #else
 440                 if (pf == PF_BRIDGE) {
 441                         /* Case 1: outdev is physical input device, we need to
 442                          * look for bridge group (when called from
 443                          * netfilter_bridge) */
 444                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 445                                          htonl(indev->ifindex)) ||
 446                         /* this is the bridge group "brX" */
 447                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 448                             nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 449                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 450                                 goto nla_put_failure;
 451                 } else {
 452                         /* Case 2: indev is bridge group, we need to look for
 453                          * physical device (when called from ipv4) */
 454                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 455                                          htonl(indev->ifindex)))
 456                                 goto nla_put_failure;
 457                         if (skb->nf_bridge && skb->nf_bridge->physindev &&
 458                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 459                                          htonl(skb->nf_bridge->physindev->ifindex)))
 460                                 goto nla_put_failure;
 461                 }
 462 #endif
 463         }
 464
 465         if (outdev) {
 466 #ifndef CONFIG_BRIDGE_NETFILTER
 467                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 468                                  htonl(outdev->ifindex)))
 469                         goto nla_put_failure;
 470 #else
 471                 if (pf == PF_BRIDGE) {
 472                         /* Case 1: outdev is physical output device, we need to
 473                          * look for bridge group (when called from
 474                          * netfilter_bridge) */
 475                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 476                                          htonl(outdev->ifindex)) ||
 477                         /* this is the bridge group "brX" */
 478                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 479                             nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 480                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 481                                 goto nla_put_failure;
 482                 } else {
 483                         /* Case 2: indev is a bridge group, we need to look
 484                          * for physical device (when called from ipv4) */
 485                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 486                                          htonl(outdev->ifindex)))
 487                                 goto nla_put_failure;
 488                         if (skb->nf_bridge && skb->nf_bridge->physoutdev &&
 489                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 490                                          htonl(skb->nf_bridge->physoutdev->ifindex)))
 491                                 goto nla_put_failure;
 492                 }
 493 #endif
 494         }
 495
 496         if (skb->mark &&
 497             nla_put_be32(inst->skb, NFULA_MARK, htonl(skb->mark)))
 498                 goto nla_put_failure;
 499
 500         if (indev && skb->dev &&
 501             skb->mac_header != skb->network_header) {
 502                 struct nfulnl_msg_packet_hw phw;
 503                 int len;
 504
 505                 memset(&phw, 0, sizeof(phw));
 506                 len = dev_parse_header(skb, phw.hw_addr);
 507                 if (len > 0) {
 508                         phw.hw_addrlen = htons(len);
 509                         if (nla_put(inst->skb, NFULA_HWADDR, sizeof(phw), &phw))
 510                                 goto nla_put_failure;
 511                 }
 512         }
 513
 514         if (indev && skb_mac_header_was_set(skb)) {
 515                 if (nla_put_be16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type)) ||
 516                     nla_put_be16(inst->skb, NFULA_HWLEN,
 517                                  htons(skb->dev->hard_header_len)))
 518                         goto nla_put_failure;
 519
 520                 hwhdrp = skb_mac_header(skb);
 521
 522                 if (skb->dev->type == ARPHRD_SIT)
 523                         hwhdrp -= ETH_HLEN;
 524
 525                 if (hwhdrp >= skb->head &&
 526                     nla_put(inst->skb, NFULA_HWHEADER,
 527                             skb->dev->hard_header_len, hwhdrp))
 528                         goto nla_put_failure;
 529         }
 530
 531         if (skb->tstamp.tv64) {
 532                 struct nfulnl_msg_packet_timestamp ts;
 533                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 534                 ts.sec = cpu_to_be64(tv.tv_sec);
 535                 ts.usec = cpu_to_be64(tv.tv_usec);
 536
 537                 if (nla_put(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts))
 538                         goto nla_put_failure;
 539         }
 540
 541         /* UID */
 542         sk = skb->sk;
 543         if (sk && sk->sk_state != TCP_TIME_WAIT) {
 544                 read_lock_bh(&sk->sk_callback_lock);
 545                 if (sk->sk_socket && sk->sk_socket->file) {
 546                         struct file *file = sk->sk_socket->file;
 547                         const struct cred *cred = file->f_cred;
 548                         struct user_namespace *user_ns = inst->peer_user_ns;
 549                         __be32 uid = htonl(from_kuid_munged(user_ns, cred->fsuid));
 550                         __be32 gid = htonl(from_kgid_munged(user_ns, cred->fsgid));
 551                         read_unlock_bh(&sk->sk_callback_lock);
 552                         if (nla_put_be32(inst->skb, NFULA_UID, uid) ||
 553                             nla_put_be32(inst->skb, NFULA_GID, gid))
 554                                 goto nla_put_failure;
 555                 } else
 556                         read_unlock_bh(&sk->sk_callback_lock);
 557         }
 558
 559         /* local sequence number */
 560         if ((inst->flags & NFULNL_CFG_F_SEQ) &&
 561             nla_put_be32(inst->skb, NFULA_SEQ, htonl(inst->seq++)))
 562                 goto nla_put_failure;
 563
 564         /* global sequence number */
 565         if ((inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) &&
 566             nla_put_be32(inst->skb, NFULA_SEQ_GLOBAL,
 567                          htonl(atomic_inc_return(&log->global_seq))))
 568                 goto nla_put_failure;
 569
 570         if (data_len) {
 571                 struct nlattr *nla;
 572                 int size = nla_attr_size(data_len);
 573
 574                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
 575                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 576                         return -1;
 577                 }
 578
 579                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 580                 nla->nla_type = NFULA_PAYLOAD;
 581                 nla->nla_len = size;
 582
 583                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 584                         BUG();
 585         }
 586
 587         nlh->nlmsg_len = inst->skb->tail - old_tail;
 588         return 0;
 589
 590 nla_put_failure:
 591         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 592         return -1;
 593 }
 594
 595 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 596
 597 static struct nf_loginfo default_loginfo = {
 598         .type =         NF_LOG_TYPE_ULOG,
 599         .u = {
 600                 .ulog = {
 601                         .copy_len       = 0xffff,
 602                         .group          = 0,
 603                         .qthreshold     = 1,
 604                 },
 605         },
 606 };
 607
 608 /* log handler for internal netfilter logging api */
 609 void
 610 nfulnl_log_packet(struct net *net,
 611                   u_int8_t pf,
 612                   unsigned int hooknum,
 613                   const struct sk_buff *skb,
 614                   const struct net_device *in,
 615                   const struct net_device *out,
 616                   const struct nf_loginfo *li_user,
 617                   const char *prefix)
 618 {
 619         unsigned int size, data_len;
 620         struct nfulnl_instance *inst;
 621         const struct nf_loginfo *li;
 622         unsigned int qthreshold;
 623         unsigned int plen;
 624         struct nfnl_log_net *log = nfnl_log_pernet(net);
 625
 626         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 627                 li = li_user;
 628         else
 629                 li = &default_loginfo;
 630
 631         inst = instance_lookup_get(log, li->u.ulog.group);
 632         if (!inst)
 633                 return;
 634
 635         plen = 0;
 636         if (prefix)
 637                 plen = strlen(prefix) + 1;
 638
 639         /* FIXME: do we want to make the size calculation conditional based on
 640          * what is actually present?  way more branches and checks, but more
 641          * memory efficient... */
 642         size =    nlmsg_total_size(sizeof(struct nfgenmsg))
 643                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 644                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 645                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 646 #ifdef CONFIG_BRIDGE_NETFILTER
 647                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 648                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 649 #endif
 650                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 651                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 652                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 653                 + nla_total_size(plen)                  /* prefix */
 654                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 655                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
 656
 657         if (in && skb_mac_header_was_set(skb)) {
 658                 size +=   nla_total_size(skb->dev->hard_header_len)
 659                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 660                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 661         }
 662
 663         spin_lock_bh(&inst->lock);
 664
 665         if (inst->flags & NFULNL_CFG_F_SEQ)
 666                 size += nla_total_size(sizeof(u_int32_t));
 667         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 668                 size += nla_total_size(sizeof(u_int32_t));
 669
 670         qthreshold = inst->qthreshold;
 671         /* per-rule qthreshold overrides per-instance */
 672         if (li->u.ulog.qthreshold)
 673                 if (qthreshold > li->u.ulog.qthreshold)
 674                         qthreshold = li->u.ulog.qthreshold;
 675
 676
 677         switch (inst->copy_mode) {
 678         case NFULNL_COPY_META:
 679         case NFULNL_COPY_NONE:
 680                 data_len = 0;
 681                 break;
 682
 683         case NFULNL_COPY_PACKET:
 684                 if (inst->copy_range == 0
 685                     || inst->copy_range > skb->len)
 686                         data_len = skb->len;
 687                 else
 688                         data_len = inst->copy_range;
 689
 690                 size += nla_total_size(data_len);
 691                 break;
 692
 693         case NFULNL_COPY_DISABLED:
 694         default:
 695                 goto unlock_and_release;
 696         }
 697
 698         if (inst->skb &&
 699             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
 700                 /* either the queue len is too high or we don't have
 701                  * enough room in the skb left. flush to userspace. */
 702                 __nfulnl_flush(inst);
 703         }
 704
 705         if (!inst->skb) {
 706                 inst->skb = nfulnl_alloc_skb(net, inst->peer_portid,
 707                                              inst->nlbufsiz, size);
 708                 if (!inst->skb)
 709                         goto alloc_failure;
 710         }
 711
 712         inst->qlen++;
 713
 714         __build_packet_message(log, inst, skb, data_len, pf,
 715                                 hooknum, in, out, prefix, plen);
 716
 717         if (inst->qlen >= qthreshold)
 718                 __nfulnl_flush(inst);
 719         /* timer_pending always called within inst->lock, so there
 720          * is no chance of a race here */
 721         else if (!timer_pending(&inst->timer)) {
 722                 instance_get(inst);
 723                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 724                 add_timer(&inst->timer);
 725         }
 726
 727 unlock_and_release:
 728         spin_unlock_bh(&inst->lock);
 729         instance_put(inst);
 730         return;
 731
 732 alloc_failure:
 733         /* FIXME: statistics */
 734         goto unlock_and_release;
 735 }
 736 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
 737
 738 static int
 739 nfulnl_rcv_nl_event(struct notifier_block *this,
 740                    unsigned long event, void *ptr)
 741 {
 742         struct netlink_notify *n = ptr;
 743         struct nfnl_log_net *log = nfnl_log_pernet(n->net);
 744
 745         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 746                 int i;
 747
 748                 /* destroy all instances for this portid */
 749                 spin_lock_bh(&log->instances_lock);
 750                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 751                         struct hlist_node *t2;
 752                         struct nfulnl_instance *inst;
 753                         struct hlist_head *head = &log->instance_table[i];
 754
 755                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 756                                 if (n->portid == inst->peer_portid)
 757                                         __instance_destroy(inst);
 758                         }
 759                 }
 760                 spin_unlock_bh(&log->instances_lock);
 761         }
 762         return NOTIFY_DONE;
 763 }
 764
 765 static struct notifier_block nfulnl_rtnl_notifier = {
 766         .notifier_call  = nfulnl_rcv_nl_event,
 767 };
 768
 769 static int
 770 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 771                    const struct nlmsghdr *nlh,
 772                    const struct nlattr * const nfqa[])
 773 {
 774         return -ENOTSUPP;
 775 }
 776
 777 static struct nf_logger nfulnl_logger __read_mostly = {
 778         .name   = "nfnetlink_log",
 779         .logfn  = &nfulnl_log_packet,
 780         .me     = THIS_MODULE,
 781 };
 782
 783 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 784         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 785         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 786         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 787         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 788         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 789         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 790 };
 791
 792 static int
 793 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 794                    const struct nlmsghdr *nlh,
 795                    const struct nlattr * const nfula[])
 796 {
 797         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 798         u_int16_t group_num = ntohs(nfmsg->res_id);
 799         struct nfulnl_instance *inst;
 800         struct nfulnl_msg_config_cmd *cmd = NULL;
 801         struct net *net = sock_net(ctnl);
 802         struct nfnl_log_net *log = nfnl_log_pernet(net);
 803         int ret = 0;
 804
 805         if (nfula[NFULA_CFG_CMD]) {
 806                 u_int8_t pf = nfmsg->nfgen_family;
 807                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 808
 809                 /* Commands without queue context */
 810                 switch (cmd->command) {
 811                 case NFULNL_CFG_CMD_PF_BIND:
 812                         return nf_log_bind_pf(net, pf, &nfulnl_logger);
 813                 case NFULNL_CFG_CMD_PF_UNBIND:
 814                         nf_log_unbind_pf(net, pf);
 815                         return 0;
 816                 }
 817         }
 818
 819         inst = instance_lookup_get(log, group_num);
 820         if (inst && inst->peer_portid != NETLINK_CB(skb).portid) {
 821                 ret = -EPERM;
 822                 goto out_put;
 823         }
 824
 825         if (cmd != NULL) {
 826                 switch (cmd->command) {
 827                 case NFULNL_CFG_CMD_BIND:
 828                         if (inst) {
 829                                 ret = -EBUSY;
 830                                 goto out_put;
 831                         }
 832
 833                         inst = instance_create(net, group_num,
 834                                                NETLINK_CB(skb).portid,
 835                                                sk_user_ns(NETLINK_CB(skb).sk));
 836                         if (IS_ERR(inst)) {
 837                                 ret = PTR_ERR(inst);
 838                                 goto out;
 839                         }
 840                         break;
 841                 case NFULNL_CFG_CMD_UNBIND:
 842                         if (!inst) {
 843                                 ret = -ENODEV;
 844                                 goto out;
 845                         }
 846
 847                         instance_destroy(log, inst);
 848                         goto out_put;
 849                 default:
 850                         ret = -ENOTSUPP;
 851                         break;
 852                 }
 853         }
 854
 855         if (nfula[NFULA_CFG_MODE]) {
 856                 struct nfulnl_msg_config_mode *params;
 857                 params = nla_data(nfula[NFULA_CFG_MODE]);
 858
 859                 if (!inst) {
 860                         ret = -ENODEV;
 861                         goto out;
 862                 }
 863                 nfulnl_set_mode(inst, params->copy_mode,
 864                                 ntohl(params->copy_range));
 865         }
 866
 867         if (nfula[NFULA_CFG_TIMEOUT]) {
 868                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 869
 870                 if (!inst) {
 871                         ret = -ENODEV;
 872                         goto out;
 873                 }
 874                 nfulnl_set_timeout(inst, ntohl(timeout));
 875         }
 876
 877         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 878                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 879
 880                 if (!inst) {
 881                         ret = -ENODEV;
 882                         goto out;
 883                 }
 884                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 885         }
 886
 887         if (nfula[NFULA_CFG_QTHRESH]) {
 888                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 889
 890                 if (!inst) {
 891                         ret = -ENODEV;
 892                         goto out;
 893                 }
 894                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 895         }
 896
 897         if (nfula[NFULA_CFG_FLAGS]) {
 898                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
 899
 900                 if (!inst) {
 901                         ret = -ENODEV;
 902                         goto out;
 903                 }
 904                 nfulnl_set_flags(inst, ntohs(flags));
 905         }
 906
 907 out_put:
 908         instance_put(inst);
 909 out:
 910         return ret;
 911 }
 912
 913 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 914         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 915                                     .attr_count = NFULA_MAX, },
 916         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 917                                     .attr_count = NFULA_CFG_MAX,
 918                                     .policy = nfula_cfg_policy },
 919 };
 920
 921 static const struct nfnetlink_subsystem nfulnl_subsys = {
 922         .name           = "log",
 923         .subsys_id      = NFNL_SUBSYS_ULOG,
 924         .cb_count       = NFULNL_MSG_MAX,
 925         .cb             = nfulnl_cb,
 926 };
 927
 928 #ifdef CONFIG_PROC_FS
 929 struct iter_state {
 930         struct seq_net_private p;
 931         unsigned int bucket;
 932 };
 933
 934 static struct hlist_node *get_first(struct net *net, struct iter_state *st)
 935 {
 936         struct nfnl_log_net *log;
 937         if (!st)
 938                 return NULL;
 939
 940         log = nfnl_log_pernet(net);
 941
 942         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 943                 struct hlist_head *head = &log->instance_table[st->bucket];
 944
 945                 if (!hlist_empty(head))
 946                         return rcu_dereference_bh(hlist_first_rcu(head));
 947         }
 948         return NULL;
 949 }
 950
 951 static struct hlist_node *get_next(struct net *net, struct iter_state *st,
 952                                    struct hlist_node *h)
 953 {
 954         h = rcu_dereference_bh(hlist_next_rcu(h));
 955         while (!h) {
 956                 struct nfnl_log_net *log;
 957                 struct hlist_head *head;
 958
 959                 if (++st->bucket >= INSTANCE_BUCKETS)
 960                         return NULL;
 961
 962                 log = nfnl_log_pernet(net);
 963                 head = &log->instance_table[st->bucket];
 964                 h = rcu_dereference_bh(hlist_first_rcu(head));
 965         }
 966         return h;
 967 }
 968
 969 static struct hlist_node *get_idx(struct net *net, struct iter_state *st,
 970                                   loff_t pos)
 971 {
 972         struct hlist_node *head;
 973         head = get_first(net, st);
 974
 975         if (head)
 976                 while (pos && (head = get_next(net, st, head)))
 977                         pos--;
 978         return pos ? NULL : head;
 979 }
 980
 981 static void *seq_start(struct seq_file *s, loff_t *pos)
 982         __acquires(rcu_bh)
 983 {
 984         rcu_read_lock_bh();
 985         return get_idx(seq_file_net(s), s->private, *pos);
 986 }
 987
 988 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 989 {
 990         (*pos)++;
 991         return get_next(seq_file_net(s), s->private, v);
 992 }
 993
 994 static void seq_stop(struct seq_file *s, void *v)
 995         __releases(rcu_bh)
 996 {
 997         rcu_read_unlock_bh();
 998 }
 999
1000 static int seq_show(struct seq_file *s, void *v)
1001 {
1002         const struct nfulnl_instance *inst = v;
1003
1004         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
1005                           inst->group_num,
1006                           inst->peer_portid, inst->qlen,
1007                           inst->copy_mode, inst->copy_range,
1008                           inst->flushtimeout, atomic_read(&inst->use));
1009 }
1010
1011 static const struct seq_operations nful_seq_ops = {
1012         .start  = seq_start,
1013         .next   = seq_next,
1014         .stop   = seq_stop,
1015         .show   = seq_show,
1016 };
1017
1018 static int nful_open(struct inode *inode, struct file *file)
1019 {
1020         return seq_open_net(inode, file, &nful_seq_ops,
1021                             sizeof(struct iter_state));
1022 }
1023
1024 static const struct file_operations nful_file_ops = {
1025         .owner   = THIS_MODULE,
1026         .open    = nful_open,
1027         .read    = seq_read,
1028         .llseek  = seq_lseek,
1029         .release = seq_release_net,
1030 };
1031
1032 #endif /* PROC_FS */
1033
1034 static int __net_init nfnl_log_net_init(struct net *net)
1035 {
1036         unsigned int i;
1037         struct nfnl_log_net *log = nfnl_log_pernet(net);
1038
1039         for (i = 0; i < INSTANCE_BUCKETS; i++)
1040                 INIT_HLIST_HEAD(&log->instance_table[i]);
1041         spin_lock_init(&log->instances_lock);
1042
1043 #ifdef CONFIG_PROC_FS
1044         if (!proc_create("nfnetlink_log", 0440,
1045                          net->nf.proc_netfilter, &nful_file_ops))
1046                 return -ENOMEM;
1047 #endif
1048         return 0;
1049 }
1050
1051 static void __net_exit nfnl_log_net_exit(struct net *net)
1052 {
1053 #ifdef CONFIG_PROC_FS
1054         remove_proc_entry("nfnetlink_log", net->nf.proc_netfilter);
1055 #endif
1056         nf_log_unset(net, &nfulnl_logger);
1057 }
1058
1059 static struct pernet_operations nfnl_log_net_ops = {
1060         .init   = nfnl_log_net_init,
1061         .exit   = nfnl_log_net_exit,
1062         .id     = &nfnl_log_net_id,
1063         .size   = sizeof(struct nfnl_log_net),
1064 };
1065
1066 static int __init nfnetlink_log_init(void)
1067 {
1068         int status = -ENOMEM;
1069
1070         /* it's not really all that important to have a random value, so
1071          * we can do this from the init function, even if there hasn't
1072          * been that much entropy yet */
1073         get_random_bytes(&hash_init, sizeof(hash_init));
1074
1075         netlink_register_notifier(&nfulnl_rtnl_notifier);
1076         status = nfnetlink_subsys_register(&nfulnl_subsys);
1077         if (status < 0) {
1078                 pr_err("log: failed to create netlink socket\n");
1079                 goto cleanup_netlink_notifier;
1080         }
1081
1082         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
1083         if (status < 0) {
1084                 pr_err("log: failed to register logger\n");
1085                 goto cleanup_subsys;
1086         }
1087
1088         status = register_pernet_subsys(&nfnl_log_net_ops);
1089         if (status < 0) {
1090                 pr_err("log: failed to register pernet ops\n");
1091                 goto cleanup_logger;
1092         }
1093         return status;
1094
1095 cleanup_logger:
1096         nf_log_unregister(&nfulnl_logger);
1097 cleanup_subsys:
1098         nfnetlink_subsys_unregister(&nfulnl_subsys);
1099 cleanup_netlink_notifier:
1100         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1101         return status;
1102 }
1103
1104 static void __exit nfnetlink_log_fini(void)
1105 {
1106         unregister_pernet_subsys(&nfnl_log_net_ops);
1107         nf_log_unregister(&nfulnl_logger);
1108         nfnetlink_subsys_unregister(&nfulnl_subsys);
1109         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1110 }
1111
1112 MODULE_DESCRIPTION("netfilter userspace logging");
1113 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1114 MODULE_LICENSE("GPL");
1115 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1116
1117 module_init(nfnetlink_log_init);
1118 module_exit(nfnetlink_log_fini);