net/netfilter/nfnetlink_log.c

   1 /*
   2  * This is a module which is used for logging packets to userspace via
   3  * nfetlink.
   4  *
   5  * (C) 2005 by Harald Welte <laforge@netfilter.org>
   6  *
   7  * Based on the old ipv4-only ipt_ULOG.c:
   8  * (C) 2000-2004 by Harald Welte <laforge@netfilter.org>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License version 2 as
  12  * published by the Free Software Foundation.
  13  */
  14 #include <linux/module.h>
  15 #include <linux/skbuff.h>
  16 #include <linux/if_arp.h>
  17 #include <linux/init.h>
  18 #include <linux/ip.h>
  19 #include <linux/ipv6.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/netfilter.h>
  22 #include <linux/netlink.h>
  23 #include <linux/netfilter/nfnetlink.h>
  24 #include <linux/netfilter/nfnetlink_log.h>
  25 #include <linux/spinlock.h>
  26 #include <linux/sysctl.h>
  27 #include <linux/proc_fs.h>
  28 #include <linux/security.h>
  29 #include <linux/list.h>
  30 #include <linux/jhash.h>
  31 #include <linux/random.h>
  32 #include <linux/slab.h>
  33 #include <net/sock.h>
  34 #include <net/netfilter/nf_log.h>
  35 #include <net/netfilter/nfnetlink_log.h>
  36
  37 #include <linux/atomic.h>
  38
  39 #ifdef CONFIG_BRIDGE_NETFILTER
  40 #include "../bridge/br_private.h"
  41 #endif
  42
  43 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
  44 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
  45 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  46 #define NFULNL_COPY_RANGE_MAX   0xFFFF  /* max packet size is limited by 16-bit struct nfattr nfa_len field */
  47
  48 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  49                                      printk(x, ## args); } while (0);
  50
  51 struct nfulnl_instance {
  52         struct hlist_node hlist;        /* global list of instances */
  53         spinlock_t lock;
  54         atomic_t use;                   /* use count */
  55
  56         unsigned int qlen;              /* number of nlmsgs in skb */
  57         struct sk_buff *skb;            /* pre-allocatd skb */
  58         struct timer_list timer;
  59         struct user_namespace *peer_user_ns;    /* User namespace of the peer process */
  60         int peer_portid;                        /* PORTID of the peer process */
  61
  62         /* configurable parameters */
  63         unsigned int flushtimeout;      /* timeout until queue flush */
  64         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  65         unsigned int qthreshold;        /* threshold of the queue */
  66         u_int32_t copy_range;
  67         u_int32_t seq;                  /* instance-local sequential counter */
  68         u_int16_t group_num;            /* number of this queue */
  69         u_int16_t flags;
  70         u_int8_t copy_mode;
  71         struct rcu_head rcu;
  72 };
  73
  74 static DEFINE_SPINLOCK(instances_lock);
  75 static atomic_t global_seq;
  76
  77 #define INSTANCE_BUCKETS        16
  78 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  79 static unsigned int hash_init;
  80
  81 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  82 {
  83         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  84 }
  85
  86 static struct nfulnl_instance *
  87 __instance_lookup(u_int16_t group_num)
  88 {
  89         struct hlist_head *head;
  90         struct nfulnl_instance *inst;
  91
  92         head = &instance_table[instance_hashfn(group_num)];
  93         hlist_for_each_entry_rcu(inst, head, hlist) {
  94                 if (inst->group_num == group_num)
  95                         return inst;
  96         }
  97         return NULL;
  98 }
  99
 100 static inline void
 101 instance_get(struct nfulnl_instance *inst)
 102 {
 103         atomic_inc(&inst->use);
 104 }
 105
 106 static struct nfulnl_instance *
 107 instance_lookup_get(u_int16_t group_num)
 108 {
 109         struct nfulnl_instance *inst;
 110
 111         rcu_read_lock_bh();
 112         inst = __instance_lookup(group_num);
 113         if (inst && !atomic_inc_not_zero(&inst->use))
 114                 inst = NULL;
 115         rcu_read_unlock_bh();
 116
 117         return inst;
 118 }
 119
 120 static void nfulnl_instance_free_rcu(struct rcu_head *head)
 121 {
 122         kfree(container_of(head, struct nfulnl_instance, rcu));
 123         module_put(THIS_MODULE);
 124 }
 125
 126 static void
 127 instance_put(struct nfulnl_instance *inst)
 128 {
 129         if (inst && atomic_dec_and_test(&inst->use))
 130                 call_rcu_bh(&inst->rcu, nfulnl_instance_free_rcu);
 131 }
 132
 133 static void nfulnl_timer(unsigned long data);
 134
 135 static struct nfulnl_instance *
 136 instance_create(u_int16_t group_num, int portid, struct user_namespace *user_ns)
 137 {
 138         struct nfulnl_instance *inst;
 139         int err;
 140
 141         spin_lock_bh(&instances_lock);
 142         if (__instance_lookup(group_num)) {
 143                 err = -EEXIST;
 144                 goto out_unlock;
 145         }
 146
 147         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 148         if (!inst) {
 149                 err = -ENOMEM;
 150                 goto out_unlock;
 151         }
 152
 153         if (!try_module_get(THIS_MODULE)) {
 154                 kfree(inst);
 155                 err = -EAGAIN;
 156                 goto out_unlock;
 157         }
 158
 159         INIT_HLIST_NODE(&inst->hlist);
 160         spin_lock_init(&inst->lock);
 161         /* needs to be two, since we _put() after creation */
 162         atomic_set(&inst->use, 2);
 163
 164         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
 165
 166         inst->peer_user_ns = user_ns;
 167         inst->peer_portid = portid;
 168         inst->group_num = group_num;
 169
 170         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 171         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 172         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 173         inst->copy_mode         = NFULNL_COPY_PACKET;
 174         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
 175
 176         hlist_add_head_rcu(&inst->hlist,
 177                        &instance_table[instance_hashfn(group_num)]);
 178
 179         spin_unlock_bh(&instances_lock);
 180
 181         return inst;
 182
 183 out_unlock:
 184         spin_unlock_bh(&instances_lock);
 185         return ERR_PTR(err);
 186 }
 187
 188 static void __nfulnl_flush(struct nfulnl_instance *inst);
 189
 190 /* called with BH disabled */
 191 static void
 192 __instance_destroy(struct nfulnl_instance *inst)
 193 {
 194         /* first pull it out of the global list */
 195         hlist_del_rcu(&inst->hlist);
 196
 197         /* then flush all pending packets from skb */
 198
 199         spin_lock(&inst->lock);
 200
 201         /* lockless readers wont be able to use us */
 202         inst->copy_mode = NFULNL_COPY_DISABLED;
 203
 204         if (inst->skb)
 205                 __nfulnl_flush(inst);
 206         spin_unlock(&inst->lock);
 207
 208         /* and finally put the refcount */
 209         instance_put(inst);
 210 }
 211
 212 static inline void
 213 instance_destroy(struct nfulnl_instance *inst)
 214 {
 215         spin_lock_bh(&instances_lock);
 216         __instance_destroy(inst);
 217         spin_unlock_bh(&instances_lock);
 218 }
 219
 220 static int
 221 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 222                   unsigned int range)
 223 {
 224         int status = 0;
 225
 226         spin_lock_bh(&inst->lock);
 227
 228         switch (mode) {
 229         case NFULNL_COPY_NONE:
 230         case NFULNL_COPY_META:
 231                 inst->copy_mode = mode;
 232                 inst->copy_range = 0;
 233                 break;
 234
 235         case NFULNL_COPY_PACKET:
 236                 inst->copy_mode = mode;
 237                 inst->copy_range = min_t(unsigned int,
 238                                          range, NFULNL_COPY_RANGE_MAX);
 239                 break;
 240
 241         default:
 242                 status = -EINVAL;
 243                 break;
 244         }
 245
 246         spin_unlock_bh(&inst->lock);
 247
 248         return status;
 249 }
 250
 251 static int
 252 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 253 {
 254         int status;
 255
 256         spin_lock_bh(&inst->lock);
 257         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 258                 status = -ERANGE;
 259         else if (nlbufsiz > 131072)
 260                 status = -ERANGE;
 261         else {
 262                 inst->nlbufsiz = nlbufsiz;
 263                 status = 0;
 264         }
 265         spin_unlock_bh(&inst->lock);
 266
 267         return status;
 268 }
 269
 270 static int
 271 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 272 {
 273         spin_lock_bh(&inst->lock);
 274         inst->flushtimeout = timeout;
 275         spin_unlock_bh(&inst->lock);
 276
 277         return 0;
 278 }
 279
 280 static int
 281 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 282 {
 283         spin_lock_bh(&inst->lock);
 284         inst->qthreshold = qthresh;
 285         spin_unlock_bh(&inst->lock);
 286
 287         return 0;
 288 }
 289
 290 static int
 291 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 292 {
 293         spin_lock_bh(&inst->lock);
 294         inst->flags = flags;
 295         spin_unlock_bh(&inst->lock);
 296
 297         return 0;
 298 }
 299
 300 static struct sk_buff *
 301 nfulnl_alloc_skb(unsigned int inst_size, unsigned int pkt_size)
 302 {
 303         struct sk_buff *skb;
 304         unsigned int n;
 305
 306         /* alloc skb which should be big enough for a whole multipart
 307          * message.  WARNING: has to be <= 128k due to slab restrictions */
 308
 309         n = max(inst_size, pkt_size);
 310         skb = alloc_skb(n, GFP_ATOMIC);
 311         if (!skb) {
 312                 if (n > pkt_size) {
 313                         /* try to allocate only as much as we need for current
 314                          * packet */
 315
 316                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 317                         if (!skb)
 318                                 pr_err("nfnetlink_log: can't even alloc %u bytes\n",
 319                                        pkt_size);
 320                 }
 321         }
 322
 323         return skb;
 324 }
 325
 326 static int
 327 __nfulnl_send(struct nfulnl_instance *inst)
 328 {
 329         int status = -1;
 330
 331         if (inst->qlen > 1) {
 332                 struct nlmsghdr *nlh = nlmsg_put(inst->skb, 0, 0,
 333                                                  NLMSG_DONE,
 334                                                  sizeof(struct nfgenmsg),
 335                                                  0);
 336                 if (!nlh)
 337                         goto out;
 338         }
 339         status = nfnetlink_unicast(inst->skb, &init_net, inst->peer_portid,
 340                                    MSG_DONTWAIT);
 341
 342         inst->qlen = 0;
 343         inst->skb = NULL;
 344 out:
 345         return status;
 346 }
 347
 348 static void
 349 __nfulnl_flush(struct nfulnl_instance *inst)
 350 {
 351         /* timer holds a reference */
 352         if (del_timer(&inst->timer))
 353                 instance_put(inst);
 354         if (inst->skb)
 355                 __nfulnl_send(inst);
 356 }
 357
 358 static void
 359 nfulnl_timer(unsigned long data)
 360 {
 361         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 362
 363         spin_lock_bh(&inst->lock);
 364         if (inst->skb)
 365                 __nfulnl_send(inst);
 366         spin_unlock_bh(&inst->lock);
 367         instance_put(inst);
 368 }
 369
 370 /* This is an inline function, we don't really care about a long
 371  * list of arguments */
 372 static inline int
 373 __build_packet_message(struct nfulnl_instance *inst,
 374                         const struct sk_buff *skb,
 375                         unsigned int data_len,
 376                         u_int8_t pf,
 377                         unsigned int hooknum,
 378                         const struct net_device *indev,
 379                         const struct net_device *outdev,
 380                         const char *prefix, unsigned int plen)
 381 {
 382         struct nfulnl_msg_packet_hdr pmsg;
 383         struct nlmsghdr *nlh;
 384         struct nfgenmsg *nfmsg;
 385         sk_buff_data_t old_tail = inst->skb->tail;
 386         struct sock *sk;
 387         const unsigned char *hwhdrp;
 388
 389         nlh = nlmsg_put(inst->skb, 0, 0,
 390                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 391                         sizeof(struct nfgenmsg), 0);
 392         if (!nlh)
 393                 return -1;
 394         nfmsg = nlmsg_data(nlh);
 395         nfmsg->nfgen_family = pf;
 396         nfmsg->version = NFNETLINK_V0;
 397         nfmsg->res_id = htons(inst->group_num);
 398
 399         pmsg.hw_protocol        = skb->protocol;
 400         pmsg.hook               = hooknum;
 401
 402         if (nla_put(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg))
 403                 goto nla_put_failure;
 404
 405         if (prefix &&
 406             nla_put(inst->skb, NFULA_PREFIX, plen, prefix))
 407                 goto nla_put_failure;
 408
 409         if (indev) {
 410 #ifndef CONFIG_BRIDGE_NETFILTER
 411                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 412                                  htonl(indev->ifindex)))
 413                         goto nla_put_failure;
 414 #else
 415                 if (pf == PF_BRIDGE) {
 416                         /* Case 1: outdev is physical input device, we need to
 417                          * look for bridge group (when called from
 418                          * netfilter_bridge) */
 419                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 420                                          htonl(indev->ifindex)) ||
 421                         /* this is the bridge group "brX" */
 422                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 423                             nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 424                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 425                                 goto nla_put_failure;
 426                 } else {
 427                         /* Case 2: indev is bridge group, we need to look for
 428                          * physical device (when called from ipv4) */
 429                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 430                                          htonl(indev->ifindex)))
 431                                 goto nla_put_failure;
 432                         if (skb->nf_bridge && skb->nf_bridge->physindev &&
 433                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 434                                          htonl(skb->nf_bridge->physindev->ifindex)))
 435                                 goto nla_put_failure;
 436                 }
 437 #endif
 438         }
 439
 440         if (outdev) {
 441 #ifndef CONFIG_BRIDGE_NETFILTER
 442                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 443                                  htonl(outdev->ifindex)))
 444                         goto nla_put_failure;
 445 #else
 446                 if (pf == PF_BRIDGE) {
 447                         /* Case 1: outdev is physical output device, we need to
 448                          * look for bridge group (when called from
 449                          * netfilter_bridge) */
 450                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 451                                          htonl(outdev->ifindex)) ||
 452                         /* this is the bridge group "brX" */
 453                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 454                             nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 455                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 456                                 goto nla_put_failure;
 457                 } else {
 458                         /* Case 2: indev is a bridge group, we need to look
 459                          * for physical device (when called from ipv4) */
 460                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 461                                          htonl(outdev->ifindex)))
 462                                 goto nla_put_failure;
 463                         if (skb->nf_bridge && skb->nf_bridge->physoutdev &&
 464                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 465                                          htonl(skb->nf_bridge->physoutdev->ifindex)))
 466                                 goto nla_put_failure;
 467                 }
 468 #endif
 469         }
 470
 471         if (skb->mark &&
 472             nla_put_be32(inst->skb, NFULA_MARK, htonl(skb->mark)))
 473                 goto nla_put_failure;
 474
 475         if (indev && skb->dev &&
 476             skb->mac_header != skb->network_header) {
 477                 struct nfulnl_msg_packet_hw phw;
 478                 int len = dev_parse_header(skb, phw.hw_addr);
 479                 if (len > 0) {
 480                         phw.hw_addrlen = htons(len);
 481                         if (nla_put(inst->skb, NFULA_HWADDR, sizeof(phw), &phw))
 482                                 goto nla_put_failure;
 483                 }
 484         }
 485
 486         if (indev && skb_mac_header_was_set(skb)) {
 487                 if (nla_put_be16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type)) ||
 488                     nla_put_be16(inst->skb, NFULA_HWLEN,
 489                                  htons(skb->dev->hard_header_len)))
 490                         goto nla_put_failure;
 491
 492                 hwhdrp = skb_mac_header(skb);
 493
 494                 if (skb->dev->type == ARPHRD_SIT)
 495                         hwhdrp -= ETH_HLEN;
 496
 497                 if (hwhdrp >= skb->head &&
 498                     nla_put(inst->skb, NFULA_HWHEADER,
 499                             skb->dev->hard_header_len, hwhdrp))
 500                         goto nla_put_failure;
 501         }
 502
 503         if (skb->tstamp.tv64) {
 504                 struct nfulnl_msg_packet_timestamp ts;
 505                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 506                 ts.sec = cpu_to_be64(tv.tv_sec);
 507                 ts.usec = cpu_to_be64(tv.tv_usec);
 508
 509                 if (nla_put(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts))
 510                         goto nla_put_failure;
 511         }
 512
 513         /* UID */
 514         sk = skb->sk;
 515         if (sk && sk->sk_state != TCP_TIME_WAIT) {
 516                 read_lock_bh(&sk->sk_callback_lock);
 517                 if (sk->sk_socket && sk->sk_socket->file) {
 518                         struct file *file = sk->sk_socket->file;
 519                         const struct cred *cred = file->f_cred;
 520                         struct user_namespace *user_ns = inst->peer_user_ns;
 521                         __be32 uid = htonl(from_kuid_munged(user_ns, cred->fsuid));
 522                         __be32 gid = htonl(from_kgid_munged(user_ns, cred->fsgid));
 523                         read_unlock_bh(&sk->sk_callback_lock);
 524                         if (nla_put_be32(inst->skb, NFULA_UID, uid) ||
 525                             nla_put_be32(inst->skb, NFULA_GID, gid))
 526                                 goto nla_put_failure;
 527                 } else
 528                         read_unlock_bh(&sk->sk_callback_lock);
 529         }
 530
 531         /* local sequence number */
 532         if ((inst->flags & NFULNL_CFG_F_SEQ) &&
 533             nla_put_be32(inst->skb, NFULA_SEQ, htonl(inst->seq++)))
 534                 goto nla_put_failure;
 535
 536         /* global sequence number */
 537         if ((inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) &&
 538             nla_put_be32(inst->skb, NFULA_SEQ_GLOBAL,
 539                          htonl(atomic_inc_return(&global_seq))))
 540                 goto nla_put_failure;
 541
 542         if (data_len) {
 543                 struct nlattr *nla;
 544                 int size = nla_attr_size(data_len);
 545
 546                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
 547                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 548                         return -1;
 549                 }
 550
 551                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 552                 nla->nla_type = NFULA_PAYLOAD;
 553                 nla->nla_len = size;
 554
 555                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 556                         BUG();
 557         }
 558
 559         nlh->nlmsg_len = inst->skb->tail - old_tail;
 560         return 0;
 561
 562 nla_put_failure:
 563         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 564         return -1;
 565 }
 566
 567 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 568
 569 static struct nf_loginfo default_loginfo = {
 570         .type =         NF_LOG_TYPE_ULOG,
 571         .u = {
 572                 .ulog = {
 573                         .copy_len       = 0xffff,
 574                         .group          = 0,
 575                         .qthreshold     = 1,
 576                 },
 577         },
 578 };
 579
 580 /* log handler for internal netfilter logging api */
 581 void
 582 nfulnl_log_packet(u_int8_t pf,
 583                   unsigned int hooknum,
 584                   const struct sk_buff *skb,
 585                   const struct net_device *in,
 586                   const struct net_device *out,
 587                   const struct nf_loginfo *li_user,
 588                   const char *prefix)
 589 {
 590         unsigned int size, data_len;
 591         struct nfulnl_instance *inst;
 592         const struct nf_loginfo *li;
 593         unsigned int qthreshold;
 594         unsigned int plen;
 595
 596         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 597                 li = li_user;
 598         else
 599                 li = &default_loginfo;
 600
 601         inst = instance_lookup_get(li->u.ulog.group);
 602         if (!inst)
 603                 return;
 604
 605         plen = 0;
 606         if (prefix)
 607                 plen = strlen(prefix) + 1;
 608
 609         /* FIXME: do we want to make the size calculation conditional based on
 610          * what is actually present?  way more branches and checks, but more
 611          * memory efficient... */
 612         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 613                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 614                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 615                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 616 #ifdef CONFIG_BRIDGE_NETFILTER
 617                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 618                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 619 #endif
 620                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 621                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 622                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 623                 + nla_total_size(plen)                  /* prefix */
 624                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 625                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
 626
 627         if (in && skb_mac_header_was_set(skb)) {
 628                 size +=   nla_total_size(skb->dev->hard_header_len)
 629                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 630                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 631         }
 632
 633         spin_lock_bh(&inst->lock);
 634
 635         if (inst->flags & NFULNL_CFG_F_SEQ)
 636                 size += nla_total_size(sizeof(u_int32_t));
 637         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 638                 size += nla_total_size(sizeof(u_int32_t));
 639
 640         qthreshold = inst->qthreshold;
 641         /* per-rule qthreshold overrides per-instance */
 642         if (li->u.ulog.qthreshold)
 643                 if (qthreshold > li->u.ulog.qthreshold)
 644                         qthreshold = li->u.ulog.qthreshold;
 645
 646
 647         switch (inst->copy_mode) {
 648         case NFULNL_COPY_META:
 649         case NFULNL_COPY_NONE:
 650                 data_len = 0;
 651                 break;
 652
 653         case NFULNL_COPY_PACKET:
 654                 if (inst->copy_range == 0
 655                     || inst->copy_range > skb->len)
 656                         data_len = skb->len;
 657                 else
 658                         data_len = inst->copy_range;
 659
 660                 size += nla_total_size(data_len);
 661                 break;
 662
 663         case NFULNL_COPY_DISABLED:
 664         default:
 665                 goto unlock_and_release;
 666         }
 667
 668         if (inst->skb &&
 669             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
 670                 /* either the queue len is too high or we don't have
 671                  * enough room in the skb left. flush to userspace. */
 672                 __nfulnl_flush(inst);
 673         }
 674
 675         if (!inst->skb) {
 676                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
 677                 if (!inst->skb)
 678                         goto alloc_failure;
 679         }
 680
 681         inst->qlen++;
 682
 683         __build_packet_message(inst, skb, data_len, pf,
 684                                 hooknum, in, out, prefix, plen);
 685
 686         if (inst->qlen >= qthreshold)
 687                 __nfulnl_flush(inst);
 688         /* timer_pending always called within inst->lock, so there
 689          * is no chance of a race here */
 690         else if (!timer_pending(&inst->timer)) {
 691                 instance_get(inst);
 692                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 693                 add_timer(&inst->timer);
 694         }
 695
 696 unlock_and_release:
 697         spin_unlock_bh(&inst->lock);
 698         instance_put(inst);
 699         return;
 700
 701 alloc_failure:
 702         /* FIXME: statistics */
 703         goto unlock_and_release;
 704 }
 705 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
 706
 707 static int
 708 nfulnl_rcv_nl_event(struct notifier_block *this,
 709                    unsigned long event, void *ptr)
 710 {
 711         struct netlink_notify *n = ptr;
 712
 713         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 714                 int i;
 715
 716                 /* destroy all instances for this portid */
 717                 spin_lock_bh(&instances_lock);
 718                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 719                         struct hlist_node *t2;
 720                         struct nfulnl_instance *inst;
 721                         struct hlist_head *head = &instance_table[i];
 722
 723                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 724                                 if ((net_eq(n->net, &init_net)) &&
 725                                     (n->portid == inst->peer_portid))
 726                                         __instance_destroy(inst);
 727                         }
 728                 }
 729                 spin_unlock_bh(&instances_lock);
 730         }
 731         return NOTIFY_DONE;
 732 }
 733
 734 static struct notifier_block nfulnl_rtnl_notifier = {
 735         .notifier_call  = nfulnl_rcv_nl_event,
 736 };
 737
 738 static int
 739 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 740                    const struct nlmsghdr *nlh,
 741                    const struct nlattr * const nfqa[])
 742 {
 743         return -ENOTSUPP;
 744 }
 745
 746 static struct nf_logger nfulnl_logger __read_mostly = {
 747         .name   = "nfnetlink_log",
 748         .logfn  = &nfulnl_log_packet,
 749         .me     = THIS_MODULE,
 750 };
 751
 752 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 753         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 754         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 755         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 756         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 757         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 758         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 759 };
 760
 761 static int
 762 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 763                    const struct nlmsghdr *nlh,
 764                    const struct nlattr * const nfula[])
 765 {
 766         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 767         u_int16_t group_num = ntohs(nfmsg->res_id);
 768         struct nfulnl_instance *inst;
 769         struct nfulnl_msg_config_cmd *cmd = NULL;
 770         struct net *net = sock_net(ctnl);
 771         int ret = 0;
 772
 773         if (nfula[NFULA_CFG_CMD]) {
 774                 u_int8_t pf = nfmsg->nfgen_family;
 775                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 776
 777                 /* Commands without queue context */
 778                 switch (cmd->command) {
 779                 case NFULNL_CFG_CMD_PF_BIND:
 780                         return nf_log_bind_pf(net, pf, &nfulnl_logger);
 781                 case NFULNL_CFG_CMD_PF_UNBIND:
 782                         nf_log_unbind_pf(net, pf);
 783                         return 0;
 784                 }
 785         }
 786
 787         inst = instance_lookup_get(group_num);
 788         if (inst && inst->peer_portid != NETLINK_CB(skb).portid) {
 789                 ret = -EPERM;
 790                 goto out_put;
 791         }
 792
 793         if (cmd != NULL) {
 794                 switch (cmd->command) {
 795                 case NFULNL_CFG_CMD_BIND:
 796                         if (inst) {
 797                                 ret = -EBUSY;
 798                                 goto out_put;
 799                         }
 800
 801                         inst = instance_create(group_num,
 802                                                NETLINK_CB(skb).portid,
 803                                                sk_user_ns(NETLINK_CB(skb).ssk));
 804                         if (IS_ERR(inst)) {
 805                                 ret = PTR_ERR(inst);
 806                                 goto out;
 807                         }
 808                         break;
 809                 case NFULNL_CFG_CMD_UNBIND:
 810                         if (!inst) {
 811                                 ret = -ENODEV;
 812                                 goto out;
 813                         }
 814
 815                         instance_destroy(inst);
 816                         goto out_put;
 817                 default:
 818                         ret = -ENOTSUPP;
 819                         break;
 820                 }
 821         }
 822
 823         if (nfula[NFULA_CFG_MODE]) {
 824                 struct nfulnl_msg_config_mode *params;
 825                 params = nla_data(nfula[NFULA_CFG_MODE]);
 826
 827                 if (!inst) {
 828                         ret = -ENODEV;
 829                         goto out;
 830                 }
 831                 nfulnl_set_mode(inst, params->copy_mode,
 832                                 ntohl(params->copy_range));
 833         }
 834
 835         if (nfula[NFULA_CFG_TIMEOUT]) {
 836                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 837
 838                 if (!inst) {
 839                         ret = -ENODEV;
 840                         goto out;
 841                 }
 842                 nfulnl_set_timeout(inst, ntohl(timeout));
 843         }
 844
 845         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 846                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 847
 848                 if (!inst) {
 849                         ret = -ENODEV;
 850                         goto out;
 851                 }
 852                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 853         }
 854
 855         if (nfula[NFULA_CFG_QTHRESH]) {
 856                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 857
 858                 if (!inst) {
 859                         ret = -ENODEV;
 860                         goto out;
 861                 }
 862                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 863         }
 864
 865         if (nfula[NFULA_CFG_FLAGS]) {
 866                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
 867
 868                 if (!inst) {
 869                         ret = -ENODEV;
 870                         goto out;
 871                 }
 872                 nfulnl_set_flags(inst, ntohs(flags));
 873         }
 874
 875 out_put:
 876         instance_put(inst);
 877 out:
 878         return ret;
 879 }
 880
 881 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 882         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 883                                     .attr_count = NFULA_MAX, },
 884         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 885                                     .attr_count = NFULA_CFG_MAX,
 886                                     .policy = nfula_cfg_policy },
 887 };
 888
 889 static const struct nfnetlink_subsystem nfulnl_subsys = {
 890         .name           = "log",
 891         .subsys_id      = NFNL_SUBSYS_ULOG,
 892         .cb_count       = NFULNL_MSG_MAX,
 893         .cb             = nfulnl_cb,
 894 };
 895
 896 #ifdef CONFIG_PROC_FS
 897 struct iter_state {
 898         unsigned int bucket;
 899 };
 900
 901 static struct hlist_node *get_first(struct iter_state *st)
 902 {
 903         if (!st)
 904                 return NULL;
 905
 906         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 907                 if (!hlist_empty(&instance_table[st->bucket]))
 908                         return rcu_dereference_bh(hlist_first_rcu(&instance_table[st->bucket]));
 909         }
 910         return NULL;
 911 }
 912
 913 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
 914 {
 915         h = rcu_dereference_bh(hlist_next_rcu(h));
 916         while (!h) {
 917                 if (++st->bucket >= INSTANCE_BUCKETS)
 918                         return NULL;
 919
 920                 h = rcu_dereference_bh(hlist_first_rcu(&instance_table[st->bucket]));
 921         }
 922         return h;
 923 }
 924
 925 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
 926 {
 927         struct hlist_node *head;
 928         head = get_first(st);
 929
 930         if (head)
 931                 while (pos && (head = get_next(st, head)))
 932                         pos--;
 933         return pos ? NULL : head;
 934 }
 935
 936 static void *seq_start(struct seq_file *seq, loff_t *pos)
 937         __acquires(rcu_bh)
 938 {
 939         rcu_read_lock_bh();
 940         return get_idx(seq->private, *pos);
 941 }
 942
 943 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 944 {
 945         (*pos)++;
 946         return get_next(s->private, v);
 947 }
 948
 949 static void seq_stop(struct seq_file *s, void *v)
 950         __releases(rcu_bh)
 951 {
 952         rcu_read_unlock_bh();
 953 }
 954
 955 static int seq_show(struct seq_file *s, void *v)
 956 {
 957         const struct nfulnl_instance *inst = v;
 958
 959         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 960                           inst->group_num,
 961                           inst->peer_portid, inst->qlen,
 962                           inst->copy_mode, inst->copy_range,
 963                           inst->flushtimeout, atomic_read(&inst->use));
 964 }
 965
 966 static const struct seq_operations nful_seq_ops = {
 967         .start  = seq_start,
 968         .next   = seq_next,
 969         .stop   = seq_stop,
 970         .show   = seq_show,
 971 };
 972
 973 static int nful_open(struct inode *inode, struct file *file)
 974 {
 975         return seq_open_private(file, &nful_seq_ops,
 976                         sizeof(struct iter_state));
 977 }
 978
 979 static const struct file_operations nful_file_ops = {
 980         .owner   = THIS_MODULE,
 981         .open    = nful_open,
 982         .read    = seq_read,
 983         .llseek  = seq_lseek,
 984         .release = seq_release_private,
 985 };
 986
 987 #endif /* PROC_FS */
 988
 989 static int __init nfnetlink_log_init(void)
 990 {
 991         int i, status = -ENOMEM;
 992
 993         for (i = 0; i < INSTANCE_BUCKETS; i++)
 994                 INIT_HLIST_HEAD(&instance_table[i]);
 995
 996         /* it's not really all that important to have a random value, so
 997          * we can do this from the init function, even if there hasn't
 998          * been that much entropy yet */
 999         get_random_bytes(&hash_init, sizeof(hash_init));
1000
1001         netlink_register_notifier(&nfulnl_rtnl_notifier);
1002         status = nfnetlink_subsys_register(&nfulnl_subsys);
1003         if (status < 0) {
1004                 printk(KERN_ERR "log: failed to create netlink socket\n");
1005                 goto cleanup_netlink_notifier;
1006         }
1007
1008         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
1009         if (status < 0) {
1010                 printk(KERN_ERR "log: failed to register logger\n");
1011                 goto cleanup_subsys;
1012         }
1013
1014 #ifdef CONFIG_PROC_FS
1015         if (!proc_create("nfnetlink_log", 0440,
1016                          proc_net_netfilter, &nful_file_ops)) {
1017                 status = -ENOMEM;
1018                 goto cleanup_logger;
1019         }
1020 #endif
1021         return status;
1022
1023 #ifdef CONFIG_PROC_FS
1024 cleanup_logger:
1025         nf_log_unregister(&nfulnl_logger);
1026 #endif
1027 cleanup_subsys:
1028         nfnetlink_subsys_unregister(&nfulnl_subsys);
1029 cleanup_netlink_notifier:
1030         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1031         return status;
1032 }
1033
1034 static void __exit nfnetlink_log_fini(void)
1035 {
1036         nf_log_unregister(&nfulnl_logger);
1037 #ifdef CONFIG_PROC_FS
1038         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1039 #endif
1040         nfnetlink_subsys_unregister(&nfulnl_subsys);
1041         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1042 }
1043
1044 MODULE_DESCRIPTION("netfilter userspace logging");
1045 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1046 MODULE_LICENSE("GPL");
1047 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1048
1049 module_init(nfnetlink_log_init);
1050 module_exit(nfnetlink_log_fini);