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