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