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/slab.h>
  32 #include <net/sock.h>
  33 #include <net/netfilter/nf_log.h>
  34 #include <net/netns/generic.h>
  35 #include <net/netfilter/nfnetlink_log.h>
  36
  37 #include <linux/atomic.h>
  38
  39 #if IS_ENABLED(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 net *net;
  60         struct user_namespace *peer_user_ns;    /* User namespace of the peer process */
  61         int peer_portid;                        /* PORTID of the peer process */
  62
  63         /* configurable parameters */
  64         unsigned int flushtimeout;      /* timeout until queue flush */
  65         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  66         unsigned int qthreshold;        /* threshold of the queue */
  67         u_int32_t copy_range;
  68         u_int32_t seq;                  /* instance-local sequential counter */
  69         u_int16_t group_num;            /* number of this queue */
  70         u_int16_t flags;
  71         u_int8_t copy_mode;
  72         struct rcu_head rcu;
  73 };
  74
  75 #define INSTANCE_BUCKETS        16
  76
  77 static int nfnl_log_net_id __read_mostly;
  78
  79 struct nfnl_log_net {
  80         spinlock_t instances_lock;
  81         struct hlist_head instance_table[INSTANCE_BUCKETS];
  82         atomic_t global_seq;
  83 };
  84
  85 static struct nfnl_log_net *nfnl_log_pernet(struct net *net)
  86 {
  87         return net_generic(net, nfnl_log_net_id);
  88 }
  89
  90 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  91 {
  92         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  93 }
  94
  95 static struct nfulnl_instance *
  96 __instance_lookup(struct nfnl_log_net *log, u_int16_t group_num)
  97 {
  98         struct hlist_head *head;
  99         struct nfulnl_instance *inst;
 100
 101         head = &log->instance_table[instance_hashfn(group_num)];
 102         hlist_for_each_entry_rcu(inst, head, hlist) {
 103                 if (inst->group_num == group_num)
 104                         return inst;
 105         }
 106         return NULL;
 107 }
 108
 109 static inline void
 110 instance_get(struct nfulnl_instance *inst)
 111 {
 112         atomic_inc(&inst->use);
 113 }
 114
 115 static struct nfulnl_instance *
 116 instance_lookup_get(struct nfnl_log_net *log, u_int16_t group_num)
 117 {
 118         struct nfulnl_instance *inst;
 119
 120         rcu_read_lock_bh();
 121         inst = __instance_lookup(log, group_num);
 122         if (inst && !atomic_inc_not_zero(&inst->use))
 123                 inst = NULL;
 124         rcu_read_unlock_bh();
 125
 126         return inst;
 127 }
 128
 129 static void nfulnl_instance_free_rcu(struct rcu_head *head)
 130 {
 131         struct nfulnl_instance *inst =
 132                 container_of(head, struct nfulnl_instance, rcu);
 133
 134         put_net(inst->net);
 135         kfree(inst);
 136         module_put(THIS_MODULE);
 137 }
 138
 139 static void
 140 instance_put(struct nfulnl_instance *inst)
 141 {
 142         if (inst && atomic_dec_and_test(&inst->use))
 143                 call_rcu_bh(&inst->rcu, nfulnl_instance_free_rcu);
 144 }
 145
 146 static void nfulnl_timer(unsigned long data);
 147
 148 static struct nfulnl_instance *
 149 instance_create(struct net *net, u_int16_t group_num,
 150                 int portid, struct user_namespace *user_ns)
 151 {
 152         struct nfulnl_instance *inst;
 153         struct nfnl_log_net *log = nfnl_log_pernet(net);
 154         int err;
 155
 156         spin_lock_bh(&log->instances_lock);
 157         if (__instance_lookup(log, group_num)) {
 158                 err = -EEXIST;
 159                 goto out_unlock;
 160         }
 161
 162         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 163         if (!inst) {
 164                 err = -ENOMEM;
 165                 goto out_unlock;
 166         }
 167
 168         if (!try_module_get(THIS_MODULE)) {
 169                 kfree(inst);
 170                 err = -EAGAIN;
 171                 goto out_unlock;
 172         }
 173
 174         INIT_HLIST_NODE(&inst->hlist);
 175         spin_lock_init(&inst->lock);
 176         /* needs to be two, since we _put() after creation */
 177         atomic_set(&inst->use, 2);
 178
 179         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
 180
 181         inst->net = get_net(net);
 182         inst->peer_user_ns = user_ns;
 183         inst->peer_portid = portid;
 184         inst->group_num = group_num;
 185
 186         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 187         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 188         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 189         inst->copy_mode         = NFULNL_COPY_PACKET;
 190         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
 191
 192         hlist_add_head_rcu(&inst->hlist,
 193                        &log->instance_table[instance_hashfn(group_num)]);
 194
 195
 196         spin_unlock_bh(&log->instances_lock);
 197
 198         return inst;
 199
 200 out_unlock:
 201         spin_unlock_bh(&log->instances_lock);
 202         return ERR_PTR(err);
 203 }
 204
 205 static void __nfulnl_flush(struct nfulnl_instance *inst);
 206
 207 /* called with BH disabled */
 208 static void
 209 __instance_destroy(struct nfulnl_instance *inst)
 210 {
 211         /* first pull it out of the global list */
 212         hlist_del_rcu(&inst->hlist);
 213
 214         /* then flush all pending packets from skb */
 215
 216         spin_lock(&inst->lock);
 217
 218         /* lockless readers wont be able to use us */
 219         inst->copy_mode = NFULNL_COPY_DISABLED;
 220
 221         if (inst->skb)
 222                 __nfulnl_flush(inst);
 223         spin_unlock(&inst->lock);
 224
 225         /* and finally put the refcount */
 226         instance_put(inst);
 227 }
 228
 229 static inline void
 230 instance_destroy(struct nfnl_log_net *log,
 231                  struct nfulnl_instance *inst)
 232 {
 233         spin_lock_bh(&log->instances_lock);
 234         __instance_destroy(inst);
 235         spin_unlock_bh(&log->instances_lock);
 236 }
 237
 238 static int
 239 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 240                   unsigned int range)
 241 {
 242         int status = 0;
 243
 244         spin_lock_bh(&inst->lock);
 245
 246         switch (mode) {
 247         case NFULNL_COPY_NONE:
 248         case NFULNL_COPY_META:
 249                 inst->copy_mode = mode;
 250                 inst->copy_range = 0;
 251                 break;
 252
 253         case NFULNL_COPY_PACKET:
 254                 inst->copy_mode = mode;
 255                 inst->copy_range = min_t(unsigned int,
 256                                          range, NFULNL_COPY_RANGE_MAX);
 257                 break;
 258
 259         default:
 260                 status = -EINVAL;
 261                 break;
 262         }
 263
 264         spin_unlock_bh(&inst->lock);
 265
 266         return status;
 267 }
 268
 269 static int
 270 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 271 {
 272         int status;
 273
 274         spin_lock_bh(&inst->lock);
 275         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 276                 status = -ERANGE;
 277         else if (nlbufsiz > 131072)
 278                 status = -ERANGE;
 279         else {
 280                 inst->nlbufsiz = nlbufsiz;
 281                 status = 0;
 282         }
 283         spin_unlock_bh(&inst->lock);
 284
 285         return status;
 286 }
 287
 288 static int
 289 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 290 {
 291         spin_lock_bh(&inst->lock);
 292         inst->flushtimeout = timeout;
 293         spin_unlock_bh(&inst->lock);
 294
 295         return 0;
 296 }
 297
 298 static int
 299 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 300 {
 301         spin_lock_bh(&inst->lock);
 302         inst->qthreshold = qthresh;
 303         spin_unlock_bh(&inst->lock);
 304
 305         return 0;
 306 }
 307
 308 static int
 309 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 310 {
 311         spin_lock_bh(&inst->lock);
 312         inst->flags = flags;
 313         spin_unlock_bh(&inst->lock);
 314
 315         return 0;
 316 }
 317
 318 static struct sk_buff *
 319 nfulnl_alloc_skb(struct net *net, u32 peer_portid, unsigned int inst_size,
 320                  unsigned int pkt_size)
 321 {
 322         struct sk_buff *skb;
 323         unsigned int n;
 324
 325         /* alloc skb which should be big enough for a whole multipart
 326          * message.  WARNING: has to be <= 128k due to slab restrictions */
 327
 328         n = max(inst_size, pkt_size);
 329         skb = nfnetlink_alloc_skb(net, n, peer_portid, GFP_ATOMIC);
 330         if (!skb) {
 331                 if (n > pkt_size) {
 332                         /* try to allocate only as much as we need for current
 333                          * packet */
 334
 335                         skb = nfnetlink_alloc_skb(net, pkt_size,
 336                                                   peer_portid, GFP_ATOMIC);
 337                         if (!skb)
 338                                 pr_err("nfnetlink_log: can't even alloc %u bytes\n",
 339                                        pkt_size);
 340                 }
 341         }
 342
 343         return skb;
 344 }
 345
 346 static int
 347 __nfulnl_send(struct nfulnl_instance *inst)
 348 {
 349         int status = -1;
 350
 351         if (inst->qlen > 1) {
 352                 struct nlmsghdr *nlh = nlmsg_put(inst->skb, 0, 0,
 353                                                  NLMSG_DONE,
 354                                                  sizeof(struct nfgenmsg),
 355                                                  0);
 356                 if (!nlh)
 357                         goto out;
 358         }
 359         status = nfnetlink_unicast(inst->skb, inst->net, inst->peer_portid,
 360                                    MSG_DONTWAIT);
 361
 362         inst->qlen = 0;
 363         inst->skb = NULL;
 364 out:
 365         return status;
 366 }
 367
 368 static void
 369 __nfulnl_flush(struct nfulnl_instance *inst)
 370 {
 371         /* timer holds a reference */
 372         if (del_timer(&inst->timer))
 373                 instance_put(inst);
 374         if (inst->skb)
 375                 __nfulnl_send(inst);
 376 }
 377
 378 static void
 379 nfulnl_timer(unsigned long data)
 380 {
 381         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 382
 383         spin_lock_bh(&inst->lock);
 384         if (inst->skb)
 385                 __nfulnl_send(inst);
 386         spin_unlock_bh(&inst->lock);
 387         instance_put(inst);
 388 }
 389
 390 /* This is an inline function, we don't really care about a long
 391  * list of arguments */
 392 static inline int
 393 __build_packet_message(struct nfnl_log_net *log,
 394                         struct nfulnl_instance *inst,
 395                         const struct sk_buff *skb,
 396                         unsigned int data_len,
 397                         u_int8_t pf,
 398                         unsigned int hooknum,
 399                         const struct net_device *indev,
 400                         const struct net_device *outdev,
 401                         const char *prefix, unsigned int plen)
 402 {
 403         struct nfulnl_msg_packet_hdr pmsg;
 404         struct nlmsghdr *nlh;
 405         struct nfgenmsg *nfmsg;
 406         sk_buff_data_t old_tail = inst->skb->tail;
 407         struct sock *sk;
 408         const unsigned char *hwhdrp;
 409
 410         nlh = nlmsg_put(inst->skb, 0, 0,
 411                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 412                         sizeof(struct nfgenmsg), 0);
 413         if (!nlh)
 414                 return -1;
 415         nfmsg = nlmsg_data(nlh);
 416         nfmsg->nfgen_family = pf;
 417         nfmsg->version = NFNETLINK_V0;
 418         nfmsg->res_id = htons(inst->group_num);
 419
 420         memset(&pmsg, 0, sizeof(pmsg));
 421         pmsg.hw_protocol        = skb->protocol;
 422         pmsg.hook               = hooknum;
 423
 424         if (nla_put(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg))
 425                 goto nla_put_failure;
 426
 427         if (prefix &&
 428             nla_put(inst->skb, NFULA_PREFIX, plen, prefix))
 429                 goto nla_put_failure;
 430
 431         if (indev) {
 432 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 433                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 434                                  htonl(indev->ifindex)))
 435                         goto nla_put_failure;
 436 #else
 437                 if (pf == PF_BRIDGE) {
 438                         /* Case 1: outdev is physical input device, we need to
 439                          * look for bridge group (when called from
 440                          * netfilter_bridge) */
 441                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 442                                          htonl(indev->ifindex)) ||
 443                         /* this is the bridge group "brX" */
 444                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 445                             nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 446                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 447                                 goto nla_put_failure;
 448                 } else {
 449                         /* Case 2: indev is bridge group, we need to look for
 450                          * physical device (when called from ipv4) */
 451                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 452                                          htonl(indev->ifindex)))
 453                                 goto nla_put_failure;
 454                         if (skb->nf_bridge && skb->nf_bridge->physindev &&
 455                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 456                                          htonl(skb->nf_bridge->physindev->ifindex)))
 457                                 goto nla_put_failure;
 458                 }
 459 #endif
 460         }
 461
 462         if (outdev) {
 463 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 464                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 465                                  htonl(outdev->ifindex)))
 466                         goto nla_put_failure;
 467 #else
 468                 if (pf == PF_BRIDGE) {
 469                         /* Case 1: outdev is physical output device, we need to
 470                          * look for bridge group (when called from
 471                          * netfilter_bridge) */
 472                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 473                                          htonl(outdev->ifindex)) ||
 474                         /* this is the bridge group "brX" */
 475                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 476                             nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 477                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 478                                 goto nla_put_failure;
 479                 } else {
 480                         /* Case 2: indev is a bridge group, we need to look
 481                          * for physical device (when called from ipv4) */
 482                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 483                                          htonl(outdev->ifindex)))
 484                                 goto nla_put_failure;
 485                         if (skb->nf_bridge && skb->nf_bridge->physoutdev &&
 486                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 487                                          htonl(skb->nf_bridge->physoutdev->ifindex)))
 488                                 goto nla_put_failure;
 489                 }
 490 #endif
 491         }
 492
 493         if (skb->mark &&
 494             nla_put_be32(inst->skb, NFULA_MARK, htonl(skb->mark)))
 495                 goto nla_put_failure;
 496
 497         if (indev && skb->dev &&
 498             skb->mac_header != skb->network_header) {
 499                 struct nfulnl_msg_packet_hw phw;
 500                 int len;
 501
 502                 memset(&phw, 0, sizeof(phw));
 503                 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 #if IS_ENABLED(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
 654         if (in && skb_mac_header_was_set(skb)) {
 655                 size +=   nla_total_size(skb->dev->hard_header_len)
 656                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 657                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 658         }
 659
 660         spin_lock_bh(&inst->lock);
 661
 662         if (inst->flags & NFULNL_CFG_F_SEQ)
 663                 size += nla_total_size(sizeof(u_int32_t));
 664         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 665                 size += nla_total_size(sizeof(u_int32_t));
 666
 667         qthreshold = inst->qthreshold;
 668         /* per-rule qthreshold overrides per-instance */
 669         if (li->u.ulog.qthreshold)
 670                 if (qthreshold > li->u.ulog.qthreshold)
 671                         qthreshold = li->u.ulog.qthreshold;
 672
 673
 674         switch (inst->copy_mode) {
 675         case NFULNL_COPY_META:
 676         case NFULNL_COPY_NONE:
 677                 data_len = 0;
 678                 break;
 679
 680         case NFULNL_COPY_PACKET:
 681                 if (inst->copy_range == 0
 682                     || 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 &&
 696             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
 697                 /* either the queue len is too high or we don't have
 698                  * enough room in the skb left. flush to userspace. */
 699                 __nfulnl_flush(inst);
 700         }
 701
 702         if (!inst->skb) {
 703                 inst->skb = nfulnl_alloc_skb(net, inst->peer_portid,
 704                                              inst->nlbufsiz, size);
 705                 if (!inst->skb)
 706                         goto alloc_failure;
 707         }
 708
 709         inst->qlen++;
 710
 711         __build_packet_message(log, inst, skb, data_len, pf,
 712                                 hooknum, in, out, prefix, plen);
 713
 714         if (inst->qlen >= qthreshold)
 715                 __nfulnl_flush(inst);
 716         /* timer_pending always called within inst->lock, so there
 717          * is no chance of a race here */
 718         else if (!timer_pending(&inst->timer)) {
 719                 instance_get(inst);
 720                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 721                 add_timer(&inst->timer);
 722         }
 723
 724 unlock_and_release:
 725         spin_unlock_bh(&inst->lock);
 726         instance_put(inst);
 727         return;
 728
 729 alloc_failure:
 730         /* FIXME: statistics */
 731         goto unlock_and_release;
 732 }
 733 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
 734
 735 static int
 736 nfulnl_rcv_nl_event(struct notifier_block *this,
 737                    unsigned long event, void *ptr)
 738 {
 739         struct netlink_notify *n = ptr;
 740         struct nfnl_log_net *log = nfnl_log_pernet(n->net);
 741
 742         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 743                 int i;
 744
 745                 /* destroy all instances for this portid */
 746                 spin_lock_bh(&log->instances_lock);
 747                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 748                         struct hlist_node *t2;
 749                         struct nfulnl_instance *inst;
 750                         struct hlist_head *head = &log->instance_table[i];
 751
 752                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 753                                 if (n->portid == inst->peer_portid)
 754                                         __instance_destroy(inst);
 755                         }
 756                 }
 757                 spin_unlock_bh(&log->instances_lock);
 758         }
 759         return NOTIFY_DONE;
 760 }
 761
 762 static struct notifier_block nfulnl_rtnl_notifier = {
 763         .notifier_call  = nfulnl_rcv_nl_event,
 764 };
 765
 766 static int
 767 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 768                    const struct nlmsghdr *nlh,
 769                    const struct nlattr * const nfqa[])
 770 {
 771         return -ENOTSUPP;
 772 }
 773
 774 static struct nf_logger nfulnl_logger __read_mostly = {
 775         .name   = "nfnetlink_log",
 776         .type   = NF_LOG_TYPE_ULOG,
 777         .logfn  = &nfulnl_log_packet,
 778         .me     = THIS_MODULE,
 779 };
 780
 781 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 782         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 783         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 784         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 785         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 786         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 787         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 788 };
 789
 790 static int
 791 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 792                    const struct nlmsghdr *nlh,
 793                    const struct nlattr * const nfula[])
 794 {
 795         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 796         u_int16_t group_num = ntohs(nfmsg->res_id);
 797         struct nfulnl_instance *inst;
 798         struct nfulnl_msg_config_cmd *cmd = NULL;
 799         struct net *net = sock_net(ctnl);
 800         struct nfnl_log_net *log = nfnl_log_pernet(net);
 801         int ret = 0;
 802
 803         if (nfula[NFULA_CFG_CMD]) {
 804                 u_int8_t pf = nfmsg->nfgen_family;
 805                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 806
 807                 /* Commands without queue context */
 808                 switch (cmd->command) {
 809                 case NFULNL_CFG_CMD_PF_BIND:
 810                         return nf_log_bind_pf(net, pf, &nfulnl_logger);
 811                 case NFULNL_CFG_CMD_PF_UNBIND:
 812                         nf_log_unbind_pf(net, pf);
 813                         return 0;
 814                 }
 815         }
 816
 817         inst = instance_lookup_get(log, group_num);
 818         if (inst && inst->peer_portid != NETLINK_CB(skb).portid) {
 819                 ret = -EPERM;
 820                 goto out_put;
 821         }
 822
 823         if (cmd != NULL) {
 824                 switch (cmd->command) {
 825                 case NFULNL_CFG_CMD_BIND:
 826                         if (inst) {
 827                                 ret = -EBUSY;
 828                                 goto out_put;
 829                         }
 830
 831                         inst = instance_create(net, group_num,
 832                                                NETLINK_CB(skb).portid,
 833                                                sk_user_ns(NETLINK_CB(skb).sk));
 834                         if (IS_ERR(inst)) {
 835                                 ret = PTR_ERR(inst);
 836                                 goto out;
 837                         }
 838                         break;
 839                 case NFULNL_CFG_CMD_UNBIND:
 840                         if (!inst) {
 841                                 ret = -ENODEV;
 842                                 goto out;
 843                         }
 844
 845                         instance_destroy(log, inst);
 846                         goto out_put;
 847                 default:
 848                         ret = -ENOTSUPP;
 849                         break;
 850                 }
 851         }
 852
 853         if (nfula[NFULA_CFG_MODE]) {
 854                 struct nfulnl_msg_config_mode *params;
 855                 params = nla_data(nfula[NFULA_CFG_MODE]);
 856
 857                 if (!inst) {
 858                         ret = -ENODEV;
 859                         goto out;
 860                 }
 861                 nfulnl_set_mode(inst, params->copy_mode,
 862                                 ntohl(params->copy_range));
 863         }
 864
 865         if (nfula[NFULA_CFG_TIMEOUT]) {
 866                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 867
 868                 if (!inst) {
 869                         ret = -ENODEV;
 870                         goto out;
 871                 }
 872                 nfulnl_set_timeout(inst, ntohl(timeout));
 873         }
 874
 875         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 876                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 877
 878                 if (!inst) {
 879                         ret = -ENODEV;
 880                         goto out;
 881                 }
 882                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 883         }
 884
 885         if (nfula[NFULA_CFG_QTHRESH]) {
 886                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 887
 888                 if (!inst) {
 889                         ret = -ENODEV;
 890                         goto out;
 891                 }
 892                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 893         }
 894
 895         if (nfula[NFULA_CFG_FLAGS]) {
 896                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
 897
 898                 if (!inst) {
 899                         ret = -ENODEV;
 900                         goto out;
 901                 }
 902                 nfulnl_set_flags(inst, ntohs(flags));
 903         }
 904
 905 out_put:
 906         instance_put(inst);
 907 out:
 908         return ret;
 909 }
 910
 911 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 912         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 913                                     .attr_count = NFULA_MAX, },
 914         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 915                                     .attr_count = NFULA_CFG_MAX,
 916                                     .policy = nfula_cfg_policy },
 917 };
 918
 919 static const struct nfnetlink_subsystem nfulnl_subsys = {
 920         .name           = "log",
 921         .subsys_id      = NFNL_SUBSYS_ULOG,
 922         .cb_count       = NFULNL_MSG_MAX,
 923         .cb             = nfulnl_cb,
 924 };
 925
 926 #ifdef CONFIG_PROC_FS
 927 struct iter_state {
 928         struct seq_net_private p;
 929         unsigned int bucket;
 930 };
 931
 932 static struct hlist_node *get_first(struct net *net, struct iter_state *st)
 933 {
 934         struct nfnl_log_net *log;
 935         if (!st)
 936                 return NULL;
 937
 938         log = nfnl_log_pernet(net);
 939
 940         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 941                 struct hlist_head *head = &log->instance_table[st->bucket];
 942
 943                 if (!hlist_empty(head))
 944                         return rcu_dereference_bh(hlist_first_rcu(head));
 945         }
 946         return NULL;
 947 }
 948
 949 static struct hlist_node *get_next(struct net *net, struct iter_state *st,
 950                                    struct hlist_node *h)
 951 {
 952         h = rcu_dereference_bh(hlist_next_rcu(h));
 953         while (!h) {
 954                 struct nfnl_log_net *log;
 955                 struct hlist_head *head;
 956
 957                 if (++st->bucket >= INSTANCE_BUCKETS)
 958                         return NULL;
 959
 960                 log = nfnl_log_pernet(net);
 961                 head = &log->instance_table[st->bucket];
 962                 h = rcu_dereference_bh(hlist_first_rcu(head));
 963         }
 964         return h;
 965 }
 966
 967 static struct hlist_node *get_idx(struct net *net, struct iter_state *st,
 968                                   loff_t pos)
 969 {
 970         struct hlist_node *head;
 971         head = get_first(net, st);
 972
 973         if (head)
 974                 while (pos && (head = get_next(net, st, head)))
 975                         pos--;
 976         return pos ? NULL : head;
 977 }
 978
 979 static void *seq_start(struct seq_file *s, loff_t *pos)
 980         __acquires(rcu_bh)
 981 {
 982         rcu_read_lock_bh();
 983         return get_idx(seq_file_net(s), s->private, *pos);
 984 }
 985
 986 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 987 {
 988         (*pos)++;
 989         return get_next(seq_file_net(s), s->private, v);
 990 }
 991
 992 static void seq_stop(struct seq_file *s, void *v)
 993         __releases(rcu_bh)
 994 {
 995         rcu_read_unlock_bh();
 996 }
 997
 998 static int seq_show(struct seq_file *s, void *v)
 999 {
1000         const struct nfulnl_instance *inst = v;
1001
1002         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
1003                           inst->group_num,
1004                           inst->peer_portid, inst->qlen,
1005                           inst->copy_mode, inst->copy_range,
1006                           inst->flushtimeout, atomic_read(&inst->use));
1007 }
1008
1009 static const struct seq_operations nful_seq_ops = {
1010         .start  = seq_start,
1011         .next   = seq_next,
1012         .stop   = seq_stop,
1013         .show   = seq_show,
1014 };
1015
1016 static int nful_open(struct inode *inode, struct file *file)
1017 {
1018         return seq_open_net(inode, file, &nful_seq_ops,
1019                             sizeof(struct iter_state));
1020 }
1021
1022 static const struct file_operations nful_file_ops = {
1023         .owner   = THIS_MODULE,
1024         .open    = nful_open,
1025         .read    = seq_read,
1026         .llseek  = seq_lseek,
1027         .release = seq_release_net,
1028 };
1029
1030 #endif /* PROC_FS */
1031
1032 static int __net_init nfnl_log_net_init(struct net *net)
1033 {
1034         unsigned int i;
1035         struct nfnl_log_net *log = nfnl_log_pernet(net);
1036
1037         for (i = 0; i < INSTANCE_BUCKETS; i++)
1038                 INIT_HLIST_HEAD(&log->instance_table[i]);
1039         spin_lock_init(&log->instances_lock);
1040
1041 #ifdef CONFIG_PROC_FS
1042         if (!proc_create("nfnetlink_log", 0440,
1043                          net->nf.proc_netfilter, &nful_file_ops))
1044                 return -ENOMEM;
1045 #endif
1046         return 0;
1047 }
1048
1049 static void __net_exit nfnl_log_net_exit(struct net *net)
1050 {
1051 #ifdef CONFIG_PROC_FS
1052         remove_proc_entry("nfnetlink_log", net->nf.proc_netfilter);
1053 #endif
1054         nf_log_unset(net, &nfulnl_logger);
1055 }
1056
1057 static struct pernet_operations nfnl_log_net_ops = {
1058         .init   = nfnl_log_net_init,
1059         .exit   = nfnl_log_net_exit,
1060         .id     = &nfnl_log_net_id,
1061         .size   = sizeof(struct nfnl_log_net),
1062 };
1063
1064 static int __init nfnetlink_log_init(void)
1065 {
1066         int status = -ENOMEM;
1067
1068         netlink_register_notifier(&nfulnl_rtnl_notifier);
1069         status = nfnetlink_subsys_register(&nfulnl_subsys);
1070         if (status < 0) {
1071                 pr_err("log: failed to create netlink socket\n");
1072                 goto cleanup_netlink_notifier;
1073         }
1074
1075         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
1076         if (status < 0) {
1077                 pr_err("log: failed to register logger\n");
1078                 goto cleanup_subsys;
1079         }
1080
1081         status = register_pernet_subsys(&nfnl_log_net_ops);
1082         if (status < 0) {
1083                 pr_err("log: failed to register pernet ops\n");
1084                 goto cleanup_logger;
1085         }
1086         return status;
1087
1088 cleanup_logger:
1089         nf_log_unregister(&nfulnl_logger);
1090 cleanup_subsys:
1091         nfnetlink_subsys_unregister(&nfulnl_subsys);
1092 cleanup_netlink_notifier:
1093         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1094         return status;
1095 }
1096
1097 static void __exit nfnetlink_log_fini(void)
1098 {
1099         unregister_pernet_subsys(&nfnl_log_net_ops);
1100         nf_log_unregister(&nfulnl_logger);
1101         nfnetlink_subsys_unregister(&nfulnl_subsys);
1102         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1103 }
1104
1105 MODULE_DESCRIPTION("netfilter userspace logging");
1106 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1107 MODULE_LICENSE("GPL");
1108 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1109 MODULE_ALIAS_NF_LOGGER(AF_INET, 1);
1110 MODULE_ALIAS_NF_LOGGER(AF_INET6, 1);
1111 MODULE_ALIAS_NF_LOGGER(AF_BRIDGE, 1);
1112
1113 module_init(nfnetlink_log_init);
1114 module_exit(nfnetlink_log_fini);