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
  16 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  17
  18 #include <linux/module.h>
  19 #include <linux/skbuff.h>
  20 #include <linux/if_arp.h>
  21 #include <linux/init.h>
  22 #include <linux/ip.h>
  23 #include <linux/ipv6.h>
  24 #include <linux/netdevice.h>
  25 #include <linux/netfilter.h>
  26 #include <linux/netfilter_bridge.h>
  27 #include <net/netlink.h>
  28 #include <linux/netfilter/nfnetlink.h>
  29 #include <linux/netfilter/nfnetlink_log.h>
  30 #include <linux/spinlock.h>
  31 #include <linux/sysctl.h>
  32 #include <linux/proc_fs.h>
  33 #include <linux/security.h>
  34 #include <linux/list.h>
  35 #include <linux/slab.h>
  36 #include <net/sock.h>
  37 #include <net/netfilter/nf_log.h>
  38 #include <net/netns/generic.h>
  39 #include <net/netfilter/nfnetlink_log.h>
  40
  41 #include <linux/atomic.h>
  42
  43 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
  44 #include "../bridge/br_private.h"
  45 #endif
  46
  47 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
  48 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
  49 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  50 /* max packet size is limited by 16-bit struct nfattr nfa_len field */
  51 #define NFULNL_COPY_RANGE_MAX   (0xFFFF - NLA_HDRLEN)
  52
  53 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  54                                      printk(x, ## args); } while (0);
  55
  56 struct nfulnl_instance {
  57         struct hlist_node hlist;        /* global list of instances */
  58         spinlock_t lock;
  59         atomic_t use;                   /* use count */
  60
  61         unsigned int qlen;              /* number of nlmsgs in skb */
  62         struct sk_buff *skb;            /* pre-allocatd skb */
  63         struct timer_list timer;
  64         struct net *net;
  65         struct user_namespace *peer_user_ns;    /* User namespace of the peer process */
  66         u32 peer_portid;                /* PORTID of the peer process */
  67
  68         /* configurable parameters */
  69         unsigned int flushtimeout;      /* timeout until queue flush */
  70         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  71         unsigned int qthreshold;        /* threshold of the queue */
  72         u_int32_t copy_range;
  73         u_int32_t seq;                  /* instance-local sequential counter */
  74         u_int16_t group_num;            /* number of this queue */
  75         u_int16_t flags;
  76         u_int8_t copy_mode;
  77         struct rcu_head rcu;
  78 };
  79
  80 #define INSTANCE_BUCKETS        16
  81
  82 static int nfnl_log_net_id __read_mostly;
  83
  84 struct nfnl_log_net {
  85         spinlock_t instances_lock;
  86         struct hlist_head instance_table[INSTANCE_BUCKETS];
  87         atomic_t global_seq;
  88 };
  89
  90 static struct nfnl_log_net *nfnl_log_pernet(struct net *net)
  91 {
  92         return net_generic(net, nfnl_log_net_id);
  93 }
  94
  95 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  96 {
  97         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  98 }
  99
 100 static struct nfulnl_instance *
 101 __instance_lookup(struct nfnl_log_net *log, u_int16_t group_num)
 102 {
 103         struct hlist_head *head;
 104         struct nfulnl_instance *inst;
 105
 106         head = &log->instance_table[instance_hashfn(group_num)];
 107         hlist_for_each_entry_rcu(inst, head, hlist) {
 108                 if (inst->group_num == group_num)
 109                         return inst;
 110         }
 111         return NULL;
 112 }
 113
 114 static inline void
 115 instance_get(struct nfulnl_instance *inst)
 116 {
 117         atomic_inc(&inst->use);
 118 }
 119
 120 static struct nfulnl_instance *
 121 instance_lookup_get(struct nfnl_log_net *log, u_int16_t group_num)
 122 {
 123         struct nfulnl_instance *inst;
 124
 125         rcu_read_lock_bh();
 126         inst = __instance_lookup(log, group_num);
 127         if (inst && !atomic_inc_not_zero(&inst->use))
 128                 inst = NULL;
 129         rcu_read_unlock_bh();
 130
 131         return inst;
 132 }
 133
 134 static void nfulnl_instance_free_rcu(struct rcu_head *head)
 135 {
 136         struct nfulnl_instance *inst =
 137                 container_of(head, struct nfulnl_instance, rcu);
 138
 139         put_net(inst->net);
 140         kfree(inst);
 141         module_put(THIS_MODULE);
 142 }
 143
 144 static void
 145 instance_put(struct nfulnl_instance *inst)
 146 {
 147         if (inst && atomic_dec_and_test(&inst->use))
 148                 call_rcu_bh(&inst->rcu, nfulnl_instance_free_rcu);
 149 }
 150
 151 static void nfulnl_timer(unsigned long data);
 152
 153 static struct nfulnl_instance *
 154 instance_create(struct net *net, u_int16_t group_num,
 155                 u32 portid, struct user_namespace *user_ns)
 156 {
 157         struct nfulnl_instance *inst;
 158         struct nfnl_log_net *log = nfnl_log_pernet(net);
 159         int err;
 160
 161         spin_lock_bh(&log->instances_lock);
 162         if (__instance_lookup(log, group_num)) {
 163                 err = -EEXIST;
 164                 goto out_unlock;
 165         }
 166
 167         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 168         if (!inst) {
 169                 err = -ENOMEM;
 170                 goto out_unlock;
 171         }
 172
 173         if (!try_module_get(THIS_MODULE)) {
 174                 kfree(inst);
 175                 err = -EAGAIN;
 176                 goto out_unlock;
 177         }
 178
 179         INIT_HLIST_NODE(&inst->hlist);
 180         spin_lock_init(&inst->lock);
 181         /* needs to be two, since we _put() after creation */
 182         atomic_set(&inst->use, 2);
 183
 184         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
 185
 186         inst->net = get_net(net);
 187         inst->peer_user_ns = user_ns;
 188         inst->peer_portid = portid;
 189         inst->group_num = group_num;
 190
 191         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 192         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 193         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 194         inst->copy_mode         = NFULNL_COPY_PACKET;
 195         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
 196
 197         hlist_add_head_rcu(&inst->hlist,
 198                        &log->instance_table[instance_hashfn(group_num)]);
 199
 200
 201         spin_unlock_bh(&log->instances_lock);
 202
 203         return inst;
 204
 205 out_unlock:
 206         spin_unlock_bh(&log->instances_lock);
 207         return ERR_PTR(err);
 208 }
 209
 210 static void __nfulnl_flush(struct nfulnl_instance *inst);
 211
 212 /* called with BH disabled */
 213 static void
 214 __instance_destroy(struct nfulnl_instance *inst)
 215 {
 216         /* first pull it out of the global list */
 217         hlist_del_rcu(&inst->hlist);
 218
 219         /* then flush all pending packets from skb */
 220
 221         spin_lock(&inst->lock);
 222
 223         /* lockless readers wont be able to use us */
 224         inst->copy_mode = NFULNL_COPY_DISABLED;
 225
 226         if (inst->skb)
 227                 __nfulnl_flush(inst);
 228         spin_unlock(&inst->lock);
 229
 230         /* and finally put the refcount */
 231         instance_put(inst);
 232 }
 233
 234 static inline void
 235 instance_destroy(struct nfnl_log_net *log,
 236                  struct nfulnl_instance *inst)
 237 {
 238         spin_lock_bh(&log->instances_lock);
 239         __instance_destroy(inst);
 240         spin_unlock_bh(&log->instances_lock);
 241 }
 242
 243 static int
 244 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 245                   unsigned int range)
 246 {
 247         int status = 0;
 248
 249         spin_lock_bh(&inst->lock);
 250
 251         switch (mode) {
 252         case NFULNL_COPY_NONE:
 253         case NFULNL_COPY_META:
 254                 inst->copy_mode = mode;
 255                 inst->copy_range = 0;
 256                 break;
 257
 258         case NFULNL_COPY_PACKET:
 259                 inst->copy_mode = mode;
 260                 if (range == 0)
 261                         range = NFULNL_COPY_RANGE_MAX;
 262                 inst->copy_range = min_t(unsigned int,
 263                                          range, NFULNL_COPY_RANGE_MAX);
 264                 break;
 265
 266         default:
 267                 status = -EINVAL;
 268                 break;
 269         }
 270
 271         spin_unlock_bh(&inst->lock);
 272
 273         return status;
 274 }
 275
 276 static int
 277 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 278 {
 279         int status;
 280
 281         spin_lock_bh(&inst->lock);
 282         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 283                 status = -ERANGE;
 284         else if (nlbufsiz > 131072)
 285                 status = -ERANGE;
 286         else {
 287                 inst->nlbufsiz = nlbufsiz;
 288                 status = 0;
 289         }
 290         spin_unlock_bh(&inst->lock);
 291
 292         return status;
 293 }
 294
 295 static int
 296 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 297 {
 298         spin_lock_bh(&inst->lock);
 299         inst->flushtimeout = timeout;
 300         spin_unlock_bh(&inst->lock);
 301
 302         return 0;
 303 }
 304
 305 static int
 306 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 307 {
 308         spin_lock_bh(&inst->lock);
 309         inst->qthreshold = qthresh;
 310         spin_unlock_bh(&inst->lock);
 311
 312         return 0;
 313 }
 314
 315 static int
 316 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 317 {
 318         spin_lock_bh(&inst->lock);
 319         inst->flags = flags;
 320         spin_unlock_bh(&inst->lock);
 321
 322         return 0;
 323 }
 324
 325 static struct sk_buff *
 326 nfulnl_alloc_skb(struct net *net, u32 peer_portid, unsigned int inst_size,
 327                  unsigned int pkt_size)
 328 {
 329         struct sk_buff *skb;
 330         unsigned int n;
 331
 332         /* alloc skb which should be big enough for a whole multipart
 333          * message.  WARNING: has to be <= 128k due to slab restrictions */
 334
 335         n = max(inst_size, pkt_size);
 336         skb = nfnetlink_alloc_skb(net, n, peer_portid, GFP_ATOMIC);
 337         if (!skb) {
 338                 if (n > pkt_size) {
 339                         /* try to allocate only as much as we need for current
 340                          * packet */
 341
 342                         skb = nfnetlink_alloc_skb(net, pkt_size,
 343                                                   peer_portid, GFP_ATOMIC);
 344                 }
 345         }
 346
 347         return skb;
 348 }
 349
 350 static void
 351 __nfulnl_send(struct nfulnl_instance *inst)
 352 {
 353         if (inst->qlen > 1) {
 354                 struct nlmsghdr *nlh = nlmsg_put(inst->skb, 0, 0,
 355                                                  NLMSG_DONE,
 356                                                  sizeof(struct nfgenmsg),
 357                                                  0);
 358                 if (WARN_ONCE(!nlh, "bad nlskb size: %u, tailroom %d\n",
 359                               inst->skb->len, skb_tailroom(inst->skb))) {
 360                         kfree_skb(inst->skb);
 361                         goto out;
 362                 }
 363         }
 364         nfnetlink_unicast(inst->skb, inst->net, inst->peer_portid,
 365                           MSG_DONTWAIT);
 366 out:
 367         inst->qlen = 0;
 368         inst->skb = NULL;
 369 }
 370
 371 static void
 372 __nfulnl_flush(struct nfulnl_instance *inst)
 373 {
 374         /* timer holds a reference */
 375         if (del_timer(&inst->timer))
 376                 instance_put(inst);
 377         if (inst->skb)
 378                 __nfulnl_send(inst);
 379 }
 380
 381 static void
 382 nfulnl_timer(unsigned long data)
 383 {
 384         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 385
 386         spin_lock_bh(&inst->lock);
 387         if (inst->skb)
 388                 __nfulnl_send(inst);
 389         spin_unlock_bh(&inst->lock);
 390         instance_put(inst);
 391 }
 392
 393 /* This is an inline function, we don't really care about a long
 394  * list of arguments */
 395 static inline int
 396 __build_packet_message(struct nfnl_log_net *log,
 397                         struct nfulnl_instance *inst,
 398                         const struct sk_buff *skb,
 399                         unsigned int data_len,
 400                         u_int8_t pf,
 401                         unsigned int hooknum,
 402                         const struct net_device *indev,
 403                         const struct net_device *outdev,
 404                         const char *prefix, unsigned int plen)
 405 {
 406         struct nfulnl_msg_packet_hdr pmsg;
 407         struct nlmsghdr *nlh;
 408         struct nfgenmsg *nfmsg;
 409         sk_buff_data_t old_tail = inst->skb->tail;
 410         struct sock *sk;
 411         const unsigned char *hwhdrp;
 412
 413         nlh = nlmsg_put(inst->skb, 0, 0,
 414                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 415                         sizeof(struct nfgenmsg), 0);
 416         if (!nlh)
 417                 return -1;
 418         nfmsg = nlmsg_data(nlh);
 419         nfmsg->nfgen_family = pf;
 420         nfmsg->version = NFNETLINK_V0;
 421         nfmsg->res_id = htons(inst->group_num);
 422
 423         memset(&pmsg, 0, sizeof(pmsg));
 424         pmsg.hw_protocol        = skb->protocol;
 425         pmsg.hook               = hooknum;
 426
 427         if (nla_put(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg))
 428                 goto nla_put_failure;
 429
 430         if (prefix &&
 431             nla_put(inst->skb, NFULA_PREFIX, plen, prefix))
 432                 goto nla_put_failure;
 433
 434         if (indev) {
 435 #if !IS_ENABLED(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                         struct net_device *physindev;
 453
 454                         /* Case 2: indev is bridge group, we need to look for
 455                          * physical device (when called from ipv4) */
 456                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 457                                          htonl(indev->ifindex)))
 458                                 goto nla_put_failure;
 459
 460                         physindev = nf_bridge_get_physindev(skb);
 461                         if (physindev &&
 462                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 463                                          htonl(physindev->ifindex)))
 464                                 goto nla_put_failure;
 465                 }
 466 #endif
 467         }
 468
 469         if (outdev) {
 470 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 471                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 472                                  htonl(outdev->ifindex)))
 473                         goto nla_put_failure;
 474 #else
 475                 if (pf == PF_BRIDGE) {
 476                         /* Case 1: outdev is physical output device, we need to
 477                          * look for bridge group (when called from
 478                          * netfilter_bridge) */
 479                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 480                                          htonl(outdev->ifindex)) ||
 481                         /* this is the bridge group "brX" */
 482                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 483                             nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 484                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 485                                 goto nla_put_failure;
 486                 } else {
 487                         struct net_device *physoutdev;
 488
 489                         /* Case 2: indev is a bridge group, we need to look
 490                          * for physical device (when called from ipv4) */
 491                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 492                                          htonl(outdev->ifindex)))
 493                                 goto nla_put_failure;
 494
 495                         physoutdev = nf_bridge_get_physoutdev(skb);
 496                         if (physoutdev &&
 497                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 498                                          htonl(physoutdev->ifindex)))
 499                                 goto nla_put_failure;
 500                 }
 501 #endif
 502         }
 503
 504         if (skb->mark &&
 505             nla_put_be32(inst->skb, NFULA_MARK, htonl(skb->mark)))
 506                 goto nla_put_failure;
 507
 508         if (indev && skb->dev &&
 509             skb->mac_header != skb->network_header) {
 510                 struct nfulnl_msg_packet_hw phw;
 511                 int len;
 512
 513                 memset(&phw, 0, sizeof(phw));
 514                 len = dev_parse_header(skb, phw.hw_addr);
 515                 if (len > 0) {
 516                         phw.hw_addrlen = htons(len);
 517                         if (nla_put(inst->skb, NFULA_HWADDR, sizeof(phw), &phw))
 518                                 goto nla_put_failure;
 519                 }
 520         }
 521
 522         if (indev && skb_mac_header_was_set(skb)) {
 523                 if (nla_put_be16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type)) ||
 524                     nla_put_be16(inst->skb, NFULA_HWLEN,
 525                                  htons(skb->dev->hard_header_len)))
 526                         goto nla_put_failure;
 527
 528                 hwhdrp = skb_mac_header(skb);
 529
 530                 if (skb->dev->type == ARPHRD_SIT)
 531                         hwhdrp -= ETH_HLEN;
 532
 533                 if (hwhdrp >= skb->head &&
 534                     nla_put(inst->skb, NFULA_HWHEADER,
 535                             skb->dev->hard_header_len, hwhdrp))
 536                         goto nla_put_failure;
 537         }
 538
 539         if (skb->tstamp.tv64) {
 540                 struct nfulnl_msg_packet_timestamp ts;
 541                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 542                 ts.sec = cpu_to_be64(tv.tv_sec);
 543                 ts.usec = cpu_to_be64(tv.tv_usec);
 544
 545                 if (nla_put(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts))
 546                         goto nla_put_failure;
 547         }
 548
 549         /* UID */
 550         sk = skb->sk;
 551         if (sk && sk_fullsock(sk)) {
 552                 read_lock_bh(&sk->sk_callback_lock);
 553                 if (sk->sk_socket && sk->sk_socket->file) {
 554                         struct file *file = sk->sk_socket->file;
 555                         const struct cred *cred = file->f_cred;
 556                         struct user_namespace *user_ns = inst->peer_user_ns;
 557                         __be32 uid = htonl(from_kuid_munged(user_ns, cred->fsuid));
 558                         __be32 gid = htonl(from_kgid_munged(user_ns, cred->fsgid));
 559                         read_unlock_bh(&sk->sk_callback_lock);
 560                         if (nla_put_be32(inst->skb, NFULA_UID, uid) ||
 561                             nla_put_be32(inst->skb, NFULA_GID, gid))
 562                                 goto nla_put_failure;
 563                 } else
 564                         read_unlock_bh(&sk->sk_callback_lock);
 565         }
 566
 567         /* local sequence number */
 568         if ((inst->flags & NFULNL_CFG_F_SEQ) &&
 569             nla_put_be32(inst->skb, NFULA_SEQ, htonl(inst->seq++)))
 570                 goto nla_put_failure;
 571
 572         /* global sequence number */
 573         if ((inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) &&
 574             nla_put_be32(inst->skb, NFULA_SEQ_GLOBAL,
 575                          htonl(atomic_inc_return(&log->global_seq))))
 576                 goto nla_put_failure;
 577
 578         if (data_len) {
 579                 struct nlattr *nla;
 580                 int size = nla_attr_size(data_len);
 581
 582                 if (skb_tailroom(inst->skb) < nla_total_size(data_len))
 583                         goto nla_put_failure;
 584
 585                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 586                 nla->nla_type = NFULA_PAYLOAD;
 587                 nla->nla_len = size;
 588
 589                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 590                         BUG();
 591         }
 592
 593         nlh->nlmsg_len = inst->skb->tail - old_tail;
 594         return 0;
 595
 596 nla_put_failure:
 597         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 598         return -1;
 599 }
 600
 601 static struct nf_loginfo default_loginfo = {
 602         .type =         NF_LOG_TYPE_ULOG,
 603         .u = {
 604                 .ulog = {
 605                         .copy_len       = 0xffff,
 606                         .group          = 0,
 607                         .qthreshold     = 1,
 608                 },
 609         },
 610 };
 611
 612 /* log handler for internal netfilter logging api */
 613 void
 614 nfulnl_log_packet(struct net *net,
 615                   u_int8_t pf,
 616                   unsigned int hooknum,
 617                   const struct sk_buff *skb,
 618                   const struct net_device *in,
 619                   const struct net_device *out,
 620                   const struct nf_loginfo *li_user,
 621                   const char *prefix)
 622 {
 623         unsigned int size, data_len;
 624         struct nfulnl_instance *inst;
 625         const struct nf_loginfo *li;
 626         unsigned int qthreshold;
 627         unsigned int plen;
 628         struct nfnl_log_net *log = nfnl_log_pernet(net);
 629
 630         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 631                 li = li_user;
 632         else
 633                 li = &default_loginfo;
 634
 635         inst = instance_lookup_get(log, li->u.ulog.group);
 636         if (!inst)
 637                 return;
 638
 639         plen = 0;
 640         if (prefix)
 641                 plen = strlen(prefix) + 1;
 642
 643         /* FIXME: do we want to make the size calculation conditional based on
 644          * what is actually present?  way more branches and checks, but more
 645          * memory efficient... */
 646         size =    nlmsg_total_size(sizeof(struct nfgenmsg))
 647                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 648                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 649                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 650 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 651                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 652                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 653 #endif
 654                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 655                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 656                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 657                 + nla_total_size(plen)                  /* prefix */
 658                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 659                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp))
 660                 + nla_total_size(sizeof(struct nfgenmsg));      /* NLMSG_DONE */
 661
 662         if (in && skb_mac_header_was_set(skb)) {
 663                 size +=   nla_total_size(skb->dev->hard_header_len)
 664                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 665                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 666         }
 667
 668         spin_lock_bh(&inst->lock);
 669
 670         if (inst->flags & NFULNL_CFG_F_SEQ)
 671                 size += nla_total_size(sizeof(u_int32_t));
 672         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 673                 size += nla_total_size(sizeof(u_int32_t));
 674
 675         qthreshold = inst->qthreshold;
 676         /* per-rule qthreshold overrides per-instance */
 677         if (li->u.ulog.qthreshold)
 678                 if (qthreshold > li->u.ulog.qthreshold)
 679                         qthreshold = li->u.ulog.qthreshold;
 680
 681
 682         switch (inst->copy_mode) {
 683         case NFULNL_COPY_META:
 684         case NFULNL_COPY_NONE:
 685                 data_len = 0;
 686                 break;
 687
 688         case NFULNL_COPY_PACKET:
 689                 if (inst->copy_range > skb->len)
 690                         data_len = skb->len;
 691                 else
 692                         data_len = inst->copy_range;
 693
 694                 size += nla_total_size(data_len);
 695                 break;
 696
 697         case NFULNL_COPY_DISABLED:
 698         default:
 699                 goto unlock_and_release;
 700         }
 701
 702         if (inst->skb && size > skb_tailroom(inst->skb)) {
 703                 /* either the queue len is too high or we don't have
 704                  * enough room in the skb left. flush to userspace. */
 705                 __nfulnl_flush(inst);
 706         }
 707
 708         if (!inst->skb) {
 709                 inst->skb = nfulnl_alloc_skb(net, inst->peer_portid,
 710                                              inst->nlbufsiz, size);
 711                 if (!inst->skb)
 712                         goto alloc_failure;
 713         }
 714
 715         inst->qlen++;
 716
 717         __build_packet_message(log, inst, skb, data_len, pf,
 718                                 hooknum, in, out, prefix, plen);
 719
 720         if (inst->qlen >= qthreshold)
 721                 __nfulnl_flush(inst);
 722         /* timer_pending always called within inst->lock, so there
 723          * is no chance of a race here */
 724         else if (!timer_pending(&inst->timer)) {
 725                 instance_get(inst);
 726                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 727                 add_timer(&inst->timer);
 728         }
 729
 730 unlock_and_release:
 731         spin_unlock_bh(&inst->lock);
 732         instance_put(inst);
 733         return;
 734
 735 alloc_failure:
 736         /* FIXME: statistics */
 737         goto unlock_and_release;
 738 }
 739 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
 740
 741 static int
 742 nfulnl_rcv_nl_event(struct notifier_block *this,
 743                    unsigned long event, void *ptr)
 744 {
 745         struct netlink_notify *n = ptr;
 746         struct nfnl_log_net *log = nfnl_log_pernet(n->net);
 747
 748         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 749                 int i;
 750
 751                 /* destroy all instances for this portid */
 752                 spin_lock_bh(&log->instances_lock);
 753                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 754                         struct hlist_node *t2;
 755                         struct nfulnl_instance *inst;
 756                         struct hlist_head *head = &log->instance_table[i];
 757
 758                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 759                                 if (n->portid == inst->peer_portid)
 760                                         __instance_destroy(inst);
 761                         }
 762                 }
 763                 spin_unlock_bh(&log->instances_lock);
 764         }
 765         return NOTIFY_DONE;
 766 }
 767
 768 static struct notifier_block nfulnl_rtnl_notifier = {
 769         .notifier_call  = nfulnl_rcv_nl_event,
 770 };
 771
 772 static int
 773 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 774                    const struct nlmsghdr *nlh,
 775                    const struct nlattr * const nfqa[])
 776 {
 777         return -ENOTSUPP;
 778 }
 779
 780 static struct nf_logger nfulnl_logger __read_mostly = {
 781         .name   = "nfnetlink_log",
 782         .type   = NF_LOG_TYPE_ULOG,
 783         .logfn  = &nfulnl_log_packet,
 784         .me     = THIS_MODULE,
 785 };
 786
 787 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 788         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 789         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 790         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 791         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 792         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 793         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 794 };
 795
 796 static int
 797 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 798                    const struct nlmsghdr *nlh,
 799                    const struct nlattr * const nfula[])
 800 {
 801         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 802         u_int16_t group_num = ntohs(nfmsg->res_id);
 803         struct nfulnl_instance *inst;
 804         struct nfulnl_msg_config_cmd *cmd = NULL;
 805         struct net *net = sock_net(ctnl);
 806         struct nfnl_log_net *log = nfnl_log_pernet(net);
 807         int ret = 0;
 808
 809         if (nfula[NFULA_CFG_CMD]) {
 810                 u_int8_t pf = nfmsg->nfgen_family;
 811                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 812
 813                 /* Commands without queue context */
 814                 switch (cmd->command) {
 815                 case NFULNL_CFG_CMD_PF_BIND:
 816                         return nf_log_bind_pf(net, pf, &nfulnl_logger);
 817                 case NFULNL_CFG_CMD_PF_UNBIND:
 818                         nf_log_unbind_pf(net, pf);
 819                         return 0;
 820                 }
 821         }
 822
 823         inst = instance_lookup_get(log, group_num);
 824         if (inst && inst->peer_portid != NETLINK_CB(skb).portid) {
 825                 ret = -EPERM;
 826                 goto out_put;
 827         }
 828
 829         if (cmd != NULL) {
 830                 switch (cmd->command) {
 831                 case NFULNL_CFG_CMD_BIND:
 832                         if (inst) {
 833                                 ret = -EBUSY;
 834                                 goto out_put;
 835                         }
 836
 837                         inst = instance_create(net, group_num,
 838                                                NETLINK_CB(skb).portid,
 839                                                sk_user_ns(NETLINK_CB(skb).sk));
 840                         if (IS_ERR(inst)) {
 841                                 ret = PTR_ERR(inst);
 842                                 goto out;
 843                         }
 844                         break;
 845                 case NFULNL_CFG_CMD_UNBIND:
 846                         if (!inst) {
 847                                 ret = -ENODEV;
 848                                 goto out;
 849                         }
 850
 851                         instance_destroy(log, inst);
 852                         goto out_put;
 853                 default:
 854                         ret = -ENOTSUPP;
 855                         break;
 856                 }
 857         }
 858
 859         if (nfula[NFULA_CFG_MODE]) {
 860                 struct nfulnl_msg_config_mode *params;
 861                 params = nla_data(nfula[NFULA_CFG_MODE]);
 862
 863                 if (!inst) {
 864                         ret = -ENODEV;
 865                         goto out;
 866                 }
 867                 nfulnl_set_mode(inst, params->copy_mode,
 868                                 ntohl(params->copy_range));
 869         }
 870
 871         if (nfula[NFULA_CFG_TIMEOUT]) {
 872                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 873
 874                 if (!inst) {
 875                         ret = -ENODEV;
 876                         goto out;
 877                 }
 878                 nfulnl_set_timeout(inst, ntohl(timeout));
 879         }
 880
 881         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 882                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 883
 884                 if (!inst) {
 885                         ret = -ENODEV;
 886                         goto out;
 887                 }
 888                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 889         }
 890
 891         if (nfula[NFULA_CFG_QTHRESH]) {
 892                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 893
 894                 if (!inst) {
 895                         ret = -ENODEV;
 896                         goto out;
 897                 }
 898                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 899         }
 900
 901         if (nfula[NFULA_CFG_FLAGS]) {
 902                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
 903
 904                 if (!inst) {
 905                         ret = -ENODEV;
 906                         goto out;
 907                 }
 908                 nfulnl_set_flags(inst, ntohs(flags));
 909         }
 910
 911 out_put:
 912         instance_put(inst);
 913 out:
 914         return ret;
 915 }
 916
 917 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 918         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 919                                     .attr_count = NFULA_MAX, },
 920         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 921                                     .attr_count = NFULA_CFG_MAX,
 922                                     .policy = nfula_cfg_policy },
 923 };
 924
 925 static const struct nfnetlink_subsystem nfulnl_subsys = {
 926         .name           = "log",
 927         .subsys_id      = NFNL_SUBSYS_ULOG,
 928         .cb_count       = NFULNL_MSG_MAX,
 929         .cb             = nfulnl_cb,
 930 };
 931
 932 #ifdef CONFIG_PROC_FS
 933 struct iter_state {
 934         struct seq_net_private p;
 935         unsigned int bucket;
 936 };
 937
 938 static struct hlist_node *get_first(struct net *net, struct iter_state *st)
 939 {
 940         struct nfnl_log_net *log;
 941         if (!st)
 942                 return NULL;
 943
 944         log = nfnl_log_pernet(net);
 945
 946         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 947                 struct hlist_head *head = &log->instance_table[st->bucket];
 948
 949                 if (!hlist_empty(head))
 950                         return rcu_dereference_bh(hlist_first_rcu(head));
 951         }
 952         return NULL;
 953 }
 954
 955 static struct hlist_node *get_next(struct net *net, struct iter_state *st,
 956                                    struct hlist_node *h)
 957 {
 958         h = rcu_dereference_bh(hlist_next_rcu(h));
 959         while (!h) {
 960                 struct nfnl_log_net *log;
 961                 struct hlist_head *head;
 962
 963                 if (++st->bucket >= INSTANCE_BUCKETS)
 964                         return NULL;
 965
 966                 log = nfnl_log_pernet(net);
 967                 head = &log->instance_table[st->bucket];
 968                 h = rcu_dereference_bh(hlist_first_rcu(head));
 969         }
 970         return h;
 971 }
 972
 973 static struct hlist_node *get_idx(struct net *net, struct iter_state *st,
 974                                   loff_t pos)
 975 {
 976         struct hlist_node *head;
 977         head = get_first(net, st);
 978
 979         if (head)
 980                 while (pos && (head = get_next(net, st, head)))
 981                         pos--;
 982         return pos ? NULL : head;
 983 }
 984
 985 static void *seq_start(struct seq_file *s, loff_t *pos)
 986         __acquires(rcu_bh)
 987 {
 988         rcu_read_lock_bh();
 989         return get_idx(seq_file_net(s), s->private, *pos);
 990 }
 991
 992 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 993 {
 994         (*pos)++;
 995         return get_next(seq_file_net(s), s->private, v);
 996 }
 997
 998 static void seq_stop(struct seq_file *s, void *v)
 999         __releases(rcu_bh)
1000 {
1001         rcu_read_unlock_bh();
1002 }
1003
1004 static int seq_show(struct seq_file *s, void *v)
1005 {
1006         const struct nfulnl_instance *inst = v;
1007
1008         seq_printf(s, "%5u %6u %5u %1u %5u %6u %2u\n",
1009                    inst->group_num,
1010                    inst->peer_portid, inst->qlen,
1011                    inst->copy_mode, inst->copy_range,
1012                    inst->flushtimeout, atomic_read(&inst->use));
1013
1014         return 0;
1015 }
1016
1017 static const struct seq_operations nful_seq_ops = {
1018         .start  = seq_start,
1019         .next   = seq_next,
1020         .stop   = seq_stop,
1021         .show   = seq_show,
1022 };
1023
1024 static int nful_open(struct inode *inode, struct file *file)
1025 {
1026         return seq_open_net(inode, file, &nful_seq_ops,
1027                             sizeof(struct iter_state));
1028 }
1029
1030 static const struct file_operations nful_file_ops = {
1031         .owner   = THIS_MODULE,
1032         .open    = nful_open,
1033         .read    = seq_read,
1034         .llseek  = seq_lseek,
1035         .release = seq_release_net,
1036 };
1037
1038 #endif /* PROC_FS */
1039
1040 static int __net_init nfnl_log_net_init(struct net *net)
1041 {
1042         unsigned int i;
1043         struct nfnl_log_net *log = nfnl_log_pernet(net);
1044
1045         for (i = 0; i < INSTANCE_BUCKETS; i++)
1046                 INIT_HLIST_HEAD(&log->instance_table[i]);
1047         spin_lock_init(&log->instances_lock);
1048
1049 #ifdef CONFIG_PROC_FS
1050         if (!proc_create("nfnetlink_log", 0440,
1051                          net->nf.proc_netfilter, &nful_file_ops))
1052                 return -ENOMEM;
1053 #endif
1054         return 0;
1055 }
1056
1057 static void __net_exit nfnl_log_net_exit(struct net *net)
1058 {
1059 #ifdef CONFIG_PROC_FS
1060         remove_proc_entry("nfnetlink_log", net->nf.proc_netfilter);
1061 #endif
1062         nf_log_unset(net, &nfulnl_logger);
1063 }
1064
1065 static struct pernet_operations nfnl_log_net_ops = {
1066         .init   = nfnl_log_net_init,
1067         .exit   = nfnl_log_net_exit,
1068         .id     = &nfnl_log_net_id,
1069         .size   = sizeof(struct nfnl_log_net),
1070 };
1071
1072 static int __init nfnetlink_log_init(void)
1073 {
1074         int status;
1075
1076         status = register_pernet_subsys(&nfnl_log_net_ops);
1077         if (status < 0) {
1078                 pr_err("failed to register pernet ops\n");
1079                 goto out;
1080         }
1081
1082         netlink_register_notifier(&nfulnl_rtnl_notifier);
1083         status = nfnetlink_subsys_register(&nfulnl_subsys);
1084         if (status < 0) {
1085                 pr_err("failed to create netlink socket\n");
1086                 goto cleanup_netlink_notifier;
1087         }
1088
1089         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
1090         if (status < 0) {
1091                 pr_err("failed to register logger\n");
1092                 goto cleanup_subsys;
1093         }
1094
1095         return status;
1096
1097 cleanup_subsys:
1098         nfnetlink_subsys_unregister(&nfulnl_subsys);
1099 cleanup_netlink_notifier:
1100         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1101         unregister_pernet_subsys(&nfnl_log_net_ops);
1102 out:
1103         return status;
1104 }
1105
1106 static void __exit nfnetlink_log_fini(void)
1107 {
1108         nf_log_unregister(&nfulnl_logger);
1109         nfnetlink_subsys_unregister(&nfulnl_subsys);
1110         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1111         unregister_pernet_subsys(&nfnl_log_net_ops);
1112 }
1113
1114 MODULE_DESCRIPTION("netfilter userspace logging");
1115 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1116 MODULE_LICENSE("GPL");
1117 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1118 MODULE_ALIAS_NF_LOGGER(AF_INET, 1);
1119 MODULE_ALIAS_NF_LOGGER(AF_INET6, 1);
1120 MODULE_ALIAS_NF_LOGGER(AF_BRIDGE, 1);
1121
1122 module_init(nfnetlink_log_init);
1123 module_exit(nfnetlink_log_fini);