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