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/netfilter/nf_conntrack_common.h>
  31 #include <linux/spinlock.h>
  32 #include <linux/sysctl.h>
  33 #include <linux/proc_fs.h>
  34 #include <linux/security.h>
  35 #include <linux/list.h>
  36 #include <linux/slab.h>
  37 #include <net/sock.h>
  38 #include <net/netfilter/nf_log.h>
  39 #include <net/netns/generic.h>
  40
  41 #include <linux/atomic.h>
  42 #include <linux/refcount.h>
  43
  44
  45 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
  46 #include "../bridge/br_private.h"
  47 #endif
  48
  49 #define NFULNL_COPY_DISABLED    0xff
  50 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
  51 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
  52 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  53 /* max packet size is limited by 16-bit struct nfattr nfa_len field */
  54 #define NFULNL_COPY_RANGE_MAX   (0xFFFF - NLA_HDRLEN)
  55
  56 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  57                                      printk(x, ## args); } while (0);
  58
  59 struct nfulnl_instance {
  60         struct hlist_node hlist;        /* global list of instances */
  61         spinlock_t lock;
  62         refcount_t use;                 /* use count */
  63
  64         unsigned int qlen;              /* number of nlmsgs in skb */
  65         struct sk_buff *skb;            /* pre-allocatd skb */
  66         struct timer_list timer;
  67         struct net *net;
  68         struct user_namespace *peer_user_ns;    /* User namespace of the peer process */
  69         u32 peer_portid;                /* PORTID of the peer process */
  70
  71         /* configurable parameters */
  72         unsigned int flushtimeout;      /* timeout until queue flush */
  73         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  74         unsigned int qthreshold;        /* threshold of the queue */
  75         u_int32_t copy_range;
  76         u_int32_t seq;                  /* instance-local sequential counter */
  77         u_int16_t group_num;            /* number of this queue */
  78         u_int16_t flags;
  79         u_int8_t copy_mode;
  80         struct rcu_head rcu;
  81 };
  82
  83 #define INSTANCE_BUCKETS        16
  84
  85 static unsigned int nfnl_log_net_id __read_mostly;
  86
  87 struct nfnl_log_net {
  88         spinlock_t instances_lock;
  89         struct hlist_head instance_table[INSTANCE_BUCKETS];
  90         atomic_t global_seq;
  91 };
  92
  93 static struct nfnl_log_net *nfnl_log_pernet(struct net *net)
  94 {
  95         return net_generic(net, nfnl_log_net_id);
  96 }
  97
  98 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  99 {
 100         return ((group_num & 0xff) % INSTANCE_BUCKETS);
 101 }
 102
 103 static struct nfulnl_instance *
 104 __instance_lookup(struct nfnl_log_net *log, u_int16_t group_num)
 105 {
 106         struct hlist_head *head;
 107         struct nfulnl_instance *inst;
 108
 109         head = &log->instance_table[instance_hashfn(group_num)];
 110         hlist_for_each_entry_rcu(inst, head, hlist) {
 111                 if (inst->group_num == group_num)
 112                         return inst;
 113         }
 114         return NULL;
 115 }
 116
 117 static inline void
 118 instance_get(struct nfulnl_instance *inst)
 119 {
 120         refcount_inc(&inst->use);
 121 }
 122
 123 static struct nfulnl_instance *
 124 instance_lookup_get(struct nfnl_log_net *log, u_int16_t group_num)
 125 {
 126         struct nfulnl_instance *inst;
 127
 128         rcu_read_lock_bh();
 129         inst = __instance_lookup(log, group_num);
 130         if (inst && !refcount_inc_not_zero(&inst->use))
 131                 inst = NULL;
 132         rcu_read_unlock_bh();
 133
 134         return inst;
 135 }
 136
 137 static void nfulnl_instance_free_rcu(struct rcu_head *head)
 138 {
 139         struct nfulnl_instance *inst =
 140                 container_of(head, struct nfulnl_instance, rcu);
 141
 142         put_net(inst->net);
 143         kfree(inst);
 144         module_put(THIS_MODULE);
 145 }
 146
 147 static void
 148 instance_put(struct nfulnl_instance *inst)
 149 {
 150         if (inst && refcount_dec_and_test(&inst->use))
 151                 call_rcu_bh(&inst->rcu, nfulnl_instance_free_rcu);
 152 }
 153
 154 static void nfulnl_timer(struct timer_list *t);
 155
 156 static struct nfulnl_instance *
 157 instance_create(struct net *net, u_int16_t group_num,
 158                 u32 portid, struct user_namespace *user_ns)
 159 {
 160         struct nfulnl_instance *inst;
 161         struct nfnl_log_net *log = nfnl_log_pernet(net);
 162         int err;
 163
 164         spin_lock_bh(&log->instances_lock);
 165         if (__instance_lookup(log, group_num)) {
 166                 err = -EEXIST;
 167                 goto out_unlock;
 168         }
 169
 170         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 171         if (!inst) {
 172                 err = -ENOMEM;
 173                 goto out_unlock;
 174         }
 175
 176         if (!try_module_get(THIS_MODULE)) {
 177                 kfree(inst);
 178                 err = -EAGAIN;
 179                 goto out_unlock;
 180         }
 181
 182         INIT_HLIST_NODE(&inst->hlist);
 183         spin_lock_init(&inst->lock);
 184         /* needs to be two, since we _put() after creation */
 185         refcount_set(&inst->use, 2);
 186
 187         timer_setup(&inst->timer, nfulnl_timer, 0);
 188
 189         inst->net = get_net(net);
 190         inst->peer_user_ns = user_ns;
 191         inst->peer_portid = portid;
 192         inst->group_num = group_num;
 193
 194         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 195         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 196         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 197         inst->copy_mode         = NFULNL_COPY_PACKET;
 198         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
 199
 200         hlist_add_head_rcu(&inst->hlist,
 201                        &log->instance_table[instance_hashfn(group_num)]);
 202
 203
 204         spin_unlock_bh(&log->instances_lock);
 205
 206         return inst;
 207
 208 out_unlock:
 209         spin_unlock_bh(&log->instances_lock);
 210         return ERR_PTR(err);
 211 }
 212
 213 static void __nfulnl_flush(struct nfulnl_instance *inst);
 214
 215 /* called with BH disabled */
 216 static void
 217 __instance_destroy(struct nfulnl_instance *inst)
 218 {
 219         /* first pull it out of the global list */
 220         hlist_del_rcu(&inst->hlist);
 221
 222         /* then flush all pending packets from skb */
 223
 224         spin_lock(&inst->lock);
 225
 226         /* lockless readers wont be able to use us */
 227         inst->copy_mode = NFULNL_COPY_DISABLED;
 228
 229         if (inst->skb)
 230                 __nfulnl_flush(inst);
 231         spin_unlock(&inst->lock);
 232
 233         /* and finally put the refcount */
 234         instance_put(inst);
 235 }
 236
 237 static inline void
 238 instance_destroy(struct nfnl_log_net *log,
 239                  struct nfulnl_instance *inst)
 240 {
 241         spin_lock_bh(&log->instances_lock);
 242         __instance_destroy(inst);
 243         spin_unlock_bh(&log->instances_lock);
 244 }
 245
 246 static int
 247 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 248                   unsigned int range)
 249 {
 250         int status = 0;
 251
 252         spin_lock_bh(&inst->lock);
 253
 254         switch (mode) {
 255         case NFULNL_COPY_NONE:
 256         case NFULNL_COPY_META:
 257                 inst->copy_mode = mode;
 258                 inst->copy_range = 0;
 259                 break;
 260
 261         case NFULNL_COPY_PACKET:
 262                 inst->copy_mode = mode;
 263                 if (range == 0)
 264                         range = NFULNL_COPY_RANGE_MAX;
 265                 inst->copy_range = min_t(unsigned int,
 266                                          range, NFULNL_COPY_RANGE_MAX);
 267                 break;
 268
 269         default:
 270                 status = -EINVAL;
 271                 break;
 272         }
 273
 274         spin_unlock_bh(&inst->lock);
 275
 276         return status;
 277 }
 278
 279 static int
 280 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 281 {
 282         int status;
 283
 284         spin_lock_bh(&inst->lock);
 285         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 286                 status = -ERANGE;
 287         else if (nlbufsiz > 131072)
 288                 status = -ERANGE;
 289         else {
 290                 inst->nlbufsiz = nlbufsiz;
 291                 status = 0;
 292         }
 293         spin_unlock_bh(&inst->lock);
 294
 295         return status;
 296 }
 297
 298 static void
 299 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 300 {
 301         spin_lock_bh(&inst->lock);
 302         inst->flushtimeout = timeout;
 303         spin_unlock_bh(&inst->lock);
 304 }
 305
 306 static void
 307 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 308 {
 309         spin_lock_bh(&inst->lock);
 310         inst->qthreshold = qthresh;
 311         spin_unlock_bh(&inst->lock);
 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 = alloc_skb(n, GFP_ATOMIC | __GFP_NOWARN);
 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 = alloc_skb(pkt_size, GFP_ATOMIC);
 342                 }
 343         }
 344
 345         return skb;
 346 }
 347
 348 static void
 349 __nfulnl_send(struct nfulnl_instance *inst)
 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 (WARN_ONCE(!nlh, "bad nlskb size: %u, tailroom %d\n",
 357                               inst->skb->len, skb_tailroom(inst->skb))) {
 358                         kfree_skb(inst->skb);
 359                         goto out;
 360                 }
 361         }
 362         nfnetlink_unicast(inst->skb, inst->net, inst->peer_portid);
 363 out:
 364         inst->qlen = 0;
 365         inst->skb = NULL;
 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(struct timer_list *t)
 380 {
 381         struct nfulnl_instance *inst = from_timer(inst, t, timer);
 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                         const struct nfnl_ct_hook *nfnl_ct,
 403                         struct nf_conn *ct, enum ip_conntrack_info ctinfo)
 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_msg_type(NFNL_SUBSYS_ULOG, 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_thresh or
 447                          * nf_log_packet.
 448                          */
 449                             nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 450                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 451                                 goto nla_put_failure;
 452                 } else {
 453                         struct net_device *physindev;
 454
 455                         /* Case 2: indev is bridge group, we need to look for
 456                          * physical device (when called from ipv4) */
 457                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 458                                          htonl(indev->ifindex)))
 459                                 goto nla_put_failure;
 460
 461                         physindev = nf_bridge_get_physindev(skb);
 462                         if (physindev &&
 463                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 464                                          htonl(physindev->ifindex)))
 465                                 goto nla_put_failure;
 466                 }
 467 #endif
 468         }
 469
 470         if (outdev) {
 471 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 472                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 473                                  htonl(outdev->ifindex)))
 474                         goto nla_put_failure;
 475 #else
 476                 if (pf == PF_BRIDGE) {
 477                         /* Case 1: outdev is physical output device, we need to
 478                          * look for bridge group (when called from
 479                          * netfilter_bridge) */
 480                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 481                                          htonl(outdev->ifindex)) ||
 482                         /* this is the bridge group "brX" */
 483                         /* rcu_read_lock()ed by nf_hook_thresh or
 484                          * nf_log_packet.
 485                          */
 486                             nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 487                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 488                                 goto nla_put_failure;
 489                 } else {
 490                         struct net_device *physoutdev;
 491
 492                         /* Case 2: indev is a bridge group, we need to look
 493                          * for physical device (when called from ipv4) */
 494                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 495                                          htonl(outdev->ifindex)))
 496                                 goto nla_put_failure;
 497
 498                         physoutdev = nf_bridge_get_physoutdev(skb);
 499                         if (physoutdev &&
 500                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 501                                          htonl(physoutdev->ifindex)))
 502                                 goto nla_put_failure;
 503                 }
 504 #endif
 505         }
 506
 507         if (skb->mark &&
 508             nla_put_be32(inst->skb, NFULA_MARK, htonl(skb->mark)))
 509                 goto nla_put_failure;
 510
 511         if (indev && skb->dev &&
 512             skb->mac_header != skb->network_header) {
 513                 struct nfulnl_msg_packet_hw phw;
 514                 int len;
 515
 516                 memset(&phw, 0, sizeof(phw));
 517                 len = dev_parse_header(skb, phw.hw_addr);
 518                 if (len > 0) {
 519                         phw.hw_addrlen = htons(len);
 520                         if (nla_put(inst->skb, NFULA_HWADDR, sizeof(phw), &phw))
 521                                 goto nla_put_failure;
 522                 }
 523         }
 524
 525         if (indev && skb_mac_header_was_set(skb)) {
 526                 if (nla_put_be16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type)) ||
 527                     nla_put_be16(inst->skb, NFULA_HWLEN,
 528                                  htons(skb->dev->hard_header_len)))
 529                         goto nla_put_failure;
 530
 531                 hwhdrp = skb_mac_header(skb);
 532
 533                 if (skb->dev->type == ARPHRD_SIT)
 534                         hwhdrp -= ETH_HLEN;
 535
 536                 if (hwhdrp >= skb->head &&
 537                     nla_put(inst->skb, NFULA_HWHEADER,
 538                             skb->dev->hard_header_len, hwhdrp))
 539                         goto nla_put_failure;
 540         }
 541
 542         if (skb->tstamp) {
 543                 struct nfulnl_msg_packet_timestamp ts;
 544                 struct timespec64 kts = ktime_to_timespec64(skb->tstamp);
 545                 ts.sec = cpu_to_be64(kts.tv_sec);
 546                 ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
 547
 548                 if (nla_put(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts))
 549                         goto nla_put_failure;
 550         }
 551
 552         /* UID */
 553         sk = skb->sk;
 554         if (sk && sk_fullsock(sk)) {
 555                 read_lock_bh(&sk->sk_callback_lock);
 556                 if (sk->sk_socket && sk->sk_socket->file) {
 557                         struct file *file = sk->sk_socket->file;
 558                         const struct cred *cred = file->f_cred;
 559                         struct user_namespace *user_ns = inst->peer_user_ns;
 560                         __be32 uid = htonl(from_kuid_munged(user_ns, cred->fsuid));
 561                         __be32 gid = htonl(from_kgid_munged(user_ns, cred->fsgid));
 562                         read_unlock_bh(&sk->sk_callback_lock);
 563                         if (nla_put_be32(inst->skb, NFULA_UID, uid) ||
 564                             nla_put_be32(inst->skb, NFULA_GID, gid))
 565                                 goto nla_put_failure;
 566                 } else
 567                         read_unlock_bh(&sk->sk_callback_lock);
 568         }
 569
 570         /* local sequence number */
 571         if ((inst->flags & NFULNL_CFG_F_SEQ) &&
 572             nla_put_be32(inst->skb, NFULA_SEQ, htonl(inst->seq++)))
 573                 goto nla_put_failure;
 574
 575         /* global sequence number */
 576         if ((inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) &&
 577             nla_put_be32(inst->skb, NFULA_SEQ_GLOBAL,
 578                          htonl(atomic_inc_return(&log->global_seq))))
 579                 goto nla_put_failure;
 580
 581         if (ct && nfnl_ct->build(inst->skb, ct, ctinfo,
 582                                  NFULA_CT, NFULA_CT_INFO) < 0)
 583                 goto nla_put_failure;
 584
 585         if (data_len) {
 586                 struct nlattr *nla;
 587                 int size = nla_attr_size(data_len);
 588
 589                 if (skb_tailroom(inst->skb) < nla_total_size(data_len))
 590                         goto nla_put_failure;
 591
 592                 nla = skb_put(inst->skb, nla_total_size(data_len));
 593                 nla->nla_type = NFULA_PAYLOAD;
 594                 nla->nla_len = size;
 595
 596                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 597                         BUG();
 598         }
 599
 600         nlh->nlmsg_len = inst->skb->tail - old_tail;
 601         return 0;
 602
 603 nla_put_failure:
 604         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 605         return -1;
 606 }
 607
 608 static const struct nf_loginfo default_loginfo = {
 609         .type =         NF_LOG_TYPE_ULOG,
 610         .u = {
 611                 .ulog = {
 612                         .copy_len       = 0xffff,
 613                         .group          = 0,
 614                         .qthreshold     = 1,
 615                 },
 616         },
 617 };
 618
 619 /* log handler for internal netfilter logging api */
 620 static void
 621 nfulnl_log_packet(struct net *net,
 622                   u_int8_t pf,
 623                   unsigned int hooknum,
 624                   const struct sk_buff *skb,
 625                   const struct net_device *in,
 626                   const struct net_device *out,
 627                   const struct nf_loginfo *li_user,
 628                   const char *prefix)
 629 {
 630         size_t size;
 631         unsigned int data_len;
 632         struct nfulnl_instance *inst;
 633         const struct nf_loginfo *li;
 634         unsigned int qthreshold;
 635         unsigned int plen = 0;
 636         struct nfnl_log_net *log = nfnl_log_pernet(net);
 637         const struct nfnl_ct_hook *nfnl_ct = NULL;
 638         struct nf_conn *ct = NULL;
 639         enum ip_conntrack_info uninitialized_var(ctinfo);
 640
 641         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 642                 li = li_user;
 643         else
 644                 li = &default_loginfo;
 645
 646         inst = instance_lookup_get(log, li->u.ulog.group);
 647         if (!inst)
 648                 return;
 649
 650         if (prefix)
 651                 plen = strlen(prefix) + 1;
 652
 653         /* FIXME: do we want to make the size calculation conditional based on
 654          * what is actually present?  way more branches and checks, but more
 655          * memory efficient... */
 656         size =    nlmsg_total_size(sizeof(struct nfgenmsg))
 657                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 658                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 659                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 660 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 661                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 662                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 663 #endif
 664                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 665                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 666                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 667                 + nla_total_size(plen)                  /* prefix */
 668                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 669                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp))
 670                 + nla_total_size(sizeof(struct nfgenmsg));      /* NLMSG_DONE */
 671
 672         if (in && skb_mac_header_was_set(skb)) {
 673                 size +=   nla_total_size(skb->dev->hard_header_len)
 674                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 675                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 676         }
 677
 678         spin_lock_bh(&inst->lock);
 679
 680         if (inst->flags & NFULNL_CFG_F_SEQ)
 681                 size += nla_total_size(sizeof(u_int32_t));
 682         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 683                 size += nla_total_size(sizeof(u_int32_t));
 684         if (inst->flags & NFULNL_CFG_F_CONNTRACK) {
 685                 nfnl_ct = rcu_dereference(nfnl_ct_hook);
 686                 if (nfnl_ct != NULL) {
 687                         ct = nfnl_ct->get_ct(skb, &ctinfo);
 688                         if (ct != NULL)
 689                                 size += nfnl_ct->build_size(ct);
 690                 }
 691         }
 692
 693         qthreshold = inst->qthreshold;
 694         /* per-rule qthreshold overrides per-instance */
 695         if (li->u.ulog.qthreshold)
 696                 if (qthreshold > li->u.ulog.qthreshold)
 697                         qthreshold = li->u.ulog.qthreshold;
 698
 699
 700         switch (inst->copy_mode) {
 701         case NFULNL_COPY_META:
 702         case NFULNL_COPY_NONE:
 703                 data_len = 0;
 704                 break;
 705
 706         case NFULNL_COPY_PACKET:
 707                 data_len = inst->copy_range;
 708                 if ((li->u.ulog.flags & NF_LOG_F_COPY_LEN) &&
 709                     (li->u.ulog.copy_len < data_len))
 710                         data_len = li->u.ulog.copy_len;
 711
 712                 if (data_len > skb->len)
 713                         data_len = skb->len;
 714
 715                 size += nla_total_size(data_len);
 716                 break;
 717
 718         case NFULNL_COPY_DISABLED:
 719         default:
 720                 goto unlock_and_release;
 721         }
 722
 723         if (inst->skb && size > skb_tailroom(inst->skb)) {
 724                 /* either the queue len is too high or we don't have
 725                  * enough room in the skb left. flush to userspace. */
 726                 __nfulnl_flush(inst);
 727         }
 728
 729         if (!inst->skb) {
 730                 inst->skb = nfulnl_alloc_skb(net, inst->peer_portid,
 731                                              inst->nlbufsiz, size);
 732                 if (!inst->skb)
 733                         goto alloc_failure;
 734         }
 735
 736         inst->qlen++;
 737
 738         __build_packet_message(log, inst, skb, data_len, pf,
 739                                 hooknum, in, out, prefix, plen,
 740                                 nfnl_ct, ct, ctinfo);
 741
 742         if (inst->qlen >= qthreshold)
 743                 __nfulnl_flush(inst);
 744         /* timer_pending always called within inst->lock, so there
 745          * is no chance of a race here */
 746         else if (!timer_pending(&inst->timer)) {
 747                 instance_get(inst);
 748                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 749                 add_timer(&inst->timer);
 750         }
 751
 752 unlock_and_release:
 753         spin_unlock_bh(&inst->lock);
 754         instance_put(inst);
 755         return;
 756
 757 alloc_failure:
 758         /* FIXME: statistics */
 759         goto unlock_and_release;
 760 }
 761
 762 static int
 763 nfulnl_rcv_nl_event(struct notifier_block *this,
 764                    unsigned long event, void *ptr)
 765 {
 766         struct netlink_notify *n = ptr;
 767         struct nfnl_log_net *log = nfnl_log_pernet(n->net);
 768
 769         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 770                 int i;
 771
 772                 /* destroy all instances for this portid */
 773                 spin_lock_bh(&log->instances_lock);
 774                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 775                         struct hlist_node *t2;
 776                         struct nfulnl_instance *inst;
 777                         struct hlist_head *head = &log->instance_table[i];
 778
 779                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 780                                 if (n->portid == inst->peer_portid)
 781                                         __instance_destroy(inst);
 782                         }
 783                 }
 784                 spin_unlock_bh(&log->instances_lock);
 785         }
 786         return NOTIFY_DONE;
 787 }
 788
 789 static struct notifier_block nfulnl_rtnl_notifier = {
 790         .notifier_call  = nfulnl_rcv_nl_event,
 791 };
 792
 793 static int nfulnl_recv_unsupp(struct net *net, struct sock *ctnl,
 794                               struct sk_buff *skb, const struct nlmsghdr *nlh,
 795                               const struct nlattr * const nfqa[],
 796                               struct netlink_ext_ack *extack)
 797 {
 798         return -ENOTSUPP;
 799 }
 800
 801 static struct nf_logger nfulnl_logger __read_mostly = {
 802         .name   = "nfnetlink_log",
 803         .type   = NF_LOG_TYPE_ULOG,
 804         .logfn  = nfulnl_log_packet,
 805         .me     = THIS_MODULE,
 806 };
 807
 808 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 809         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 810         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 811         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 812         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 813         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 814         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 815 };
 816
 817 static int nfulnl_recv_config(struct net *net, struct sock *ctnl,
 818                               struct sk_buff *skb, const struct nlmsghdr *nlh,
 819                               const struct nlattr * const nfula[],
 820                               struct netlink_ext_ack *extack)
 821 {
 822         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 823         u_int16_t group_num = ntohs(nfmsg->res_id);
 824         struct nfulnl_instance *inst;
 825         struct nfulnl_msg_config_cmd *cmd = NULL;
 826         struct nfnl_log_net *log = nfnl_log_pernet(net);
 827         int ret = 0;
 828         u16 flags = 0;
 829
 830         if (nfula[NFULA_CFG_CMD]) {
 831                 u_int8_t pf = nfmsg->nfgen_family;
 832                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 833
 834                 /* Commands without queue context */
 835                 switch (cmd->command) {
 836                 case NFULNL_CFG_CMD_PF_BIND:
 837                         return nf_log_bind_pf(net, pf, &nfulnl_logger);
 838                 case NFULNL_CFG_CMD_PF_UNBIND:
 839                         nf_log_unbind_pf(net, pf);
 840                         return 0;
 841                 }
 842         }
 843
 844         inst = instance_lookup_get(log, group_num);
 845         if (inst && inst->peer_portid != NETLINK_CB(skb).portid) {
 846                 ret = -EPERM;
 847                 goto out_put;
 848         }
 849
 850         /* Check if we support these flags in first place, dependencies should
 851          * be there too not to break atomicity.
 852          */
 853         if (nfula[NFULA_CFG_FLAGS]) {
 854                 flags = ntohs(nla_get_be16(nfula[NFULA_CFG_FLAGS]));
 855
 856                 if ((flags & NFULNL_CFG_F_CONNTRACK) &&
 857                     !rcu_access_pointer(nfnl_ct_hook)) {
 858 #ifdef CONFIG_MODULES
 859                         nfnl_unlock(NFNL_SUBSYS_ULOG);
 860                         request_module("ip_conntrack_netlink");
 861                         nfnl_lock(NFNL_SUBSYS_ULOG);
 862                         if (rcu_access_pointer(nfnl_ct_hook)) {
 863                                 ret = -EAGAIN;
 864                                 goto out_put;
 865                         }
 866 #endif
 867                         ret = -EOPNOTSUPP;
 868                         goto out_put;
 869                 }
 870         }
 871
 872         if (cmd != NULL) {
 873                 switch (cmd->command) {
 874                 case NFULNL_CFG_CMD_BIND:
 875                         if (inst) {
 876                                 ret = -EBUSY;
 877                                 goto out_put;
 878                         }
 879
 880                         inst = instance_create(net, group_num,
 881                                                NETLINK_CB(skb).portid,
 882                                                sk_user_ns(NETLINK_CB(skb).sk));
 883                         if (IS_ERR(inst)) {
 884                                 ret = PTR_ERR(inst);
 885                                 goto out;
 886                         }
 887                         break;
 888                 case NFULNL_CFG_CMD_UNBIND:
 889                         if (!inst) {
 890                                 ret = -ENODEV;
 891                                 goto out;
 892                         }
 893
 894                         instance_destroy(log, inst);
 895                         goto out_put;
 896                 default:
 897                         ret = -ENOTSUPP;
 898                         goto out_put;
 899                 }
 900         } else if (!inst) {
 901                 ret = -ENODEV;
 902                 goto out;
 903         }
 904
 905         if (nfula[NFULA_CFG_MODE]) {
 906                 struct nfulnl_msg_config_mode *params =
 907                         nla_data(nfula[NFULA_CFG_MODE]);
 908
 909                 nfulnl_set_mode(inst, params->copy_mode,
 910                                 ntohl(params->copy_range));
 911         }
 912
 913         if (nfula[NFULA_CFG_TIMEOUT]) {
 914                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 915
 916                 nfulnl_set_timeout(inst, ntohl(timeout));
 917         }
 918
 919         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 920                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 921
 922                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 923         }
 924
 925         if (nfula[NFULA_CFG_QTHRESH]) {
 926                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 927
 928                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 929         }
 930
 931         if (nfula[NFULA_CFG_FLAGS])
 932                 nfulnl_set_flags(inst, flags);
 933
 934 out_put:
 935         instance_put(inst);
 936 out:
 937         return ret;
 938 }
 939
 940 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 941         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 942                                     .attr_count = NFULA_MAX, },
 943         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 944                                     .attr_count = NFULA_CFG_MAX,
 945                                     .policy = nfula_cfg_policy },
 946 };
 947
 948 static const struct nfnetlink_subsystem nfulnl_subsys = {
 949         .name           = "log",
 950         .subsys_id      = NFNL_SUBSYS_ULOG,
 951         .cb_count       = NFULNL_MSG_MAX,
 952         .cb             = nfulnl_cb,
 953 };
 954
 955 #ifdef CONFIG_PROC_FS
 956 struct iter_state {
 957         struct seq_net_private p;
 958         unsigned int bucket;
 959 };
 960
 961 static struct hlist_node *get_first(struct net *net, struct iter_state *st)
 962 {
 963         struct nfnl_log_net *log;
 964         if (!st)
 965                 return NULL;
 966
 967         log = nfnl_log_pernet(net);
 968
 969         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 970                 struct hlist_head *head = &log->instance_table[st->bucket];
 971
 972                 if (!hlist_empty(head))
 973                         return rcu_dereference_bh(hlist_first_rcu(head));
 974         }
 975         return NULL;
 976 }
 977
 978 static struct hlist_node *get_next(struct net *net, struct iter_state *st,
 979                                    struct hlist_node *h)
 980 {
 981         h = rcu_dereference_bh(hlist_next_rcu(h));
 982         while (!h) {
 983                 struct nfnl_log_net *log;
 984                 struct hlist_head *head;
 985
 986                 if (++st->bucket >= INSTANCE_BUCKETS)
 987                         return NULL;
 988
 989                 log = nfnl_log_pernet(net);
 990                 head = &log->instance_table[st->bucket];
 991                 h = rcu_dereference_bh(hlist_first_rcu(head));
 992         }
 993         return h;
 994 }
 995
 996 static struct hlist_node *get_idx(struct net *net, struct iter_state *st,
 997                                   loff_t pos)
 998 {
 999         struct hlist_node *head;
1000         head = get_first(net, st);
1001
1002         if (head)
1003                 while (pos && (head = get_next(net, st, head)))
1004                         pos--;
1005         return pos ? NULL : head;
1006 }
1007
1008 static void *seq_start(struct seq_file *s, loff_t *pos)
1009         __acquires(rcu_bh)
1010 {
1011         rcu_read_lock_bh();
1012         return get_idx(seq_file_net(s), s->private, *pos);
1013 }
1014
1015 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1016 {
1017         (*pos)++;
1018         return get_next(seq_file_net(s), s->private, v);
1019 }
1020
1021 static void seq_stop(struct seq_file *s, void *v)
1022         __releases(rcu_bh)
1023 {
1024         rcu_read_unlock_bh();
1025 }
1026
1027 static int seq_show(struct seq_file *s, void *v)
1028 {
1029         const struct nfulnl_instance *inst = v;
1030
1031         seq_printf(s, "%5u %6u %5u %1u %5u %6u %2u\n",
1032                    inst->group_num,
1033                    inst->peer_portid, inst->qlen,
1034                    inst->copy_mode, inst->copy_range,
1035                    inst->flushtimeout, refcount_read(&inst->use));
1036
1037         return 0;
1038 }
1039
1040 static const struct seq_operations nful_seq_ops = {
1041         .start  = seq_start,
1042         .next   = seq_next,
1043         .stop   = seq_stop,
1044         .show   = seq_show,
1045 };
1046 #endif /* PROC_FS */
1047
1048 static int __net_init nfnl_log_net_init(struct net *net)
1049 {
1050         unsigned int i;
1051         struct nfnl_log_net *log = nfnl_log_pernet(net);
1052 #ifdef CONFIG_PROC_FS
1053         struct proc_dir_entry *proc;
1054         kuid_t root_uid;
1055         kgid_t root_gid;
1056 #endif
1057
1058         for (i = 0; i < INSTANCE_BUCKETS; i++)
1059                 INIT_HLIST_HEAD(&log->instance_table[i]);
1060         spin_lock_init(&log->instances_lock);
1061
1062 #ifdef CONFIG_PROC_FS
1063         proc = proc_create_net("nfnetlink_log", 0440, net->nf.proc_netfilter,
1064                         &nful_seq_ops, sizeof(struct iter_state));
1065         if (!proc)
1066                 return -ENOMEM;
1067
1068         root_uid = make_kuid(net->user_ns, 0);
1069         root_gid = make_kgid(net->user_ns, 0);
1070         if (uid_valid(root_uid) && gid_valid(root_gid))
1071                 proc_set_user(proc, root_uid, root_gid);
1072 #endif
1073         return 0;
1074 }
1075
1076 static void __net_exit nfnl_log_net_exit(struct net *net)
1077 {
1078         struct nfnl_log_net *log = nfnl_log_pernet(net);
1079         unsigned int i;
1080
1081 #ifdef CONFIG_PROC_FS
1082         remove_proc_entry("nfnetlink_log", net->nf.proc_netfilter);
1083 #endif
1084         nf_log_unset(net, &nfulnl_logger);
1085         for (i = 0; i < INSTANCE_BUCKETS; i++)
1086                 WARN_ON_ONCE(!hlist_empty(&log->instance_table[i]));
1087 }
1088
1089 static struct pernet_operations nfnl_log_net_ops = {
1090         .init   = nfnl_log_net_init,
1091         .exit   = nfnl_log_net_exit,
1092         .id     = &nfnl_log_net_id,
1093         .size   = sizeof(struct nfnl_log_net),
1094 };
1095
1096 static int __init nfnetlink_log_init(void)
1097 {
1098         int status;
1099
1100         status = register_pernet_subsys(&nfnl_log_net_ops);
1101         if (status < 0) {
1102                 pr_err("failed to register pernet ops\n");
1103                 goto out;
1104         }
1105
1106         netlink_register_notifier(&nfulnl_rtnl_notifier);
1107         status = nfnetlink_subsys_register(&nfulnl_subsys);
1108         if (status < 0) {
1109                 pr_err("failed to create netlink socket\n");
1110                 goto cleanup_netlink_notifier;
1111         }
1112
1113         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
1114         if (status < 0) {
1115                 pr_err("failed to register logger\n");
1116                 goto cleanup_subsys;
1117         }
1118
1119         return status;
1120
1121 cleanup_subsys:
1122         nfnetlink_subsys_unregister(&nfulnl_subsys);
1123 cleanup_netlink_notifier:
1124         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1125         unregister_pernet_subsys(&nfnl_log_net_ops);
1126 out:
1127         return status;
1128 }
1129
1130 static void __exit nfnetlink_log_fini(void)
1131 {
1132         nfnetlink_subsys_unregister(&nfulnl_subsys);
1133         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1134         unregister_pernet_subsys(&nfnl_log_net_ops);
1135         nf_log_unregister(&nfulnl_logger);
1136 }
1137
1138 MODULE_DESCRIPTION("netfilter userspace logging");
1139 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1140 MODULE_LICENSE("GPL");
1141 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1142 MODULE_ALIAS_NF_LOGGER(AF_INET, 1);
1143 MODULE_ALIAS_NF_LOGGER(AF_INET6, 1);
1144 MODULE_ALIAS_NF_LOGGER(AF_BRIDGE, 1);
1145 MODULE_ALIAS_NF_LOGGER(3, 1); /* NFPROTO_ARP */
1146 MODULE_ALIAS_NF_LOGGER(5, 1); /* NFPROTO_NETDEV */
1147
1148 module_init(nfnetlink_log_init);
1149 module_exit(nfnetlink_log_fini);