net/netfilter/nfnetlink_log.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * This is a module which is used for logging packets to userspace via
   4  * nfetlink.
   5  *
   6  * (C) 2005 by Harald Welte <laforge@netfilter.org>
   7  * (C) 2006-2012 Patrick McHardy <kaber@trash.net>
   8  *
   9  * Based on the old ipv4-only ipt_ULOG.c:
  10  * (C) 2000-2004 by Harald Welte <laforge@netfilter.org>
  11  */
  12
  13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  14
  15 #include <linux/module.h>
  16 #include <linux/skbuff.h>
  17 #include <linux/if_arp.h>
  18 #include <linux/init.h>
  19 #include <linux/ip.h>
  20 #include <linux/ipv6.h>
  21 #include <linux/netdevice.h>
  22 #include <linux/netfilter.h>
  23 #include <linux/netfilter_bridge.h>
  24 #include <net/netlink.h>
  25 #include <linux/netfilter/nfnetlink.h>
  26 #include <linux/netfilter/nfnetlink_log.h>
  27 #include <linux/netfilter/nf_conntrack_common.h>
  28 #include <linux/spinlock.h>
  29 #include <linux/sysctl.h>
  30 #include <linux/proc_fs.h>
  31 #include <linux/security.h>
  32 #include <linux/list.h>
  33 #include <linux/slab.h>
  34 #include <net/sock.h>
  35 #include <net/netfilter/nf_log.h>
  36 #include <net/netns/generic.h>
  37
  38 #include <linux/atomic.h>
  39 #include <linux/refcount.h>
  40
  41
  42 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
  43 #include "../bridge/br_private.h"
  44 #endif
  45
  46 #define NFULNL_COPY_DISABLED    0xff
  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         refcount_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 unsigned 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         refcount_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 && !refcount_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 && refcount_dec_and_test(&inst->use))
 148                 call_rcu(&inst->rcu, nfulnl_instance_free_rcu);
 149 }
 150
 151 static void nfulnl_timer(struct timer_list *t);
 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         refcount_set(&inst->use, 2);
 183
 184         timer_setup(&inst->timer, nfulnl_timer, 0);
 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 void
 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
 303 static void
 304 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 305 {
 306         spin_lock_bh(&inst->lock);
 307         inst->qthreshold = qthresh;
 308         spin_unlock_bh(&inst->lock);
 309 }
 310
 311 static int
 312 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 313 {
 314         spin_lock_bh(&inst->lock);
 315         inst->flags = flags;
 316         spin_unlock_bh(&inst->lock);
 317
 318         return 0;
 319 }
 320
 321 static struct sk_buff *
 322 nfulnl_alloc_skb(struct net *net, u32 peer_portid, unsigned int inst_size,
 323                  unsigned int pkt_size)
 324 {
 325         struct sk_buff *skb;
 326         unsigned int n;
 327
 328         /* alloc skb which should be big enough for a whole multipart
 329          * message.  WARNING: has to be <= 128k due to slab restrictions */
 330
 331         n = max(inst_size, pkt_size);
 332         skb = alloc_skb(n, GFP_ATOMIC | __GFP_NOWARN);
 333         if (!skb) {
 334                 if (n > pkt_size) {
 335                         /* try to allocate only as much as we need for current
 336                          * packet */
 337
 338                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 339                 }
 340         }
 341
 342         return skb;
 343 }
 344
 345 static void
 346 __nfulnl_send(struct nfulnl_instance *inst)
 347 {
 348         if (inst->qlen > 1) {
 349                 struct nlmsghdr *nlh = nlmsg_put(inst->skb, 0, 0,
 350                                                  NLMSG_DONE,
 351                                                  sizeof(struct nfgenmsg),
 352                                                  0);
 353                 if (WARN_ONCE(!nlh, "bad nlskb size: %u, tailroom %d\n",
 354                               inst->skb->len, skb_tailroom(inst->skb))) {
 355                         kfree_skb(inst->skb);
 356                         goto out;
 357                 }
 358         }
 359         nfnetlink_unicast(inst->skb, inst->net, inst->peer_portid,
 360                           MSG_DONTWAIT);
 361 out:
 362         inst->qlen = 0;
 363         inst->skb = NULL;
 364 }
 365
 366 static void
 367 __nfulnl_flush(struct nfulnl_instance *inst)
 368 {
 369         /* timer holds a reference */
 370         if (del_timer(&inst->timer))
 371                 instance_put(inst);
 372         if (inst->skb)
 373                 __nfulnl_send(inst);
 374 }
 375
 376 static void
 377 nfulnl_timer(struct timer_list *t)
 378 {
 379         struct nfulnl_instance *inst = from_timer(inst, t, timer);
 380
 381         spin_lock_bh(&inst->lock);
 382         if (inst->skb)
 383                 __nfulnl_send(inst);
 384         spin_unlock_bh(&inst->lock);
 385         instance_put(inst);
 386 }
 387
 388 /* This is an inline function, we don't really care about a long
 389  * list of arguments */
 390 static inline int
 391 __build_packet_message(struct nfnl_log_net *log,
 392                         struct nfulnl_instance *inst,
 393                         const struct sk_buff *skb,
 394                         unsigned int data_len,
 395                         u_int8_t pf,
 396                         unsigned int hooknum,
 397                         const struct net_device *indev,
 398                         const struct net_device *outdev,
 399                         const char *prefix, unsigned int plen,
 400                         const struct nfnl_ct_hook *nfnl_ct,
 401                         struct nf_conn *ct, enum ip_conntrack_info ctinfo)
 402 {
 403         struct nfulnl_msg_packet_hdr pmsg;
 404         struct nlmsghdr *nlh;
 405         struct nfgenmsg *nfmsg;
 406         sk_buff_data_t old_tail = inst->skb->tail;
 407         struct sock *sk;
 408         const unsigned char *hwhdrp;
 409
 410         nlh = nlmsg_put(inst->skb, 0, 0,
 411                         nfnl_msg_type(NFNL_SUBSYS_ULOG, NFULNL_MSG_PACKET),
 412                         sizeof(struct nfgenmsg), 0);
 413         if (!nlh)
 414                 return -1;
 415         nfmsg = nlmsg_data(nlh);
 416         nfmsg->nfgen_family = pf;
 417         nfmsg->version = NFNETLINK_V0;
 418         nfmsg->res_id = htons(inst->group_num);
 419
 420         memset(&pmsg, 0, sizeof(pmsg));
 421         pmsg.hw_protocol        = skb->protocol;
 422         pmsg.hook               = hooknum;
 423
 424         if (nla_put(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg))
 425                 goto nla_put_failure;
 426
 427         if (prefix &&
 428             nla_put(inst->skb, NFULA_PREFIX, plen, prefix))
 429                 goto nla_put_failure;
 430
 431         if (indev) {
 432 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 433                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 434                                  htonl(indev->ifindex)))
 435                         goto nla_put_failure;
 436 #else
 437                 if (pf == PF_BRIDGE) {
 438                         /* Case 1: outdev is physical input device, we need to
 439                          * look for bridge group (when called from
 440                          * netfilter_bridge) */
 441                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 442                                          htonl(indev->ifindex)) ||
 443                         /* this is the bridge group "brX" */
 444                         /* rcu_read_lock()ed by nf_hook_thresh or
 445                          * nf_log_packet.
 446                          */
 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                         struct net_device *physindev;
 452
 453                         /* Case 2: indev is bridge group, we need to look for
 454                          * physical device (when called from ipv4) */
 455                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 456                                          htonl(indev->ifindex)))
 457                                 goto nla_put_failure;
 458
 459                         physindev = nf_bridge_get_physindev(skb);
 460                         if (physindev &&
 461                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 462                                          htonl(physindev->ifindex)))
 463                                 goto nla_put_failure;
 464                 }
 465 #endif
 466         }
 467
 468         if (outdev) {
 469 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 470                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 471                                  htonl(outdev->ifindex)))
 472                         goto nla_put_failure;
 473 #else
 474                 if (pf == PF_BRIDGE) {
 475                         /* Case 1: outdev is physical output device, we need to
 476                          * look for bridge group (when called from
 477                          * netfilter_bridge) */
 478                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 479                                          htonl(outdev->ifindex)) ||
 480                         /* this is the bridge group "brX" */
 481                         /* rcu_read_lock()ed by nf_hook_thresh or
 482                          * nf_log_packet.
 483                          */
 484                             nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 485                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 486                                 goto nla_put_failure;
 487                 } else {
 488                         struct net_device *physoutdev;
 489
 490                         /* Case 2: indev is a bridge group, we need to look
 491                          * for physical device (when called from ipv4) */
 492                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 493                                          htonl(outdev->ifindex)))
 494                                 goto nla_put_failure;
 495
 496                         physoutdev = nf_bridge_get_physoutdev(skb);
 497                         if (physoutdev &&
 498                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 499                                          htonl(physoutdev->ifindex)))
 500                                 goto nla_put_failure;
 501                 }
 502 #endif
 503         }
 504
 505         if (skb->mark &&
 506             nla_put_be32(inst->skb, NFULA_MARK, htonl(skb->mark)))
 507                 goto nla_put_failure;
 508
 509         if (indev && skb->dev &&
 510             skb->mac_header != skb->network_header) {
 511                 struct nfulnl_msg_packet_hw phw;
 512                 int len;
 513
 514                 memset(&phw, 0, sizeof(phw));
 515                 len = dev_parse_header(skb, phw.hw_addr);
 516                 if (len > 0) {
 517                         phw.hw_addrlen = htons(len);
 518                         if (nla_put(inst->skb, NFULA_HWADDR, sizeof(phw), &phw))
 519                                 goto nla_put_failure;
 520                 }
 521         }
 522
 523         if (indev && skb_mac_header_was_set(skb)) {
 524                 if (nla_put_be16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type)) ||
 525                     nla_put_be16(inst->skb, NFULA_HWLEN,
 526                                  htons(skb->dev->hard_header_len)))
 527                         goto nla_put_failure;
 528
 529                 hwhdrp = skb_mac_header(skb);
 530
 531                 if (skb->dev->type == ARPHRD_SIT)
 532                         hwhdrp -= ETH_HLEN;
 533
 534                 if (hwhdrp >= skb->head &&
 535                     nla_put(inst->skb, NFULA_HWHEADER,
 536                             skb->dev->hard_header_len, hwhdrp))
 537                         goto nla_put_failure;
 538         }
 539
 540         if (hooknum <= NF_INET_FORWARD && skb->tstamp) {
 541                 struct nfulnl_msg_packet_timestamp ts;
 542                 struct timespec64 kts = ktime_to_timespec64(skb->tstamp);
 543                 ts.sec = cpu_to_be64(kts.tv_sec);
 544                 ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
 545
 546                 if (nla_put(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts))
 547                         goto nla_put_failure;
 548         }
 549
 550         /* UID */
 551         sk = skb->sk;
 552         if (sk && sk_fullsock(sk)) {
 553                 read_lock_bh(&sk->sk_callback_lock);
 554                 if (sk->sk_socket && sk->sk_socket->file) {
 555                         struct file *file = sk->sk_socket->file;
 556                         const struct cred *cred = file->f_cred;
 557                         struct user_namespace *user_ns = inst->peer_user_ns;
 558                         __be32 uid = htonl(from_kuid_munged(user_ns, cred->fsuid));
 559                         __be32 gid = htonl(from_kgid_munged(user_ns, cred->fsgid));
 560                         read_unlock_bh(&sk->sk_callback_lock);
 561                         if (nla_put_be32(inst->skb, NFULA_UID, uid) ||
 562                             nla_put_be32(inst->skb, NFULA_GID, gid))
 563                                 goto nla_put_failure;
 564                 } else
 565                         read_unlock_bh(&sk->sk_callback_lock);
 566         }
 567
 568         /* local sequence number */
 569         if ((inst->flags & NFULNL_CFG_F_SEQ) &&
 570             nla_put_be32(inst->skb, NFULA_SEQ, htonl(inst->seq++)))
 571                 goto nla_put_failure;
 572
 573         /* global sequence number */
 574         if ((inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) &&
 575             nla_put_be32(inst->skb, NFULA_SEQ_GLOBAL,
 576                          htonl(atomic_inc_return(&log->global_seq))))
 577                 goto nla_put_failure;
 578
 579         if (ct && nfnl_ct->build(inst->skb, ct, ctinfo,
 580                                  NFULA_CT, NFULA_CT_INFO) < 0)
 581                 goto nla_put_failure;
 582
 583         if (data_len) {
 584                 struct nlattr *nla;
 585                 int size = nla_attr_size(data_len);
 586
 587                 if (skb_tailroom(inst->skb) < nla_total_size(data_len))
 588                         goto nla_put_failure;
 589
 590                 nla = skb_put(inst->skb, nla_total_size(data_len));
 591                 nla->nla_type = NFULA_PAYLOAD;
 592                 nla->nla_len = size;
 593
 594                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 595                         BUG();
 596         }
 597
 598         nlh->nlmsg_len = inst->skb->tail - old_tail;
 599         return 0;
 600
 601 nla_put_failure:
 602         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 603         return -1;
 604 }
 605
 606 static const struct nf_loginfo default_loginfo = {
 607         .type =         NF_LOG_TYPE_ULOG,
 608         .u = {
 609                 .ulog = {
 610                         .copy_len       = 0xffff,
 611                         .group          = 0,
 612                         .qthreshold     = 1,
 613                 },
 614         },
 615 };
 616
 617 /* log handler for internal netfilter logging api */
 618 static void
 619 nfulnl_log_packet(struct net *net,
 620                   u_int8_t pf,
 621                   unsigned int hooknum,
 622                   const struct sk_buff *skb,
 623                   const struct net_device *in,
 624                   const struct net_device *out,
 625                   const struct nf_loginfo *li_user,
 626                   const char *prefix)
 627 {
 628         size_t size;
 629         unsigned int data_len;
 630         struct nfulnl_instance *inst;
 631         const struct nf_loginfo *li;
 632         unsigned int qthreshold;
 633         unsigned int plen = 0;
 634         struct nfnl_log_net *log = nfnl_log_pernet(net);
 635         const struct nfnl_ct_hook *nfnl_ct = NULL;
 636         struct nf_conn *ct = NULL;
 637         enum ip_conntrack_info uninitialized_var(ctinfo);
 638
 639         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 640                 li = li_user;
 641         else
 642                 li = &default_loginfo;
 643
 644         inst = instance_lookup_get(log, li->u.ulog.group);
 645         if (!inst)
 646                 return;
 647
 648         if (prefix)
 649                 plen = strlen(prefix) + 1;
 650
 651         /* FIXME: do we want to make the size calculation conditional based on
 652          * what is actually present?  way more branches and checks, but more
 653          * memory efficient... */
 654         size =    nlmsg_total_size(sizeof(struct nfgenmsg))
 655                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 656                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 657                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 658 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 659                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 660                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 661 #endif
 662                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 663                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 664                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 665                 + nla_total_size(plen)                  /* prefix */
 666                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 667                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp))
 668                 + nla_total_size(sizeof(struct nfgenmsg));      /* NLMSG_DONE */
 669
 670         if (in && skb_mac_header_was_set(skb)) {
 671                 size +=   nla_total_size(skb->dev->hard_header_len)
 672                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 673                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 674         }
 675
 676         spin_lock_bh(&inst->lock);
 677
 678         if (inst->flags & NFULNL_CFG_F_SEQ)
 679                 size += nla_total_size(sizeof(u_int32_t));
 680         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 681                 size += nla_total_size(sizeof(u_int32_t));
 682         if (inst->flags & NFULNL_CFG_F_CONNTRACK) {
 683                 nfnl_ct = rcu_dereference(nfnl_ct_hook);
 684                 if (nfnl_ct != NULL) {
 685                         ct = nfnl_ct->get_ct(skb, &ctinfo);
 686                         if (ct != NULL)
 687                                 size += nfnl_ct->build_size(ct);
 688                 }
 689         }
 690
 691         qthreshold = inst->qthreshold;
 692         /* per-rule qthreshold overrides per-instance */
 693         if (li->u.ulog.qthreshold)
 694                 if (qthreshold > li->u.ulog.qthreshold)
 695                         qthreshold = li->u.ulog.qthreshold;
 696
 697
 698         switch (inst->copy_mode) {
 699         case NFULNL_COPY_META:
 700         case NFULNL_COPY_NONE:
 701                 data_len = 0;
 702                 break;
 703
 704         case NFULNL_COPY_PACKET:
 705                 data_len = inst->copy_range;
 706                 if ((li->u.ulog.flags & NF_LOG_F_COPY_LEN) &&
 707                     (li->u.ulog.copy_len < data_len))
 708                         data_len = li->u.ulog.copy_len;
 709
 710                 if (data_len > skb->len)
 711                         data_len = skb->len;
 712
 713                 size += nla_total_size(data_len);
 714                 break;
 715
 716         case NFULNL_COPY_DISABLED:
 717         default:
 718                 goto unlock_and_release;
 719         }
 720
 721         if (inst->skb && size > skb_tailroom(inst->skb)) {
 722                 /* either the queue len is too high or we don't have
 723                  * enough room in the skb left. flush to userspace. */
 724                 __nfulnl_flush(inst);
 725         }
 726
 727         if (!inst->skb) {
 728                 inst->skb = nfulnl_alloc_skb(net, inst->peer_portid,
 729                                              inst->nlbufsiz, size);
 730                 if (!inst->skb)
 731                         goto alloc_failure;
 732         }
 733
 734         inst->qlen++;
 735
 736         __build_packet_message(log, inst, skb, data_len, pf,
 737                                 hooknum, in, out, prefix, plen,
 738                                 nfnl_ct, ct, ctinfo);
 739
 740         if (inst->qlen >= qthreshold)
 741                 __nfulnl_flush(inst);
 742         /* timer_pending always called within inst->lock, so there
 743          * is no chance of a race here */
 744         else if (!timer_pending(&inst->timer)) {
 745                 instance_get(inst);
 746                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 747                 add_timer(&inst->timer);
 748         }
 749
 750 unlock_and_release:
 751         spin_unlock_bh(&inst->lock);
 752         instance_put(inst);
 753         return;
 754
 755 alloc_failure:
 756         /* FIXME: statistics */
 757         goto unlock_and_release;
 758 }
 759
 760 static int
 761 nfulnl_rcv_nl_event(struct notifier_block *this,
 762                    unsigned long event, void *ptr)
 763 {
 764         struct netlink_notify *n = ptr;
 765         struct nfnl_log_net *log = nfnl_log_pernet(n->net);
 766
 767         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 768                 int i;
 769
 770                 /* destroy all instances for this portid */
 771                 spin_lock_bh(&log->instances_lock);
 772                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 773                         struct hlist_node *t2;
 774                         struct nfulnl_instance *inst;
 775                         struct hlist_head *head = &log->instance_table[i];
 776
 777                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 778                                 if (n->portid == inst->peer_portid)
 779                                         __instance_destroy(inst);
 780                         }
 781                 }
 782                 spin_unlock_bh(&log->instances_lock);
 783         }
 784         return NOTIFY_DONE;
 785 }
 786
 787 static struct notifier_block nfulnl_rtnl_notifier = {
 788         .notifier_call  = nfulnl_rcv_nl_event,
 789 };
 790
 791 static int nfulnl_recv_unsupp(struct net *net, struct sock *ctnl,
 792                               struct sk_buff *skb, const struct nlmsghdr *nlh,
 793                               const struct nlattr * const nfqa[],
 794                               struct netlink_ext_ack *extack)
 795 {
 796         return -ENOTSUPP;
 797 }
 798
 799 static struct nf_logger nfulnl_logger __read_mostly = {
 800         .name   = "nfnetlink_log",
 801         .type   = NF_LOG_TYPE_ULOG,
 802         .logfn  = nfulnl_log_packet,
 803         .me     = THIS_MODULE,
 804 };
 805
 806 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 807         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 808         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 809         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 810         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 811         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 812         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 813 };
 814
 815 static int nfulnl_recv_config(struct net *net, struct sock *ctnl,
 816                               struct sk_buff *skb, const struct nlmsghdr *nlh,
 817                               const struct nlattr * const nfula[],
 818                               struct netlink_ext_ack *extack)
 819 {
 820         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 821         u_int16_t group_num = ntohs(nfmsg->res_id);
 822         struct nfulnl_instance *inst;
 823         struct nfulnl_msg_config_cmd *cmd = NULL;
 824         struct nfnl_log_net *log = nfnl_log_pernet(net);
 825         int ret = 0;
 826         u16 flags = 0;
 827
 828         if (nfula[NFULA_CFG_CMD]) {
 829                 u_int8_t pf = nfmsg->nfgen_family;
 830                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 831
 832                 /* Commands without queue context */
 833                 switch (cmd->command) {
 834                 case NFULNL_CFG_CMD_PF_BIND:
 835                         return nf_log_bind_pf(net, pf, &nfulnl_logger);
 836                 case NFULNL_CFG_CMD_PF_UNBIND:
 837                         nf_log_unbind_pf(net, pf);
 838                         return 0;
 839                 }
 840         }
 841
 842         inst = instance_lookup_get(log, group_num);
 843         if (inst && inst->peer_portid != NETLINK_CB(skb).portid) {
 844                 ret = -EPERM;
 845                 goto out_put;
 846         }
 847
 848         /* Check if we support these flags in first place, dependencies should
 849          * be there too not to break atomicity.
 850          */
 851         if (nfula[NFULA_CFG_FLAGS]) {
 852                 flags = ntohs(nla_get_be16(nfula[NFULA_CFG_FLAGS]));
 853
 854                 if ((flags & NFULNL_CFG_F_CONNTRACK) &&
 855                     !rcu_access_pointer(nfnl_ct_hook)) {
 856 #ifdef CONFIG_MODULES
 857                         nfnl_unlock(NFNL_SUBSYS_ULOG);
 858                         request_module("ip_conntrack_netlink");
 859                         nfnl_lock(NFNL_SUBSYS_ULOG);
 860                         if (rcu_access_pointer(nfnl_ct_hook)) {
 861                                 ret = -EAGAIN;
 862                                 goto out_put;
 863                         }
 864 #endif
 865                         ret = -EOPNOTSUPP;
 866                         goto out_put;
 867                 }
 868         }
 869
 870         if (cmd != NULL) {
 871                 switch (cmd->command) {
 872                 case NFULNL_CFG_CMD_BIND:
 873                         if (inst) {
 874                                 ret = -EBUSY;
 875                                 goto out_put;
 876                         }
 877
 878                         inst = instance_create(net, group_num,
 879                                                NETLINK_CB(skb).portid,
 880                                                sk_user_ns(NETLINK_CB(skb).sk));
 881                         if (IS_ERR(inst)) {
 882                                 ret = PTR_ERR(inst);
 883                                 goto out;
 884                         }
 885                         break;
 886                 case NFULNL_CFG_CMD_UNBIND:
 887                         if (!inst) {
 888                                 ret = -ENODEV;
 889                                 goto out;
 890                         }
 891
 892                         instance_destroy(log, inst);
 893                         goto out_put;
 894                 default:
 895                         ret = -ENOTSUPP;
 896                         goto out_put;
 897                 }
 898         } else if (!inst) {
 899                 ret = -ENODEV;
 900                 goto out;
 901         }
 902
 903         if (nfula[NFULA_CFG_MODE]) {
 904                 struct nfulnl_msg_config_mode *params =
 905                         nla_data(nfula[NFULA_CFG_MODE]);
 906
 907                 nfulnl_set_mode(inst, params->copy_mode,
 908                                 ntohl(params->copy_range));
 909         }
 910
 911         if (nfula[NFULA_CFG_TIMEOUT]) {
 912                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 913
 914                 nfulnl_set_timeout(inst, ntohl(timeout));
 915         }
 916
 917         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 918                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 919
 920                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 921         }
 922
 923         if (nfula[NFULA_CFG_QTHRESH]) {
 924                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 925
 926                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 927         }
 928
 929         if (nfula[NFULA_CFG_FLAGS])
 930                 nfulnl_set_flags(inst, flags);
 931
 932 out_put:
 933         instance_put(inst);
 934 out:
 935         return ret;
 936 }
 937
 938 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 939         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 940                                     .attr_count = NFULA_MAX, },
 941         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 942                                     .attr_count = NFULA_CFG_MAX,
 943                                     .policy = nfula_cfg_policy },
 944 };
 945
 946 static const struct nfnetlink_subsystem nfulnl_subsys = {
 947         .name           = "log",
 948         .subsys_id      = NFNL_SUBSYS_ULOG,
 949         .cb_count       = NFULNL_MSG_MAX,
 950         .cb             = nfulnl_cb,
 951 };
 952
 953 #ifdef CONFIG_PROC_FS
 954 struct iter_state {
 955         struct seq_net_private p;
 956         unsigned int bucket;
 957 };
 958
 959 static struct hlist_node *get_first(struct net *net, struct iter_state *st)
 960 {
 961         struct nfnl_log_net *log;
 962         if (!st)
 963                 return NULL;
 964
 965         log = nfnl_log_pernet(net);
 966
 967         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 968                 struct hlist_head *head = &log->instance_table[st->bucket];
 969
 970                 if (!hlist_empty(head))
 971                         return rcu_dereference_bh(hlist_first_rcu(head));
 972         }
 973         return NULL;
 974 }
 975
 976 static struct hlist_node *get_next(struct net *net, struct iter_state *st,
 977                                    struct hlist_node *h)
 978 {
 979         h = rcu_dereference_bh(hlist_next_rcu(h));
 980         while (!h) {
 981                 struct nfnl_log_net *log;
 982                 struct hlist_head *head;
 983
 984                 if (++st->bucket >= INSTANCE_BUCKETS)
 985                         return NULL;
 986
 987                 log = nfnl_log_pernet(net);
 988                 head = &log->instance_table[st->bucket];
 989                 h = rcu_dereference_bh(hlist_first_rcu(head));
 990         }
 991         return h;
 992 }
 993
 994 static struct hlist_node *get_idx(struct net *net, struct iter_state *st,
 995                                   loff_t pos)
 996 {
 997         struct hlist_node *head;
 998         head = get_first(net, st);
 999
1000         if (head)
1001                 while (pos && (head = get_next(net, st, head)))
1002                         pos--;
1003         return pos ? NULL : head;
1004 }
1005
1006 static void *seq_start(struct seq_file *s, loff_t *pos)
1007         __acquires(rcu_bh)
1008 {
1009         rcu_read_lock_bh();
1010         return get_idx(seq_file_net(s), s->private, *pos);
1011 }
1012
1013 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1014 {
1015         (*pos)++;
1016         return get_next(seq_file_net(s), s->private, v);
1017 }
1018
1019 static void seq_stop(struct seq_file *s, void *v)
1020         __releases(rcu_bh)
1021 {
1022         rcu_read_unlock_bh();
1023 }
1024
1025 static int seq_show(struct seq_file *s, void *v)
1026 {
1027         const struct nfulnl_instance *inst = v;
1028
1029         seq_printf(s, "%5u %6u %5u %1u %5u %6u %2u\n",
1030                    inst->group_num,
1031                    inst->peer_portid, inst->qlen,
1032                    inst->copy_mode, inst->copy_range,
1033                    inst->flushtimeout, refcount_read(&inst->use));
1034
1035         return 0;
1036 }
1037
1038 static const struct seq_operations nful_seq_ops = {
1039         .start  = seq_start,
1040         .next   = seq_next,
1041         .stop   = seq_stop,
1042         .show   = seq_show,
1043 };
1044 #endif /* PROC_FS */
1045
1046 static int __net_init nfnl_log_net_init(struct net *net)
1047 {
1048         unsigned int i;
1049         struct nfnl_log_net *log = nfnl_log_pernet(net);
1050 #ifdef CONFIG_PROC_FS
1051         struct proc_dir_entry *proc;
1052         kuid_t root_uid;
1053         kgid_t root_gid;
1054 #endif
1055
1056         for (i = 0; i < INSTANCE_BUCKETS; i++)
1057                 INIT_HLIST_HEAD(&log->instance_table[i]);
1058         spin_lock_init(&log->instances_lock);
1059
1060 #ifdef CONFIG_PROC_FS
1061         proc = proc_create_net("nfnetlink_log", 0440, net->nf.proc_netfilter,
1062                         &nful_seq_ops, sizeof(struct iter_state));
1063         if (!proc)
1064                 return -ENOMEM;
1065
1066         root_uid = make_kuid(net->user_ns, 0);
1067         root_gid = make_kgid(net->user_ns, 0);
1068         if (uid_valid(root_uid) && gid_valid(root_gid))
1069                 proc_set_user(proc, root_uid, root_gid);
1070 #endif
1071         return 0;
1072 }
1073
1074 static void __net_exit nfnl_log_net_exit(struct net *net)
1075 {
1076         struct nfnl_log_net *log = nfnl_log_pernet(net);
1077         unsigned int i;
1078
1079 #ifdef CONFIG_PROC_FS
1080         remove_proc_entry("nfnetlink_log", net->nf.proc_netfilter);
1081 #endif
1082         nf_log_unset(net, &nfulnl_logger);
1083         for (i = 0; i < INSTANCE_BUCKETS; i++)
1084                 WARN_ON_ONCE(!hlist_empty(&log->instance_table[i]));
1085 }
1086
1087 static struct pernet_operations nfnl_log_net_ops = {
1088         .init   = nfnl_log_net_init,
1089         .exit   = nfnl_log_net_exit,
1090         .id     = &nfnl_log_net_id,
1091         .size   = sizeof(struct nfnl_log_net),
1092 };
1093
1094 static int __init nfnetlink_log_init(void)
1095 {
1096         int status;
1097
1098         status = register_pernet_subsys(&nfnl_log_net_ops);
1099         if (status < 0) {
1100                 pr_err("failed to register pernet ops\n");
1101                 goto out;
1102         }
1103
1104         netlink_register_notifier(&nfulnl_rtnl_notifier);
1105         status = nfnetlink_subsys_register(&nfulnl_subsys);
1106         if (status < 0) {
1107                 pr_err("failed to create netlink socket\n");
1108                 goto cleanup_netlink_notifier;
1109         }
1110
1111         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
1112         if (status < 0) {
1113                 pr_err("failed to register logger\n");
1114                 goto cleanup_subsys;
1115         }
1116
1117         return status;
1118
1119 cleanup_subsys:
1120         nfnetlink_subsys_unregister(&nfulnl_subsys);
1121 cleanup_netlink_notifier:
1122         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1123         unregister_pernet_subsys(&nfnl_log_net_ops);
1124 out:
1125         return status;
1126 }
1127
1128 static void __exit nfnetlink_log_fini(void)
1129 {
1130         nfnetlink_subsys_unregister(&nfulnl_subsys);
1131         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1132         unregister_pernet_subsys(&nfnl_log_net_ops);
1133         nf_log_unregister(&nfulnl_logger);
1134 }
1135
1136 MODULE_DESCRIPTION("netfilter userspace logging");
1137 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1138 MODULE_LICENSE("GPL");
1139 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1140 MODULE_ALIAS_NF_LOGGER(AF_INET, 1);
1141 MODULE_ALIAS_NF_LOGGER(AF_INET6, 1);
1142 MODULE_ALIAS_NF_LOGGER(AF_BRIDGE, 1);
1143 MODULE_ALIAS_NF_LOGGER(3, 1); /* NFPROTO_ARP */
1144 MODULE_ALIAS_NF_LOGGER(5, 1); /* NFPROTO_NETDEV */
1145
1146 module_init(nfnetlink_log_init);
1147 module_exit(nfnetlink_log_fini);