net/netfilter/nfnetlink_log.c

   1 /*
   2  * This is a module which is used for logging packets to userspace via
   3  * nfetlink.
   4  *
   5  * (C) 2005 by Harald Welte <laforge@netfilter.org>
   6  * (C) 2006-2012 Patrick McHardy <kaber@trash.net>
   7  *
   8  * Based on the old ipv4-only ipt_ULOG.c:
   9  * (C) 2000-2004 by Harald Welte <laforge@netfilter.org>
  10  *
  11  * This program is free software; you can redistribute it and/or modify
  12  * it under the terms of the GNU General Public License version 2 as
  13  * published by the Free Software Foundation.
  14  */
  15
  16 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  17
  18 #include <linux/module.h>
  19 #include <linux/skbuff.h>
  20 #include <linux/if_arp.h>
  21 #include <linux/init.h>
  22 #include <linux/ip.h>
  23 #include <linux/ipv6.h>
  24 #include <linux/netdevice.h>
  25 #include <linux/netfilter.h>
  26 #include <linux/netfilter_bridge.h>
  27 #include <net/netlink.h>
  28 #include <linux/netfilter/nfnetlink.h>
  29 #include <linux/netfilter/nfnetlink_log.h>
  30 #include <linux/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 #include <net/netfilter/nfnetlink_log.h>
  41
  42 #include <linux/atomic.h>
  43 #include <linux/refcount.h>
  44
  45
  46 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
  47 #include "../bridge/br_private.h"
  48 #endif
  49
  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(unsigned long data);
 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         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
 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                           MSG_DONTWAIT);
 364 out:
 365         inst->qlen = 0;
 366         inst->skb = NULL;
 367 }
 368
 369 static void
 370 __nfulnl_flush(struct nfulnl_instance *inst)
 371 {
 372         /* timer holds a reference */
 373         if (del_timer(&inst->timer))
 374                 instance_put(inst);
 375         if (inst->skb)
 376                 __nfulnl_send(inst);
 377 }
 378
 379 static void
 380 nfulnl_timer(unsigned long data)
 381 {
 382         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 383
 384         spin_lock_bh(&inst->lock);
 385         if (inst->skb)
 386                 __nfulnl_send(inst);
 387         spin_unlock_bh(&inst->lock);
 388         instance_put(inst);
 389 }
 390
 391 /* This is an inline function, we don't really care about a long
 392  * list of arguments */
 393 static inline int
 394 __build_packet_message(struct nfnl_log_net *log,
 395                         struct nfulnl_instance *inst,
 396                         const struct sk_buff *skb,
 397                         unsigned int data_len,
 398                         u_int8_t pf,
 399                         unsigned int hooknum,
 400                         const struct net_device *indev,
 401                         const struct net_device *outdev,
 402                         const char *prefix, unsigned int plen,
 403                         const struct nfnl_ct_hook *nfnl_ct,
 404                         struct nf_conn *ct, enum ip_conntrack_info ctinfo)
 405 {
 406         struct nfulnl_msg_packet_hdr pmsg;
 407         struct nlmsghdr *nlh;
 408         struct nfgenmsg *nfmsg;
 409         sk_buff_data_t old_tail = inst->skb->tail;
 410         struct sock *sk;
 411         const unsigned char *hwhdrp;
 412
 413         nlh = nlmsg_put(inst->skb, 0, 0,
 414                         nfnl_msg_type(NFNL_SUBSYS_ULOG, NFULNL_MSG_PACKET),
 415                         sizeof(struct nfgenmsg), 0);
 416         if (!nlh)
 417                 return -1;
 418         nfmsg = nlmsg_data(nlh);
 419         nfmsg->nfgen_family = pf;
 420         nfmsg->version = NFNETLINK_V0;
 421         nfmsg->res_id = htons(inst->group_num);
 422
 423         memset(&pmsg, 0, sizeof(pmsg));
 424         pmsg.hw_protocol        = skb->protocol;
 425         pmsg.hook               = hooknum;
 426
 427         if (nla_put(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg))
 428                 goto nla_put_failure;
 429
 430         if (prefix &&
 431             nla_put(inst->skb, NFULA_PREFIX, plen, prefix))
 432                 goto nla_put_failure;
 433
 434         if (indev) {
 435 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 436                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 437                                  htonl(indev->ifindex)))
 438                         goto nla_put_failure;
 439 #else
 440                 if (pf == PF_BRIDGE) {
 441                         /* Case 1: outdev is physical input device, we need to
 442                          * look for bridge group (when called from
 443                          * netfilter_bridge) */
 444                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 445                                          htonl(indev->ifindex)) ||
 446                         /* this is the bridge group "brX" */
 447                         /* rcu_read_lock()ed by nf_hook_thresh or
 448                          * nf_log_packet.
 449                          */
 450                             nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 451                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 452                                 goto nla_put_failure;
 453                 } else {
 454                         struct net_device *physindev;
 455
 456                         /* Case 2: indev is bridge group, we need to look for
 457                          * physical device (when called from ipv4) */
 458                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 459                                          htonl(indev->ifindex)))
 460                                 goto nla_put_failure;
 461
 462                         physindev = nf_bridge_get_physindev(skb);
 463                         if (physindev &&
 464                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 465                                          htonl(physindev->ifindex)))
 466                                 goto nla_put_failure;
 467                 }
 468 #endif
 469         }
 470
 471         if (outdev) {
 472 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 473                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 474                                  htonl(outdev->ifindex)))
 475                         goto nla_put_failure;
 476 #else
 477                 if (pf == PF_BRIDGE) {
 478                         /* Case 1: outdev is physical output device, we need to
 479                          * look for bridge group (when called from
 480                          * netfilter_bridge) */
 481                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 482                                          htonl(outdev->ifindex)) ||
 483                         /* this is the bridge group "brX" */
 484                         /* rcu_read_lock()ed by nf_hook_thresh or
 485                          * nf_log_packet.
 486                          */
 487                             nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 488                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 489                                 goto nla_put_failure;
 490                 } else {
 491                         struct net_device *physoutdev;
 492
 493                         /* Case 2: indev is a bridge group, we need to look
 494                          * for physical device (when called from ipv4) */
 495                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 496                                          htonl(outdev->ifindex)))
 497                                 goto nla_put_failure;
 498
 499                         physoutdev = nf_bridge_get_physoutdev(skb);
 500                         if (physoutdev &&
 501                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 502                                          htonl(physoutdev->ifindex)))
 503                                 goto nla_put_failure;
 504                 }
 505 #endif
 506         }
 507
 508         if (skb->mark &&
 509             nla_put_be32(inst->skb, NFULA_MARK, htonl(skb->mark)))
 510                 goto nla_put_failure;
 511
 512         if (indev && skb->dev &&
 513             skb->mac_header != skb->network_header) {
 514                 struct nfulnl_msg_packet_hw phw;
 515                 int len;
 516
 517                 memset(&phw, 0, sizeof(phw));
 518                 len = dev_parse_header(skb, phw.hw_addr);
 519                 if (len > 0) {
 520                         phw.hw_addrlen = htons(len);
 521                         if (nla_put(inst->skb, NFULA_HWADDR, sizeof(phw), &phw))
 522                                 goto nla_put_failure;
 523                 }
 524         }
 525
 526         if (indev && skb_mac_header_was_set(skb)) {
 527                 if (nla_put_be16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type)) ||
 528                     nla_put_be16(inst->skb, NFULA_HWLEN,
 529                                  htons(skb->dev->hard_header_len)))
 530                         goto nla_put_failure;
 531
 532                 hwhdrp = skb_mac_header(skb);
 533
 534                 if (skb->dev->type == ARPHRD_SIT)
 535                         hwhdrp -= ETH_HLEN;
 536
 537                 if (hwhdrp >= skb->head &&
 538                     nla_put(inst->skb, NFULA_HWHEADER,
 539                             skb->dev->hard_header_len, hwhdrp))
 540                         goto nla_put_failure;
 541         }
 542
 543         if (skb->tstamp) {
 544                 struct nfulnl_msg_packet_timestamp ts;
 545                 struct timespec64 kts = ktime_to_timespec64(skb->tstamp);
 546                 ts.sec = cpu_to_be64(kts.tv_sec);
 547                 ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
 548
 549                 if (nla_put(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts))
 550                         goto nla_put_failure;
 551         }
 552
 553         /* UID */
 554         sk = skb->sk;
 555         if (sk && sk_fullsock(sk)) {
 556                 read_lock_bh(&sk->sk_callback_lock);
 557                 if (sk->sk_socket && sk->sk_socket->file) {
 558                         struct file *file = sk->sk_socket->file;
 559                         const struct cred *cred = file->f_cred;
 560                         struct user_namespace *user_ns = inst->peer_user_ns;
 561                         __be32 uid = htonl(from_kuid_munged(user_ns, cred->fsuid));
 562                         __be32 gid = htonl(from_kgid_munged(user_ns, cred->fsgid));
 563                         read_unlock_bh(&sk->sk_callback_lock);
 564                         if (nla_put_be32(inst->skb, NFULA_UID, uid) ||
 565                             nla_put_be32(inst->skb, NFULA_GID, gid))
 566                                 goto nla_put_failure;
 567                 } else
 568                         read_unlock_bh(&sk->sk_callback_lock);
 569         }
 570
 571         /* local sequence number */
 572         if ((inst->flags & NFULNL_CFG_F_SEQ) &&
 573             nla_put_be32(inst->skb, NFULA_SEQ, htonl(inst->seq++)))
 574                 goto nla_put_failure;
 575
 576         /* global sequence number */
 577         if ((inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) &&
 578             nla_put_be32(inst->skb, NFULA_SEQ_GLOBAL,
 579                          htonl(atomic_inc_return(&log->global_seq))))
 580                 goto nla_put_failure;
 581
 582         if (ct && nfnl_ct->build(inst->skb, ct, ctinfo,
 583                                  NFULA_CT, NFULA_CT_INFO) < 0)
 584                 goto nla_put_failure;
 585
 586         if (data_len) {
 587                 struct nlattr *nla;
 588                 int size = nla_attr_size(data_len);
 589
 590                 if (skb_tailroom(inst->skb) < nla_total_size(data_len))
 591                         goto nla_put_failure;
 592
 593                 nla = skb_put(inst->skb, nla_total_size(data_len));
 594                 nla->nla_type = NFULA_PAYLOAD;
 595                 nla->nla_len = size;
 596
 597                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 598                         BUG();
 599         }
 600
 601         nlh->nlmsg_len = inst->skb->tail - old_tail;
 602         return 0;
 603
 604 nla_put_failure:
 605         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 606         return -1;
 607 }
 608
 609 static struct nf_loginfo default_loginfo = {
 610         .type =         NF_LOG_TYPE_ULOG,
 611         .u = {
 612                 .ulog = {
 613                         .copy_len       = 0xffff,
 614                         .group          = 0,
 615                         .qthreshold     = 1,
 616                 },
 617         },
 618 };
 619
 620 /* log handler for internal netfilter logging api */
 621 void
 622 nfulnl_log_packet(struct net *net,
 623                   u_int8_t pf,
 624                   unsigned int hooknum,
 625                   const struct sk_buff *skb,
 626                   const struct net_device *in,
 627                   const struct net_device *out,
 628                   const struct nf_loginfo *li_user,
 629                   const char *prefix)
 630 {
 631         size_t size;
 632         unsigned int data_len;
 633         struct nfulnl_instance *inst;
 634         const struct nf_loginfo *li;
 635         unsigned int qthreshold;
 636         unsigned int plen;
 637         struct nfnl_log_net *log = nfnl_log_pernet(net);
 638         const struct nfnl_ct_hook *nfnl_ct = NULL;
 639         struct nf_conn *ct = NULL;
 640         enum ip_conntrack_info uninitialized_var(ctinfo);
 641
 642         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 643                 li = li_user;
 644         else
 645                 li = &default_loginfo;
 646
 647         inst = instance_lookup_get(log, li->u.ulog.group);
 648         if (!inst)
 649                 return;
 650
 651         plen = 0;
 652         if (prefix)
 653                 plen = strlen(prefix) + 1;
 654
 655         /* FIXME: do we want to make the size calculation conditional based on
 656          * what is actually present?  way more branches and checks, but more
 657          * memory efficient... */
 658         size =    nlmsg_total_size(sizeof(struct nfgenmsg))
 659                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 660                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 661                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 662 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 663                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 664                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 665 #endif
 666                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 667                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 668                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 669                 + nla_total_size(plen)                  /* prefix */
 670                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 671                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp))
 672                 + nla_total_size(sizeof(struct nfgenmsg));      /* NLMSG_DONE */
 673
 674         if (in && skb_mac_header_was_set(skb)) {
 675                 size +=   nla_total_size(skb->dev->hard_header_len)
 676                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 677                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 678         }
 679
 680         spin_lock_bh(&inst->lock);
 681
 682         if (inst->flags & NFULNL_CFG_F_SEQ)
 683                 size += nla_total_size(sizeof(u_int32_t));
 684         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 685                 size += nla_total_size(sizeof(u_int32_t));
 686         if (inst->flags & NFULNL_CFG_F_CONNTRACK) {
 687                 nfnl_ct = rcu_dereference(nfnl_ct_hook);
 688                 if (nfnl_ct != NULL) {
 689                         ct = nfnl_ct->get_ct(skb, &ctinfo);
 690                         if (ct != NULL)
 691                                 size += nfnl_ct->build_size(ct);
 692                 }
 693         }
 694
 695         qthreshold = inst->qthreshold;
 696         /* per-rule qthreshold overrides per-instance */
 697         if (li->u.ulog.qthreshold)
 698                 if (qthreshold > li->u.ulog.qthreshold)
 699                         qthreshold = li->u.ulog.qthreshold;
 700
 701
 702         switch (inst->copy_mode) {
 703         case NFULNL_COPY_META:
 704         case NFULNL_COPY_NONE:
 705                 data_len = 0;
 706                 break;
 707
 708         case NFULNL_COPY_PACKET:
 709                 data_len = inst->copy_range;
 710                 if ((li->u.ulog.flags & NF_LOG_F_COPY_LEN) &&
 711                     (li->u.ulog.copy_len < data_len))
 712                         data_len = li->u.ulog.copy_len;
 713
 714                 if (data_len > skb->len)
 715                         data_len = skb->len;
 716
 717                 size += nla_total_size(data_len);
 718                 break;
 719
 720         case NFULNL_COPY_DISABLED:
 721         default:
 722                 goto unlock_and_release;
 723         }
 724
 725         if (inst->skb && size > skb_tailroom(inst->skb)) {
 726                 /* either the queue len is too high or we don't have
 727                  * enough room in the skb left. flush to userspace. */
 728                 __nfulnl_flush(inst);
 729         }
 730
 731         if (!inst->skb) {
 732                 inst->skb = nfulnl_alloc_skb(net, inst->peer_portid,
 733                                              inst->nlbufsiz, size);
 734                 if (!inst->skb)
 735                         goto alloc_failure;
 736         }
 737
 738         inst->qlen++;
 739
 740         __build_packet_message(log, inst, skb, data_len, pf,
 741                                 hooknum, in, out, prefix, plen,
 742                                 nfnl_ct, ct, ctinfo);
 743
 744         if (inst->qlen >= qthreshold)
 745                 __nfulnl_flush(inst);
 746         /* timer_pending always called within inst->lock, so there
 747          * is no chance of a race here */
 748         else if (!timer_pending(&inst->timer)) {
 749                 instance_get(inst);
 750                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 751                 add_timer(&inst->timer);
 752         }
 753
 754 unlock_and_release:
 755         spin_unlock_bh(&inst->lock);
 756         instance_put(inst);
 757         return;
 758
 759 alloc_failure:
 760         /* FIXME: statistics */
 761         goto unlock_and_release;
 762 }
 763 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
 764
 765 static int
 766 nfulnl_rcv_nl_event(struct notifier_block *this,
 767                    unsigned long event, void *ptr)
 768 {
 769         struct netlink_notify *n = ptr;
 770         struct nfnl_log_net *log = nfnl_log_pernet(n->net);
 771
 772         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 773                 int i;
 774
 775                 /* destroy all instances for this portid */
 776                 spin_lock_bh(&log->instances_lock);
 777                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 778                         struct hlist_node *t2;
 779                         struct nfulnl_instance *inst;
 780                         struct hlist_head *head = &log->instance_table[i];
 781
 782                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 783                                 if (n->portid == inst->peer_portid)
 784                                         __instance_destroy(inst);
 785                         }
 786                 }
 787                 spin_unlock_bh(&log->instances_lock);
 788         }
 789         return NOTIFY_DONE;
 790 }
 791
 792 static struct notifier_block nfulnl_rtnl_notifier = {
 793         .notifier_call  = nfulnl_rcv_nl_event,
 794 };
 795
 796 static int nfulnl_recv_unsupp(struct net *net, struct sock *ctnl,
 797                               struct sk_buff *skb, const struct nlmsghdr *nlh,
 798                               const struct nlattr * const nfqa[],
 799                               struct netlink_ext_ack *extack)
 800 {
 801         return -ENOTSUPP;
 802 }
 803
 804 static struct nf_logger nfulnl_logger __read_mostly = {
 805         .name   = "nfnetlink_log",
 806         .type   = NF_LOG_TYPE_ULOG,
 807         .logfn  = nfulnl_log_packet,
 808         .me     = THIS_MODULE,
 809 };
 810
 811 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 812         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 813         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 814         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 815         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 816         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 817         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 818 };
 819
 820 static int nfulnl_recv_config(struct net *net, struct sock *ctnl,
 821                               struct sk_buff *skb, const struct nlmsghdr *nlh,
 822                               const struct nlattr * const nfula[],
 823                               struct netlink_ext_ack *extack)
 824 {
 825         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 826         u_int16_t group_num = ntohs(nfmsg->res_id);
 827         struct nfulnl_instance *inst;
 828         struct nfulnl_msg_config_cmd *cmd = NULL;
 829         struct nfnl_log_net *log = nfnl_log_pernet(net);
 830         int ret = 0;
 831         u16 flags = 0;
 832
 833         if (nfula[NFULA_CFG_CMD]) {
 834                 u_int8_t pf = nfmsg->nfgen_family;
 835                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 836
 837                 /* Commands without queue context */
 838                 switch (cmd->command) {
 839                 case NFULNL_CFG_CMD_PF_BIND:
 840                         return nf_log_bind_pf(net, pf, &nfulnl_logger);
 841                 case NFULNL_CFG_CMD_PF_UNBIND:
 842                         nf_log_unbind_pf(net, pf);
 843                         return 0;
 844                 }
 845         }
 846
 847         inst = instance_lookup_get(log, group_num);
 848         if (inst && inst->peer_portid != NETLINK_CB(skb).portid) {
 849                 ret = -EPERM;
 850                 goto out_put;
 851         }
 852
 853         /* Check if we support these flags in first place, dependencies should
 854          * be there too not to break atomicity.
 855          */
 856         if (nfula[NFULA_CFG_FLAGS]) {
 857                 flags = ntohs(nla_get_be16(nfula[NFULA_CFG_FLAGS]));
 858
 859                 if ((flags & NFULNL_CFG_F_CONNTRACK) &&
 860                     !rcu_access_pointer(nfnl_ct_hook)) {
 861 #ifdef CONFIG_MODULES
 862                         nfnl_unlock(NFNL_SUBSYS_ULOG);
 863                         request_module("ip_conntrack_netlink");
 864                         nfnl_lock(NFNL_SUBSYS_ULOG);
 865                         if (rcu_access_pointer(nfnl_ct_hook)) {
 866                                 ret = -EAGAIN;
 867                                 goto out_put;
 868                         }
 869 #endif
 870                         ret = -EOPNOTSUPP;
 871                         goto out_put;
 872                 }
 873         }
 874
 875         if (cmd != NULL) {
 876                 switch (cmd->command) {
 877                 case NFULNL_CFG_CMD_BIND:
 878                         if (inst) {
 879                                 ret = -EBUSY;
 880                                 goto out_put;
 881                         }
 882
 883                         inst = instance_create(net, group_num,
 884                                                NETLINK_CB(skb).portid,
 885                                                sk_user_ns(NETLINK_CB(skb).sk));
 886                         if (IS_ERR(inst)) {
 887                                 ret = PTR_ERR(inst);
 888                                 goto out;
 889                         }
 890                         break;
 891                 case NFULNL_CFG_CMD_UNBIND:
 892                         if (!inst) {
 893                                 ret = -ENODEV;
 894                                 goto out;
 895                         }
 896
 897                         instance_destroy(log, inst);
 898                         goto out_put;
 899                 default:
 900                         ret = -ENOTSUPP;
 901                         goto out_put;
 902                 }
 903         } else if (!inst) {
 904                 ret = -ENODEV;
 905                 goto out;
 906         }
 907
 908         if (nfula[NFULA_CFG_MODE]) {
 909                 struct nfulnl_msg_config_mode *params =
 910                         nla_data(nfula[NFULA_CFG_MODE]);
 911
 912                 nfulnl_set_mode(inst, params->copy_mode,
 913                                 ntohl(params->copy_range));
 914         }
 915
 916         if (nfula[NFULA_CFG_TIMEOUT]) {
 917                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 918
 919                 nfulnl_set_timeout(inst, ntohl(timeout));
 920         }
 921
 922         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 923                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 924
 925                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 926         }
 927
 928         if (nfula[NFULA_CFG_QTHRESH]) {
 929                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 930
 931                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 932         }
 933
 934         if (nfula[NFULA_CFG_FLAGS])
 935                 nfulnl_set_flags(inst, flags);
 936
 937 out_put:
 938         instance_put(inst);
 939 out:
 940         return ret;
 941 }
 942
 943 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 944         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 945                                     .attr_count = NFULA_MAX, },
 946         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 947                                     .attr_count = NFULA_CFG_MAX,
 948                                     .policy = nfula_cfg_policy },
 949 };
 950
 951 static const struct nfnetlink_subsystem nfulnl_subsys = {
 952         .name           = "log",
 953         .subsys_id      = NFNL_SUBSYS_ULOG,
 954         .cb_count       = NFULNL_MSG_MAX,
 955         .cb             = nfulnl_cb,
 956 };
 957
 958 #ifdef CONFIG_PROC_FS
 959 struct iter_state {
 960         struct seq_net_private p;
 961         unsigned int bucket;
 962 };
 963
 964 static struct hlist_node *get_first(struct net *net, struct iter_state *st)
 965 {
 966         struct nfnl_log_net *log;
 967         if (!st)
 968                 return NULL;
 969
 970         log = nfnl_log_pernet(net);
 971
 972         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 973                 struct hlist_head *head = &log->instance_table[st->bucket];
 974
 975                 if (!hlist_empty(head))
 976                         return rcu_dereference_bh(hlist_first_rcu(head));
 977         }
 978         return NULL;
 979 }
 980
 981 static struct hlist_node *get_next(struct net *net, struct iter_state *st,
 982                                    struct hlist_node *h)
 983 {
 984         h = rcu_dereference_bh(hlist_next_rcu(h));
 985         while (!h) {
 986                 struct nfnl_log_net *log;
 987                 struct hlist_head *head;
 988
 989                 if (++st->bucket >= INSTANCE_BUCKETS)
 990                         return NULL;
 991
 992                 log = nfnl_log_pernet(net);
 993                 head = &log->instance_table[st->bucket];
 994                 h = rcu_dereference_bh(hlist_first_rcu(head));
 995         }
 996         return h;
 997 }
 998
 999 static struct hlist_node *get_idx(struct net *net, struct iter_state *st,
1000                                   loff_t pos)
1001 {
1002         struct hlist_node *head;
1003         head = get_first(net, st);
1004
1005         if (head)
1006                 while (pos && (head = get_next(net, st, head)))
1007                         pos--;
1008         return pos ? NULL : head;
1009 }
1010
1011 static void *seq_start(struct seq_file *s, loff_t *pos)
1012         __acquires(rcu_bh)
1013 {
1014         rcu_read_lock_bh();
1015         return get_idx(seq_file_net(s), s->private, *pos);
1016 }
1017
1018 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1019 {
1020         (*pos)++;
1021         return get_next(seq_file_net(s), s->private, v);
1022 }
1023
1024 static void seq_stop(struct seq_file *s, void *v)
1025         __releases(rcu_bh)
1026 {
1027         rcu_read_unlock_bh();
1028 }
1029
1030 static int seq_show(struct seq_file *s, void *v)
1031 {
1032         const struct nfulnl_instance *inst = v;
1033
1034         seq_printf(s, "%5u %6u %5u %1u %5u %6u %2u\n",
1035                    inst->group_num,
1036                    inst->peer_portid, inst->qlen,
1037                    inst->copy_mode, inst->copy_range,
1038                    inst->flushtimeout, refcount_read(&inst->use));
1039
1040         return 0;
1041 }
1042
1043 static const struct seq_operations nful_seq_ops = {
1044         .start  = seq_start,
1045         .next   = seq_next,
1046         .stop   = seq_stop,
1047         .show   = seq_show,
1048 };
1049
1050 static int nful_open(struct inode *inode, struct file *file)
1051 {
1052         return seq_open_net(inode, file, &nful_seq_ops,
1053                             sizeof(struct iter_state));
1054 }
1055
1056 static const struct file_operations nful_file_ops = {
1057         .owner   = THIS_MODULE,
1058         .open    = nful_open,
1059         .read    = seq_read,
1060         .llseek  = seq_lseek,
1061         .release = seq_release_net,
1062 };
1063
1064 #endif /* PROC_FS */
1065
1066 static int __net_init nfnl_log_net_init(struct net *net)
1067 {
1068         unsigned int i;
1069         struct nfnl_log_net *log = nfnl_log_pernet(net);
1070 #ifdef CONFIG_PROC_FS
1071         struct proc_dir_entry *proc;
1072         kuid_t root_uid;
1073         kgid_t root_gid;
1074 #endif
1075
1076         for (i = 0; i < INSTANCE_BUCKETS; i++)
1077                 INIT_HLIST_HEAD(&log->instance_table[i]);
1078         spin_lock_init(&log->instances_lock);
1079
1080 #ifdef CONFIG_PROC_FS
1081         proc = proc_create("nfnetlink_log", 0440,
1082                            net->nf.proc_netfilter, &nful_file_ops);
1083         if (!proc)
1084                 return -ENOMEM;
1085
1086         root_uid = make_kuid(net->user_ns, 0);
1087         root_gid = make_kgid(net->user_ns, 0);
1088         if (uid_valid(root_uid) && gid_valid(root_gid))
1089                 proc_set_user(proc, root_uid, root_gid);
1090 #endif
1091         return 0;
1092 }
1093
1094 static void __net_exit nfnl_log_net_exit(struct net *net)
1095 {
1096 #ifdef CONFIG_PROC_FS
1097         remove_proc_entry("nfnetlink_log", net->nf.proc_netfilter);
1098 #endif
1099         nf_log_unset(net, &nfulnl_logger);
1100 }
1101
1102 static struct pernet_operations nfnl_log_net_ops = {
1103         .init   = nfnl_log_net_init,
1104         .exit   = nfnl_log_net_exit,
1105         .id     = &nfnl_log_net_id,
1106         .size   = sizeof(struct nfnl_log_net),
1107 };
1108
1109 static int __init nfnetlink_log_init(void)
1110 {
1111         int status;
1112
1113         status = register_pernet_subsys(&nfnl_log_net_ops);
1114         if (status < 0) {
1115                 pr_err("failed to register pernet ops\n");
1116                 goto out;
1117         }
1118
1119         netlink_register_notifier(&nfulnl_rtnl_notifier);
1120         status = nfnetlink_subsys_register(&nfulnl_subsys);
1121         if (status < 0) {
1122                 pr_err("failed to create netlink socket\n");
1123                 goto cleanup_netlink_notifier;
1124         }
1125
1126         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
1127         if (status < 0) {
1128                 pr_err("failed to register logger\n");
1129                 goto cleanup_subsys;
1130         }
1131
1132         return status;
1133
1134 cleanup_subsys:
1135         nfnetlink_subsys_unregister(&nfulnl_subsys);
1136 cleanup_netlink_notifier:
1137         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1138         unregister_pernet_subsys(&nfnl_log_net_ops);
1139 out:
1140         return status;
1141 }
1142
1143 static void __exit nfnetlink_log_fini(void)
1144 {
1145         nfnetlink_subsys_unregister(&nfulnl_subsys);
1146         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1147         unregister_pernet_subsys(&nfnl_log_net_ops);
1148         nf_log_unregister(&nfulnl_logger);
1149 }
1150
1151 MODULE_DESCRIPTION("netfilter userspace logging");
1152 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1153 MODULE_LICENSE("GPL");
1154 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1155 MODULE_ALIAS_NF_LOGGER(AF_INET, 1);
1156 MODULE_ALIAS_NF_LOGGER(AF_INET6, 1);
1157 MODULE_ALIAS_NF_LOGGER(AF_BRIDGE, 1);
1158 MODULE_ALIAS_NF_LOGGER(3, 1); /* NFPROTO_ARP */
1159 MODULE_ALIAS_NF_LOGGER(5, 1); /* NFPROTO_NETDEV */
1160
1161 module_init(nfnetlink_log_init);
1162 module_exit(nfnetlink_log_fini);