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 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
  47 #include <net/netfilter/nf_conntrack.h>
  48 #endif
  49
  50 #define NFULNL_COPY_DISABLED    0xff
  51 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
  52 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
  53 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  54 /* max packet size is limited by 16-bit struct nfattr nfa_len field */
  55 #define NFULNL_COPY_RANGE_MAX   (0xFFFF - NLA_HDRLEN)
  56
  57 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  58                                      printk(x, ## args); } while (0);
  59
  60 struct nfulnl_instance {
  61         struct hlist_node hlist;        /* global list of instances */
  62         spinlock_t lock;
  63         refcount_t use;                 /* use count */
  64
  65         unsigned int qlen;              /* number of nlmsgs in skb */
  66         struct sk_buff *skb;            /* pre-allocatd skb */
  67         struct timer_list timer;
  68         struct net *net;
  69         struct user_namespace *peer_user_ns;    /* User namespace of the peer process */
  70         u32 peer_portid;                /* PORTID of the peer process */
  71
  72         /* configurable parameters */
  73         unsigned int flushtimeout;      /* timeout until queue flush */
  74         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  75         unsigned int qthreshold;        /* threshold of the queue */
  76         u_int32_t copy_range;
  77         u_int32_t seq;                  /* instance-local sequential counter */
  78         u_int16_t group_num;            /* number of this queue */
  79         u_int16_t flags;
  80         u_int8_t copy_mode;
  81         struct rcu_head rcu;
  82 };
  83
  84 #define INSTANCE_BUCKETS        16
  85
  86 static unsigned int nfnl_log_net_id __read_mostly;
  87
  88 struct nfnl_log_net {
  89         spinlock_t instances_lock;
  90         struct hlist_head instance_table[INSTANCE_BUCKETS];
  91         atomic_t global_seq;
  92 };
  93
  94 static struct nfnl_log_net *nfnl_log_pernet(struct net *net)
  95 {
  96         return net_generic(net, nfnl_log_net_id);
  97 }
  98
  99 static inline u_int8_t instance_hashfn(u_int16_t group_num)
 100 {
 101         return ((group_num & 0xff) % INSTANCE_BUCKETS);
 102 }
 103
 104 static struct nfulnl_instance *
 105 __instance_lookup(struct nfnl_log_net *log, u_int16_t group_num)
 106 {
 107         struct hlist_head *head;
 108         struct nfulnl_instance *inst;
 109
 110         head = &log->instance_table[instance_hashfn(group_num)];
 111         hlist_for_each_entry_rcu(inst, head, hlist) {
 112                 if (inst->group_num == group_num)
 113                         return inst;
 114         }
 115         return NULL;
 116 }
 117
 118 static inline void
 119 instance_get(struct nfulnl_instance *inst)
 120 {
 121         refcount_inc(&inst->use);
 122 }
 123
 124 static struct nfulnl_instance *
 125 instance_lookup_get(struct nfnl_log_net *log, u_int16_t group_num)
 126 {
 127         struct nfulnl_instance *inst;
 128
 129         rcu_read_lock_bh();
 130         inst = __instance_lookup(log, group_num);
 131         if (inst && !refcount_inc_not_zero(&inst->use))
 132                 inst = NULL;
 133         rcu_read_unlock_bh();
 134
 135         return inst;
 136 }
 137
 138 static void nfulnl_instance_free_rcu(struct rcu_head *head)
 139 {
 140         struct nfulnl_instance *inst =
 141                 container_of(head, struct nfulnl_instance, rcu);
 142
 143         put_net(inst->net);
 144         kfree(inst);
 145         module_put(THIS_MODULE);
 146 }
 147
 148 static void
 149 instance_put(struct nfulnl_instance *inst)
 150 {
 151         if (inst && refcount_dec_and_test(&inst->use))
 152                 call_rcu(&inst->rcu, nfulnl_instance_free_rcu);
 153 }
 154
 155 static void nfulnl_timer(struct timer_list *t);
 156
 157 static struct nfulnl_instance *
 158 instance_create(struct net *net, u_int16_t group_num,
 159                 u32 portid, struct user_namespace *user_ns)
 160 {
 161         struct nfulnl_instance *inst;
 162         struct nfnl_log_net *log = nfnl_log_pernet(net);
 163         int err;
 164
 165         spin_lock_bh(&log->instances_lock);
 166         if (__instance_lookup(log, group_num)) {
 167                 err = -EEXIST;
 168                 goto out_unlock;
 169         }
 170
 171         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 172         if (!inst) {
 173                 err = -ENOMEM;
 174                 goto out_unlock;
 175         }
 176
 177         if (!try_module_get(THIS_MODULE)) {
 178                 kfree(inst);
 179                 err = -EAGAIN;
 180                 goto out_unlock;
 181         }
 182
 183         INIT_HLIST_NODE(&inst->hlist);
 184         spin_lock_init(&inst->lock);
 185         /* needs to be two, since we _put() after creation */
 186         refcount_set(&inst->use, 2);
 187
 188         timer_setup(&inst->timer, nfulnl_timer, 0);
 189
 190         inst->net = get_net(net);
 191         inst->peer_user_ns = user_ns;
 192         inst->peer_portid = portid;
 193         inst->group_num = group_num;
 194
 195         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 196         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 197         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 198         inst->copy_mode         = NFULNL_COPY_PACKET;
 199         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
 200
 201         hlist_add_head_rcu(&inst->hlist,
 202                        &log->instance_table[instance_hashfn(group_num)]);
 203
 204
 205         spin_unlock_bh(&log->instances_lock);
 206
 207         return inst;
 208
 209 out_unlock:
 210         spin_unlock_bh(&log->instances_lock);
 211         return ERR_PTR(err);
 212 }
 213
 214 static void __nfulnl_flush(struct nfulnl_instance *inst);
 215
 216 /* called with BH disabled */
 217 static void
 218 __instance_destroy(struct nfulnl_instance *inst)
 219 {
 220         /* first pull it out of the global list */
 221         hlist_del_rcu(&inst->hlist);
 222
 223         /* then flush all pending packets from skb */
 224
 225         spin_lock(&inst->lock);
 226
 227         /* lockless readers wont be able to use us */
 228         inst->copy_mode = NFULNL_COPY_DISABLED;
 229
 230         if (inst->skb)
 231                 __nfulnl_flush(inst);
 232         spin_unlock(&inst->lock);
 233
 234         /* and finally put the refcount */
 235         instance_put(inst);
 236 }
 237
 238 static inline void
 239 instance_destroy(struct nfnl_log_net *log,
 240                  struct nfulnl_instance *inst)
 241 {
 242         spin_lock_bh(&log->instances_lock);
 243         __instance_destroy(inst);
 244         spin_unlock_bh(&log->instances_lock);
 245 }
 246
 247 static int
 248 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 249                   unsigned int range)
 250 {
 251         int status = 0;
 252
 253         spin_lock_bh(&inst->lock);
 254
 255         switch (mode) {
 256         case NFULNL_COPY_NONE:
 257         case NFULNL_COPY_META:
 258                 inst->copy_mode = mode;
 259                 inst->copy_range = 0;
 260                 break;
 261
 262         case NFULNL_COPY_PACKET:
 263                 inst->copy_mode = mode;
 264                 if (range == 0)
 265                         range = NFULNL_COPY_RANGE_MAX;
 266                 inst->copy_range = min_t(unsigned int,
 267                                          range, NFULNL_COPY_RANGE_MAX);
 268                 break;
 269
 270         default:
 271                 status = -EINVAL;
 272                 break;
 273         }
 274
 275         spin_unlock_bh(&inst->lock);
 276
 277         return status;
 278 }
 279
 280 static int
 281 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 282 {
 283         int status;
 284
 285         spin_lock_bh(&inst->lock);
 286         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 287                 status = -ERANGE;
 288         else if (nlbufsiz > 131072)
 289                 status = -ERANGE;
 290         else {
 291                 inst->nlbufsiz = nlbufsiz;
 292                 status = 0;
 293         }
 294         spin_unlock_bh(&inst->lock);
 295
 296         return status;
 297 }
 298
 299 static void
 300 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 301 {
 302         spin_lock_bh(&inst->lock);
 303         inst->flushtimeout = timeout;
 304         spin_unlock_bh(&inst->lock);
 305 }
 306
 307 static void
 308 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 309 {
 310         spin_lock_bh(&inst->lock);
 311         inst->qthreshold = qthresh;
 312         spin_unlock_bh(&inst->lock);
 313 }
 314
 315 static int
 316 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 317 {
 318         spin_lock_bh(&inst->lock);
 319         inst->flags = flags;
 320         spin_unlock_bh(&inst->lock);
 321
 322         return 0;
 323 }
 324
 325 static struct sk_buff *
 326 nfulnl_alloc_skb(struct net *net, u32 peer_portid, unsigned int inst_size,
 327                  unsigned int pkt_size)
 328 {
 329         struct sk_buff *skb;
 330         unsigned int n;
 331
 332         /* alloc skb which should be big enough for a whole multipart
 333          * message.  WARNING: has to be <= 128k due to slab restrictions */
 334
 335         n = max(inst_size, pkt_size);
 336         skb = alloc_skb(n, GFP_ATOMIC | __GFP_NOWARN);
 337         if (!skb) {
 338                 if (n > pkt_size) {
 339                         /* try to allocate only as much as we need for current
 340                          * packet */
 341
 342                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 343                 }
 344         }
 345
 346         return skb;
 347 }
 348
 349 static void
 350 __nfulnl_send(struct nfulnl_instance *inst)
 351 {
 352         if (inst->qlen > 1) {
 353                 struct nlmsghdr *nlh = nlmsg_put(inst->skb, 0, 0,
 354                                                  NLMSG_DONE,
 355                                                  sizeof(struct nfgenmsg),
 356                                                  0);
 357                 if (WARN_ONCE(!nlh, "bad nlskb size: %u, tailroom %d\n",
 358                               inst->skb->len, skb_tailroom(inst->skb))) {
 359                         kfree_skb(inst->skb);
 360                         goto out;
 361                 }
 362         }
 363         nfnetlink_unicast(inst->skb, inst->net, inst->peer_portid);
 364 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(struct timer_list *t)
 381 {
 382         struct nfulnl_instance *inst = from_timer(inst, t, timer);
 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 static u32 nfulnl_get_bridge_size(const struct sk_buff *skb)
 392 {
 393         u32 size = 0;
 394
 395         if (!skb_mac_header_was_set(skb))
 396                 return 0;
 397
 398         if (skb_vlan_tag_present(skb)) {
 399                 size += nla_total_size(0); /* nested */
 400                 size += nla_total_size(sizeof(u16)); /* id */
 401                 size += nla_total_size(sizeof(u16)); /* tag */
 402         }
 403
 404         if (skb->network_header > skb->mac_header)
 405                 size += nla_total_size(skb->network_header - skb->mac_header);
 406
 407         return size;
 408 }
 409
 410 static int nfulnl_put_bridge(struct nfulnl_instance *inst, const struct sk_buff *skb)
 411 {
 412         if (!skb_mac_header_was_set(skb))
 413                 return 0;
 414
 415         if (skb_vlan_tag_present(skb)) {
 416                 struct nlattr *nest;
 417
 418                 nest = nla_nest_start(inst->skb, NFULA_VLAN);
 419                 if (!nest)
 420                         goto nla_put_failure;
 421
 422                 if (nla_put_be16(inst->skb, NFULA_VLAN_TCI, htons(skb->vlan_tci)) ||
 423                     nla_put_be16(inst->skb, NFULA_VLAN_PROTO, skb->vlan_proto))
 424                         goto nla_put_failure;
 425
 426                 nla_nest_end(inst->skb, nest);
 427         }
 428
 429         if (skb->mac_header < skb->network_header) {
 430                 int len = (int)(skb->network_header - skb->mac_header);
 431
 432                 if (nla_put(inst->skb, NFULA_L2HDR, len, skb_mac_header(skb)))
 433                         goto nla_put_failure;
 434         }
 435
 436         return 0;
 437
 438 nla_put_failure:
 439         return -1;
 440 }
 441
 442 /* This is an inline function, we don't really care about a long
 443  * list of arguments */
 444 static inline int
 445 __build_packet_message(struct nfnl_log_net *log,
 446                         struct nfulnl_instance *inst,
 447                         const struct sk_buff *skb,
 448                         unsigned int data_len,
 449                         u_int8_t pf,
 450                         unsigned int hooknum,
 451                         const struct net_device *indev,
 452                         const struct net_device *outdev,
 453                         const char *prefix, unsigned int plen,
 454                         const struct nfnl_ct_hook *nfnl_ct,
 455                         struct nf_conn *ct, enum ip_conntrack_info ctinfo)
 456 {
 457         struct nfulnl_msg_packet_hdr pmsg;
 458         struct nlmsghdr *nlh;
 459         sk_buff_data_t old_tail = inst->skb->tail;
 460         struct sock *sk;
 461         const unsigned char *hwhdrp;
 462
 463         nlh = nfnl_msg_put(inst->skb, 0, 0,
 464                            nfnl_msg_type(NFNL_SUBSYS_ULOG, NFULNL_MSG_PACKET),
 465                            0, pf, NFNETLINK_V0, htons(inst->group_num));
 466         if (!nlh)
 467                 return -1;
 468
 469         memset(&pmsg, 0, sizeof(pmsg));
 470         pmsg.hw_protocol        = skb->protocol;
 471         pmsg.hook               = hooknum;
 472
 473         if (nla_put(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg))
 474                 goto nla_put_failure;
 475
 476         if (prefix &&
 477             nla_put(inst->skb, NFULA_PREFIX, plen, prefix))
 478                 goto nla_put_failure;
 479
 480         if (indev) {
 481 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 482                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 483                                  htonl(indev->ifindex)))
 484                         goto nla_put_failure;
 485 #else
 486                 if (pf == PF_BRIDGE) {
 487                         /* Case 1: outdev is physical input device, we need to
 488                          * look for bridge group (when called from
 489                          * netfilter_bridge) */
 490                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 491                                          htonl(indev->ifindex)) ||
 492                         /* this is the bridge group "brX" */
 493                         /* rcu_read_lock()ed by nf_hook_thresh or
 494                          * nf_log_packet.
 495                          */
 496                             nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 497                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 498                                 goto nla_put_failure;
 499                 } else {
 500                         struct net_device *physindev;
 501
 502                         /* Case 2: indev is bridge group, we need to look for
 503                          * physical device (when called from ipv4) */
 504                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 505                                          htonl(indev->ifindex)))
 506                                 goto nla_put_failure;
 507
 508                         physindev = nf_bridge_get_physindev(skb);
 509                         if (physindev &&
 510                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 511                                          htonl(physindev->ifindex)))
 512                                 goto nla_put_failure;
 513                 }
 514 #endif
 515         }
 516
 517         if (outdev) {
 518 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 519                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 520                                  htonl(outdev->ifindex)))
 521                         goto nla_put_failure;
 522 #else
 523                 if (pf == PF_BRIDGE) {
 524                         /* Case 1: outdev is physical output device, we need to
 525                          * look for bridge group (when called from
 526                          * netfilter_bridge) */
 527                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 528                                          htonl(outdev->ifindex)) ||
 529                         /* this is the bridge group "brX" */
 530                         /* rcu_read_lock()ed by nf_hook_thresh or
 531                          * nf_log_packet.
 532                          */
 533                             nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 534                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 535                                 goto nla_put_failure;
 536                 } else {
 537                         struct net_device *physoutdev;
 538
 539                         /* Case 2: indev is a bridge group, we need to look
 540                          * for physical device (when called from ipv4) */
 541                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 542                                          htonl(outdev->ifindex)))
 543                                 goto nla_put_failure;
 544
 545                         physoutdev = nf_bridge_get_physoutdev(skb);
 546                         if (physoutdev &&
 547                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 548                                          htonl(physoutdev->ifindex)))
 549                                 goto nla_put_failure;
 550                 }
 551 #endif
 552         }
 553
 554         if (skb->mark &&
 555             nla_put_be32(inst->skb, NFULA_MARK, htonl(skb->mark)))
 556                 goto nla_put_failure;
 557
 558         if (indev && skb->dev &&
 559             skb->mac_header != skb->network_header) {
 560                 struct nfulnl_msg_packet_hw phw;
 561                 int len;
 562
 563                 memset(&phw, 0, sizeof(phw));
 564                 len = dev_parse_header(skb, phw.hw_addr);
 565                 if (len > 0) {
 566                         phw.hw_addrlen = htons(len);
 567                         if (nla_put(inst->skb, NFULA_HWADDR, sizeof(phw), &phw))
 568                                 goto nla_put_failure;
 569                 }
 570         }
 571
 572         if (indev && skb_mac_header_was_set(skb)) {
 573                 if (nla_put_be16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type)) ||
 574                     nla_put_be16(inst->skb, NFULA_HWLEN,
 575                                  htons(skb->dev->hard_header_len)))
 576                         goto nla_put_failure;
 577
 578                 hwhdrp = skb_mac_header(skb);
 579
 580                 if (skb->dev->type == ARPHRD_SIT)
 581                         hwhdrp -= ETH_HLEN;
 582
 583                 if (hwhdrp >= skb->head &&
 584                     nla_put(inst->skb, NFULA_HWHEADER,
 585                             skb->dev->hard_header_len, hwhdrp))
 586                         goto nla_put_failure;
 587         }
 588
 589         if (hooknum <= NF_INET_FORWARD && skb->tstamp) {
 590                 struct nfulnl_msg_packet_timestamp ts;
 591                 struct timespec64 kts = ktime_to_timespec64(skb->tstamp);
 592                 ts.sec = cpu_to_be64(kts.tv_sec);
 593                 ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
 594
 595                 if (nla_put(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts))
 596                         goto nla_put_failure;
 597         }
 598
 599         /* UID */
 600         sk = skb->sk;
 601         if (sk && sk_fullsock(sk)) {
 602                 read_lock_bh(&sk->sk_callback_lock);
 603                 if (sk->sk_socket && sk->sk_socket->file) {
 604                         struct file *file = sk->sk_socket->file;
 605                         const struct cred *cred = file->f_cred;
 606                         struct user_namespace *user_ns = inst->peer_user_ns;
 607                         __be32 uid = htonl(from_kuid_munged(user_ns, cred->fsuid));
 608                         __be32 gid = htonl(from_kgid_munged(user_ns, cred->fsgid));
 609                         read_unlock_bh(&sk->sk_callback_lock);
 610                         if (nla_put_be32(inst->skb, NFULA_UID, uid) ||
 611                             nla_put_be32(inst->skb, NFULA_GID, gid))
 612                                 goto nla_put_failure;
 613                 } else
 614                         read_unlock_bh(&sk->sk_callback_lock);
 615         }
 616
 617         /* local sequence number */
 618         if ((inst->flags & NFULNL_CFG_F_SEQ) &&
 619             nla_put_be32(inst->skb, NFULA_SEQ, htonl(inst->seq++)))
 620                 goto nla_put_failure;
 621
 622         /* global sequence number */
 623         if ((inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) &&
 624             nla_put_be32(inst->skb, NFULA_SEQ_GLOBAL,
 625                          htonl(atomic_inc_return(&log->global_seq))))
 626                 goto nla_put_failure;
 627
 628         if (ct && nfnl_ct->build(inst->skb, ct, ctinfo,
 629                                  NFULA_CT, NFULA_CT_INFO) < 0)
 630                 goto nla_put_failure;
 631
 632         if ((pf == NFPROTO_NETDEV || pf == NFPROTO_BRIDGE) &&
 633             nfulnl_put_bridge(inst, skb) < 0)
 634                 goto nla_put_failure;
 635
 636         if (data_len) {
 637                 struct nlattr *nla;
 638                 int size = nla_attr_size(data_len);
 639
 640                 if (skb_tailroom(inst->skb) < nla_total_size(data_len))
 641                         goto nla_put_failure;
 642
 643                 nla = skb_put(inst->skb, nla_total_size(data_len));
 644                 nla->nla_type = NFULA_PAYLOAD;
 645                 nla->nla_len = size;
 646
 647                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 648                         BUG();
 649         }
 650
 651         nlh->nlmsg_len = inst->skb->tail - old_tail;
 652         return 0;
 653
 654 nla_put_failure:
 655         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 656         return -1;
 657 }
 658
 659 static const struct nf_loginfo default_loginfo = {
 660         .type =         NF_LOG_TYPE_ULOG,
 661         .u = {
 662                 .ulog = {
 663                         .copy_len       = 0xffff,
 664                         .group          = 0,
 665                         .qthreshold     = 1,
 666                 },
 667         },
 668 };
 669
 670 /* log handler for internal netfilter logging api */
 671 static void
 672 nfulnl_log_packet(struct net *net,
 673                   u_int8_t pf,
 674                   unsigned int hooknum,
 675                   const struct sk_buff *skb,
 676                   const struct net_device *in,
 677                   const struct net_device *out,
 678                   const struct nf_loginfo *li_user,
 679                   const char *prefix)
 680 {
 681         size_t size;
 682         unsigned int data_len;
 683         struct nfulnl_instance *inst;
 684         const struct nf_loginfo *li;
 685         unsigned int qthreshold;
 686         unsigned int plen = 0;
 687         struct nfnl_log_net *log = nfnl_log_pernet(net);
 688         const struct nfnl_ct_hook *nfnl_ct = NULL;
 689         struct nf_conn *ct = NULL;
 690         enum ip_conntrack_info ctinfo;
 691
 692         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 693                 li = li_user;
 694         else
 695                 li = &default_loginfo;
 696
 697         inst = instance_lookup_get(log, li->u.ulog.group);
 698         if (!inst)
 699                 return;
 700
 701         if (prefix)
 702                 plen = strlen(prefix) + 1;
 703
 704         /* FIXME: do we want to make the size calculation conditional based on
 705          * what is actually present?  way more branches and checks, but more
 706          * memory efficient... */
 707         size = nlmsg_total_size(sizeof(struct nfgenmsg))
 708                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 709                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 710                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 711 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 712                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 713                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 714 #endif
 715                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 716                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 717                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 718                 + nla_total_size(plen)                  /* prefix */
 719                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 720                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp))
 721                 + nla_total_size(sizeof(struct nfgenmsg));      /* NLMSG_DONE */
 722
 723         if (in && skb_mac_header_was_set(skb)) {
 724                 size += nla_total_size(skb->dev->hard_header_len)
 725                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 726                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 727         }
 728
 729         spin_lock_bh(&inst->lock);
 730
 731         if (inst->flags & NFULNL_CFG_F_SEQ)
 732                 size += nla_total_size(sizeof(u_int32_t));
 733         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 734                 size += nla_total_size(sizeof(u_int32_t));
 735 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 736         if (inst->flags & NFULNL_CFG_F_CONNTRACK) {
 737                 nfnl_ct = rcu_dereference(nfnl_ct_hook);
 738                 if (nfnl_ct != NULL) {
 739                         ct = nf_ct_get(skb, &ctinfo);
 740                         if (ct != NULL)
 741                                 size += nfnl_ct->build_size(ct);
 742                 }
 743         }
 744 #endif
 745         if (pf == NFPROTO_NETDEV || pf == NFPROTO_BRIDGE)
 746                 size += nfulnl_get_bridge_size(skb);
 747
 748         qthreshold = inst->qthreshold;
 749         /* per-rule qthreshold overrides per-instance */
 750         if (li->u.ulog.qthreshold)
 751                 if (qthreshold > li->u.ulog.qthreshold)
 752                         qthreshold = li->u.ulog.qthreshold;
 753
 754
 755         switch (inst->copy_mode) {
 756         case NFULNL_COPY_META:
 757         case NFULNL_COPY_NONE:
 758                 data_len = 0;
 759                 break;
 760
 761         case NFULNL_COPY_PACKET:
 762                 data_len = inst->copy_range;
 763                 if ((li->u.ulog.flags & NF_LOG_F_COPY_LEN) &&
 764                     (li->u.ulog.copy_len < data_len))
 765                         data_len = li->u.ulog.copy_len;
 766
 767                 if (data_len > skb->len)
 768                         data_len = skb->len;
 769
 770                 size += nla_total_size(data_len);
 771                 break;
 772
 773         case NFULNL_COPY_DISABLED:
 774         default:
 775                 goto unlock_and_release;
 776         }
 777
 778         if (inst->skb && size > skb_tailroom(inst->skb)) {
 779                 /* either the queue len is too high or we don't have
 780                  * enough room in the skb left. flush to userspace. */
 781                 __nfulnl_flush(inst);
 782         }
 783
 784         if (!inst->skb) {
 785                 inst->skb = nfulnl_alloc_skb(net, inst->peer_portid,
 786                                              inst->nlbufsiz, size);
 787                 if (!inst->skb)
 788                         goto alloc_failure;
 789         }
 790
 791         inst->qlen++;
 792
 793         __build_packet_message(log, inst, skb, data_len, pf,
 794                                 hooknum, in, out, prefix, plen,
 795                                 nfnl_ct, ct, ctinfo);
 796
 797         if (inst->qlen >= qthreshold)
 798                 __nfulnl_flush(inst);
 799         /* timer_pending always called within inst->lock, so there
 800          * is no chance of a race here */
 801         else if (!timer_pending(&inst->timer)) {
 802                 instance_get(inst);
 803                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 804                 add_timer(&inst->timer);
 805         }
 806
 807 unlock_and_release:
 808         spin_unlock_bh(&inst->lock);
 809         instance_put(inst);
 810         return;
 811
 812 alloc_failure:
 813         /* FIXME: statistics */
 814         goto unlock_and_release;
 815 }
 816
 817 static int
 818 nfulnl_rcv_nl_event(struct notifier_block *this,
 819                    unsigned long event, void *ptr)
 820 {
 821         struct netlink_notify *n = ptr;
 822         struct nfnl_log_net *log = nfnl_log_pernet(n->net);
 823
 824         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 825                 int i;
 826
 827                 /* destroy all instances for this portid */
 828                 spin_lock_bh(&log->instances_lock);
 829                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 830                         struct hlist_node *t2;
 831                         struct nfulnl_instance *inst;
 832                         struct hlist_head *head = &log->instance_table[i];
 833
 834                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 835                                 if (n->portid == inst->peer_portid)
 836                                         __instance_destroy(inst);
 837                         }
 838                 }
 839                 spin_unlock_bh(&log->instances_lock);
 840         }
 841         return NOTIFY_DONE;
 842 }
 843
 844 static struct notifier_block nfulnl_rtnl_notifier = {
 845         .notifier_call  = nfulnl_rcv_nl_event,
 846 };
 847
 848 static int nfulnl_recv_unsupp(struct sk_buff *skb, const struct nfnl_info *info,
 849                               const struct nlattr * const nfula[])
 850 {
 851         return -ENOTSUPP;
 852 }
 853
 854 static struct nf_logger nfulnl_logger __read_mostly = {
 855         .name   = "nfnetlink_log",
 856         .type   = NF_LOG_TYPE_ULOG,
 857         .logfn  = nfulnl_log_packet,
 858         .me     = THIS_MODULE,
 859 };
 860
 861 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 862         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 863         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 864         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 865         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 866         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 867         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 868 };
 869
 870 static int nfulnl_recv_config(struct sk_buff *skb, const struct nfnl_info *info,
 871                               const struct nlattr * const nfula[])
 872 {
 873         struct nfnl_log_net *log = nfnl_log_pernet(info->net);
 874         struct nfgenmsg *nfmsg = nlmsg_data(info->nlh);
 875         u_int16_t group_num = ntohs(nfmsg->res_id);
 876         struct nfulnl_msg_config_cmd *cmd = NULL;
 877         struct nfulnl_instance *inst;
 878         u16 flags = 0;
 879         int ret = 0;
 880
 881         if (nfula[NFULA_CFG_CMD]) {
 882                 u_int8_t pf = nfmsg->nfgen_family;
 883                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 884
 885                 /* Commands without queue context */
 886                 switch (cmd->command) {
 887                 case NFULNL_CFG_CMD_PF_BIND:
 888                         return nf_log_bind_pf(info->net, pf, &nfulnl_logger);
 889                 case NFULNL_CFG_CMD_PF_UNBIND:
 890                         nf_log_unbind_pf(info->net, pf);
 891                         return 0;
 892                 }
 893         }
 894
 895         inst = instance_lookup_get(log, group_num);
 896         if (inst && inst->peer_portid != NETLINK_CB(skb).portid) {
 897                 ret = -EPERM;
 898                 goto out_put;
 899         }
 900
 901         /* Check if we support these flags in first place, dependencies should
 902          * be there too not to break atomicity.
 903          */
 904         if (nfula[NFULA_CFG_FLAGS]) {
 905                 flags = ntohs(nla_get_be16(nfula[NFULA_CFG_FLAGS]));
 906
 907                 if ((flags & NFULNL_CFG_F_CONNTRACK) &&
 908                     !rcu_access_pointer(nfnl_ct_hook)) {
 909 #ifdef CONFIG_MODULES
 910                         nfnl_unlock(NFNL_SUBSYS_ULOG);
 911                         request_module("ip_conntrack_netlink");
 912                         nfnl_lock(NFNL_SUBSYS_ULOG);
 913                         if (rcu_access_pointer(nfnl_ct_hook)) {
 914                                 ret = -EAGAIN;
 915                                 goto out_put;
 916                         }
 917 #endif
 918                         ret = -EOPNOTSUPP;
 919                         goto out_put;
 920                 }
 921         }
 922
 923         if (cmd != NULL) {
 924                 switch (cmd->command) {
 925                 case NFULNL_CFG_CMD_BIND:
 926                         if (inst) {
 927                                 ret = -EBUSY;
 928                                 goto out_put;
 929                         }
 930
 931                         inst = instance_create(info->net, group_num,
 932                                                NETLINK_CB(skb).portid,
 933                                                sk_user_ns(NETLINK_CB(skb).sk));
 934                         if (IS_ERR(inst)) {
 935                                 ret = PTR_ERR(inst);
 936                                 goto out;
 937                         }
 938                         break;
 939                 case NFULNL_CFG_CMD_UNBIND:
 940                         if (!inst) {
 941                                 ret = -ENODEV;
 942                                 goto out;
 943                         }
 944
 945                         instance_destroy(log, inst);
 946                         goto out_put;
 947                 default:
 948                         ret = -ENOTSUPP;
 949                         goto out_put;
 950                 }
 951         } else if (!inst) {
 952                 ret = -ENODEV;
 953                 goto out;
 954         }
 955
 956         if (nfula[NFULA_CFG_MODE]) {
 957                 struct nfulnl_msg_config_mode *params =
 958                         nla_data(nfula[NFULA_CFG_MODE]);
 959
 960                 nfulnl_set_mode(inst, params->copy_mode,
 961                                 ntohl(params->copy_range));
 962         }
 963
 964         if (nfula[NFULA_CFG_TIMEOUT]) {
 965                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 966
 967                 nfulnl_set_timeout(inst, ntohl(timeout));
 968         }
 969
 970         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 971                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 972
 973                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 974         }
 975
 976         if (nfula[NFULA_CFG_QTHRESH]) {
 977                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 978
 979                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 980         }
 981
 982         if (nfula[NFULA_CFG_FLAGS])
 983                 nfulnl_set_flags(inst, flags);
 984
 985 out_put:
 986         instance_put(inst);
 987 out:
 988         return ret;
 989 }
 990
 991 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 992         [NFULNL_MSG_PACKET]     = {
 993                 .call           = nfulnl_recv_unsupp,
 994                 .type           = NFNL_CB_MUTEX,
 995                 .attr_count     = NFULA_MAX,
 996         },
 997         [NFULNL_MSG_CONFIG]     = {
 998                 .call           = nfulnl_recv_config,
 999                 .type           = NFNL_CB_MUTEX,
1000                 .attr_count     = NFULA_CFG_MAX,
1001                 .policy         = nfula_cfg_policy
1002         },
1003 };
1004
1005 static const struct nfnetlink_subsystem nfulnl_subsys = {
1006         .name           = "log",
1007         .subsys_id      = NFNL_SUBSYS_ULOG,
1008         .cb_count       = NFULNL_MSG_MAX,
1009         .cb             = nfulnl_cb,
1010 };
1011
1012 #ifdef CONFIG_PROC_FS
1013 struct iter_state {
1014         struct seq_net_private p;
1015         unsigned int bucket;
1016 };
1017
1018 static struct hlist_node *get_first(struct net *net, struct iter_state *st)
1019 {
1020         struct nfnl_log_net *log;
1021         if (!st)
1022                 return NULL;
1023
1024         log = nfnl_log_pernet(net);
1025
1026         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1027                 struct hlist_head *head = &log->instance_table[st->bucket];
1028
1029                 if (!hlist_empty(head))
1030                         return rcu_dereference_bh(hlist_first_rcu(head));
1031         }
1032         return NULL;
1033 }
1034
1035 static struct hlist_node *get_next(struct net *net, struct iter_state *st,
1036                                    struct hlist_node *h)
1037 {
1038         h = rcu_dereference_bh(hlist_next_rcu(h));
1039         while (!h) {
1040                 struct nfnl_log_net *log;
1041                 struct hlist_head *head;
1042
1043                 if (++st->bucket >= INSTANCE_BUCKETS)
1044                         return NULL;
1045
1046                 log = nfnl_log_pernet(net);
1047                 head = &log->instance_table[st->bucket];
1048                 h = rcu_dereference_bh(hlist_first_rcu(head));
1049         }
1050         return h;
1051 }
1052
1053 static struct hlist_node *get_idx(struct net *net, struct iter_state *st,
1054                                   loff_t pos)
1055 {
1056         struct hlist_node *head;
1057         head = get_first(net, st);
1058
1059         if (head)
1060                 while (pos && (head = get_next(net, st, head)))
1061                         pos--;
1062         return pos ? NULL : head;
1063 }
1064
1065 static void *seq_start(struct seq_file *s, loff_t *pos)
1066         __acquires(rcu_bh)
1067 {
1068         rcu_read_lock_bh();
1069         return get_idx(seq_file_net(s), s->private, *pos);
1070 }
1071
1072 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1073 {
1074         (*pos)++;
1075         return get_next(seq_file_net(s), s->private, v);
1076 }
1077
1078 static void seq_stop(struct seq_file *s, void *v)
1079         __releases(rcu_bh)
1080 {
1081         rcu_read_unlock_bh();
1082 }
1083
1084 static int seq_show(struct seq_file *s, void *v)
1085 {
1086         const struct nfulnl_instance *inst = v;
1087
1088         seq_printf(s, "%5u %6u %5u %1u %5u %6u %2u\n",
1089                    inst->group_num,
1090                    inst->peer_portid, inst->qlen,
1091                    inst->copy_mode, inst->copy_range,
1092                    inst->flushtimeout, refcount_read(&inst->use));
1093
1094         return 0;
1095 }
1096
1097 static const struct seq_operations nful_seq_ops = {
1098         .start  = seq_start,
1099         .next   = seq_next,
1100         .stop   = seq_stop,
1101         .show   = seq_show,
1102 };
1103 #endif /* PROC_FS */
1104
1105 static int __net_init nfnl_log_net_init(struct net *net)
1106 {
1107         unsigned int i;
1108         struct nfnl_log_net *log = nfnl_log_pernet(net);
1109 #ifdef CONFIG_PROC_FS
1110         struct proc_dir_entry *proc;
1111         kuid_t root_uid;
1112         kgid_t root_gid;
1113 #endif
1114
1115         for (i = 0; i < INSTANCE_BUCKETS; i++)
1116                 INIT_HLIST_HEAD(&log->instance_table[i]);
1117         spin_lock_init(&log->instances_lock);
1118
1119 #ifdef CONFIG_PROC_FS
1120         proc = proc_create_net("nfnetlink_log", 0440, net->nf.proc_netfilter,
1121                         &nful_seq_ops, sizeof(struct iter_state));
1122         if (!proc)
1123                 return -ENOMEM;
1124
1125         root_uid = make_kuid(net->user_ns, 0);
1126         root_gid = make_kgid(net->user_ns, 0);
1127         if (uid_valid(root_uid) && gid_valid(root_gid))
1128                 proc_set_user(proc, root_uid, root_gid);
1129 #endif
1130         return 0;
1131 }
1132
1133 static void __net_exit nfnl_log_net_exit(struct net *net)
1134 {
1135         struct nfnl_log_net *log = nfnl_log_pernet(net);
1136         unsigned int i;
1137
1138 #ifdef CONFIG_PROC_FS
1139         remove_proc_entry("nfnetlink_log", net->nf.proc_netfilter);
1140 #endif
1141         nf_log_unset(net, &nfulnl_logger);
1142         for (i = 0; i < INSTANCE_BUCKETS; i++)
1143                 WARN_ON_ONCE(!hlist_empty(&log->instance_table[i]));
1144 }
1145
1146 static struct pernet_operations nfnl_log_net_ops = {
1147         .init   = nfnl_log_net_init,
1148         .exit   = nfnl_log_net_exit,
1149         .id     = &nfnl_log_net_id,
1150         .size   = sizeof(struct nfnl_log_net),
1151 };
1152
1153 static int __init nfnetlink_log_init(void)
1154 {
1155         int status;
1156
1157         status = register_pernet_subsys(&nfnl_log_net_ops);
1158         if (status < 0) {
1159                 pr_err("failed to register pernet ops\n");
1160                 goto out;
1161         }
1162
1163         netlink_register_notifier(&nfulnl_rtnl_notifier);
1164         status = nfnetlink_subsys_register(&nfulnl_subsys);
1165         if (status < 0) {
1166                 pr_err("failed to create netlink socket\n");
1167                 goto cleanup_netlink_notifier;
1168         }
1169
1170         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
1171         if (status < 0) {
1172                 pr_err("failed to register logger\n");
1173                 goto cleanup_subsys;
1174         }
1175
1176         return status;
1177
1178 cleanup_subsys:
1179         nfnetlink_subsys_unregister(&nfulnl_subsys);
1180 cleanup_netlink_notifier:
1181         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1182         unregister_pernet_subsys(&nfnl_log_net_ops);
1183 out:
1184         return status;
1185 }
1186
1187 static void __exit nfnetlink_log_fini(void)
1188 {
1189         nfnetlink_subsys_unregister(&nfulnl_subsys);
1190         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1191         unregister_pernet_subsys(&nfnl_log_net_ops);
1192         nf_log_unregister(&nfulnl_logger);
1193 }
1194
1195 MODULE_DESCRIPTION("netfilter userspace logging");
1196 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1197 MODULE_LICENSE("GPL");
1198 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1199 MODULE_ALIAS_NF_LOGGER(AF_INET, 1);
1200 MODULE_ALIAS_NF_LOGGER(AF_INET6, 1);
1201 MODULE_ALIAS_NF_LOGGER(AF_BRIDGE, 1);
1202 MODULE_ALIAS_NF_LOGGER(3, 1); /* NFPROTO_ARP */
1203 MODULE_ALIAS_NF_LOGGER(5, 1); /* NFPROTO_NETDEV */
1204
1205 module_init(nfnetlink_log_init);
1206 module_exit(nfnetlink_log_fini);