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         netns_tracker ns_tracker;
  70         struct user_namespace *peer_user_ns;    /* User namespace of the peer process */
  71         u32 peer_portid;                /* PORTID of the peer process */
  72
  73         /* configurable parameters */
  74         unsigned int flushtimeout;      /* timeout until queue flush */
  75         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  76         unsigned int qthreshold;        /* threshold of the queue */
  77         u_int32_t copy_range;
  78         u_int32_t seq;                  /* instance-local sequential counter */
  79         u_int16_t group_num;            /* number of this queue */
  80         u_int16_t flags;
  81         u_int8_t copy_mode;
  82         struct rcu_head rcu;
  83 };
  84
  85 #define INSTANCE_BUCKETS        16
  86
  87 static unsigned int nfnl_log_net_id __read_mostly;
  88
  89 struct nfnl_log_net {
  90         spinlock_t instances_lock;
  91         struct hlist_head instance_table[INSTANCE_BUCKETS];
  92         atomic_t global_seq;
  93 };
  94
  95 static struct nfnl_log_net *nfnl_log_pernet(struct net *net)
  96 {
  97         return net_generic(net, nfnl_log_net_id);
  98 }
  99
 100 static inline u_int8_t instance_hashfn(u_int16_t group_num)
 101 {
 102         return ((group_num & 0xff) % INSTANCE_BUCKETS);
 103 }
 104
 105 static struct nfulnl_instance *
 106 __instance_lookup(const struct nfnl_log_net *log, u16 group_num)
 107 {
 108         const struct hlist_head *head;
 109         struct nfulnl_instance *inst;
 110
 111         head = &log->instance_table[instance_hashfn(group_num)];
 112         hlist_for_each_entry_rcu(inst, head, hlist) {
 113                 if (inst->group_num == group_num)
 114                         return inst;
 115         }
 116         return NULL;
 117 }
 118
 119 static inline void
 120 instance_get(struct nfulnl_instance *inst)
 121 {
 122         refcount_inc(&inst->use);
 123 }
 124
 125 static struct nfulnl_instance *
 126 instance_lookup_get_rcu(const struct nfnl_log_net *log, u16 group_num)
 127 {
 128         struct nfulnl_instance *inst;
 129
 130         inst = __instance_lookup(log, group_num);
 131         if (inst && !refcount_inc_not_zero(&inst->use))
 132                 inst = NULL;
 133
 134         return inst;
 135 }
 136
 137 static struct nfulnl_instance *
 138 instance_lookup_get(const struct nfnl_log_net *log, u16 group_num)
 139 {
 140         struct nfulnl_instance *inst;
 141
 142         rcu_read_lock();
 143         inst = instance_lookup_get_rcu(log, group_num);
 144         rcu_read_unlock();
 145
 146         return inst;
 147 }
 148
 149 static void nfulnl_instance_free_rcu(struct rcu_head *head)
 150 {
 151         struct nfulnl_instance *inst =
 152                 container_of(head, struct nfulnl_instance, rcu);
 153
 154         put_net_track(inst->net, &inst->ns_tracker);
 155         kfree(inst);
 156         module_put(THIS_MODULE);
 157 }
 158
 159 static void
 160 instance_put(struct nfulnl_instance *inst)
 161 {
 162         if (inst && refcount_dec_and_test(&inst->use))
 163                 call_rcu(&inst->rcu, nfulnl_instance_free_rcu);
 164 }
 165
 166 static void nfulnl_timer(struct timer_list *t);
 167
 168 static struct nfulnl_instance *
 169 instance_create(struct net *net, u_int16_t group_num,
 170                 u32 portid, struct user_namespace *user_ns)
 171 {
 172         struct nfulnl_instance *inst;
 173         struct nfnl_log_net *log = nfnl_log_pernet(net);
 174         int err;
 175
 176         spin_lock_bh(&log->instances_lock);
 177         if (__instance_lookup(log, group_num)) {
 178                 err = -EEXIST;
 179                 goto out_unlock;
 180         }
 181
 182         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 183         if (!inst) {
 184                 err = -ENOMEM;
 185                 goto out_unlock;
 186         }
 187
 188         if (!try_module_get(THIS_MODULE)) {
 189                 kfree(inst);
 190                 err = -EAGAIN;
 191                 goto out_unlock;
 192         }
 193
 194         INIT_HLIST_NODE(&inst->hlist);
 195         spin_lock_init(&inst->lock);
 196         /* needs to be two, since we _put() after creation */
 197         refcount_set(&inst->use, 2);
 198
 199         timer_setup(&inst->timer, nfulnl_timer, 0);
 200
 201         inst->net = get_net_track(net, &inst->ns_tracker, GFP_ATOMIC);
 202         inst->peer_user_ns = user_ns;
 203         inst->peer_portid = portid;
 204         inst->group_num = group_num;
 205
 206         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 207         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 208         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 209         inst->copy_mode         = NFULNL_COPY_PACKET;
 210         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
 211
 212         hlist_add_head_rcu(&inst->hlist,
 213                        &log->instance_table[instance_hashfn(group_num)]);
 214
 215
 216         spin_unlock_bh(&log->instances_lock);
 217
 218         return inst;
 219
 220 out_unlock:
 221         spin_unlock_bh(&log->instances_lock);
 222         return ERR_PTR(err);
 223 }
 224
 225 static void __nfulnl_flush(struct nfulnl_instance *inst);
 226
 227 /* called with BH disabled */
 228 static void
 229 __instance_destroy(struct nfulnl_instance *inst)
 230 {
 231         /* first pull it out of the global list */
 232         hlist_del_rcu(&inst->hlist);
 233
 234         /* then flush all pending packets from skb */
 235
 236         spin_lock(&inst->lock);
 237
 238         /* lockless readers wont be able to use us */
 239         inst->copy_mode = NFULNL_COPY_DISABLED;
 240
 241         if (inst->skb)
 242                 __nfulnl_flush(inst);
 243         spin_unlock(&inst->lock);
 244
 245         /* and finally put the refcount */
 246         instance_put(inst);
 247 }
 248
 249 static inline void
 250 instance_destroy(struct nfnl_log_net *log,
 251                  struct nfulnl_instance *inst)
 252 {
 253         spin_lock_bh(&log->instances_lock);
 254         __instance_destroy(inst);
 255         spin_unlock_bh(&log->instances_lock);
 256 }
 257
 258 static int
 259 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 260                   unsigned int range)
 261 {
 262         int status = 0;
 263
 264         spin_lock_bh(&inst->lock);
 265
 266         switch (mode) {
 267         case NFULNL_COPY_NONE:
 268         case NFULNL_COPY_META:
 269                 inst->copy_mode = mode;
 270                 inst->copy_range = 0;
 271                 break;
 272
 273         case NFULNL_COPY_PACKET:
 274                 inst->copy_mode = mode;
 275                 if (range == 0)
 276                         range = NFULNL_COPY_RANGE_MAX;
 277                 inst->copy_range = min_t(unsigned int,
 278                                          range, NFULNL_COPY_RANGE_MAX);
 279                 break;
 280
 281         default:
 282                 status = -EINVAL;
 283                 break;
 284         }
 285
 286         spin_unlock_bh(&inst->lock);
 287
 288         return status;
 289 }
 290
 291 static int
 292 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 293 {
 294         int status;
 295
 296         spin_lock_bh(&inst->lock);
 297         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 298                 status = -ERANGE;
 299         else if (nlbufsiz > 131072)
 300                 status = -ERANGE;
 301         else {
 302                 inst->nlbufsiz = nlbufsiz;
 303                 status = 0;
 304         }
 305         spin_unlock_bh(&inst->lock);
 306
 307         return status;
 308 }
 309
 310 static void
 311 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 312 {
 313         spin_lock_bh(&inst->lock);
 314         inst->flushtimeout = timeout;
 315         spin_unlock_bh(&inst->lock);
 316 }
 317
 318 static void
 319 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 320 {
 321         spin_lock_bh(&inst->lock);
 322         inst->qthreshold = qthresh;
 323         spin_unlock_bh(&inst->lock);
 324 }
 325
 326 static int
 327 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 328 {
 329         spin_lock_bh(&inst->lock);
 330         inst->flags = flags;
 331         spin_unlock_bh(&inst->lock);
 332
 333         return 0;
 334 }
 335
 336 static struct sk_buff *
 337 nfulnl_alloc_skb(struct net *net, u32 peer_portid, unsigned int inst_size,
 338                  unsigned int pkt_size)
 339 {
 340         struct sk_buff *skb;
 341         unsigned int n;
 342
 343         /* alloc skb which should be big enough for a whole multipart
 344          * message.  WARNING: has to be <= 128k due to slab restrictions */
 345
 346         n = max(inst_size, pkt_size);
 347         skb = alloc_skb(n, GFP_ATOMIC | __GFP_NOWARN);
 348         if (!skb) {
 349                 if (n > pkt_size) {
 350                         /* try to allocate only as much as we need for current
 351                          * packet */
 352
 353                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 354                 }
 355         }
 356
 357         return skb;
 358 }
 359
 360 static void
 361 __nfulnl_send(struct nfulnl_instance *inst)
 362 {
 363         if (inst->qlen > 1) {
 364                 struct nlmsghdr *nlh = nlmsg_put(inst->skb, 0, 0,
 365                                                  NLMSG_DONE,
 366                                                  sizeof(struct nfgenmsg),
 367                                                  0);
 368                 if (WARN_ONCE(!nlh, "bad nlskb size: %u, tailroom %d\n",
 369                               inst->skb->len, skb_tailroom(inst->skb))) {
 370                         kfree_skb(inst->skb);
 371                         goto out;
 372                 }
 373         }
 374         nfnetlink_unicast(inst->skb, inst->net, inst->peer_portid);
 375 out:
 376         inst->qlen = 0;
 377         inst->skb = NULL;
 378 }
 379
 380 static void
 381 __nfulnl_flush(struct nfulnl_instance *inst)
 382 {
 383         /* timer holds a reference */
 384         if (del_timer(&inst->timer))
 385                 instance_put(inst);
 386         if (inst->skb)
 387                 __nfulnl_send(inst);
 388 }
 389
 390 static void
 391 nfulnl_timer(struct timer_list *t)
 392 {
 393         struct nfulnl_instance *inst = from_timer(inst, t, timer);
 394
 395         spin_lock_bh(&inst->lock);
 396         if (inst->skb)
 397                 __nfulnl_send(inst);
 398         spin_unlock_bh(&inst->lock);
 399         instance_put(inst);
 400 }
 401
 402 static u32 nfulnl_get_bridge_size(const struct sk_buff *skb)
 403 {
 404         u32 size = 0;
 405
 406         if (!skb_mac_header_was_set(skb))
 407                 return 0;
 408
 409         if (skb_vlan_tag_present(skb)) {
 410                 size += nla_total_size(0); /* nested */
 411                 size += nla_total_size(sizeof(u16)); /* id */
 412                 size += nla_total_size(sizeof(u16)); /* tag */
 413         }
 414
 415         if (skb->network_header > skb->mac_header)
 416                 size += nla_total_size(skb->network_header - skb->mac_header);
 417
 418         return size;
 419 }
 420
 421 static int nfulnl_put_bridge(struct nfulnl_instance *inst, const struct sk_buff *skb)
 422 {
 423         if (!skb_mac_header_was_set(skb))
 424                 return 0;
 425
 426         if (skb_vlan_tag_present(skb)) {
 427                 struct nlattr *nest;
 428
 429                 nest = nla_nest_start(inst->skb, NFULA_VLAN);
 430                 if (!nest)
 431                         goto nla_put_failure;
 432
 433                 if (nla_put_be16(inst->skb, NFULA_VLAN_TCI, htons(skb->vlan_tci)) ||
 434                     nla_put_be16(inst->skb, NFULA_VLAN_PROTO, skb->vlan_proto))
 435                         goto nla_put_failure;
 436
 437                 nla_nest_end(inst->skb, nest);
 438         }
 439
 440         if (skb->mac_header < skb->network_header) {
 441                 int len = (int)(skb->network_header - skb->mac_header);
 442
 443                 if (nla_put(inst->skb, NFULA_L2HDR, len, skb_mac_header(skb)))
 444                         goto nla_put_failure;
 445         }
 446
 447         return 0;
 448
 449 nla_put_failure:
 450         return -1;
 451 }
 452
 453 /* This is an inline function, we don't really care about a long
 454  * list of arguments */
 455 static inline int
 456 __build_packet_message(struct nfnl_log_net *log,
 457                         struct nfulnl_instance *inst,
 458                         const struct sk_buff *skb,
 459                         unsigned int data_len,
 460                         u_int8_t pf,
 461                         unsigned int hooknum,
 462                         const struct net_device *indev,
 463                         const struct net_device *outdev,
 464                         const char *prefix, unsigned int plen,
 465                         const struct nfnl_ct_hook *nfnl_ct,
 466                         struct nf_conn *ct, enum ip_conntrack_info ctinfo)
 467 {
 468         struct nfulnl_msg_packet_hdr pmsg;
 469         struct nlmsghdr *nlh;
 470         sk_buff_data_t old_tail = inst->skb->tail;
 471         struct sock *sk;
 472         const unsigned char *hwhdrp;
 473         ktime_t tstamp;
 474
 475         nlh = nfnl_msg_put(inst->skb, 0, 0,
 476                            nfnl_msg_type(NFNL_SUBSYS_ULOG, NFULNL_MSG_PACKET),
 477                            0, pf, NFNETLINK_V0, htons(inst->group_num));
 478         if (!nlh)
 479                 return -1;
 480
 481         memset(&pmsg, 0, sizeof(pmsg));
 482         pmsg.hw_protocol        = skb->protocol;
 483         pmsg.hook               = hooknum;
 484
 485         if (nla_put(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg))
 486                 goto nla_put_failure;
 487
 488         if (prefix &&
 489             nla_put(inst->skb, NFULA_PREFIX, plen, prefix))
 490                 goto nla_put_failure;
 491
 492         if (indev) {
 493 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 494                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 495                                  htonl(indev->ifindex)))
 496                         goto nla_put_failure;
 497 #else
 498                 if (pf == PF_BRIDGE) {
 499                         /* Case 1: outdev is physical input device, we need to
 500                          * look for bridge group (when called from
 501                          * netfilter_bridge) */
 502                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 503                                          htonl(indev->ifindex)) ||
 504                         /* this is the bridge group "brX" */
 505                         /* rcu_read_lock()ed by nf_hook_thresh or
 506                          * nf_log_packet.
 507                          */
 508                             nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 509                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 510                                 goto nla_put_failure;
 511                 } else {
 512                         struct net_device *physindev;
 513
 514                         /* Case 2: indev is bridge group, we need to look for
 515                          * physical device (when called from ipv4) */
 516                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 517                                          htonl(indev->ifindex)))
 518                                 goto nla_put_failure;
 519
 520                         physindev = nf_bridge_get_physindev(skb);
 521                         if (physindev &&
 522                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 523                                          htonl(physindev->ifindex)))
 524                                 goto nla_put_failure;
 525                 }
 526 #endif
 527         }
 528
 529         if (outdev) {
 530 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 531                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 532                                  htonl(outdev->ifindex)))
 533                         goto nla_put_failure;
 534 #else
 535                 if (pf == PF_BRIDGE) {
 536                         /* Case 1: outdev is physical output device, we need to
 537                          * look for bridge group (when called from
 538                          * netfilter_bridge) */
 539                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 540                                          htonl(outdev->ifindex)) ||
 541                         /* this is the bridge group "brX" */
 542                         /* rcu_read_lock()ed by nf_hook_thresh or
 543                          * nf_log_packet.
 544                          */
 545                             nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 546                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 547                                 goto nla_put_failure;
 548                 } else {
 549                         struct net_device *physoutdev;
 550
 551                         /* Case 2: indev is a bridge group, we need to look
 552                          * for physical device (when called from ipv4) */
 553                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 554                                          htonl(outdev->ifindex)))
 555                                 goto nla_put_failure;
 556
 557                         physoutdev = nf_bridge_get_physoutdev(skb);
 558                         if (physoutdev &&
 559                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 560                                          htonl(physoutdev->ifindex)))
 561                                 goto nla_put_failure;
 562                 }
 563 #endif
 564         }
 565
 566         if (skb->mark &&
 567             nla_put_be32(inst->skb, NFULA_MARK, htonl(skb->mark)))
 568                 goto nla_put_failure;
 569
 570         if (indev && skb->dev &&
 571             skb_mac_header_was_set(skb) &&
 572             skb_mac_header_len(skb) != 0) {
 573                 struct nfulnl_msg_packet_hw phw;
 574                 int len;
 575
 576                 memset(&phw, 0, sizeof(phw));
 577                 len = dev_parse_header(skb, phw.hw_addr);
 578                 if (len > 0) {
 579                         phw.hw_addrlen = htons(len);
 580                         if (nla_put(inst->skb, NFULA_HWADDR, sizeof(phw), &phw))
 581                                 goto nla_put_failure;
 582                 }
 583         }
 584
 585         if (indev && skb_mac_header_was_set(skb)) {
 586                 if (nla_put_be16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type)) ||
 587                     nla_put_be16(inst->skb, NFULA_HWLEN,
 588                                  htons(skb->dev->hard_header_len)))
 589                         goto nla_put_failure;
 590
 591                 hwhdrp = skb_mac_header(skb);
 592
 593                 if (skb->dev->type == ARPHRD_SIT)
 594                         hwhdrp -= ETH_HLEN;
 595
 596                 if (hwhdrp >= skb->head &&
 597                     nla_put(inst->skb, NFULA_HWHEADER,
 598                             skb->dev->hard_header_len, hwhdrp))
 599                         goto nla_put_failure;
 600         }
 601
 602         tstamp = skb_tstamp_cond(skb, false);
 603         if (hooknum <= NF_INET_FORWARD && tstamp) {
 604                 struct nfulnl_msg_packet_timestamp ts;
 605                 struct timespec64 kts = ktime_to_timespec64(tstamp);
 606                 ts.sec = cpu_to_be64(kts.tv_sec);
 607                 ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
 608
 609                 if (nla_put(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts))
 610                         goto nla_put_failure;
 611         }
 612
 613         /* UID */
 614         sk = skb->sk;
 615         if (sk && sk_fullsock(sk)) {
 616                 read_lock_bh(&sk->sk_callback_lock);
 617                 if (sk->sk_socket && sk->sk_socket->file) {
 618                         struct file *file = sk->sk_socket->file;
 619                         const struct cred *cred = file->f_cred;
 620                         struct user_namespace *user_ns = inst->peer_user_ns;
 621                         __be32 uid = htonl(from_kuid_munged(user_ns, cred->fsuid));
 622                         __be32 gid = htonl(from_kgid_munged(user_ns, cred->fsgid));
 623                         read_unlock_bh(&sk->sk_callback_lock);
 624                         if (nla_put_be32(inst->skb, NFULA_UID, uid) ||
 625                             nla_put_be32(inst->skb, NFULA_GID, gid))
 626                                 goto nla_put_failure;
 627                 } else
 628                         read_unlock_bh(&sk->sk_callback_lock);
 629         }
 630
 631         /* local sequence number */
 632         if ((inst->flags & NFULNL_CFG_F_SEQ) &&
 633             nla_put_be32(inst->skb, NFULA_SEQ, htonl(inst->seq++)))
 634                 goto nla_put_failure;
 635
 636         /* global sequence number */
 637         if ((inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) &&
 638             nla_put_be32(inst->skb, NFULA_SEQ_GLOBAL,
 639                          htonl(atomic_inc_return(&log->global_seq))))
 640                 goto nla_put_failure;
 641
 642         if (ct && nfnl_ct->build(inst->skb, ct, ctinfo,
 643                                  NFULA_CT, NFULA_CT_INFO) < 0)
 644                 goto nla_put_failure;
 645
 646         if ((pf == NFPROTO_NETDEV || pf == NFPROTO_BRIDGE) &&
 647             nfulnl_put_bridge(inst, skb) < 0)
 648                 goto nla_put_failure;
 649
 650         if (data_len) {
 651                 struct nlattr *nla;
 652                 int size = nla_attr_size(data_len);
 653
 654                 if (skb_tailroom(inst->skb) < nla_total_size(data_len))
 655                         goto nla_put_failure;
 656
 657                 nla = skb_put(inst->skb, nla_total_size(data_len));
 658                 nla->nla_type = NFULA_PAYLOAD;
 659                 nla->nla_len = size;
 660
 661                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 662                         BUG();
 663         }
 664
 665         nlh->nlmsg_len = inst->skb->tail - old_tail;
 666         return 0;
 667
 668 nla_put_failure:
 669         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 670         return -1;
 671 }
 672
 673 static const struct nf_loginfo default_loginfo = {
 674         .type =         NF_LOG_TYPE_ULOG,
 675         .u = {
 676                 .ulog = {
 677                         .copy_len       = 0xffff,
 678                         .group          = 0,
 679                         .qthreshold     = 1,
 680                 },
 681         },
 682 };
 683
 684 /* log handler for internal netfilter logging api */
 685 static void
 686 nfulnl_log_packet(struct net *net,
 687                   u_int8_t pf,
 688                   unsigned int hooknum,
 689                   const struct sk_buff *skb,
 690                   const struct net_device *in,
 691                   const struct net_device *out,
 692                   const struct nf_loginfo *li_user,
 693                   const char *prefix)
 694 {
 695         size_t size;
 696         unsigned int data_len;
 697         struct nfulnl_instance *inst;
 698         const struct nf_loginfo *li;
 699         unsigned int qthreshold;
 700         unsigned int plen = 0;
 701         struct nfnl_log_net *log = nfnl_log_pernet(net);
 702         const struct nfnl_ct_hook *nfnl_ct = NULL;
 703         struct nf_conn *ct = NULL;
 704         enum ip_conntrack_info ctinfo;
 705
 706         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 707                 li = li_user;
 708         else
 709                 li = &default_loginfo;
 710
 711         inst = instance_lookup_get_rcu(log, li->u.ulog.group);
 712         if (!inst)
 713                 return;
 714
 715         if (prefix)
 716                 plen = strlen(prefix) + 1;
 717
 718         /* FIXME: do we want to make the size calculation conditional based on
 719          * what is actually present?  way more branches and checks, but more
 720          * memory efficient... */
 721         size = nlmsg_total_size(sizeof(struct nfgenmsg))
 722                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 723                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 724                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 725 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 726                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 727                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 728 #endif
 729                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 730                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 731                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 732                 + nla_total_size(plen)                  /* prefix */
 733                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 734                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp))
 735                 + nla_total_size(sizeof(struct nfgenmsg));      /* NLMSG_DONE */
 736
 737         if (in && skb_mac_header_was_set(skb)) {
 738                 size += nla_total_size(skb->dev->hard_header_len)
 739                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 740                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 741         }
 742
 743         spin_lock_bh(&inst->lock);
 744
 745         if (inst->flags & NFULNL_CFG_F_SEQ)
 746                 size += nla_total_size(sizeof(u_int32_t));
 747         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 748                 size += nla_total_size(sizeof(u_int32_t));
 749 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 750         if (inst->flags & NFULNL_CFG_F_CONNTRACK) {
 751                 nfnl_ct = rcu_dereference(nfnl_ct_hook);
 752                 if (nfnl_ct != NULL) {
 753                         ct = nf_ct_get(skb, &ctinfo);
 754                         if (ct != NULL)
 755                                 size += nfnl_ct->build_size(ct);
 756                 }
 757         }
 758 #endif
 759         if (pf == NFPROTO_NETDEV || pf == NFPROTO_BRIDGE)
 760                 size += nfulnl_get_bridge_size(skb);
 761
 762         qthreshold = inst->qthreshold;
 763         /* per-rule qthreshold overrides per-instance */
 764         if (li->u.ulog.qthreshold)
 765                 if (qthreshold > li->u.ulog.qthreshold)
 766                         qthreshold = li->u.ulog.qthreshold;
 767
 768
 769         switch (inst->copy_mode) {
 770         case NFULNL_COPY_META:
 771         case NFULNL_COPY_NONE:
 772                 data_len = 0;
 773                 break;
 774
 775         case NFULNL_COPY_PACKET:
 776                 data_len = inst->copy_range;
 777                 if ((li->u.ulog.flags & NF_LOG_F_COPY_LEN) &&
 778                     (li->u.ulog.copy_len < data_len))
 779                         data_len = li->u.ulog.copy_len;
 780
 781                 if (data_len > skb->len)
 782                         data_len = skb->len;
 783
 784                 size += nla_total_size(data_len);
 785                 break;
 786
 787         case NFULNL_COPY_DISABLED:
 788         default:
 789                 goto unlock_and_release;
 790         }
 791
 792         if (inst->skb && size > skb_tailroom(inst->skb)) {
 793                 /* either the queue len is too high or we don't have
 794                  * enough room in the skb left. flush to userspace. */
 795                 __nfulnl_flush(inst);
 796         }
 797
 798         if (!inst->skb) {
 799                 inst->skb = nfulnl_alloc_skb(net, inst->peer_portid,
 800                                              inst->nlbufsiz, size);
 801                 if (!inst->skb)
 802                         goto alloc_failure;
 803         }
 804
 805         inst->qlen++;
 806
 807         __build_packet_message(log, inst, skb, data_len, pf,
 808                                 hooknum, in, out, prefix, plen,
 809                                 nfnl_ct, ct, ctinfo);
 810
 811         if (inst->qlen >= qthreshold)
 812                 __nfulnl_flush(inst);
 813         /* timer_pending always called within inst->lock, so there
 814          * is no chance of a race here */
 815         else if (!timer_pending(&inst->timer)) {
 816                 instance_get(inst);
 817                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 818                 add_timer(&inst->timer);
 819         }
 820
 821 unlock_and_release:
 822         spin_unlock_bh(&inst->lock);
 823         instance_put(inst);
 824         return;
 825
 826 alloc_failure:
 827         /* FIXME: statistics */
 828         goto unlock_and_release;
 829 }
 830
 831 static int
 832 nfulnl_rcv_nl_event(struct notifier_block *this,
 833                    unsigned long event, void *ptr)
 834 {
 835         struct netlink_notify *n = ptr;
 836         struct nfnl_log_net *log = nfnl_log_pernet(n->net);
 837
 838         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 839                 int i;
 840
 841                 /* destroy all instances for this portid */
 842                 spin_lock_bh(&log->instances_lock);
 843                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 844                         struct hlist_node *t2;
 845                         struct nfulnl_instance *inst;
 846                         struct hlist_head *head = &log->instance_table[i];
 847
 848                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 849                                 if (n->portid == inst->peer_portid)
 850                                         __instance_destroy(inst);
 851                         }
 852                 }
 853                 spin_unlock_bh(&log->instances_lock);
 854         }
 855         return NOTIFY_DONE;
 856 }
 857
 858 static struct notifier_block nfulnl_rtnl_notifier = {
 859         .notifier_call  = nfulnl_rcv_nl_event,
 860 };
 861
 862 static int nfulnl_recv_unsupp(struct sk_buff *skb, const struct nfnl_info *info,
 863                               const struct nlattr * const nfula[])
 864 {
 865         return -ENOTSUPP;
 866 }
 867
 868 static struct nf_logger nfulnl_logger __read_mostly = {
 869         .name   = "nfnetlink_log",
 870         .type   = NF_LOG_TYPE_ULOG,
 871         .logfn  = nfulnl_log_packet,
 872         .me     = THIS_MODULE,
 873 };
 874
 875 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 876         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 877         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 878         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 879         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 880         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 881         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 882 };
 883
 884 static int nfulnl_recv_config(struct sk_buff *skb, const struct nfnl_info *info,
 885                               const struct nlattr * const nfula[])
 886 {
 887         struct nfnl_log_net *log = nfnl_log_pernet(info->net);
 888         u_int16_t group_num = ntohs(info->nfmsg->res_id);
 889         struct nfulnl_msg_config_cmd *cmd = NULL;
 890         struct nfulnl_instance *inst;
 891         u16 flags = 0;
 892         int ret = 0;
 893
 894         if (nfula[NFULA_CFG_CMD]) {
 895                 u_int8_t pf = info->nfmsg->nfgen_family;
 896                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 897
 898                 /* Commands without queue context */
 899                 switch (cmd->command) {
 900                 case NFULNL_CFG_CMD_PF_BIND:
 901                         return nf_log_bind_pf(info->net, pf, &nfulnl_logger);
 902                 case NFULNL_CFG_CMD_PF_UNBIND:
 903                         nf_log_unbind_pf(info->net, pf);
 904                         return 0;
 905                 }
 906         }
 907
 908         inst = instance_lookup_get(log, group_num);
 909         if (inst && inst->peer_portid != NETLINK_CB(skb).portid) {
 910                 ret = -EPERM;
 911                 goto out_put;
 912         }
 913
 914         /* Check if we support these flags in first place, dependencies should
 915          * be there too not to break atomicity.
 916          */
 917         if (nfula[NFULA_CFG_FLAGS]) {
 918                 flags = ntohs(nla_get_be16(nfula[NFULA_CFG_FLAGS]));
 919
 920                 if ((flags & NFULNL_CFG_F_CONNTRACK) &&
 921                     !rcu_access_pointer(nfnl_ct_hook)) {
 922 #ifdef CONFIG_MODULES
 923                         nfnl_unlock(NFNL_SUBSYS_ULOG);
 924                         request_module("ip_conntrack_netlink");
 925                         nfnl_lock(NFNL_SUBSYS_ULOG);
 926                         if (rcu_access_pointer(nfnl_ct_hook)) {
 927                                 ret = -EAGAIN;
 928                                 goto out_put;
 929                         }
 930 #endif
 931                         ret = -EOPNOTSUPP;
 932                         goto out_put;
 933                 }
 934         }
 935
 936         if (cmd != NULL) {
 937                 switch (cmd->command) {
 938                 case NFULNL_CFG_CMD_BIND:
 939                         if (inst) {
 940                                 ret = -EBUSY;
 941                                 goto out_put;
 942                         }
 943
 944                         inst = instance_create(info->net, group_num,
 945                                                NETLINK_CB(skb).portid,
 946                                                sk_user_ns(NETLINK_CB(skb).sk));
 947                         if (IS_ERR(inst)) {
 948                                 ret = PTR_ERR(inst);
 949                                 goto out;
 950                         }
 951                         break;
 952                 case NFULNL_CFG_CMD_UNBIND:
 953                         if (!inst) {
 954                                 ret = -ENODEV;
 955                                 goto out;
 956                         }
 957
 958                         instance_destroy(log, inst);
 959                         goto out_put;
 960                 default:
 961                         ret = -ENOTSUPP;
 962                         goto out_put;
 963                 }
 964         } else if (!inst) {
 965                 ret = -ENODEV;
 966                 goto out;
 967         }
 968
 969         if (nfula[NFULA_CFG_MODE]) {
 970                 struct nfulnl_msg_config_mode *params =
 971                         nla_data(nfula[NFULA_CFG_MODE]);
 972
 973                 nfulnl_set_mode(inst, params->copy_mode,
 974                                 ntohl(params->copy_range));
 975         }
 976
 977         if (nfula[NFULA_CFG_TIMEOUT]) {
 978                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 979
 980                 nfulnl_set_timeout(inst, ntohl(timeout));
 981         }
 982
 983         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 984                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 985
 986                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 987         }
 988
 989         if (nfula[NFULA_CFG_QTHRESH]) {
 990                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 991
 992                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 993         }
 994
 995         if (nfula[NFULA_CFG_FLAGS])
 996                 nfulnl_set_flags(inst, flags);
 997
 998 out_put:
 999         instance_put(inst);
1000 out:
1001         return ret;
1002 }
1003
1004 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
1005         [NFULNL_MSG_PACKET]     = {
1006                 .call           = nfulnl_recv_unsupp,
1007                 .type           = NFNL_CB_MUTEX,
1008                 .attr_count     = NFULA_MAX,
1009         },
1010         [NFULNL_MSG_CONFIG]     = {
1011                 .call           = nfulnl_recv_config,
1012                 .type           = NFNL_CB_MUTEX,
1013                 .attr_count     = NFULA_CFG_MAX,
1014                 .policy         = nfula_cfg_policy
1015         },
1016 };
1017
1018 static const struct nfnetlink_subsystem nfulnl_subsys = {
1019         .name           = "log",
1020         .subsys_id      = NFNL_SUBSYS_ULOG,
1021         .cb_count       = NFULNL_MSG_MAX,
1022         .cb             = nfulnl_cb,
1023 };
1024
1025 #ifdef CONFIG_PROC_FS
1026 struct iter_state {
1027         struct seq_net_private p;
1028         unsigned int bucket;
1029 };
1030
1031 static struct hlist_node *get_first(struct net *net, struct iter_state *st)
1032 {
1033         struct nfnl_log_net *log;
1034         if (!st)
1035                 return NULL;
1036
1037         log = nfnl_log_pernet(net);
1038
1039         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1040                 struct hlist_head *head = &log->instance_table[st->bucket];
1041
1042                 if (!hlist_empty(head))
1043                         return rcu_dereference(hlist_first_rcu(head));
1044         }
1045         return NULL;
1046 }
1047
1048 static struct hlist_node *get_next(struct net *net, struct iter_state *st,
1049                                    struct hlist_node *h)
1050 {
1051         h = rcu_dereference(hlist_next_rcu(h));
1052         while (!h) {
1053                 struct nfnl_log_net *log;
1054                 struct hlist_head *head;
1055
1056                 if (++st->bucket >= INSTANCE_BUCKETS)
1057                         return NULL;
1058
1059                 log = nfnl_log_pernet(net);
1060                 head = &log->instance_table[st->bucket];
1061                 h = rcu_dereference(hlist_first_rcu(head));
1062         }
1063         return h;
1064 }
1065
1066 static struct hlist_node *get_idx(struct net *net, struct iter_state *st,
1067                                   loff_t pos)
1068 {
1069         struct hlist_node *head;
1070         head = get_first(net, st);
1071
1072         if (head)
1073                 while (pos && (head = get_next(net, st, head)))
1074                         pos--;
1075         return pos ? NULL : head;
1076 }
1077
1078 static void *seq_start(struct seq_file *s, loff_t *pos)
1079         __acquires(rcu)
1080 {
1081         rcu_read_lock();
1082         return get_idx(seq_file_net(s), s->private, *pos);
1083 }
1084
1085 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1086 {
1087         (*pos)++;
1088         return get_next(seq_file_net(s), s->private, v);
1089 }
1090
1091 static void seq_stop(struct seq_file *s, void *v)
1092         __releases(rcu)
1093 {
1094         rcu_read_unlock();
1095 }
1096
1097 static int seq_show(struct seq_file *s, void *v)
1098 {
1099         const struct nfulnl_instance *inst = v;
1100
1101         seq_printf(s, "%5u %6u %5u %1u %5u %6u %2u\n",
1102                    inst->group_num,
1103                    inst->peer_portid, inst->qlen,
1104                    inst->copy_mode, inst->copy_range,
1105                    inst->flushtimeout, refcount_read(&inst->use));
1106
1107         return 0;
1108 }
1109
1110 static const struct seq_operations nful_seq_ops = {
1111         .start  = seq_start,
1112         .next   = seq_next,
1113         .stop   = seq_stop,
1114         .show   = seq_show,
1115 };
1116 #endif /* PROC_FS */
1117
1118 static int __net_init nfnl_log_net_init(struct net *net)
1119 {
1120         unsigned int i;
1121         struct nfnl_log_net *log = nfnl_log_pernet(net);
1122 #ifdef CONFIG_PROC_FS
1123         struct proc_dir_entry *proc;
1124         kuid_t root_uid;
1125         kgid_t root_gid;
1126 #endif
1127
1128         for (i = 0; i < INSTANCE_BUCKETS; i++)
1129                 INIT_HLIST_HEAD(&log->instance_table[i]);
1130         spin_lock_init(&log->instances_lock);
1131
1132 #ifdef CONFIG_PROC_FS
1133         proc = proc_create_net("nfnetlink_log", 0440, net->nf.proc_netfilter,
1134                         &nful_seq_ops, sizeof(struct iter_state));
1135         if (!proc)
1136                 return -ENOMEM;
1137
1138         root_uid = make_kuid(net->user_ns, 0);
1139         root_gid = make_kgid(net->user_ns, 0);
1140         if (uid_valid(root_uid) && gid_valid(root_gid))
1141                 proc_set_user(proc, root_uid, root_gid);
1142 #endif
1143         return 0;
1144 }
1145
1146 static void __net_exit nfnl_log_net_exit(struct net *net)
1147 {
1148         struct nfnl_log_net *log = nfnl_log_pernet(net);
1149         unsigned int i;
1150
1151 #ifdef CONFIG_PROC_FS
1152         remove_proc_entry("nfnetlink_log", net->nf.proc_netfilter);
1153 #endif
1154         nf_log_unset(net, &nfulnl_logger);
1155         for (i = 0; i < INSTANCE_BUCKETS; i++)
1156                 WARN_ON_ONCE(!hlist_empty(&log->instance_table[i]));
1157 }
1158
1159 static struct pernet_operations nfnl_log_net_ops = {
1160         .init   = nfnl_log_net_init,
1161         .exit   = nfnl_log_net_exit,
1162         .id     = &nfnl_log_net_id,
1163         .size   = sizeof(struct nfnl_log_net),
1164 };
1165
1166 static int __init nfnetlink_log_init(void)
1167 {
1168         int status;
1169
1170         status = register_pernet_subsys(&nfnl_log_net_ops);
1171         if (status < 0) {
1172                 pr_err("failed to register pernet ops\n");
1173                 goto out;
1174         }
1175
1176         netlink_register_notifier(&nfulnl_rtnl_notifier);
1177         status = nfnetlink_subsys_register(&nfulnl_subsys);
1178         if (status < 0) {
1179                 pr_err("failed to create netlink socket\n");
1180                 goto cleanup_netlink_notifier;
1181         }
1182
1183         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
1184         if (status < 0) {
1185                 pr_err("failed to register logger\n");
1186                 goto cleanup_subsys;
1187         }
1188
1189         return status;
1190
1191 cleanup_subsys:
1192         nfnetlink_subsys_unregister(&nfulnl_subsys);
1193 cleanup_netlink_notifier:
1194         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1195         unregister_pernet_subsys(&nfnl_log_net_ops);
1196 out:
1197         return status;
1198 }
1199
1200 static void __exit nfnetlink_log_fini(void)
1201 {
1202         nfnetlink_subsys_unregister(&nfulnl_subsys);
1203         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1204         unregister_pernet_subsys(&nfnl_log_net_ops);
1205         nf_log_unregister(&nfulnl_logger);
1206 }
1207
1208 MODULE_DESCRIPTION("netfilter userspace logging");
1209 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1210 MODULE_LICENSE("GPL");
1211 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1212 MODULE_ALIAS_NF_LOGGER(AF_INET, 1);
1213 MODULE_ALIAS_NF_LOGGER(AF_INET6, 1);
1214 MODULE_ALIAS_NF_LOGGER(AF_BRIDGE, 1);
1215 MODULE_ALIAS_NF_LOGGER(3, 1); /* NFPROTO_ARP */
1216 MODULE_ALIAS_NF_LOGGER(5, 1); /* NFPROTO_NETDEV */
1217
1218 module_init(nfnetlink_log_init);
1219 module_exit(nfnetlink_log_fini);