net/ipv4/netfilter/nf_conntrack_proto_icmp.c

   1 /* (C) 1999-2001 Paul `Rusty' Russell
   2  * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  */
   8
   9 #include <linux/types.h>
  10 #include <linux/timer.h>
  11 #include <linux/netfilter.h>
  12 #include <linux/in.h>
  13 #include <linux/icmp.h>
  14 #include <linux/seq_file.h>
  15 #include <net/ip.h>
  16 #include <net/checksum.h>
  17 #include <linux/netfilter_ipv4.h>
  18 #include <net/netfilter/nf_conntrack_tuple.h>
  19 #include <net/netfilter/nf_conntrack_l4proto.h>
  20 #include <net/netfilter/nf_conntrack_core.h>
  21 #include <net/netfilter/nf_conntrack_zones.h>
  22 #include <net/netfilter/nf_log.h>
  23
  24 static unsigned int nf_ct_icmp_timeout __read_mostly = 30*HZ;
  25
  26 static inline struct nf_icmp_net *icmp_pernet(struct net *net)
  27 {
  28         return &net->ct.nf_ct_proto.icmp;
  29 }
  30
  31 static bool icmp_pkt_to_tuple(const struct sk_buff *skb, unsigned int dataoff,
  32                               struct nf_conntrack_tuple *tuple)
  33 {
  34         const struct icmphdr *hp;
  35         struct icmphdr _hdr;
  36
  37         hp = skb_header_pointer(skb, dataoff, sizeof(_hdr), &_hdr);
  38         if (hp == NULL)
  39                 return false;
  40
  41         tuple->dst.u.icmp.type = hp->type;
  42         tuple->src.u.icmp.id = hp->un.echo.id;
  43         tuple->dst.u.icmp.code = hp->code;
  44
  45         return true;
  46 }
  47
  48 /* Add 1; spaces filled with 0. */
  49 static const u_int8_t invmap[] = {
  50         [ICMP_ECHO] = ICMP_ECHOREPLY + 1,
  51         [ICMP_ECHOREPLY] = ICMP_ECHO + 1,
  52         [ICMP_TIMESTAMP] = ICMP_TIMESTAMPREPLY + 1,
  53         [ICMP_TIMESTAMPREPLY] = ICMP_TIMESTAMP + 1,
  54         [ICMP_INFO_REQUEST] = ICMP_INFO_REPLY + 1,
  55         [ICMP_INFO_REPLY] = ICMP_INFO_REQUEST + 1,
  56         [ICMP_ADDRESS] = ICMP_ADDRESSREPLY + 1,
  57         [ICMP_ADDRESSREPLY] = ICMP_ADDRESS + 1
  58 };
  59
  60 static bool icmp_invert_tuple(struct nf_conntrack_tuple *tuple,
  61                               const struct nf_conntrack_tuple *orig)
  62 {
  63         if (orig->dst.u.icmp.type >= sizeof(invmap) ||
  64             !invmap[orig->dst.u.icmp.type])
  65                 return false;
  66
  67         tuple->src.u.icmp.id = orig->src.u.icmp.id;
  68         tuple->dst.u.icmp.type = invmap[orig->dst.u.icmp.type] - 1;
  69         tuple->dst.u.icmp.code = orig->dst.u.icmp.code;
  70         return true;
  71 }
  72
  73 /* Print out the per-protocol part of the tuple. */
  74 static int icmp_print_tuple(struct seq_file *s,
  75                             const struct nf_conntrack_tuple *tuple)
  76 {
  77         return seq_printf(s, "type=%u code=%u id=%u ",
  78                           tuple->dst.u.icmp.type,
  79                           tuple->dst.u.icmp.code,
  80                           ntohs(tuple->src.u.icmp.id));
  81 }
  82
  83 static unsigned int *icmp_get_timeouts(struct net *net)
  84 {
  85         return &icmp_pernet(net)->timeout;
  86 }
  87
  88 /* Returns verdict for packet, or -1 for invalid. */
  89 static int icmp_packet(struct nf_conn *ct,
  90                        const struct sk_buff *skb,
  91                        unsigned int dataoff,
  92                        enum ip_conntrack_info ctinfo,
  93                        u_int8_t pf,
  94                        unsigned int hooknum,
  95                        unsigned int *timeout)
  96 {
  97         /* Do not immediately delete the connection after the first
  98            successful reply to avoid excessive conntrackd traffic
  99            and also to handle correctly ICMP echo reply duplicates. */
 100         nf_ct_refresh_acct(ct, ctinfo, skb, *timeout);
 101
 102         return NF_ACCEPT;
 103 }
 104
 105 /* Called when a new connection for this protocol found. */
 106 static bool icmp_new(struct nf_conn *ct, const struct sk_buff *skb,
 107                      unsigned int dataoff, unsigned int *timeouts)
 108 {
 109         static const u_int8_t valid_new[] = {
 110                 [ICMP_ECHO] = 1,
 111                 [ICMP_TIMESTAMP] = 1,
 112                 [ICMP_INFO_REQUEST] = 1,
 113                 [ICMP_ADDRESS] = 1
 114         };
 115
 116         if (ct->tuplehash[0].tuple.dst.u.icmp.type >= sizeof(valid_new) ||
 117             !valid_new[ct->tuplehash[0].tuple.dst.u.icmp.type]) {
 118                 /* Can't create a new ICMP `conn' with this. */
 119                 pr_debug("icmp: can't create new conn with type %u\n",
 120                          ct->tuplehash[0].tuple.dst.u.icmp.type);
 121                 nf_ct_dump_tuple_ip(&ct->tuplehash[0].tuple);
 122                 return false;
 123         }
 124         return true;
 125 }
 126
 127 /* Returns conntrack if it dealt with ICMP, and filled in skb fields */
 128 static int
 129 icmp_error_message(struct net *net, struct nf_conn *tmpl, struct sk_buff *skb,
 130                  enum ip_conntrack_info *ctinfo,
 131                  unsigned int hooknum)
 132 {
 133         struct nf_conntrack_tuple innertuple, origtuple;
 134         const struct nf_conntrack_l4proto *innerproto;
 135         const struct nf_conntrack_tuple_hash *h;
 136         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 137
 138         NF_CT_ASSERT(skb->nfct == NULL);
 139
 140         /* Are they talking about one of our connections? */
 141         if (!nf_ct_get_tuplepr(skb,
 142                                skb_network_offset(skb) + ip_hdrlen(skb)
 143                                                        + sizeof(struct icmphdr),
 144                                PF_INET, &origtuple)) {
 145                 pr_debug("icmp_error_message: failed to get tuple\n");
 146                 return -NF_ACCEPT;
 147         }
 148
 149         /* rcu_read_lock()ed by nf_hook_slow */
 150         innerproto = __nf_ct_l4proto_find(PF_INET, origtuple.dst.protonum);
 151
 152         /* Ordinarily, we'd expect the inverted tupleproto, but it's
 153            been preserved inside the ICMP. */
 154         if (!nf_ct_invert_tuple(&innertuple, &origtuple,
 155                                 &nf_conntrack_l3proto_ipv4, innerproto)) {
 156                 pr_debug("icmp_error_message: no match\n");
 157                 return -NF_ACCEPT;
 158         }
 159
 160         *ctinfo = IP_CT_RELATED;
 161
 162         h = nf_conntrack_find_get(net, zone, &innertuple);
 163         if (!h) {
 164                 pr_debug("icmp_error_message: no match\n");
 165                 return -NF_ACCEPT;
 166         }
 167
 168         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 169                 *ctinfo += IP_CT_IS_REPLY;
 170
 171         /* Update skb to refer to this connection */
 172         skb->nfct = &nf_ct_tuplehash_to_ctrack(h)->ct_general;
 173         skb->nfctinfo = *ctinfo;
 174         return NF_ACCEPT;
 175 }
 176
 177 /* Small and modified version of icmp_rcv */
 178 static int
 179 icmp_error(struct net *net, struct nf_conn *tmpl,
 180            struct sk_buff *skb, unsigned int dataoff,
 181            enum ip_conntrack_info *ctinfo, u_int8_t pf, unsigned int hooknum)
 182 {
 183         const struct icmphdr *icmph;
 184         struct icmphdr _ih;
 185
 186         /* Not enough header? */
 187         icmph = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_ih), &_ih);
 188         if (icmph == NULL) {
 189                 if (LOG_INVALID(net, IPPROTO_ICMP))
 190                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
 191                                       "nf_ct_icmp: short packet ");
 192                 return -NF_ACCEPT;
 193         }
 194
 195         /* See ip_conntrack_proto_tcp.c */
 196         if (net->ct.sysctl_checksum && hooknum == NF_INET_PRE_ROUTING &&
 197             nf_ip_checksum(skb, hooknum, dataoff, 0)) {
 198                 if (LOG_INVALID(net, IPPROTO_ICMP))
 199                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
 200                                       "nf_ct_icmp: bad HW ICMP checksum ");
 201                 return -NF_ACCEPT;
 202         }
 203
 204         /*
 205          *      18 is the highest 'known' ICMP type. Anything else is a mystery
 206          *
 207          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
 208          *                discarded.
 209          */
 210         if (icmph->type > NR_ICMP_TYPES) {
 211                 if (LOG_INVALID(net, IPPROTO_ICMP))
 212                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
 213                                       "nf_ct_icmp: invalid ICMP type ");
 214                 return -NF_ACCEPT;
 215         }
 216
 217         /* Need to track icmp error message? */
 218         if (icmph->type != ICMP_DEST_UNREACH &&
 219             icmph->type != ICMP_SOURCE_QUENCH &&
 220             icmph->type != ICMP_TIME_EXCEEDED &&
 221             icmph->type != ICMP_PARAMETERPROB &&
 222             icmph->type != ICMP_REDIRECT)
 223                 return NF_ACCEPT;
 224
 225         return icmp_error_message(net, tmpl, skb, ctinfo, hooknum);
 226 }
 227
 228 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 229
 230 #include <linux/netfilter/nfnetlink.h>
 231 #include <linux/netfilter/nfnetlink_conntrack.h>
 232
 233 static int icmp_tuple_to_nlattr(struct sk_buff *skb,
 234                                 const struct nf_conntrack_tuple *t)
 235 {
 236         if (nla_put_be16(skb, CTA_PROTO_ICMP_ID, t->src.u.icmp.id) ||
 237             nla_put_u8(skb, CTA_PROTO_ICMP_TYPE, t->dst.u.icmp.type) ||
 238             nla_put_u8(skb, CTA_PROTO_ICMP_CODE, t->dst.u.icmp.code))
 239                 goto nla_put_failure;
 240         return 0;
 241
 242 nla_put_failure:
 243         return -1;
 244 }
 245
 246 static const struct nla_policy icmp_nla_policy[CTA_PROTO_MAX+1] = {
 247         [CTA_PROTO_ICMP_TYPE]   = { .type = NLA_U8 },
 248         [CTA_PROTO_ICMP_CODE]   = { .type = NLA_U8 },
 249         [CTA_PROTO_ICMP_ID]     = { .type = NLA_U16 },
 250 };
 251
 252 static int icmp_nlattr_to_tuple(struct nlattr *tb[],
 253                                 struct nf_conntrack_tuple *tuple)
 254 {
 255         if (!tb[CTA_PROTO_ICMP_TYPE] ||
 256             !tb[CTA_PROTO_ICMP_CODE] ||
 257             !tb[CTA_PROTO_ICMP_ID])
 258                 return -EINVAL;
 259
 260         tuple->dst.u.icmp.type = nla_get_u8(tb[CTA_PROTO_ICMP_TYPE]);
 261         tuple->dst.u.icmp.code = nla_get_u8(tb[CTA_PROTO_ICMP_CODE]);
 262         tuple->src.u.icmp.id = nla_get_be16(tb[CTA_PROTO_ICMP_ID]);
 263
 264         if (tuple->dst.u.icmp.type >= sizeof(invmap) ||
 265             !invmap[tuple->dst.u.icmp.type])
 266                 return -EINVAL;
 267
 268         return 0;
 269 }
 270
 271 static int icmp_nlattr_tuple_size(void)
 272 {
 273         return nla_policy_len(icmp_nla_policy, CTA_PROTO_MAX + 1);
 274 }
 275 #endif
 276
 277 #if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT)
 278
 279 #include <linux/netfilter/nfnetlink.h>
 280 #include <linux/netfilter/nfnetlink_cttimeout.h>
 281
 282 static int icmp_timeout_nlattr_to_obj(struct nlattr *tb[],
 283                                       struct net *net, void *data)
 284 {
 285         unsigned int *timeout = data;
 286         struct nf_icmp_net *in = icmp_pernet(net);
 287
 288         if (tb[CTA_TIMEOUT_ICMP_TIMEOUT]) {
 289                 *timeout =
 290                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_ICMP_TIMEOUT])) * HZ;
 291         } else {
 292                 /* Set default ICMP timeout. */
 293                 *timeout = in->timeout;
 294         }
 295         return 0;
 296 }
 297
 298 static int
 299 icmp_timeout_obj_to_nlattr(struct sk_buff *skb, const void *data)
 300 {
 301         const unsigned int *timeout = data;
 302
 303         if (nla_put_be32(skb, CTA_TIMEOUT_ICMP_TIMEOUT, htonl(*timeout / HZ)))
 304                 goto nla_put_failure;
 305         return 0;
 306
 307 nla_put_failure:
 308         return -ENOSPC;
 309 }
 310
 311 static const struct nla_policy
 312 icmp_timeout_nla_policy[CTA_TIMEOUT_ICMP_MAX+1] = {
 313         [CTA_TIMEOUT_ICMP_TIMEOUT]      = { .type = NLA_U32 },
 314 };
 315 #endif /* CONFIG_NF_CT_NETLINK_TIMEOUT */
 316
 317 #ifdef CONFIG_SYSCTL
 318 static struct ctl_table icmp_sysctl_table[] = {
 319         {
 320                 .procname       = "nf_conntrack_icmp_timeout",
 321                 .maxlen         = sizeof(unsigned int),
 322                 .mode           = 0644,
 323                 .proc_handler   = proc_dointvec_jiffies,
 324         },
 325         { }
 326 };
 327 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
 328 static struct ctl_table icmp_compat_sysctl_table[] = {
 329         {
 330                 .procname       = "ip_conntrack_icmp_timeout",
 331                 .maxlen         = sizeof(unsigned int),
 332                 .mode           = 0644,
 333                 .proc_handler   = proc_dointvec_jiffies,
 334         },
 335         { }
 336 };
 337 #endif /* CONFIG_NF_CONNTRACK_PROC_COMPAT */
 338 #endif /* CONFIG_SYSCTL */
 339
 340 static int icmp_kmemdup_sysctl_table(struct nf_proto_net *pn,
 341                                      struct nf_icmp_net *in)
 342 {
 343 #ifdef CONFIG_SYSCTL
 344         pn->ctl_table = kmemdup(icmp_sysctl_table,
 345                                 sizeof(icmp_sysctl_table),
 346                                 GFP_KERNEL);
 347         if (!pn->ctl_table)
 348                 return -ENOMEM;
 349
 350         pn->ctl_table[0].data = &in->timeout;
 351 #endif
 352         return 0;
 353 }
 354
 355 static int icmp_kmemdup_compat_sysctl_table(struct nf_proto_net *pn,
 356                                             struct nf_icmp_net *in)
 357 {
 358 #ifdef CONFIG_SYSCTL
 359 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
 360         pn->ctl_compat_table = kmemdup(icmp_compat_sysctl_table,
 361                                        sizeof(icmp_compat_sysctl_table),
 362                                        GFP_KERNEL);
 363         if (!pn->ctl_compat_table)
 364                 return -ENOMEM;
 365
 366         pn->ctl_compat_table[0].data = &in->timeout;
 367 #endif
 368 #endif
 369         return 0;
 370 }
 371
 372 static int icmp_init_net(struct net *net, u_int16_t proto)
 373 {
 374         int ret;
 375         struct nf_icmp_net *in = icmp_pernet(net);
 376         struct nf_proto_net *pn = &in->pn;
 377
 378         in->timeout = nf_ct_icmp_timeout;
 379
 380         ret = icmp_kmemdup_compat_sysctl_table(pn, in);
 381         if (ret < 0)
 382                 return ret;
 383
 384         ret = icmp_kmemdup_sysctl_table(pn, in);
 385         if (ret < 0)
 386                 nf_ct_kfree_compat_sysctl_table(pn);
 387
 388         return ret;
 389 }
 390
 391 static struct nf_proto_net *icmp_get_net_proto(struct net *net)
 392 {
 393         return &net->ct.nf_ct_proto.icmp.pn;
 394 }
 395
 396 struct nf_conntrack_l4proto nf_conntrack_l4proto_icmp __read_mostly =
 397 {
 398         .l3proto                = PF_INET,
 399         .l4proto                = IPPROTO_ICMP,
 400         .name                   = "icmp",
 401         .pkt_to_tuple           = icmp_pkt_to_tuple,
 402         .invert_tuple           = icmp_invert_tuple,
 403         .print_tuple            = icmp_print_tuple,
 404         .packet                 = icmp_packet,
 405         .get_timeouts           = icmp_get_timeouts,
 406         .new                    = icmp_new,
 407         .error                  = icmp_error,
 408         .destroy                = NULL,
 409         .me                     = NULL,
 410 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 411         .tuple_to_nlattr        = icmp_tuple_to_nlattr,
 412         .nlattr_tuple_size      = icmp_nlattr_tuple_size,
 413         .nlattr_to_tuple        = icmp_nlattr_to_tuple,
 414         .nla_policy             = icmp_nla_policy,
 415 #endif
 416 #if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT)
 417         .ctnl_timeout           = {
 418                 .nlattr_to_obj  = icmp_timeout_nlattr_to_obj,
 419                 .obj_to_nlattr  = icmp_timeout_obj_to_nlattr,
 420                 .nlattr_max     = CTA_TIMEOUT_ICMP_MAX,
 421                 .obj_size       = sizeof(unsigned int),
 422                 .nla_policy     = icmp_timeout_nla_policy,
 423         },
 424 #endif /* CONFIG_NF_CT_NETLINK_TIMEOUT */
 425         .init_net               = icmp_init_net,
 426         .get_net_proto          = icmp_get_net_proto,
 427 };