net/netfilter/nf_conntrack_proto_icmp.c

   1 /* (C) 1999-2001 Paul `Rusty' Russell
   2  * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
   3  * (C) 2006-2010 Patrick McHardy <kaber@trash.net>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License version 2 as
   7  * published by the Free Software Foundation.
   8  */
   9
  10 #include <linux/types.h>
  11 #include <linux/timer.h>
  12 #include <linux/netfilter.h>
  13 #include <linux/in.h>
  14 #include <linux/icmp.h>
  15 #include <linux/seq_file.h>
  16 #include <net/ip.h>
  17 #include <net/checksum.h>
  18 #include <linux/netfilter_ipv4.h>
  19 #include <net/netfilter/nf_conntrack_tuple.h>
  20 #include <net/netfilter/nf_conntrack_l4proto.h>
  21 #include <net/netfilter/nf_conntrack_core.h>
  22 #include <net/netfilter/nf_conntrack_timeout.h>
  23 #include <net/netfilter/nf_conntrack_zones.h>
  24 #include <net/netfilter/nf_log.h>
  25
  26 static const unsigned int nf_ct_icmp_timeout = 30*HZ;
  27
  28 static inline struct nf_icmp_net *icmp_pernet(struct net *net)
  29 {
  30         return &net->ct.nf_ct_proto.icmp;
  31 }
  32
  33 static bool icmp_pkt_to_tuple(const struct sk_buff *skb, unsigned int dataoff,
  34                               struct net *net, struct nf_conntrack_tuple *tuple)
  35 {
  36         const struct icmphdr *hp;
  37         struct icmphdr _hdr;
  38
  39         hp = skb_header_pointer(skb, dataoff, sizeof(_hdr), &_hdr);
  40         if (hp == NULL)
  41                 return false;
  42
  43         tuple->dst.u.icmp.type = hp->type;
  44         tuple->src.u.icmp.id = hp->un.echo.id;
  45         tuple->dst.u.icmp.code = hp->code;
  46
  47         return true;
  48 }
  49
  50 /* Add 1; spaces filled with 0. */
  51 static const u_int8_t invmap[] = {
  52         [ICMP_ECHO] = ICMP_ECHOREPLY + 1,
  53         [ICMP_ECHOREPLY] = ICMP_ECHO + 1,
  54         [ICMP_TIMESTAMP] = ICMP_TIMESTAMPREPLY + 1,
  55         [ICMP_TIMESTAMPREPLY] = ICMP_TIMESTAMP + 1,
  56         [ICMP_INFO_REQUEST] = ICMP_INFO_REPLY + 1,
  57         [ICMP_INFO_REPLY] = ICMP_INFO_REQUEST + 1,
  58         [ICMP_ADDRESS] = ICMP_ADDRESSREPLY + 1,
  59         [ICMP_ADDRESSREPLY] = ICMP_ADDRESS + 1
  60 };
  61
  62 static bool icmp_invert_tuple(struct nf_conntrack_tuple *tuple,
  63                               const struct nf_conntrack_tuple *orig)
  64 {
  65         if (orig->dst.u.icmp.type >= sizeof(invmap) ||
  66             !invmap[orig->dst.u.icmp.type])
  67                 return false;
  68
  69         tuple->src.u.icmp.id = orig->src.u.icmp.id;
  70         tuple->dst.u.icmp.type = invmap[orig->dst.u.icmp.type] - 1;
  71         tuple->dst.u.icmp.code = orig->dst.u.icmp.code;
  72         return true;
  73 }
  74
  75 /* Returns verdict for packet, or -1 for invalid. */
  76 static int icmp_packet(struct nf_conn *ct,
  77                        struct sk_buff *skb,
  78                        unsigned int dataoff,
  79                        enum ip_conntrack_info ctinfo,
  80                        const struct nf_hook_state *state)
  81 {
  82         /* Do not immediately delete the connection after the first
  83            successful reply to avoid excessive conntrackd traffic
  84            and also to handle correctly ICMP echo reply duplicates. */
  85         unsigned int *timeout = nf_ct_timeout_lookup(ct);
  86         static const u_int8_t valid_new[] = {
  87                 [ICMP_ECHO] = 1,
  88                 [ICMP_TIMESTAMP] = 1,
  89                 [ICMP_INFO_REQUEST] = 1,
  90                 [ICMP_ADDRESS] = 1
  91         };
  92
  93         if (state->pf != NFPROTO_IPV4)
  94                 return -NF_ACCEPT;
  95
  96         if (ct->tuplehash[0].tuple.dst.u.icmp.type >= sizeof(valid_new) ||
  97             !valid_new[ct->tuplehash[0].tuple.dst.u.icmp.type]) {
  98                 /* Can't create a new ICMP `conn' with this. */
  99                 pr_debug("icmp: can't create new conn with type %u\n",
 100                          ct->tuplehash[0].tuple.dst.u.icmp.type);
 101                 nf_ct_dump_tuple_ip(&ct->tuplehash[0].tuple);
 102                 return -NF_ACCEPT;
 103         }
 104
 105         if (!timeout)
 106                 timeout = &icmp_pernet(nf_ct_net(ct))->timeout;
 107
 108         nf_ct_refresh_acct(ct, ctinfo, skb, *timeout);
 109         return NF_ACCEPT;
 110 }
 111
 112 /* Returns conntrack if it dealt with ICMP, and filled in skb fields */
 113 static int
 114 icmp_error_message(struct nf_conn *tmpl, struct sk_buff *skb,
 115                    const struct nf_hook_state *state)
 116 {
 117         struct nf_conntrack_tuple innertuple, origtuple;
 118         const struct nf_conntrack_l4proto *innerproto;
 119         const struct nf_conntrack_tuple_hash *h;
 120         const struct nf_conntrack_zone *zone;
 121         enum ip_conntrack_info ctinfo;
 122         struct nf_conntrack_zone tmp;
 123
 124         WARN_ON(skb_nfct(skb));
 125         zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
 126
 127         /* Are they talking about one of our connections? */
 128         if (!nf_ct_get_tuplepr(skb,
 129                                skb_network_offset(skb) + ip_hdrlen(skb)
 130                                                        + sizeof(struct icmphdr),
 131                                PF_INET, state->net, &origtuple)) {
 132                 pr_debug("icmp_error_message: failed to get tuple\n");
 133                 return -NF_ACCEPT;
 134         }
 135
 136         /* rcu_read_lock()ed by nf_hook_thresh */
 137         innerproto = __nf_ct_l4proto_find(origtuple.dst.protonum);
 138
 139         /* Ordinarily, we'd expect the inverted tupleproto, but it's
 140            been preserved inside the ICMP. */
 141         if (!nf_ct_invert_tuple(&innertuple, &origtuple, innerproto)) {
 142                 pr_debug("icmp_error_message: no match\n");
 143                 return -NF_ACCEPT;
 144         }
 145
 146         ctinfo = IP_CT_RELATED;
 147
 148         h = nf_conntrack_find_get(state->net, zone, &innertuple);
 149         if (!h) {
 150                 pr_debug("icmp_error_message: no match\n");
 151                 return -NF_ACCEPT;
 152         }
 153
 154         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 155                 ctinfo += IP_CT_IS_REPLY;
 156
 157         /* Update skb to refer to this connection */
 158         nf_ct_set(skb, nf_ct_tuplehash_to_ctrack(h), ctinfo);
 159         return NF_ACCEPT;
 160 }
 161
 162 static void icmp_error_log(const struct sk_buff *skb,
 163                            const struct nf_hook_state *state,
 164                            const char *msg)
 165 {
 166         nf_l4proto_log_invalid(skb, state->net, state->pf,
 167                                IPPROTO_ICMP, "%s", msg);
 168 }
 169
 170 /* Small and modified version of icmp_rcv */
 171 int nf_conntrack_icmpv4_error(struct nf_conn *tmpl,
 172                               struct sk_buff *skb, unsigned int dataoff,
 173                               const struct nf_hook_state *state)
 174 {
 175         const struct icmphdr *icmph;
 176         struct icmphdr _ih;
 177
 178         /* Not enough header? */
 179         icmph = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_ih), &_ih);
 180         if (icmph == NULL) {
 181                 icmp_error_log(skb, state, "short packet");
 182                 return -NF_ACCEPT;
 183         }
 184
 185         /* See ip_conntrack_proto_tcp.c */
 186         if (state->net->ct.sysctl_checksum &&
 187             state->hook == NF_INET_PRE_ROUTING &&
 188             nf_ip_checksum(skb, state->hook, dataoff, 0)) {
 189                 icmp_error_log(skb, state, "bad hw icmp checksum");
 190                 return -NF_ACCEPT;
 191         }
 192
 193         /*
 194          *      18 is the highest 'known' ICMP type. Anything else is a mystery
 195          *
 196          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
 197          *                discarded.
 198          */
 199         if (icmph->type > NR_ICMP_TYPES) {
 200                 icmp_error_log(skb, state, "invalid icmp type");
 201                 return -NF_ACCEPT;
 202         }
 203
 204         /* Need to track icmp error message? */
 205         if (icmph->type != ICMP_DEST_UNREACH &&
 206             icmph->type != ICMP_SOURCE_QUENCH &&
 207             icmph->type != ICMP_TIME_EXCEEDED &&
 208             icmph->type != ICMP_PARAMETERPROB &&
 209             icmph->type != ICMP_REDIRECT)
 210                 return NF_ACCEPT;
 211
 212         return icmp_error_message(tmpl, skb, state);
 213 }
 214
 215 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 216
 217 #include <linux/netfilter/nfnetlink.h>
 218 #include <linux/netfilter/nfnetlink_conntrack.h>
 219
 220 static int icmp_tuple_to_nlattr(struct sk_buff *skb,
 221                                 const struct nf_conntrack_tuple *t)
 222 {
 223         if (nla_put_be16(skb, CTA_PROTO_ICMP_ID, t->src.u.icmp.id) ||
 224             nla_put_u8(skb, CTA_PROTO_ICMP_TYPE, t->dst.u.icmp.type) ||
 225             nla_put_u8(skb, CTA_PROTO_ICMP_CODE, t->dst.u.icmp.code))
 226                 goto nla_put_failure;
 227         return 0;
 228
 229 nla_put_failure:
 230         return -1;
 231 }
 232
 233 static const struct nla_policy icmp_nla_policy[CTA_PROTO_MAX+1] = {
 234         [CTA_PROTO_ICMP_TYPE]   = { .type = NLA_U8 },
 235         [CTA_PROTO_ICMP_CODE]   = { .type = NLA_U8 },
 236         [CTA_PROTO_ICMP_ID]     = { .type = NLA_U16 },
 237 };
 238
 239 static int icmp_nlattr_to_tuple(struct nlattr *tb[],
 240                                 struct nf_conntrack_tuple *tuple)
 241 {
 242         if (!tb[CTA_PROTO_ICMP_TYPE] ||
 243             !tb[CTA_PROTO_ICMP_CODE] ||
 244             !tb[CTA_PROTO_ICMP_ID])
 245                 return -EINVAL;
 246
 247         tuple->dst.u.icmp.type = nla_get_u8(tb[CTA_PROTO_ICMP_TYPE]);
 248         tuple->dst.u.icmp.code = nla_get_u8(tb[CTA_PROTO_ICMP_CODE]);
 249         tuple->src.u.icmp.id = nla_get_be16(tb[CTA_PROTO_ICMP_ID]);
 250
 251         if (tuple->dst.u.icmp.type >= sizeof(invmap) ||
 252             !invmap[tuple->dst.u.icmp.type])
 253                 return -EINVAL;
 254
 255         return 0;
 256 }
 257
 258 static unsigned int icmp_nlattr_tuple_size(void)
 259 {
 260         static unsigned int size __read_mostly;
 261
 262         if (!size)
 263                 size = nla_policy_len(icmp_nla_policy, CTA_PROTO_MAX + 1);
 264
 265         return size;
 266 }
 267 #endif
 268
 269 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
 270
 271 #include <linux/netfilter/nfnetlink.h>
 272 #include <linux/netfilter/nfnetlink_cttimeout.h>
 273
 274 static int icmp_timeout_nlattr_to_obj(struct nlattr *tb[],
 275                                       struct net *net, void *data)
 276 {
 277         unsigned int *timeout = data;
 278         struct nf_icmp_net *in = icmp_pernet(net);
 279
 280         if (tb[CTA_TIMEOUT_ICMP_TIMEOUT]) {
 281                 if (!timeout)
 282                         timeout = &in->timeout;
 283                 *timeout =
 284                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_ICMP_TIMEOUT])) * HZ;
 285         } else if (timeout) {
 286                 /* Set default ICMP timeout. */
 287                 *timeout = in->timeout;
 288         }
 289         return 0;
 290 }
 291
 292 static int
 293 icmp_timeout_obj_to_nlattr(struct sk_buff *skb, const void *data)
 294 {
 295         const unsigned int *timeout = data;
 296
 297         if (nla_put_be32(skb, CTA_TIMEOUT_ICMP_TIMEOUT, htonl(*timeout / HZ)))
 298                 goto nla_put_failure;
 299         return 0;
 300
 301 nla_put_failure:
 302         return -ENOSPC;
 303 }
 304
 305 static const struct nla_policy
 306 icmp_timeout_nla_policy[CTA_TIMEOUT_ICMP_MAX+1] = {
 307         [CTA_TIMEOUT_ICMP_TIMEOUT]      = { .type = NLA_U32 },
 308 };
 309 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
 310
 311 #ifdef CONFIG_SYSCTL
 312 static struct ctl_table icmp_sysctl_table[] = {
 313         {
 314                 .procname       = "nf_conntrack_icmp_timeout",
 315                 .maxlen         = sizeof(unsigned int),
 316                 .mode           = 0644,
 317                 .proc_handler   = proc_dointvec_jiffies,
 318         },
 319         { }
 320 };
 321 #endif /* CONFIG_SYSCTL */
 322
 323 static int icmp_kmemdup_sysctl_table(struct nf_proto_net *pn,
 324                                      struct nf_icmp_net *in)
 325 {
 326 #ifdef CONFIG_SYSCTL
 327         pn->ctl_table = kmemdup(icmp_sysctl_table,
 328                                 sizeof(icmp_sysctl_table),
 329                                 GFP_KERNEL);
 330         if (!pn->ctl_table)
 331                 return -ENOMEM;
 332
 333         pn->ctl_table[0].data = &in->timeout;
 334 #endif
 335         return 0;
 336 }
 337
 338 static int icmp_init_net(struct net *net)
 339 {
 340         struct nf_icmp_net *in = icmp_pernet(net);
 341         struct nf_proto_net *pn = &in->pn;
 342
 343         in->timeout = nf_ct_icmp_timeout;
 344
 345         return icmp_kmemdup_sysctl_table(pn, in);
 346 }
 347
 348 static struct nf_proto_net *icmp_get_net_proto(struct net *net)
 349 {
 350         return &net->ct.nf_ct_proto.icmp.pn;
 351 }
 352
 353 const struct nf_conntrack_l4proto nf_conntrack_l4proto_icmp =
 354 {
 355         .l4proto                = IPPROTO_ICMP,
 356         .pkt_to_tuple           = icmp_pkt_to_tuple,
 357         .invert_tuple           = icmp_invert_tuple,
 358         .packet                 = icmp_packet,
 359         .destroy                = NULL,
 360         .me                     = NULL,
 361 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 362         .tuple_to_nlattr        = icmp_tuple_to_nlattr,
 363         .nlattr_tuple_size      = icmp_nlattr_tuple_size,
 364         .nlattr_to_tuple        = icmp_nlattr_to_tuple,
 365         .nla_policy             = icmp_nla_policy,
 366 #endif
 367 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
 368         .ctnl_timeout           = {
 369                 .nlattr_to_obj  = icmp_timeout_nlattr_to_obj,
 370                 .obj_to_nlattr  = icmp_timeout_obj_to_nlattr,
 371                 .nlattr_max     = CTA_TIMEOUT_ICMP_MAX,
 372                 .obj_size       = sizeof(unsigned int),
 373                 .nla_policy     = icmp_timeout_nla_policy,
 374         },
 375 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
 376         .init_net               = icmp_init_net,
 377         .get_net_proto          = icmp_get_net_proto,
 378 };