net/netfilter/ipvs/ip_vs_proto_udp.c

   1 /*
   2  * ip_vs_proto_udp.c:   UDP load balancing support for IPVS
   3  *
   4  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
   5  *              Julian Anastasov <ja@ssi.bg>
   6  *
   7  *              This program is free software; you can redistribute it and/or
   8  *              modify it under the terms of the GNU General Public License
   9  *              as published by the Free Software Foundation; either version
  10  *              2 of the License, or (at your option) any later version.
  11  *
  12  * Changes:     Hans Schillstrom <hans.schillstrom@ericsson.com>
  13  *              Network name space (netns) aware.
  14  *
  15  */
  16
  17 #define KMSG_COMPONENT "IPVS"
  18 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  19
  20 #include <linux/in.h>
  21 #include <linux/ip.h>
  22 #include <linux/kernel.h>
  23 #include <linux/netfilter.h>
  24 #include <linux/netfilter_ipv4.h>
  25 #include <linux/udp.h>
  26
  27 #include <net/ip_vs.h>
  28 #include <net/ip.h>
  29 #include <net/ip6_checksum.h>
  30
  31 static int
  32 udp_csum_check(int af, struct sk_buff *skb, struct ip_vs_protocol *pp);
  33
  34 static int
  35 udp_conn_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb,
  36                   struct ip_vs_proto_data *pd,
  37                   int *verdict, struct ip_vs_conn **cpp,
  38                   struct ip_vs_iphdr *iph)
  39 {
  40         struct ip_vs_service *svc;
  41         struct udphdr _udph, *uh;
  42         __be16 _ports[2], *ports = NULL;
  43
  44         if (likely(!ip_vs_iph_icmp(iph))) {
  45                 /* IPv6 fragments, only first fragment will hit this */
  46                 uh = skb_header_pointer(skb, iph->len, sizeof(_udph), &_udph);
  47                 if (uh)
  48                         ports = &uh->source;
  49         } else {
  50                 ports = skb_header_pointer(
  51                         skb, iph->len, sizeof(_ports), &_ports);
  52         }
  53
  54         if (!ports) {
  55                 *verdict = NF_DROP;
  56                 return 0;
  57         }
  58
  59         if (likely(!ip_vs_iph_inverse(iph)))
  60                 svc = ip_vs_service_find(ipvs, af, skb->mark, iph->protocol,
  61                                          &iph->daddr, ports[1]);
  62         else
  63                 svc = ip_vs_service_find(ipvs, af, skb->mark, iph->protocol,
  64                                          &iph->saddr, ports[0]);
  65
  66         if (svc) {
  67                 int ignored;
  68
  69                 if (ip_vs_todrop(ipvs)) {
  70                         /*
  71                          * It seems that we are very loaded.
  72                          * We have to drop this packet :(
  73                          */
  74                         *verdict = NF_DROP;
  75                         return 0;
  76                 }
  77
  78                 /*
  79                  * Let the virtual server select a real server for the
  80                  * incoming connection, and create a connection entry.
  81                  */
  82                 *cpp = ip_vs_schedule(svc, skb, pd, &ignored, iph);
  83                 if (!*cpp && ignored <= 0) {
  84                         if (!ignored)
  85                                 *verdict = ip_vs_leave(svc, skb, pd, iph);
  86                         else
  87                                 *verdict = NF_DROP;
  88                         return 0;
  89                 }
  90         }
  91         /* NF_ACCEPT */
  92         return 1;
  93 }
  94
  95
  96 static inline void
  97 udp_fast_csum_update(int af, struct udphdr *uhdr,
  98                      const union nf_inet_addr *oldip,
  99                      const union nf_inet_addr *newip,
 100                      __be16 oldport, __be16 newport)
 101 {
 102 #ifdef CONFIG_IP_VS_IPV6
 103         if (af == AF_INET6)
 104                 uhdr->check =
 105                         csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
 106                                          ip_vs_check_diff2(oldport, newport,
 107                                                 ~csum_unfold(uhdr->check))));
 108         else
 109 #endif
 110                 uhdr->check =
 111                         csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
 112                                          ip_vs_check_diff2(oldport, newport,
 113                                                 ~csum_unfold(uhdr->check))));
 114         if (!uhdr->check)
 115                 uhdr->check = CSUM_MANGLED_0;
 116 }
 117
 118 static inline void
 119 udp_partial_csum_update(int af, struct udphdr *uhdr,
 120                      const union nf_inet_addr *oldip,
 121                      const union nf_inet_addr *newip,
 122                      __be16 oldlen, __be16 newlen)
 123 {
 124 #ifdef CONFIG_IP_VS_IPV6
 125         if (af == AF_INET6)
 126                 uhdr->check =
 127                         ~csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
 128                                          ip_vs_check_diff2(oldlen, newlen,
 129                                                 csum_unfold(uhdr->check))));
 130         else
 131 #endif
 132         uhdr->check =
 133                 ~csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
 134                                 ip_vs_check_diff2(oldlen, newlen,
 135                                                 csum_unfold(uhdr->check))));
 136 }
 137
 138
 139 static int
 140 udp_snat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp,
 141                  struct ip_vs_conn *cp, struct ip_vs_iphdr *iph)
 142 {
 143         struct udphdr *udph;
 144         unsigned int udphoff = iph->len;
 145         int oldlen;
 146         int payload_csum = 0;
 147
 148 #ifdef CONFIG_IP_VS_IPV6
 149         if (cp->af == AF_INET6 && iph->fragoffs)
 150                 return 1;
 151 #endif
 152         oldlen = skb->len - udphoff;
 153
 154         /* csum_check requires unshared skb */
 155         if (!skb_make_writable(skb, udphoff+sizeof(*udph)))
 156                 return 0;
 157
 158         if (unlikely(cp->app != NULL)) {
 159                 int ret;
 160
 161                 /* Some checks before mangling */
 162                 if (!udp_csum_check(cp->af, skb, pp))
 163                         return 0;
 164
 165                 /*
 166                  *      Call application helper if needed
 167                  */
 168                 if (!(ret = ip_vs_app_pkt_out(cp, skb, iph)))
 169                         return 0;
 170                 /* ret=2: csum update is needed after payload mangling */
 171                 if (ret == 1)
 172                         oldlen = skb->len - udphoff;
 173                 else
 174                         payload_csum = 1;
 175         }
 176
 177         udph = (void *)skb_network_header(skb) + udphoff;
 178         udph->source = cp->vport;
 179
 180         /*
 181          *      Adjust UDP checksums
 182          */
 183         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 184                 udp_partial_csum_update(cp->af, udph, &cp->daddr, &cp->vaddr,
 185                                         htons(oldlen),
 186                                         htons(skb->len - udphoff));
 187         } else if (!payload_csum && (udph->check != 0)) {
 188                 /* Only port and addr are changed, do fast csum update */
 189                 udp_fast_csum_update(cp->af, udph, &cp->daddr, &cp->vaddr,
 190                                      cp->dport, cp->vport);
 191                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 192                         skb->ip_summed = cp->app ?
 193                                          CHECKSUM_UNNECESSARY : CHECKSUM_NONE;
 194         } else {
 195                 /* full checksum calculation */
 196                 udph->check = 0;
 197                 skb->csum = skb_checksum(skb, udphoff, skb->len - udphoff, 0);
 198 #ifdef CONFIG_IP_VS_IPV6
 199                 if (cp->af == AF_INET6)
 200                         udph->check = csum_ipv6_magic(&cp->vaddr.in6,
 201                                                       &cp->caddr.in6,
 202                                                       skb->len - udphoff,
 203                                                       cp->protocol, skb->csum);
 204                 else
 205 #endif
 206                         udph->check = csum_tcpudp_magic(cp->vaddr.ip,
 207                                                         cp->caddr.ip,
 208                                                         skb->len - udphoff,
 209                                                         cp->protocol,
 210                                                         skb->csum);
 211                 if (udph->check == 0)
 212                         udph->check = CSUM_MANGLED_0;
 213                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 214                 IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%zd)\n",
 215                           pp->name, udph->check,
 216                           (char*)&(udph->check) - (char*)udph);
 217         }
 218         return 1;
 219 }
 220
 221
 222 static int
 223 udp_dnat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp,
 224                  struct ip_vs_conn *cp, struct ip_vs_iphdr *iph)
 225 {
 226         struct udphdr *udph;
 227         unsigned int udphoff = iph->len;
 228         int oldlen;
 229         int payload_csum = 0;
 230
 231 #ifdef CONFIG_IP_VS_IPV6
 232         if (cp->af == AF_INET6 && iph->fragoffs)
 233                 return 1;
 234 #endif
 235         oldlen = skb->len - udphoff;
 236
 237         /* csum_check requires unshared skb */
 238         if (!skb_make_writable(skb, udphoff+sizeof(*udph)))
 239                 return 0;
 240
 241         if (unlikely(cp->app != NULL)) {
 242                 int ret;
 243
 244                 /* Some checks before mangling */
 245                 if (!udp_csum_check(cp->af, skb, pp))
 246                         return 0;
 247
 248                 /*
 249                  *      Attempt ip_vs_app call.
 250                  *      It will fix ip_vs_conn
 251                  */
 252                 if (!(ret = ip_vs_app_pkt_in(cp, skb, iph)))
 253                         return 0;
 254                 /* ret=2: csum update is needed after payload mangling */
 255                 if (ret == 1)
 256                         oldlen = skb->len - udphoff;
 257                 else
 258                         payload_csum = 1;
 259         }
 260
 261         udph = (void *)skb_network_header(skb) + udphoff;
 262         udph->dest = cp->dport;
 263
 264         /*
 265          *      Adjust UDP checksums
 266          */
 267         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 268                 udp_partial_csum_update(cp->af, udph, &cp->vaddr, &cp->daddr,
 269                                         htons(oldlen),
 270                                         htons(skb->len - udphoff));
 271         } else if (!payload_csum && (udph->check != 0)) {
 272                 /* Only port and addr are changed, do fast csum update */
 273                 udp_fast_csum_update(cp->af, udph, &cp->vaddr, &cp->daddr,
 274                                      cp->vport, cp->dport);
 275                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 276                         skb->ip_summed = cp->app ?
 277                                          CHECKSUM_UNNECESSARY : CHECKSUM_NONE;
 278         } else {
 279                 /* full checksum calculation */
 280                 udph->check = 0;
 281                 skb->csum = skb_checksum(skb, udphoff, skb->len - udphoff, 0);
 282 #ifdef CONFIG_IP_VS_IPV6
 283                 if (cp->af == AF_INET6)
 284                         udph->check = csum_ipv6_magic(&cp->caddr.in6,
 285                                                       &cp->daddr.in6,
 286                                                       skb->len - udphoff,
 287                                                       cp->protocol, skb->csum);
 288                 else
 289 #endif
 290                         udph->check = csum_tcpudp_magic(cp->caddr.ip,
 291                                                         cp->daddr.ip,
 292                                                         skb->len - udphoff,
 293                                                         cp->protocol,
 294                                                         skb->csum);
 295                 if (udph->check == 0)
 296                         udph->check = CSUM_MANGLED_0;
 297                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 298         }
 299         return 1;
 300 }
 301
 302
 303 static int
 304 udp_csum_check(int af, struct sk_buff *skb, struct ip_vs_protocol *pp)
 305 {
 306         struct udphdr _udph, *uh;
 307         unsigned int udphoff;
 308
 309 #ifdef CONFIG_IP_VS_IPV6
 310         if (af == AF_INET6)
 311                 udphoff = sizeof(struct ipv6hdr);
 312         else
 313 #endif
 314                 udphoff = ip_hdrlen(skb);
 315
 316         uh = skb_header_pointer(skb, udphoff, sizeof(_udph), &_udph);
 317         if (uh == NULL)
 318                 return 0;
 319
 320         if (uh->check != 0) {
 321                 switch (skb->ip_summed) {
 322                 case CHECKSUM_NONE:
 323                         skb->csum = skb_checksum(skb, udphoff,
 324                                                  skb->len - udphoff, 0);
 325                         /* fall through */
 326                 case CHECKSUM_COMPLETE:
 327 #ifdef CONFIG_IP_VS_IPV6
 328                         if (af == AF_INET6) {
 329                                 if (csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 330                                                     &ipv6_hdr(skb)->daddr,
 331                                                     skb->len - udphoff,
 332                                                     ipv6_hdr(skb)->nexthdr,
 333                                                     skb->csum)) {
 334                                         IP_VS_DBG_RL_PKT(0, af, pp, skb, 0,
 335                                                          "Failed checksum for");
 336                                         return 0;
 337                                 }
 338                         } else
 339 #endif
 340                                 if (csum_tcpudp_magic(ip_hdr(skb)->saddr,
 341                                                       ip_hdr(skb)->daddr,
 342                                                       skb->len - udphoff,
 343                                                       ip_hdr(skb)->protocol,
 344                                                       skb->csum)) {
 345                                         IP_VS_DBG_RL_PKT(0, af, pp, skb, 0,
 346                                                          "Failed checksum for");
 347                                         return 0;
 348                                 }
 349                         break;
 350                 default:
 351                         /* No need to checksum. */
 352                         break;
 353                 }
 354         }
 355         return 1;
 356 }
 357
 358 static inline __u16 udp_app_hashkey(__be16 port)
 359 {
 360         return (((__force u16)port >> UDP_APP_TAB_BITS) ^ (__force u16)port)
 361                 & UDP_APP_TAB_MASK;
 362 }
 363
 364
 365 static int udp_register_app(struct netns_ipvs *ipvs, struct ip_vs_app *inc)
 366 {
 367         struct ip_vs_app *i;
 368         __u16 hash;
 369         __be16 port = inc->port;
 370         int ret = 0;
 371         struct ip_vs_proto_data *pd = ip_vs_proto_data_get(ipvs, IPPROTO_UDP);
 372
 373         hash = udp_app_hashkey(port);
 374
 375         list_for_each_entry(i, &ipvs->udp_apps[hash], p_list) {
 376                 if (i->port == port) {
 377                         ret = -EEXIST;
 378                         goto out;
 379                 }
 380         }
 381         list_add_rcu(&inc->p_list, &ipvs->udp_apps[hash]);
 382         atomic_inc(&pd->appcnt);
 383
 384   out:
 385         return ret;
 386 }
 387
 388
 389 static void
 390 udp_unregister_app(struct netns_ipvs *ipvs, struct ip_vs_app *inc)
 391 {
 392         struct ip_vs_proto_data *pd = ip_vs_proto_data_get(ipvs, IPPROTO_UDP);
 393
 394         atomic_dec(&pd->appcnt);
 395         list_del_rcu(&inc->p_list);
 396 }
 397
 398
 399 static int udp_app_conn_bind(struct ip_vs_conn *cp)
 400 {
 401         struct netns_ipvs *ipvs = cp->ipvs;
 402         int hash;
 403         struct ip_vs_app *inc;
 404         int result = 0;
 405
 406         /* Default binding: bind app only for NAT */
 407         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
 408                 return 0;
 409
 410         /* Lookup application incarnations and bind the right one */
 411         hash = udp_app_hashkey(cp->vport);
 412
 413         list_for_each_entry_rcu(inc, &ipvs->udp_apps[hash], p_list) {
 414                 if (inc->port == cp->vport) {
 415                         if (unlikely(!ip_vs_app_inc_get(inc)))
 416                                 break;
 417
 418                         IP_VS_DBG_BUF(9, "%s(): Binding conn %s:%u->"
 419                                       "%s:%u to app %s on port %u\n",
 420                                       __func__,
 421                                       IP_VS_DBG_ADDR(cp->af, &cp->caddr),
 422                                       ntohs(cp->cport),
 423                                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr),
 424                                       ntohs(cp->vport),
 425                                       inc->name, ntohs(inc->port));
 426
 427                         cp->app = inc;
 428                         if (inc->init_conn)
 429                                 result = inc->init_conn(inc, cp);
 430                         break;
 431                 }
 432         }
 433
 434         return result;
 435 }
 436
 437
 438 static const int udp_timeouts[IP_VS_UDP_S_LAST+1] = {
 439         [IP_VS_UDP_S_NORMAL]            =       5*60*HZ,
 440         [IP_VS_UDP_S_LAST]              =       2*HZ,
 441 };
 442
 443 static const char *const udp_state_name_table[IP_VS_UDP_S_LAST+1] = {
 444         [IP_VS_UDP_S_NORMAL]            =       "UDP",
 445         [IP_VS_UDP_S_LAST]              =       "BUG!",
 446 };
 447
 448 static const char * udp_state_name(int state)
 449 {
 450         if (state >= IP_VS_UDP_S_LAST)
 451                 return "ERR!";
 452         return udp_state_name_table[state] ? udp_state_name_table[state] : "?";
 453 }
 454
 455 static void
 456 udp_state_transition(struct ip_vs_conn *cp, int direction,
 457                      const struct sk_buff *skb,
 458                      struct ip_vs_proto_data *pd)
 459 {
 460         if (unlikely(!pd)) {
 461                 pr_err("UDP no ns data\n");
 462                 return;
 463         }
 464
 465         cp->timeout = pd->timeout_table[IP_VS_UDP_S_NORMAL];
 466         if (direction == IP_VS_DIR_OUTPUT)
 467                 ip_vs_control_assure_ct(cp);
 468 }
 469
 470 static int __udp_init(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd)
 471 {
 472         ip_vs_init_hash_table(ipvs->udp_apps, UDP_APP_TAB_SIZE);
 473         pd->timeout_table = ip_vs_create_timeout_table((int *)udp_timeouts,
 474                                                         sizeof(udp_timeouts));
 475         if (!pd->timeout_table)
 476                 return -ENOMEM;
 477         return 0;
 478 }
 479
 480 static void __udp_exit(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd)
 481 {
 482         kfree(pd->timeout_table);
 483 }
 484
 485
 486 struct ip_vs_protocol ip_vs_protocol_udp = {
 487         .name =                 "UDP",
 488         .protocol =             IPPROTO_UDP,
 489         .num_states =           IP_VS_UDP_S_LAST,
 490         .dont_defrag =          0,
 491         .init =                 NULL,
 492         .exit =                 NULL,
 493         .init_netns =           __udp_init,
 494         .exit_netns =           __udp_exit,
 495         .conn_schedule =        udp_conn_schedule,
 496         .conn_in_get =          ip_vs_conn_in_get_proto,
 497         .conn_out_get =         ip_vs_conn_out_get_proto,
 498         .snat_handler =         udp_snat_handler,
 499         .dnat_handler =         udp_dnat_handler,
 500         .state_transition =     udp_state_transition,
 501         .state_name =           udp_state_name,
 502         .register_app =         udp_register_app,
 503         .unregister_app =       udp_unregister_app,
 504         .app_conn_bind =        udp_app_conn_bind,
 505         .debug_packet =         ip_vs_tcpudp_debug_packet,
 506         .timeout_change =       NULL,
 507 };