net/ipv4/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:
  13  *
  14  */
  15
  16 #include <linux/in.h>
  17 #include <linux/ip.h>
  18 #include <linux/kernel.h>
  19 #include <linux/netfilter.h>
  20 #include <linux/netfilter_ipv4.h>
  21 #include <linux/udp.h>
  22
  23 #include <net/ip_vs.h>
  24 #include <net/ip.h>
  25
  26 static struct ip_vs_conn *
  27 udp_conn_in_get(const struct sk_buff *skb, struct ip_vs_protocol *pp,
  28                 const struct iphdr *iph, unsigned int proto_off, int inverse)
  29 {
  30         struct ip_vs_conn *cp;
  31         __be16 _ports[2], *pptr;
  32
  33         pptr = skb_header_pointer(skb, proto_off, sizeof(_ports), _ports);
  34         if (pptr == NULL)
  35                 return NULL;
  36
  37         if (likely(!inverse)) {
  38                 cp = ip_vs_conn_in_get(iph->protocol,
  39                                        iph->saddr, pptr[0],
  40                                        iph->daddr, pptr[1]);
  41         } else {
  42                 cp = ip_vs_conn_in_get(iph->protocol,
  43                                        iph->daddr, pptr[1],
  44                                        iph->saddr, pptr[0]);
  45         }
  46
  47         return cp;
  48 }
  49
  50
  51 static struct ip_vs_conn *
  52 udp_conn_out_get(const struct sk_buff *skb, struct ip_vs_protocol *pp,
  53                  const struct iphdr *iph, unsigned int proto_off, int inverse)
  54 {
  55         struct ip_vs_conn *cp;
  56         __be16 _ports[2], *pptr;
  57
  58         pptr = skb_header_pointer(skb, ip_hdrlen(skb),
  59                                   sizeof(_ports), _ports);
  60         if (pptr == NULL)
  61                 return NULL;
  62
  63         if (likely(!inverse)) {
  64                 cp = ip_vs_conn_out_get(iph->protocol,
  65                                         iph->saddr, pptr[0],
  66                                         iph->daddr, pptr[1]);
  67         } else {
  68                 cp = ip_vs_conn_out_get(iph->protocol,
  69                                         iph->daddr, pptr[1],
  70                                         iph->saddr, pptr[0]);
  71         }
  72
  73         return cp;
  74 }
  75
  76
  77 static int
  78 udp_conn_schedule(struct sk_buff *skb, struct ip_vs_protocol *pp,
  79                   int *verdict, struct ip_vs_conn **cpp)
  80 {
  81         struct ip_vs_service *svc;
  82         struct udphdr _udph, *uh;
  83
  84         uh = skb_header_pointer(skb, ip_hdrlen(skb),
  85                                 sizeof(_udph), &_udph);
  86         if (uh == NULL) {
  87                 *verdict = NF_DROP;
  88                 return 0;
  89         }
  90
  91         if ((svc = ip_vs_service_get(skb->mark, ip_hdr(skb)->protocol,
  92                                      ip_hdr(skb)->daddr, uh->dest))) {
  93                 if (ip_vs_todrop()) {
  94                         /*
  95                          * It seems that we are very loaded.
  96                          * We have to drop this packet :(
  97                          */
  98                         ip_vs_service_put(svc);
  99                         *verdict = NF_DROP;
 100                         return 0;
 101                 }
 102
 103                 /*
 104                  * Let the virtual server select a real server for the
 105                  * incoming connection, and create a connection entry.
 106                  */
 107                 *cpp = ip_vs_schedule(svc, skb);
 108                 if (!*cpp) {
 109                         *verdict = ip_vs_leave(svc, skb, pp);
 110                         return 0;
 111                 }
 112                 ip_vs_service_put(svc);
 113         }
 114         return 1;
 115 }
 116
 117
 118 static inline void
 119 udp_fast_csum_update(struct udphdr *uhdr, __be32 oldip, __be32 newip,
 120                      __be16 oldport, __be16 newport)
 121 {
 122         uhdr->check =
 123                 csum_fold(ip_vs_check_diff4(oldip, newip,
 124                                  ip_vs_check_diff2(oldport, newport,
 125                                         ~csum_unfold(uhdr->check))));
 126         if (!uhdr->check)
 127                 uhdr->check = CSUM_MANGLED_0;
 128 }
 129
 130 static int
 131 udp_snat_handler(struct sk_buff *skb,
 132                  struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
 133 {
 134         struct udphdr *udph;
 135         const unsigned int udphoff = ip_hdrlen(skb);
 136
 137         /* csum_check requires unshared skb */
 138         if (!skb_make_writable(skb, udphoff+sizeof(*udph)))
 139                 return 0;
 140
 141         if (unlikely(cp->app != NULL)) {
 142                 /* Some checks before mangling */
 143                 if (pp->csum_check && !pp->csum_check(skb, pp))
 144                         return 0;
 145
 146                 /*
 147                  *      Call application helper if needed
 148                  */
 149                 if (!ip_vs_app_pkt_out(cp, skb))
 150                         return 0;
 151         }
 152
 153         udph = (void *)ip_hdr(skb) + udphoff;
 154         udph->source = cp->vport;
 155
 156         /*
 157          *      Adjust UDP checksums
 158          */
 159         if (!cp->app && (udph->check != 0)) {
 160                 /* Only port and addr are changed, do fast csum update */
 161                 udp_fast_csum_update(udph, cp->daddr, cp->vaddr,
 162                                      cp->dport, cp->vport);
 163                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 164                         skb->ip_summed = CHECKSUM_NONE;
 165         } else {
 166                 /* full checksum calculation */
 167                 udph->check = 0;
 168                 skb->csum = skb_checksum(skb, udphoff, skb->len - udphoff, 0);
 169                 udph->check = csum_tcpudp_magic(cp->vaddr, cp->caddr,
 170                                                 skb->len - udphoff,
 171                                                 cp->protocol, skb->csum);
 172                 if (udph->check == 0)
 173                         udph->check = CSUM_MANGLED_0;
 174                 IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%zd)\n",
 175                           pp->name, udph->check,
 176                           (char*)&(udph->check) - (char*)udph);
 177         }
 178         return 1;
 179 }
 180
 181
 182 static int
 183 udp_dnat_handler(struct sk_buff *skb,
 184                  struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
 185 {
 186         struct udphdr *udph;
 187         unsigned int udphoff = ip_hdrlen(skb);
 188
 189         /* csum_check requires unshared skb */
 190         if (!skb_make_writable(skb, udphoff+sizeof(*udph)))
 191                 return 0;
 192
 193         if (unlikely(cp->app != NULL)) {
 194                 /* Some checks before mangling */
 195                 if (pp->csum_check && !pp->csum_check(skb, pp))
 196                         return 0;
 197
 198                 /*
 199                  *      Attempt ip_vs_app call.
 200                  *      It will fix ip_vs_conn
 201                  */
 202                 if (!ip_vs_app_pkt_in(cp, skb))
 203                         return 0;
 204         }
 205
 206         udph = (void *)ip_hdr(skb) + udphoff;
 207         udph->dest = cp->dport;
 208
 209         /*
 210          *      Adjust UDP checksums
 211          */
 212         if (!cp->app && (udph->check != 0)) {
 213                 /* Only port and addr are changed, do fast csum update */
 214                 udp_fast_csum_update(udph, cp->vaddr, cp->daddr,
 215                                      cp->vport, cp->dport);
 216                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 217                         skb->ip_summed = CHECKSUM_NONE;
 218         } else {
 219                 /* full checksum calculation */
 220                 udph->check = 0;
 221                 skb->csum = skb_checksum(skb, udphoff, skb->len - udphoff, 0);
 222                 udph->check = csum_tcpudp_magic(cp->caddr, cp->daddr,
 223                                                 skb->len - udphoff,
 224                                                 cp->protocol, skb->csum);
 225                 if (udph->check == 0)
 226                         udph->check = CSUM_MANGLED_0;
 227                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 228         }
 229         return 1;
 230 }
 231
 232
 233 static int
 234 udp_csum_check(struct sk_buff *skb, struct ip_vs_protocol *pp)
 235 {
 236         struct udphdr _udph, *uh;
 237         const unsigned int udphoff = ip_hdrlen(skb);
 238
 239         uh = skb_header_pointer(skb, udphoff, sizeof(_udph), &_udph);
 240         if (uh == NULL)
 241                 return 0;
 242
 243         if (uh->check != 0) {
 244                 switch (skb->ip_summed) {
 245                 case CHECKSUM_NONE:
 246                         skb->csum = skb_checksum(skb, udphoff,
 247                                                  skb->len - udphoff, 0);
 248                 case CHECKSUM_COMPLETE:
 249                         if (csum_tcpudp_magic(ip_hdr(skb)->saddr,
 250                                               ip_hdr(skb)->daddr,
 251                                               skb->len - udphoff,
 252                                               ip_hdr(skb)->protocol,
 253                                               skb->csum)) {
 254                                 IP_VS_DBG_RL_PKT(0, pp, skb, 0,
 255                                                  "Failed checksum for");
 256                                 return 0;
 257                         }
 258                         break;
 259                 default:
 260                         /* No need to checksum. */
 261                         break;
 262                 }
 263         }
 264         return 1;
 265 }
 266
 267
 268 /*
 269  *      Note: the caller guarantees that only one of register_app,
 270  *      unregister_app or app_conn_bind is called each time.
 271  */
 272
 273 #define UDP_APP_TAB_BITS        4
 274 #define UDP_APP_TAB_SIZE        (1 << UDP_APP_TAB_BITS)
 275 #define UDP_APP_TAB_MASK        (UDP_APP_TAB_SIZE - 1)
 276
 277 static struct list_head udp_apps[UDP_APP_TAB_SIZE];
 278 static DEFINE_SPINLOCK(udp_app_lock);
 279
 280 static inline __u16 udp_app_hashkey(__be16 port)
 281 {
 282         return (((__force u16)port >> UDP_APP_TAB_BITS) ^ (__force u16)port)
 283                 & UDP_APP_TAB_MASK;
 284 }
 285
 286
 287 static int udp_register_app(struct ip_vs_app *inc)
 288 {
 289         struct ip_vs_app *i;
 290         __u16 hash;
 291         __be16 port = inc->port;
 292         int ret = 0;
 293
 294         hash = udp_app_hashkey(port);
 295
 296
 297         spin_lock_bh(&udp_app_lock);
 298         list_for_each_entry(i, &udp_apps[hash], p_list) {
 299                 if (i->port == port) {
 300                         ret = -EEXIST;
 301                         goto out;
 302                 }
 303         }
 304         list_add(&inc->p_list, &udp_apps[hash]);
 305         atomic_inc(&ip_vs_protocol_udp.appcnt);
 306
 307   out:
 308         spin_unlock_bh(&udp_app_lock);
 309         return ret;
 310 }
 311
 312
 313 static void
 314 udp_unregister_app(struct ip_vs_app *inc)
 315 {
 316         spin_lock_bh(&udp_app_lock);
 317         atomic_dec(&ip_vs_protocol_udp.appcnt);
 318         list_del(&inc->p_list);
 319         spin_unlock_bh(&udp_app_lock);
 320 }
 321
 322
 323 static int udp_app_conn_bind(struct ip_vs_conn *cp)
 324 {
 325         int hash;
 326         struct ip_vs_app *inc;
 327         int result = 0;
 328
 329         /* Default binding: bind app only for NAT */
 330         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
 331                 return 0;
 332
 333         /* Lookup application incarnations and bind the right one */
 334         hash = udp_app_hashkey(cp->vport);
 335
 336         spin_lock(&udp_app_lock);
 337         list_for_each_entry(inc, &udp_apps[hash], p_list) {
 338                 if (inc->port == cp->vport) {
 339                         if (unlikely(!ip_vs_app_inc_get(inc)))
 340                                 break;
 341                         spin_unlock(&udp_app_lock);
 342
 343                         IP_VS_DBG(9, "%s: Binding conn %u.%u.%u.%u:%u->"
 344                                   "%u.%u.%u.%u:%u to app %s on port %u\n",
 345                                   __func__,
 346                                   NIPQUAD(cp->caddr), ntohs(cp->cport),
 347                                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
 348                                   inc->name, ntohs(inc->port));
 349                         cp->app = inc;
 350                         if (inc->init_conn)
 351                                 result = inc->init_conn(inc, cp);
 352                         goto out;
 353                 }
 354         }
 355         spin_unlock(&udp_app_lock);
 356
 357   out:
 358         return result;
 359 }
 360
 361
 362 static int udp_timeouts[IP_VS_UDP_S_LAST+1] = {
 363         [IP_VS_UDP_S_NORMAL]            =       5*60*HZ,
 364         [IP_VS_UDP_S_LAST]              =       2*HZ,
 365 };
 366
 367 static char * udp_state_name_table[IP_VS_UDP_S_LAST+1] = {
 368         [IP_VS_UDP_S_NORMAL]            =       "UDP",
 369         [IP_VS_UDP_S_LAST]              =       "BUG!",
 370 };
 371
 372
 373 static int
 374 udp_set_state_timeout(struct ip_vs_protocol *pp, char *sname, int to)
 375 {
 376         return ip_vs_set_state_timeout(pp->timeout_table, IP_VS_UDP_S_LAST,
 377                                        udp_state_name_table, sname, to);
 378 }
 379
 380 static const char * udp_state_name(int state)
 381 {
 382         if (state >= IP_VS_UDP_S_LAST)
 383                 return "ERR!";
 384         return udp_state_name_table[state] ? udp_state_name_table[state] : "?";
 385 }
 386
 387 static int
 388 udp_state_transition(struct ip_vs_conn *cp, int direction,
 389                      const struct sk_buff *skb,
 390                      struct ip_vs_protocol *pp)
 391 {
 392         cp->timeout = pp->timeout_table[IP_VS_UDP_S_NORMAL];
 393         return 1;
 394 }
 395
 396 static void udp_init(struct ip_vs_protocol *pp)
 397 {
 398         IP_VS_INIT_HASH_TABLE(udp_apps);
 399         pp->timeout_table = udp_timeouts;
 400 }
 401
 402 static void udp_exit(struct ip_vs_protocol *pp)
 403 {
 404 }
 405
 406
 407 struct ip_vs_protocol ip_vs_protocol_udp = {
 408         .name =                 "UDP",
 409         .protocol =             IPPROTO_UDP,
 410         .num_states =           IP_VS_UDP_S_LAST,
 411         .dont_defrag =          0,
 412         .init =                 udp_init,
 413         .exit =                 udp_exit,
 414         .conn_schedule =        udp_conn_schedule,
 415         .conn_in_get =          udp_conn_in_get,
 416         .conn_out_get =         udp_conn_out_get,
 417         .snat_handler =         udp_snat_handler,
 418         .dnat_handler =         udp_dnat_handler,
 419         .csum_check =           udp_csum_check,
 420         .state_transition =     udp_state_transition,
 421         .state_name =           udp_state_name,
 422         .register_app =         udp_register_app,
 423         .unregister_app =       udp_unregister_app,
 424         .app_conn_bind =        udp_app_conn_bind,
 425         .debug_packet =         ip_vs_tcpudp_debug_packet,
 426         .timeout_change =       NULL,
 427         .set_state_timeout =    udp_set_state_timeout,
 428 };