net/ipv4/ipvs/ip_vs_xmit.c

   1 /*
   2  * ip_vs_xmit.c: various packet transmitters for IPVS
   3  *
   4  * Version:     $Id: ip_vs_xmit.c,v 1.2 2002/11/30 01:50:35 wensong Exp $
   5  *
   6  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
   7  *              Julian Anastasov <ja@ssi.bg>
   8  *
   9  *              This program is free software; you can redistribute it and/or
  10  *              modify it under the terms of the GNU General Public License
  11  *              as published by the Free Software Foundation; either version
  12  *              2 of the License, or (at your option) any later version.
  13  *
  14  * Changes:
  15  *
  16  */
  17
  18 #include <linux/kernel.h>
  19 #include <linux/ip.h>
  20 #include <linux/tcp.h>                  /* for tcphdr */
  21 #include <net/tcp.h>                    /* for csum_tcpudp_magic */
  22 #include <net/udp.h>
  23 #include <net/icmp.h>                   /* for icmp_send */
  24 #include <net/route.h>                  /* for ip_route_output */
  25 #include <linux/netfilter.h>
  26 #include <linux/netfilter_ipv4.h>
  27
  28 #include <net/ip_vs.h>
  29
  30
  31 /*
  32  *      Destination cache to speed up outgoing route lookup
  33  */
  34 static inline void
  35 __ip_vs_dst_set(struct ip_vs_dest *dest, u32 rtos, struct dst_entry *dst)
  36 {
  37         struct dst_entry *old_dst;
  38
  39         old_dst = dest->dst_cache;
  40         dest->dst_cache = dst;
  41         dest->dst_rtos = rtos;
  42         dst_release(old_dst);
  43 }
  44
  45 static inline struct dst_entry *
  46 __ip_vs_dst_check(struct ip_vs_dest *dest, u32 rtos, u32 cookie)
  47 {
  48         struct dst_entry *dst = dest->dst_cache;
  49
  50         if (!dst)
  51                 return NULL;
  52         if ((dst->obsolete || rtos != dest->dst_rtos) &&
  53             dst->ops->check(dst, cookie) == NULL) {
  54                 dest->dst_cache = 0;
  55                 return NULL;
  56         }
  57         dst_hold(dst);
  58         return dst;
  59 }
  60
  61 static inline struct rtable *
  62 __ip_vs_get_out_rt(struct ip_vs_conn *cp, u32 rtos)
  63 {
  64         struct rtable *rt;                      /* Route to the other host */
  65         struct ip_vs_dest *dest = cp->dest;
  66
  67         if (dest) {
  68                 spin_lock(&dest->dst_lock);
  69                 if (!(rt = (struct rtable *)
  70                       __ip_vs_dst_check(dest, rtos, 0))) {
  71                         struct flowi fl = {
  72                                 .oif = 0,
  73                                 .nl_u = {
  74                                         .ip4_u = {
  75                                                 .daddr = dest->addr,
  76                                                 .saddr = 0,
  77                                                 .tos = rtos, } },
  78                         };
  79
  80                         if (ip_route_output_key(&rt, &fl)) {
  81                                 spin_unlock(&dest->dst_lock);
  82                                 IP_VS_DBG_RL("ip_route_output error, "
  83                                              "dest: %u.%u.%u.%u\n",
  84                                              NIPQUAD(dest->addr));
  85                                 return NULL;
  86                         }
  87                         __ip_vs_dst_set(dest, rtos, dst_clone(&rt->u.dst));
  88                         IP_VS_DBG(10, "new dst %u.%u.%u.%u, refcnt=%d, rtos=%X\n",
  89                                   NIPQUAD(dest->addr),
  90                                   atomic_read(&rt->u.dst.__refcnt), rtos);
  91                 }
  92                 spin_unlock(&dest->dst_lock);
  93         } else {
  94                 struct flowi fl = {
  95                         .oif = 0,
  96                         .nl_u = {
  97                                 .ip4_u = {
  98                                         .daddr = cp->daddr,
  99                                         .saddr = 0,
 100                                         .tos = rtos, } },
 101                 };
 102
 103                 if (ip_route_output_key(&rt, &fl)) {
 104                         IP_VS_DBG_RL("ip_route_output error, dest: "
 105                                      "%u.%u.%u.%u\n", NIPQUAD(cp->daddr));
 106                         return NULL;
 107                 }
 108         }
 109
 110         return rt;
 111 }
 112
 113
 114 /*
 115  *      Release dest->dst_cache before a dest is removed
 116  */
 117 void
 118 ip_vs_dst_reset(struct ip_vs_dest *dest)
 119 {
 120         struct dst_entry *old_dst;
 121
 122         old_dst = dest->dst_cache;
 123         dest->dst_cache = NULL;
 124         dst_release(old_dst);
 125 }
 126
 127
 128 #define IP_VS_XMIT(skb, rt)                             \
 129 do {                                                    \
 130         (skb)->nfcache |= NFC_IPVS_PROPERTY;            \
 131         NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, (skb), NULL,  \
 132                 (rt)->u.dst.dev, dst_output);           \
 133 } while (0)
 134
 135
 136 /*
 137  *      NULL transmitter (do nothing except return NF_ACCEPT)
 138  */
 139 int
 140 ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 141                 struct ip_vs_protocol *pp)
 142 {
 143         /* we do not touch skb and do not need pskb ptr */
 144         return NF_ACCEPT;
 145 }
 146
 147
 148 /*
 149  *      Bypass transmitter
 150  *      Let packets bypass the destination when the destination is not
 151  *      available, it may be only used in transparent cache cluster.
 152  */
 153 int
 154 ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 155                   struct ip_vs_protocol *pp)
 156 {
 157         struct rtable *rt;                      /* Route to the other host */
 158         struct iphdr  *iph = skb->nh.iph;
 159         u8     tos = iph->tos;
 160         int    mtu;
 161         struct flowi fl = {
 162                 .oif = 0,
 163                 .nl_u = {
 164                         .ip4_u = {
 165                                 .daddr = iph->daddr,
 166                                 .saddr = 0,
 167                                 .tos = RT_TOS(tos), } },
 168         };
 169
 170         EnterFunction(10);
 171
 172         if (ip_route_output_key(&rt, &fl)) {
 173                 IP_VS_DBG_RL("ip_vs_bypass_xmit(): ip_route_output error, "
 174                              "dest: %u.%u.%u.%u\n", NIPQUAD(iph->daddr));
 175                 goto tx_error_icmp;
 176         }
 177
 178         /* MTU checking */
 179         mtu = dst_pmtu(&rt->u.dst);
 180         if ((skb->len > mtu) && (iph->frag_off&__constant_htons(IP_DF))) {
 181                 ip_rt_put(rt);
 182                 icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu));
 183                 IP_VS_DBG_RL("ip_vs_bypass_xmit(): frag needed\n");
 184                 goto tx_error;
 185         }
 186
 187         /*
 188          * Call ip_send_check because we are not sure it is called
 189          * after ip_defrag. Is copy-on-write needed?
 190          */
 191         if (unlikely((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)) {
 192                 ip_rt_put(rt);
 193                 return NF_STOLEN;
 194         }
 195         ip_send_check(skb->nh.iph);
 196
 197         /* drop old route */
 198         dst_release(skb->dst);
 199         skb->dst = &rt->u.dst;
 200
 201         /* Another hack: avoid icmp_send in ip_fragment */
 202         skb->local_df = 1;
 203
 204 #ifdef CONFIG_NETFILTER_DEBUG
 205         skb->nf_debug = 0;
 206 #endif /* CONFIG_NETFILTER_DEBUG */
 207         IP_VS_XMIT(skb, rt);
 208
 209         LeaveFunction(10);
 210         return NF_STOLEN;
 211
 212  tx_error_icmp:
 213         dst_link_failure(skb);
 214  tx_error:
 215         kfree_skb(skb);
 216         LeaveFunction(10);
 217         return NF_STOLEN;
 218 }
 219
 220
 221 /*
 222  *      NAT transmitter (only for outside-to-inside nat forwarding)
 223  *      Not used for related ICMP
 224  */
 225 int
 226 ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 227                struct ip_vs_protocol *pp)
 228 {
 229         struct rtable *rt;              /* Route to the other host */
 230         int mtu;
 231         struct iphdr *iph = skb->nh.iph;
 232
 233         EnterFunction(10);
 234
 235         /* check if it is a connection of no-client-port */
 236         if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) {
 237                 __u16 pt;
 238                 if (skb_copy_bits(skb, iph->ihl*4, &pt, sizeof(pt)) < 0)
 239                         goto tx_error;
 240                 ip_vs_conn_fill_cport(cp, pt);
 241                 IP_VS_DBG(10, "filled cport=%d\n", ntohs(pt));
 242         }
 243
 244         if (!(rt = __ip_vs_get_out_rt(cp, RT_TOS(iph->tos))))
 245                 goto tx_error_icmp;
 246
 247         /* MTU checking */
 248         mtu = dst_pmtu(&rt->u.dst);
 249         if ((skb->len > mtu) && (iph->frag_off&__constant_htons(IP_DF))) {
 250                 ip_rt_put(rt);
 251                 icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu));
 252                 IP_VS_DBG_RL_PKT(0, pp, skb, 0, "ip_vs_nat_xmit(): frag needed for");
 253                 goto tx_error;
 254         }
 255
 256         /* copy-on-write the packet before mangling it */
 257         if (!ip_vs_make_skb_writable(&skb, sizeof(struct iphdr)))
 258                 goto tx_error_put;
 259
 260         if (skb_cow(skb, rt->u.dst.dev->hard_header_len))
 261                 goto tx_error_put;
 262
 263         /* drop old route */
 264         dst_release(skb->dst);
 265         skb->dst = &rt->u.dst;
 266
 267         /* mangle the packet */
 268         if (pp->dnat_handler && !pp->dnat_handler(&skb, pp, cp))
 269                 goto tx_error;
 270         skb->nh.iph->daddr = cp->daddr;
 271         ip_send_check(skb->nh.iph);
 272
 273         IP_VS_DBG_PKT(10, pp, skb, 0, "After DNAT");
 274
 275         /* FIXME: when application helper enlarges the packet and the length
 276            is larger than the MTU of outgoing device, there will be still
 277            MTU problem. */
 278
 279         /* Another hack: avoid icmp_send in ip_fragment */
 280         skb->local_df = 1;
 281
 282 #ifdef CONFIG_NETFILTER_DEBUG
 283         skb->nf_debug = 0;
 284 #endif /* CONFIG_NETFILTER_DEBUG */
 285         IP_VS_XMIT(skb, rt);
 286
 287         LeaveFunction(10);
 288         return NF_STOLEN;
 289
 290   tx_error_icmp:
 291         dst_link_failure(skb);
 292   tx_error:
 293         LeaveFunction(10);
 294         kfree_skb(skb);
 295         return NF_STOLEN;
 296   tx_error_put:
 297         ip_rt_put(rt);
 298         goto tx_error;
 299 }
 300
 301
 302 /*
 303  *   IP Tunneling transmitter
 304  *
 305  *   This function encapsulates the packet in a new IP packet, its
 306  *   destination will be set to cp->daddr. Most code of this function
 307  *   is taken from ipip.c.
 308  *
 309  *   It is used in VS/TUN cluster. The load balancer selects a real
 310  *   server from a cluster based on a scheduling algorithm,
 311  *   encapsulates the request packet and forwards it to the selected
 312  *   server. For example, all real servers are configured with
 313  *   "ifconfig tunl0 <Virtual IP Address> up". When the server receives
 314  *   the encapsulated packet, it will decapsulate the packet, processe
 315  *   the request and return the response packets directly to the client
 316  *   without passing the load balancer. This can greatly increase the
 317  *   scalability of virtual server.
 318  *
 319  *   Used for ANY protocol
 320  */
 321 int
 322 ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 323                   struct ip_vs_protocol *pp)
 324 {
 325         struct rtable *rt;                      /* Route to the other host */
 326         struct net_device *tdev;                /* Device to other host */
 327         struct iphdr  *old_iph = skb->nh.iph;
 328         u8     tos = old_iph->tos;
 329         u16    df = old_iph->frag_off;
 330         struct iphdr  *iph;                     /* Our new IP header */
 331         int    max_headroom;                    /* The extra header space needed */
 332         int    mtu;
 333
 334         EnterFunction(10);
 335
 336         if (skb->protocol != __constant_htons(ETH_P_IP)) {
 337                 IP_VS_DBG_RL("ip_vs_tunnel_xmit(): protocol error, "
 338                              "ETH_P_IP: %d, skb protocol: %d\n",
 339                              __constant_htons(ETH_P_IP), skb->protocol);
 340                 goto tx_error;
 341         }
 342
 343         if (!(rt = __ip_vs_get_out_rt(cp, RT_TOS(tos))))
 344                 goto tx_error_icmp;
 345
 346         tdev = rt->u.dst.dev;
 347
 348         mtu = dst_pmtu(&rt->u.dst) - sizeof(struct iphdr);
 349         if (mtu < 68) {
 350                 ip_rt_put(rt);
 351                 IP_VS_DBG_RL("ip_vs_tunnel_xmit(): mtu less than 68\n");
 352                 goto tx_error;
 353         }
 354         if (skb->dst)
 355                 skb->dst->ops->update_pmtu(skb->dst, mtu);
 356
 357         df |= (old_iph->frag_off&__constant_htons(IP_DF));
 358
 359         if ((old_iph->frag_off&__constant_htons(IP_DF))
 360             && mtu < ntohs(old_iph->tot_len)) {
 361                 icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu));
 362                 ip_rt_put(rt);
 363                 IP_VS_DBG_RL("ip_vs_tunnel_xmit(): frag needed\n");
 364                 goto tx_error;
 365         }
 366
 367         /*
 368          * Okay, now see if we can stuff it in the buffer as-is.
 369          */
 370         max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr);
 371
 372         if (skb_headroom(skb) < max_headroom
 373             || skb_cloned(skb) || skb_shared(skb)) {
 374                 struct sk_buff *new_skb =
 375                         skb_realloc_headroom(skb, max_headroom);
 376                 if (!new_skb) {
 377                         ip_rt_put(rt);
 378                         kfree_skb(skb);
 379                         IP_VS_ERR_RL("ip_vs_tunnel_xmit(): no memory\n");
 380                         return NF_STOLEN;
 381                 }
 382                 kfree_skb(skb);
 383                 skb = new_skb;
 384                 old_iph = skb->nh.iph;
 385         }
 386
 387         skb->h.raw = (void *) old_iph;
 388
 389         /* fix old IP header checksum */
 390         ip_send_check(old_iph);
 391
 392         skb->nh.raw = skb_push(skb, sizeof(struct iphdr));
 393         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 394
 395         /* drop old route */
 396         dst_release(skb->dst);
 397         skb->dst = &rt->u.dst;
 398
 399         /*
 400          *      Push down and install the IPIP header.
 401          */
 402         iph                     =       skb->nh.iph;
 403         iph->version            =       4;
 404         iph->ihl                =       sizeof(struct iphdr)>>2;
 405         iph->frag_off           =       df;
 406         iph->protocol           =       IPPROTO_IPIP;
 407         iph->tos                =       tos;
 408         iph->daddr              =       rt->rt_dst;
 409         iph->saddr              =       rt->rt_src;
 410         iph->ttl                =       old_iph->ttl;
 411         iph->tot_len            =       htons(skb->len);
 412         ip_select_ident(iph, &rt->u.dst, NULL);
 413         ip_send_check(iph);
 414
 415         skb->ip_summed = CHECKSUM_NONE;
 416
 417         /* Another hack: avoid icmp_send in ip_fragment */
 418         skb->local_df = 1;
 419
 420 #ifdef CONFIG_NETFILTER_DEBUG
 421         skb->nf_debug = 0;
 422 #endif /* CONFIG_NETFILTER_DEBUG */
 423
 424         IP_VS_XMIT(skb, rt);
 425
 426         LeaveFunction(10);
 427
 428         return NF_STOLEN;
 429
 430   tx_error_icmp:
 431         dst_link_failure(skb);
 432   tx_error:
 433         kfree_skb(skb);
 434         LeaveFunction(10);
 435         return NF_STOLEN;
 436 }
 437
 438
 439 /*
 440  *      Direct Routing transmitter
 441  *      Used for ANY protocol
 442  */
 443 int
 444 ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 445               struct ip_vs_protocol *pp)
 446 {
 447         struct rtable *rt;                      /* Route to the other host */
 448         struct iphdr  *iph = skb->nh.iph;
 449         int    mtu;
 450
 451         EnterFunction(10);
 452
 453         if (!(rt = __ip_vs_get_out_rt(cp, RT_TOS(iph->tos))))
 454                 goto tx_error_icmp;
 455
 456         /* MTU checking */
 457         mtu = dst_pmtu(&rt->u.dst);
 458         if ((iph->frag_off&__constant_htons(IP_DF)) && skb->len > mtu) {
 459                 icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu));
 460                 ip_rt_put(rt);
 461                 IP_VS_DBG_RL("ip_vs_dr_xmit(): frag needed\n");
 462                 goto tx_error;
 463         }
 464
 465         /*
 466          * Call ip_send_check because we are not sure it is called
 467          * after ip_defrag. Is copy-on-write needed?
 468          */
 469         if (unlikely((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)) {
 470                 ip_rt_put(rt);
 471                 return NF_STOLEN;
 472         }
 473         ip_send_check(skb->nh.iph);
 474
 475         /* drop old route */
 476         dst_release(skb->dst);
 477         skb->dst = &rt->u.dst;
 478
 479         /* Another hack: avoid icmp_send in ip_fragment */
 480         skb->local_df = 1;
 481
 482 #ifdef CONFIG_NETFILTER_DEBUG
 483         skb->nf_debug = 0;
 484 #endif /* CONFIG_NETFILTER_DEBUG */
 485         IP_VS_XMIT(skb, rt);
 486
 487         LeaveFunction(10);
 488         return NF_STOLEN;
 489
 490   tx_error_icmp:
 491         dst_link_failure(skb);
 492   tx_error:
 493         kfree_skb(skb);
 494         LeaveFunction(10);
 495         return NF_STOLEN;
 496 }
 497
 498
 499 /*
 500  *      ICMP packet transmitter
 501  *      called by the ip_vs_in_icmp
 502  */
 503 int
 504 ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 505                 struct ip_vs_protocol *pp, int offset)
 506 {
 507         struct rtable   *rt;    /* Route to the other host */
 508         int mtu;
 509         int rc;
 510
 511         EnterFunction(10);
 512
 513         /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
 514            forwarded directly here, because there is no need to
 515            translate address/port back */
 516         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
 517                 if (cp->packet_xmit)
 518                         rc = cp->packet_xmit(skb, cp, pp);
 519                 else
 520                         rc = NF_ACCEPT;
 521                 /* do not touch skb anymore */
 522                 atomic_inc(&cp->in_pkts);
 523                 __ip_vs_conn_put(cp);
 524                 goto out;
 525         }
 526
 527         /*
 528          * mangle and send the packet here (only for VS/NAT)
 529          */
 530
 531         if (!(rt = __ip_vs_get_out_rt(cp, RT_TOS(skb->nh.iph->tos))))
 532                 goto tx_error_icmp;
 533
 534         /* MTU checking */
 535         mtu = dst_pmtu(&rt->u.dst);
 536         if ((skb->len > mtu) && (skb->nh.iph->frag_off&__constant_htons(IP_DF))) {
 537                 ip_rt_put(rt);
 538                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
 539                 IP_VS_DBG_RL("ip_vs_in_icmp(): frag needed\n");
 540                 goto tx_error;
 541         }
 542
 543         /* copy-on-write the packet before mangling it */
 544         if (!ip_vs_make_skb_writable(&skb, offset))
 545                 goto tx_error_put;
 546
 547         if (skb_cow(skb, rt->u.dst.dev->hard_header_len))
 548                 goto tx_error_put;
 549
 550         /* drop the old route when skb is not shared */
 551         dst_release(skb->dst);
 552         skb->dst = &rt->u.dst;
 553
 554         ip_vs_nat_icmp(skb, pp, cp, 0);
 555
 556         /* Another hack: avoid icmp_send in ip_fragment */
 557         skb->local_df = 1;
 558
 559 #ifdef CONFIG_NETFILTER_DEBUG
 560         skb->nf_debug = 0;
 561 #endif /* CONFIG_NETFILTER_DEBUG */
 562         IP_VS_XMIT(skb, rt);
 563
 564         rc = NF_STOLEN;
 565         goto out;
 566
 567   tx_error_icmp:
 568         dst_link_failure(skb);
 569   tx_error:
 570         dev_kfree_skb(skb);
 571         rc = NF_STOLEN;
 572   out:
 573         LeaveFunction(10);
 574         return rc;
 575   tx_error_put:
 576         ip_rt_put(rt);
 577         goto tx_error;
 578 }