net/ipv4/ipvs/ip_vs_sync.c

   1 /*
   2  * IPVS         An implementation of the IP virtual server support for the
   3  *              LINUX operating system.  IPVS is now implemented as a module
   4  *              over the NetFilter framework. IPVS can be used to build a
   5  *              high-performance and highly available server based on a
   6  *              cluster of servers.
   7  *
   8  * Version:     $Id: ip_vs_sync.c,v 1.13 2003/06/08 09:31:19 wensong Exp $
   9  *
  10  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
  11  *
  12  * ip_vs_sync:  sync connection info from master load balancer to backups
  13  *              through multicast
  14  *
  15  * Changes:
  16  *      Alexandre Cassen        :       Added master & backup support at a time.
  17  *      Alexandre Cassen        :       Added SyncID support for incoming sync
  18  *                                      messages filtering.
  19  *      Justin Ossevoort        :       Fix endian problem on sync message size.
  20  */
  21
  22 #include <linux/module.h>
  23 #include <linux/slab.h>
  24 #include <linux/inetdevice.h>
  25 #include <linux/net.h>
  26 #include <linux/completion.h>
  27 #include <linux/delay.h>
  28 #include <linux/skbuff.h>
  29 #include <linux/in.h>
  30 #include <linux/igmp.h>                 /* for ip_mc_join_group */
  31 #include <linux/udp.h>
  32
  33 #include <net/ip.h>
  34 #include <net/sock.h>
  35 #include <asm/uaccess.h>                /* for get_fs and set_fs */
  36
  37 #include <net/ip_vs.h>
  38
  39 #define IP_VS_SYNC_GROUP 0xe0000051    /* multicast addr - 224.0.0.81 */
  40 #define IP_VS_SYNC_PORT  8848          /* multicast port */
  41
  42
  43 /*
  44  *      IPVS sync connection entry
  45  */
  46 struct ip_vs_sync_conn {
  47         __u8                    reserved;
  48
  49         /* Protocol, addresses and port numbers */
  50         __u8                    protocol;       /* Which protocol (TCP/UDP) */
  51         __be16                  cport;
  52         __be16                  vport;
  53         __be16                  dport;
  54         __be32                  caddr;          /* client address */
  55         __be32                  vaddr;          /* virtual address */
  56         __be32                  daddr;          /* destination address */
  57
  58         /* Flags and state transition */
  59         __be16                  flags;          /* status flags */
  60         __be16                  state;          /* state info */
  61
  62         /* The sequence options start here */
  63 };
  64
  65 struct ip_vs_sync_conn_options {
  66         struct ip_vs_seq        in_seq;         /* incoming seq. struct */
  67         struct ip_vs_seq        out_seq;        /* outgoing seq. struct */
  68 };
  69
  70 struct ip_vs_sync_thread_data {
  71         struct completion *startup;
  72         int state;
  73 };
  74
  75 #define SIMPLE_CONN_SIZE  (sizeof(struct ip_vs_sync_conn))
  76 #define FULL_CONN_SIZE  \
  77 (sizeof(struct ip_vs_sync_conn) + sizeof(struct ip_vs_sync_conn_options))
  78
  79
  80 /*
  81   The master mulitcasts messages to the backup load balancers in the
  82   following format.
  83
  84        0                   1                   2                   3
  85        0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
  86       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  87       |  Count Conns  |    SyncID     |            Size               |
  88       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  89       |                                                               |
  90       |                    IPVS Sync Connection (1)                   |
  91       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  92       |                            .                                  |
  93       |                            .                                  |
  94       |                            .                                  |
  95       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  96       |                                                               |
  97       |                    IPVS Sync Connection (n)                   |
  98       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  99 */
 100
 101 #define SYNC_MESG_HEADER_LEN    4
 102
 103 struct ip_vs_sync_mesg {
 104         __u8                    nr_conns;
 105         __u8                    syncid;
 106         __u16                   size;
 107
 108         /* ip_vs_sync_conn entries start here */
 109 };
 110
 111 /* the maximum length of sync (sending/receiving) message */
 112 static int sync_send_mesg_maxlen;
 113 static int sync_recv_mesg_maxlen;
 114
 115 struct ip_vs_sync_buff {
 116         struct list_head        list;
 117         unsigned long           firstuse;
 118
 119         /* pointers for the message data */
 120         struct ip_vs_sync_mesg  *mesg;
 121         unsigned char           *head;
 122         unsigned char           *end;
 123 };
 124
 125
 126 /* the sync_buff list head and the lock */
 127 static LIST_HEAD(ip_vs_sync_queue);
 128 static DEFINE_SPINLOCK(ip_vs_sync_lock);
 129
 130 /* current sync_buff for accepting new conn entries */
 131 static struct ip_vs_sync_buff   *curr_sb = NULL;
 132 static DEFINE_SPINLOCK(curr_sb_lock);
 133
 134 /* ipvs sync daemon state */
 135 volatile int ip_vs_sync_state = IP_VS_STATE_NONE;
 136 volatile int ip_vs_master_syncid = 0;
 137 volatile int ip_vs_backup_syncid = 0;
 138
 139 /* multicast interface name */
 140 char ip_vs_master_mcast_ifn[IP_VS_IFNAME_MAXLEN];
 141 char ip_vs_backup_mcast_ifn[IP_VS_IFNAME_MAXLEN];
 142
 143 /* multicast addr */
 144 static struct sockaddr_in mcast_addr;
 145
 146
 147 static inline void sb_queue_tail(struct ip_vs_sync_buff *sb)
 148 {
 149         spin_lock(&ip_vs_sync_lock);
 150         list_add_tail(&sb->list, &ip_vs_sync_queue);
 151         spin_unlock(&ip_vs_sync_lock);
 152 }
 153
 154 static inline struct ip_vs_sync_buff * sb_dequeue(void)
 155 {
 156         struct ip_vs_sync_buff *sb;
 157
 158         spin_lock_bh(&ip_vs_sync_lock);
 159         if (list_empty(&ip_vs_sync_queue)) {
 160                 sb = NULL;
 161         } else {
 162                 sb = list_entry(ip_vs_sync_queue.next,
 163                                 struct ip_vs_sync_buff,
 164                                 list);
 165                 list_del(&sb->list);
 166         }
 167         spin_unlock_bh(&ip_vs_sync_lock);
 168
 169         return sb;
 170 }
 171
 172 static inline struct ip_vs_sync_buff * ip_vs_sync_buff_create(void)
 173 {
 174         struct ip_vs_sync_buff *sb;
 175
 176         if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
 177                 return NULL;
 178
 179         if (!(sb->mesg=kmalloc(sync_send_mesg_maxlen, GFP_ATOMIC))) {
 180                 kfree(sb);
 181                 return NULL;
 182         }
 183         sb->mesg->nr_conns = 0;
 184         sb->mesg->syncid = ip_vs_master_syncid;
 185         sb->mesg->size = 4;
 186         sb->head = (unsigned char *)sb->mesg + 4;
 187         sb->end = (unsigned char *)sb->mesg + sync_send_mesg_maxlen;
 188         sb->firstuse = jiffies;
 189         return sb;
 190 }
 191
 192 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
 193 {
 194         kfree(sb->mesg);
 195         kfree(sb);
 196 }
 197
 198 /*
 199  *      Get the current sync buffer if it has been created for more
 200  *      than the specified time or the specified time is zero.
 201  */
 202 static inline struct ip_vs_sync_buff *
 203 get_curr_sync_buff(unsigned long time)
 204 {
 205         struct ip_vs_sync_buff *sb;
 206
 207         spin_lock_bh(&curr_sb_lock);
 208         if (curr_sb && (time == 0 ||
 209                         time_before(jiffies - curr_sb->firstuse, time))) {
 210                 sb = curr_sb;
 211                 curr_sb = NULL;
 212         } else
 213                 sb = NULL;
 214         spin_unlock_bh(&curr_sb_lock);
 215         return sb;
 216 }
 217
 218
 219 /*
 220  *      Add an ip_vs_conn information into the current sync_buff.
 221  *      Called by ip_vs_in.
 222  */
 223 void ip_vs_sync_conn(struct ip_vs_conn *cp)
 224 {
 225         struct ip_vs_sync_mesg *m;
 226         struct ip_vs_sync_conn *s;
 227         int len;
 228
 229         spin_lock(&curr_sb_lock);
 230         if (!curr_sb) {
 231                 if (!(curr_sb=ip_vs_sync_buff_create())) {
 232                         spin_unlock(&curr_sb_lock);
 233                         IP_VS_ERR("ip_vs_sync_buff_create failed.\n");
 234                         return;
 235                 }
 236         }
 237
 238         len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
 239                 SIMPLE_CONN_SIZE;
 240         m = curr_sb->mesg;
 241         s = (struct ip_vs_sync_conn *)curr_sb->head;
 242
 243         /* copy members */
 244         s->protocol = cp->protocol;
 245         s->cport = cp->cport;
 246         s->vport = cp->vport;
 247         s->dport = cp->dport;
 248         s->caddr = cp->caddr;
 249         s->vaddr = cp->vaddr;
 250         s->daddr = cp->daddr;
 251         s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
 252         s->state = htons(cp->state);
 253         if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
 254                 struct ip_vs_sync_conn_options *opt =
 255                         (struct ip_vs_sync_conn_options *)&s[1];
 256                 memcpy(opt, &cp->in_seq, sizeof(*opt));
 257         }
 258
 259         m->nr_conns++;
 260         m->size += len;
 261         curr_sb->head += len;
 262
 263         /* check if there is a space for next one */
 264         if (curr_sb->head+FULL_CONN_SIZE > curr_sb->end) {
 265                 sb_queue_tail(curr_sb);
 266                 curr_sb = NULL;
 267         }
 268         spin_unlock(&curr_sb_lock);
 269
 270         /* synchronize its controller if it has */
 271         if (cp->control)
 272                 ip_vs_sync_conn(cp->control);
 273 }
 274
 275
 276 /*
 277  *      Process received multicast message and create the corresponding
 278  *      ip_vs_conn entries.
 279  */
 280 static void ip_vs_process_message(const char *buffer, const size_t buflen)
 281 {
 282         struct ip_vs_sync_mesg *m = (struct ip_vs_sync_mesg *)buffer;
 283         struct ip_vs_sync_conn *s;
 284         struct ip_vs_sync_conn_options *opt;
 285         struct ip_vs_conn *cp;
 286         struct ip_vs_protocol *pp;
 287         struct ip_vs_dest *dest;
 288         char *p;
 289         int i;
 290
 291         if (buflen < sizeof(struct ip_vs_sync_mesg)) {
 292                 IP_VS_ERR_RL("sync message header too short\n");
 293                 return;
 294         }
 295
 296         /* Convert size back to host byte order */
 297         m->size = ntohs(m->size);
 298
 299         if (buflen != m->size) {
 300                 IP_VS_ERR_RL("bogus sync message size\n");
 301                 return;
 302         }
 303
 304         /* SyncID sanity check */
 305         if (ip_vs_backup_syncid != 0 && m->syncid != ip_vs_backup_syncid) {
 306                 IP_VS_DBG(7, "Ignoring incoming msg with syncid = %d\n",
 307                           m->syncid);
 308                 return;
 309         }
 310
 311         p = (char *)buffer + sizeof(struct ip_vs_sync_mesg);
 312         for (i=0; i<m->nr_conns; i++) {
 313                 unsigned flags, state;
 314
 315                 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
 316                         IP_VS_ERR_RL("bogus conn in sync message\n");
 317                         return;
 318                 }
 319                 s = (struct ip_vs_sync_conn *) p;
 320                 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
 321                 flags &= ~IP_VS_CONN_F_HASHED;
 322                 if (flags & IP_VS_CONN_F_SEQ_MASK) {
 323                         opt = (struct ip_vs_sync_conn_options *)&s[1];
 324                         p += FULL_CONN_SIZE;
 325                         if (p > buffer+buflen) {
 326                                 IP_VS_ERR_RL("bogus conn options in sync message\n");
 327                                 return;
 328                         }
 329                 } else {
 330                         opt = NULL;
 331                         p += SIMPLE_CONN_SIZE;
 332                 }
 333
 334                 state = ntohs(s->state);
 335                 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
 336                         pp = ip_vs_proto_get(s->protocol);
 337                         if (!pp) {
 338                                 IP_VS_ERR_RL("Unsupported protocol %u in sync msg\n",
 339                                         s->protocol);
 340                                 continue;
 341                         }
 342                         if (state >= pp->num_states) {
 343                                 IP_VS_DBG(2, "Invalid %s state %u in sync msg\n",
 344                                         pp->name, state);
 345                                 continue;
 346                         }
 347                 } else {
 348                         /* protocol in templates is not used for state/timeout */
 349                         pp = NULL;
 350                         if (state > 0) {
 351                                 IP_VS_DBG(2, "Invalid template state %u in sync msg\n",
 352                                         state);
 353                                 state = 0;
 354                         }
 355                 }
 356
 357                 if (!(flags & IP_VS_CONN_F_TEMPLATE))
 358                         cp = ip_vs_conn_in_get(s->protocol,
 359                                                s->caddr, s->cport,
 360                                                s->vaddr, s->vport);
 361                 else
 362                         cp = ip_vs_ct_in_get(s->protocol,
 363                                                s->caddr, s->cport,
 364                                                s->vaddr, s->vport);
 365                 if (!cp) {
 366                         /*
 367                          * Find the appropriate destination for the connection.
 368                          * If it is not found the connection will remain unbound
 369                          * but still handled.
 370                          */
 371                         dest = ip_vs_find_dest(s->daddr, s->dport,
 372                                                s->vaddr, s->vport,
 373                                                s->protocol);
 374                         /*  Set the approprite ativity flag */
 375                         if (s->protocol == IPPROTO_TCP) {
 376                                 if (state != IP_VS_TCP_S_ESTABLISHED)
 377                                         flags |= IP_VS_CONN_F_INACTIVE;
 378                                 else
 379                                         flags &= ~IP_VS_CONN_F_INACTIVE;
 380                         }
 381                         cp = ip_vs_conn_new(s->protocol,
 382                                             s->caddr, s->cport,
 383                                             s->vaddr, s->vport,
 384                                             s->daddr, s->dport,
 385                                             flags, dest);
 386                         if (dest)
 387                                 atomic_dec(&dest->refcnt);
 388                         if (!cp) {
 389                                 IP_VS_ERR("ip_vs_conn_new failed\n");
 390                                 return;
 391                         }
 392                 } else if (!cp->dest) {
 393                         dest = ip_vs_try_bind_dest(cp);
 394                         if (dest)
 395                                 atomic_dec(&dest->refcnt);
 396                 } else if ((cp->dest) && (cp->protocol == IPPROTO_TCP) &&
 397                            (cp->state != state)) {
 398                         /* update active/inactive flag for the connection */
 399                         dest = cp->dest;
 400                         if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
 401                                 (state != IP_VS_TCP_S_ESTABLISHED)) {
 402                                 atomic_dec(&dest->activeconns);
 403                                 atomic_inc(&dest->inactconns);
 404                                 cp->flags |= IP_VS_CONN_F_INACTIVE;
 405                         } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) &&
 406                                 (state == IP_VS_TCP_S_ESTABLISHED)) {
 407                                 atomic_inc(&dest->activeconns);
 408                                 atomic_dec(&dest->inactconns);
 409                                 cp->flags &= ~IP_VS_CONN_F_INACTIVE;
 410                         }
 411                 }
 412
 413                 if (opt)
 414                         memcpy(&cp->in_seq, opt, sizeof(*opt));
 415                 atomic_set(&cp->in_pkts, sysctl_ip_vs_sync_threshold[0]);
 416                 cp->state = state;
 417                 cp->old_state = cp->state;
 418                 /*
 419                  * We can not recover the right timeout for templates
 420                  * in all cases, we can not find the right fwmark
 421                  * virtual service. If needed, we can do it for
 422                  * non-fwmark persistent services.
 423                  */
 424                 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pp->timeout_table)
 425                         cp->timeout = pp->timeout_table[state];
 426                 else
 427                         cp->timeout = (3*60*HZ);
 428                 ip_vs_conn_put(cp);
 429         }
 430 }
 431
 432
 433 /*
 434  *      Setup loopback of outgoing multicasts on a sending socket
 435  */
 436 static void set_mcast_loop(struct sock *sk, u_char loop)
 437 {
 438         struct inet_sock *inet = inet_sk(sk);
 439
 440         /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
 441         lock_sock(sk);
 442         inet->mc_loop = loop ? 1 : 0;
 443         release_sock(sk);
 444 }
 445
 446 /*
 447  *      Specify TTL for outgoing multicasts on a sending socket
 448  */
 449 static void set_mcast_ttl(struct sock *sk, u_char ttl)
 450 {
 451         struct inet_sock *inet = inet_sk(sk);
 452
 453         /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
 454         lock_sock(sk);
 455         inet->mc_ttl = ttl;
 456         release_sock(sk);
 457 }
 458
 459 /*
 460  *      Specifiy default interface for outgoing multicasts
 461  */
 462 static int set_mcast_if(struct sock *sk, char *ifname)
 463 {
 464         struct net_device *dev;
 465         struct inet_sock *inet = inet_sk(sk);
 466
 467         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
 468                 return -ENODEV;
 469
 470         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
 471                 return -EINVAL;
 472
 473         lock_sock(sk);
 474         inet->mc_index = dev->ifindex;
 475         /*  inet->mc_addr  = 0; */
 476         release_sock(sk);
 477
 478         return 0;
 479 }
 480
 481
 482 /*
 483  *      Set the maximum length of sync message according to the
 484  *      specified interface's MTU.
 485  */
 486 static int set_sync_mesg_maxlen(int sync_state)
 487 {
 488         struct net_device *dev;
 489         int num;
 490
 491         if (sync_state == IP_VS_STATE_MASTER) {
 492                 if ((dev = __dev_get_by_name(&init_net, ip_vs_master_mcast_ifn)) == NULL)
 493                         return -ENODEV;
 494
 495                 num = (dev->mtu - sizeof(struct iphdr) -
 496                        sizeof(struct udphdr) -
 497                        SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE;
 498                 sync_send_mesg_maxlen =
 499                         SYNC_MESG_HEADER_LEN + SIMPLE_CONN_SIZE * num;
 500                 IP_VS_DBG(7, "setting the maximum length of sync sending "
 501                           "message %d.\n", sync_send_mesg_maxlen);
 502         } else if (sync_state == IP_VS_STATE_BACKUP) {
 503                 if ((dev = __dev_get_by_name(&init_net, ip_vs_backup_mcast_ifn)) == NULL)
 504                         return -ENODEV;
 505
 506                 sync_recv_mesg_maxlen = dev->mtu -
 507                         sizeof(struct iphdr) - sizeof(struct udphdr);
 508                 IP_VS_DBG(7, "setting the maximum length of sync receiving "
 509                           "message %d.\n", sync_recv_mesg_maxlen);
 510         }
 511
 512         return 0;
 513 }
 514
 515
 516 /*
 517  *      Join a multicast group.
 518  *      the group is specified by a class D multicast address 224.0.0.0/8
 519  *      in the in_addr structure passed in as a parameter.
 520  */
 521 static int
 522 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
 523 {
 524         struct ip_mreqn mreq;
 525         struct net_device *dev;
 526         int ret;
 527
 528         memset(&mreq, 0, sizeof(mreq));
 529         memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
 530
 531         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
 532                 return -ENODEV;
 533         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
 534                 return -EINVAL;
 535
 536         mreq.imr_ifindex = dev->ifindex;
 537
 538         lock_sock(sk);
 539         ret = ip_mc_join_group(sk, &mreq);
 540         release_sock(sk);
 541
 542         return ret;
 543 }
 544
 545
 546 static int bind_mcastif_addr(struct socket *sock, char *ifname)
 547 {
 548         struct net_device *dev;
 549         __be32 addr;
 550         struct sockaddr_in sin;
 551
 552         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
 553                 return -ENODEV;
 554
 555         addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
 556         if (!addr)
 557                 IP_VS_ERR("You probably need to specify IP address on "
 558                           "multicast interface.\n");
 559
 560         IP_VS_DBG(7, "binding socket with (%s) %u.%u.%u.%u\n",
 561                   ifname, NIPQUAD(addr));
 562
 563         /* Now bind the socket with the address of multicast interface */
 564         sin.sin_family       = AF_INET;
 565         sin.sin_addr.s_addr  = addr;
 566         sin.sin_port         = 0;
 567
 568         return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
 569 }
 570
 571 /*
 572  *      Set up sending multicast socket over UDP
 573  */
 574 static struct socket * make_send_sock(void)
 575 {
 576         struct socket *sock;
 577
 578         /* First create a socket */
 579         if (sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock) < 0) {
 580                 IP_VS_ERR("Error during creation of socket; terminating\n");
 581                 return NULL;
 582         }
 583
 584         if (set_mcast_if(sock->sk, ip_vs_master_mcast_ifn) < 0) {
 585                 IP_VS_ERR("Error setting outbound mcast interface\n");
 586                 goto error;
 587         }
 588
 589         set_mcast_loop(sock->sk, 0);
 590         set_mcast_ttl(sock->sk, 1);
 591
 592         if (bind_mcastif_addr(sock, ip_vs_master_mcast_ifn) < 0) {
 593                 IP_VS_ERR("Error binding address of the mcast interface\n");
 594                 goto error;
 595         }
 596
 597         if (sock->ops->connect(sock,
 598                                (struct sockaddr*)&mcast_addr,
 599                                sizeof(struct sockaddr), 0) < 0) {
 600                 IP_VS_ERR("Error connecting to the multicast addr\n");
 601                 goto error;
 602         }
 603
 604         return sock;
 605
 606   error:
 607         sock_release(sock);
 608         return NULL;
 609 }
 610
 611
 612 /*
 613  *      Set up receiving multicast socket over UDP
 614  */
 615 static struct socket * make_receive_sock(void)
 616 {
 617         struct socket *sock;
 618
 619         /* First create a socket */
 620         if (sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock) < 0) {
 621                 IP_VS_ERR("Error during creation of socket; terminating\n");
 622                 return NULL;
 623         }
 624
 625         /* it is equivalent to the REUSEADDR option in user-space */
 626         sock->sk->sk_reuse = 1;
 627
 628         if (sock->ops->bind(sock,
 629                             (struct sockaddr*)&mcast_addr,
 630                             sizeof(struct sockaddr)) < 0) {
 631                 IP_VS_ERR("Error binding to the multicast addr\n");
 632                 goto error;
 633         }
 634
 635         /* join the multicast group */
 636         if (join_mcast_group(sock->sk,
 637                              (struct in_addr*)&mcast_addr.sin_addr,
 638                              ip_vs_backup_mcast_ifn) < 0) {
 639                 IP_VS_ERR("Error joining to the multicast group\n");
 640                 goto error;
 641         }
 642
 643         return sock;
 644
 645   error:
 646         sock_release(sock);
 647         return NULL;
 648 }
 649
 650
 651 static int
 652 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
 653 {
 654         struct msghdr   msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
 655         struct kvec     iov;
 656         int             len;
 657
 658         EnterFunction(7);
 659         iov.iov_base     = (void *)buffer;
 660         iov.iov_len      = length;
 661
 662         len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
 663
 664         LeaveFunction(7);
 665         return len;
 666 }
 667
 668 static void
 669 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
 670 {
 671         int msize;
 672
 673         msize = msg->size;
 674
 675         /* Put size in network byte order */
 676         msg->size = htons(msg->size);
 677
 678         if (ip_vs_send_async(sock, (char *)msg, msize) != msize)
 679                 IP_VS_ERR("ip_vs_send_async error\n");
 680 }
 681
 682 static int
 683 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
 684 {
 685         struct msghdr           msg = {NULL,};
 686         struct kvec             iov;
 687         int                     len;
 688
 689         EnterFunction(7);
 690
 691         /* Receive a packet */
 692         iov.iov_base     = buffer;
 693         iov.iov_len      = (size_t)buflen;
 694
 695         len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, 0);
 696
 697         if (len < 0)
 698                 return -1;
 699
 700         LeaveFunction(7);
 701         return len;
 702 }
 703
 704
 705 static DECLARE_WAIT_QUEUE_HEAD(sync_wait);
 706 static pid_t sync_master_pid = 0;
 707 static pid_t sync_backup_pid = 0;
 708
 709 static DECLARE_WAIT_QUEUE_HEAD(stop_sync_wait);
 710 static int stop_master_sync = 0;
 711 static int stop_backup_sync = 0;
 712
 713 static void sync_master_loop(void)
 714 {
 715         struct socket *sock;
 716         struct ip_vs_sync_buff *sb;
 717
 718         /* create the sending multicast socket */
 719         sock = make_send_sock();
 720         if (!sock)
 721                 return;
 722
 723         IP_VS_INFO("sync thread started: state = MASTER, mcast_ifn = %s, "
 724                    "syncid = %d\n",
 725                    ip_vs_master_mcast_ifn, ip_vs_master_syncid);
 726
 727         for (;;) {
 728                 while ((sb=sb_dequeue())) {
 729                         ip_vs_send_sync_msg(sock, sb->mesg);
 730                         ip_vs_sync_buff_release(sb);
 731                 }
 732
 733                 /* check if entries stay in curr_sb for 2 seconds */
 734                 if ((sb = get_curr_sync_buff(2*HZ))) {
 735                         ip_vs_send_sync_msg(sock, sb->mesg);
 736                         ip_vs_sync_buff_release(sb);
 737                 }
 738
 739                 if (stop_master_sync)
 740                         break;
 741
 742                 msleep_interruptible(1000);
 743         }
 744
 745         /* clean up the sync_buff queue */
 746         while ((sb=sb_dequeue())) {
 747                 ip_vs_sync_buff_release(sb);
 748         }
 749
 750         /* clean up the current sync_buff */
 751         if ((sb = get_curr_sync_buff(0))) {
 752                 ip_vs_sync_buff_release(sb);
 753         }
 754
 755         /* release the sending multicast socket */
 756         sock_release(sock);
 757 }
 758
 759
 760 static void sync_backup_loop(void)
 761 {
 762         struct socket *sock;
 763         char *buf;
 764         int len;
 765
 766         if (!(buf = kmalloc(sync_recv_mesg_maxlen, GFP_ATOMIC))) {
 767                 IP_VS_ERR("sync_backup_loop: kmalloc error\n");
 768                 return;
 769         }
 770
 771         /* create the receiving multicast socket */
 772         sock = make_receive_sock();
 773         if (!sock)
 774                 goto out;
 775
 776         IP_VS_INFO("sync thread started: state = BACKUP, mcast_ifn = %s, "
 777                    "syncid = %d\n",
 778                    ip_vs_backup_mcast_ifn, ip_vs_backup_syncid);
 779
 780         for (;;) {
 781                 /* do you have data now? */
 782                 while (!skb_queue_empty(&(sock->sk->sk_receive_queue))) {
 783                         if ((len =
 784                              ip_vs_receive(sock, buf,
 785                                            sync_recv_mesg_maxlen)) <= 0) {
 786                                 IP_VS_ERR("receiving message error\n");
 787                                 break;
 788                         }
 789                         /* disable bottom half, because it accessed the data
 790                            shared by softirq while getting/creating conns */
 791                         local_bh_disable();
 792                         ip_vs_process_message(buf, len);
 793                         local_bh_enable();
 794                 }
 795
 796                 if (stop_backup_sync)
 797                         break;
 798
 799                 msleep_interruptible(1000);
 800         }
 801
 802         /* release the sending multicast socket */
 803         sock_release(sock);
 804
 805   out:
 806         kfree(buf);
 807 }
 808
 809
 810 static void set_sync_pid(int sync_state, pid_t sync_pid)
 811 {
 812         if (sync_state == IP_VS_STATE_MASTER)
 813                 sync_master_pid = sync_pid;
 814         else if (sync_state == IP_VS_STATE_BACKUP)
 815                 sync_backup_pid = sync_pid;
 816 }
 817
 818 static void set_stop_sync(int sync_state, int set)
 819 {
 820         if (sync_state == IP_VS_STATE_MASTER)
 821                 stop_master_sync = set;
 822         else if (sync_state == IP_VS_STATE_BACKUP)
 823                 stop_backup_sync = set;
 824         else {
 825                 stop_master_sync = set;
 826                 stop_backup_sync = set;
 827         }
 828 }
 829
 830 static int sync_thread(void *startup)
 831 {
 832         DECLARE_WAITQUEUE(wait, current);
 833         mm_segment_t oldmm;
 834         int state;
 835         const char *name;
 836         struct ip_vs_sync_thread_data *tinfo = startup;
 837
 838         /* increase the module use count */
 839         ip_vs_use_count_inc();
 840
 841         if (ip_vs_sync_state & IP_VS_STATE_MASTER && !sync_master_pid) {
 842                 state = IP_VS_STATE_MASTER;
 843                 name = "ipvs_syncmaster";
 844         } else if (ip_vs_sync_state & IP_VS_STATE_BACKUP && !sync_backup_pid) {
 845                 state = IP_VS_STATE_BACKUP;
 846                 name = "ipvs_syncbackup";
 847         } else {
 848                 IP_VS_BUG();
 849                 ip_vs_use_count_dec();
 850                 return -EINVAL;
 851         }
 852
 853         daemonize(name);
 854
 855         oldmm = get_fs();
 856         set_fs(KERNEL_DS);
 857
 858         /* Block all signals */
 859         spin_lock_irq(&current->sighand->siglock);
 860         siginitsetinv(&current->blocked, 0);
 861         recalc_sigpending();
 862         spin_unlock_irq(&current->sighand->siglock);
 863
 864         /* set the maximum length of sync message */
 865         set_sync_mesg_maxlen(state);
 866
 867         /* set up multicast address */
 868         mcast_addr.sin_family = AF_INET;
 869         mcast_addr.sin_port = htons(IP_VS_SYNC_PORT);
 870         mcast_addr.sin_addr.s_addr = htonl(IP_VS_SYNC_GROUP);
 871
 872         add_wait_queue(&sync_wait, &wait);
 873
 874         set_sync_pid(state, task_pid_nr(current));
 875         complete(tinfo->startup);
 876
 877         /*
 878          * once we call the completion queue above, we should
 879          * null out that reference, since its allocated on the
 880          * stack of the creating kernel thread
 881          */
 882         tinfo->startup = NULL;
 883
 884         /* processing master/backup loop here */
 885         if (state == IP_VS_STATE_MASTER)
 886                 sync_master_loop();
 887         else if (state == IP_VS_STATE_BACKUP)
 888                 sync_backup_loop();
 889         else IP_VS_BUG();
 890
 891         remove_wait_queue(&sync_wait, &wait);
 892
 893         /* thread exits */
 894
 895         /*
 896          * If we weren't explicitly stopped, then we
 897          * exited in error, and should undo our state
 898          */
 899         if ((!stop_master_sync) && (!stop_backup_sync))
 900                 ip_vs_sync_state -= tinfo->state;
 901
 902         set_sync_pid(state, 0);
 903         IP_VS_INFO("sync thread stopped!\n");
 904
 905         set_fs(oldmm);
 906
 907         /* decrease the module use count */
 908         ip_vs_use_count_dec();
 909
 910         set_stop_sync(state, 0);
 911         wake_up(&stop_sync_wait);
 912
 913         /*
 914          * we need to free the structure that was allocated
 915          * for us in start_sync_thread
 916          */
 917         kfree(tinfo);
 918         return 0;
 919 }
 920
 921
 922 static int fork_sync_thread(void *startup)
 923 {
 924         pid_t pid;
 925
 926         /* fork the sync thread here, then the parent process of the
 927            sync thread is the init process after this thread exits. */
 928   repeat:
 929         if ((pid = kernel_thread(sync_thread, startup, 0)) < 0) {
 930                 IP_VS_ERR("could not create sync_thread due to %d... "
 931                           "retrying.\n", pid);
 932                 msleep_interruptible(1000);
 933                 goto repeat;
 934         }
 935
 936         return 0;
 937 }
 938
 939
 940 int start_sync_thread(int state, char *mcast_ifn, __u8 syncid)
 941 {
 942         DECLARE_COMPLETION_ONSTACK(startup);
 943         pid_t pid;
 944         struct ip_vs_sync_thread_data *tinfo;
 945
 946         if ((state == IP_VS_STATE_MASTER && sync_master_pid) ||
 947             (state == IP_VS_STATE_BACKUP && sync_backup_pid))
 948                 return -EEXIST;
 949
 950         /*
 951          * Note that tinfo will be freed in sync_thread on exit
 952          */
 953         tinfo = kmalloc(sizeof(struct ip_vs_sync_thread_data), GFP_KERNEL);
 954         if (!tinfo)
 955                 return -ENOMEM;
 956
 957         IP_VS_DBG(7, "%s: pid %d\n", __func__, task_pid_nr(current));
 958         IP_VS_DBG(7, "Each ip_vs_sync_conn entry need %Zd bytes\n",
 959                   sizeof(struct ip_vs_sync_conn));
 960
 961         ip_vs_sync_state |= state;
 962         if (state == IP_VS_STATE_MASTER) {
 963                 strlcpy(ip_vs_master_mcast_ifn, mcast_ifn,
 964                         sizeof(ip_vs_master_mcast_ifn));
 965                 ip_vs_master_syncid = syncid;
 966         } else {
 967                 strlcpy(ip_vs_backup_mcast_ifn, mcast_ifn,
 968                         sizeof(ip_vs_backup_mcast_ifn));
 969                 ip_vs_backup_syncid = syncid;
 970         }
 971
 972         tinfo->state = state;
 973         tinfo->startup = &startup;
 974
 975   repeat:
 976         if ((pid = kernel_thread(fork_sync_thread, tinfo, 0)) < 0) {
 977                 IP_VS_ERR("could not create fork_sync_thread due to %d... "
 978                           "retrying.\n", pid);
 979                 msleep_interruptible(1000);
 980                 goto repeat;
 981         }
 982
 983         wait_for_completion(&startup);
 984
 985         return 0;
 986 }
 987
 988
 989 int stop_sync_thread(int state)
 990 {
 991         DECLARE_WAITQUEUE(wait, current);
 992
 993         if ((state == IP_VS_STATE_MASTER && !sync_master_pid) ||
 994             (state == IP_VS_STATE_BACKUP && !sync_backup_pid))
 995                 return -ESRCH;
 996
 997         IP_VS_DBG(7, "%s: pid %d\n", __func__, task_pid_nr(current));
 998         IP_VS_INFO("stopping sync thread %d ...\n",
 999                    (state == IP_VS_STATE_MASTER) ?
1000                    sync_master_pid : sync_backup_pid);
1001
1002         __set_current_state(TASK_UNINTERRUPTIBLE);
1003         add_wait_queue(&stop_sync_wait, &wait);
1004         set_stop_sync(state, 1);
1005         ip_vs_sync_state -= state;
1006         wake_up(&sync_wait);
1007         schedule();
1008         __set_current_state(TASK_RUNNING);
1009         remove_wait_queue(&stop_sync_wait, &wait);
1010
1011         /* Note: no need to reap the sync thread, because its parent
1012            process is the init process */
1013
1014         if ((state == IP_VS_STATE_MASTER && stop_master_sync) ||
1015             (state == IP_VS_STATE_BACKUP && stop_backup_sync))
1016                 IP_VS_BUG();
1017
1018         return 0;
1019 }