git.oblomov.eu Git - linux-2.6/blob - 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  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
   9  *
  10  * ip_vs_sync:  sync connection info from master load balancer to backups
  11  *              through multicast
  12  *
  13  * Changes:
  14  *      Alexandre Cassen        :       Added master & backup support at a time.
  15  *      Alexandre Cassen        :       Added SyncID support for incoming sync
  16  *                                      messages filtering.
  17  *      Justin Ossevoort        :       Fix endian problem on sync message size.
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/slab.h>
  22 #include <linux/inetdevice.h>
  23 #include <linux/net.h>
  24 #include <linux/completion.h>
  25 #include <linux/delay.h>
  26 #include <linux/skbuff.h>
  27 #include <linux/in.h>
  28 #include <linux/igmp.h>                 /* for ip_mc_join_group */
  29 #include <linux/udp.h>
  30
  31 #include <net/ip.h>
  32 #include <net/sock.h>
  33 #include <asm/uaccess.h>                /* for get_fs and set_fs */
  34
  35 #include <net/ip_vs.h>
  36
  37 #define IP_VS_SYNC_GROUP 0xe0000051    /* multicast addr - 224.0.0.81 */
  38 #define IP_VS_SYNC_PORT  8848          /* multicast port */
  39
  40
  41 /*
  42  *      IPVS sync connection entry
  43  */
  44 struct ip_vs_sync_conn {
  45         __u8                    reserved;
  46
  47         /* Protocol, addresses and port numbers */
  48         __u8                    protocol;       /* Which protocol (TCP/UDP) */
  49         __be16                  cport;
  50         __be16                  vport;
  51         __be16                  dport;
  52         __be32                  caddr;          /* client address */
  53         __be32                  vaddr;          /* virtual address */
  54         __be32                  daddr;          /* destination address */
  55
  56         /* Flags and state transition */
  57         __be16                  flags;          /* status flags */
  58         __be16                  state;          /* state info */
  59
  60         /* The sequence options start here */
  61 };
  62
  63 struct ip_vs_sync_conn_options {
  64         struct ip_vs_seq        in_seq;         /* incoming seq. struct */
  65         struct ip_vs_seq        out_seq;        /* outgoing seq. struct */
  66 };
  67
  68 struct ip_vs_sync_thread_data {
  69         struct completion *startup;
  70         int state;
  71 };
  72
  73 #define SIMPLE_CONN_SIZE  (sizeof(struct ip_vs_sync_conn))
  74 #define FULL_CONN_SIZE  \
  75 (sizeof(struct ip_vs_sync_conn) + sizeof(struct ip_vs_sync_conn_options))
  76
  77
  78 /*
  79   The master mulitcasts messages to the backup load balancers in the
  80   following format.
  81
  82        0                   1                   2                   3
  83        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
  84       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  85       |  Count Conns  |    SyncID     |            Size               |
  86       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  87       |                                                               |
  88       |                    IPVS Sync Connection (1)                   |
  89       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  90       |                            .                                  |
  91       |                            .                                  |
  92       |                            .                                  |
  93       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  94       |                                                               |
  95       |                    IPVS Sync Connection (n)                   |
  96       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  97 */
  98
  99 #define SYNC_MESG_HEADER_LEN    4
 100
 101 struct ip_vs_sync_mesg {
 102         __u8                    nr_conns;
 103         __u8                    syncid;
 104         __u16                   size;
 105
 106         /* ip_vs_sync_conn entries start here */
 107 };
 108
 109 /* the maximum length of sync (sending/receiving) message */
 110 static int sync_send_mesg_maxlen;
 111 static int sync_recv_mesg_maxlen;
 112
 113 struct ip_vs_sync_buff {
 114         struct list_head        list;
 115         unsigned long           firstuse;
 116
 117         /* pointers for the message data */
 118         struct ip_vs_sync_mesg  *mesg;
 119         unsigned char           *head;
 120         unsigned char           *end;
 121 };
 122
 123
 124 /* the sync_buff list head and the lock */
 125 static LIST_HEAD(ip_vs_sync_queue);
 126 static DEFINE_SPINLOCK(ip_vs_sync_lock);
 127
 128 /* current sync_buff for accepting new conn entries */
 129 static struct ip_vs_sync_buff   *curr_sb = NULL;
 130 static DEFINE_SPINLOCK(curr_sb_lock);
 131
 132 /* ipvs sync daemon state */
 133 volatile int ip_vs_sync_state = IP_VS_STATE_NONE;
 134 volatile int ip_vs_master_syncid = 0;
 135 volatile int ip_vs_backup_syncid = 0;
 136
 137 /* multicast interface name */
 138 char ip_vs_master_mcast_ifn[IP_VS_IFNAME_MAXLEN];
 139 char ip_vs_backup_mcast_ifn[IP_VS_IFNAME_MAXLEN];
 140
 141 /* multicast addr */
 142 static struct sockaddr_in mcast_addr;
 143
 144
 145 static inline void sb_queue_tail(struct ip_vs_sync_buff *sb)
 146 {
 147         spin_lock(&ip_vs_sync_lock);
 148         list_add_tail(&sb->list, &ip_vs_sync_queue);
 149         spin_unlock(&ip_vs_sync_lock);
 150 }
 151
 152 static inline struct ip_vs_sync_buff * sb_dequeue(void)
 153 {
 154         struct ip_vs_sync_buff *sb;
 155
 156         spin_lock_bh(&ip_vs_sync_lock);
 157         if (list_empty(&ip_vs_sync_queue)) {
 158                 sb = NULL;
 159         } else {
 160                 sb = list_entry(ip_vs_sync_queue.next,
 161                                 struct ip_vs_sync_buff,
 162                                 list);
 163                 list_del(&sb->list);
 164         }
 165         spin_unlock_bh(&ip_vs_sync_lock);
 166
 167         return sb;
 168 }
 169
 170 static inline struct ip_vs_sync_buff * ip_vs_sync_buff_create(void)
 171 {
 172         struct ip_vs_sync_buff *sb;
 173
 174         if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
 175                 return NULL;
 176
 177         if (!(sb->mesg=kmalloc(sync_send_mesg_maxlen, GFP_ATOMIC))) {
 178                 kfree(sb);
 179                 return NULL;
 180         }
 181         sb->mesg->nr_conns = 0;
 182         sb->mesg->syncid = ip_vs_master_syncid;
 183         sb->mesg->size = 4;
 184         sb->head = (unsigned char *)sb->mesg + 4;
 185         sb->end = (unsigned char *)sb->mesg + sync_send_mesg_maxlen;
 186         sb->firstuse = jiffies;
 187         return sb;
 188 }
 189
 190 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
 191 {
 192         kfree(sb->mesg);
 193         kfree(sb);
 194 }
 195
 196 /*
 197  *      Get the current sync buffer if it has been created for more
 198  *      than the specified time or the specified time is zero.
 199  */
 200 static inline struct ip_vs_sync_buff *
 201 get_curr_sync_buff(unsigned long time)
 202 {
 203         struct ip_vs_sync_buff *sb;
 204
 205         spin_lock_bh(&curr_sb_lock);
 206         if (curr_sb && (time == 0 ||
 207                         time_before(jiffies - curr_sb->firstuse, time))) {
 208                 sb = curr_sb;
 209                 curr_sb = NULL;
 210         } else
 211                 sb = NULL;
 212         spin_unlock_bh(&curr_sb_lock);
 213         return sb;
 214 }
 215
 216
 217 /*
 218  *      Add an ip_vs_conn information into the current sync_buff.
 219  *      Called by ip_vs_in.
 220  */
 221 void ip_vs_sync_conn(struct ip_vs_conn *cp)
 222 {
 223         struct ip_vs_sync_mesg *m;
 224         struct ip_vs_sync_conn *s;
 225         int len;
 226
 227         spin_lock(&curr_sb_lock);
 228         if (!curr_sb) {
 229                 if (!(curr_sb=ip_vs_sync_buff_create())) {
 230                         spin_unlock(&curr_sb_lock);
 231                         IP_VS_ERR("ip_vs_sync_buff_create failed.\n");
 232                         return;
 233                 }
 234         }
 235
 236         len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
 237                 SIMPLE_CONN_SIZE;
 238         m = curr_sb->mesg;
 239         s = (struct ip_vs_sync_conn *)curr_sb->head;
 240
 241         /* copy members */
 242         s->protocol = cp->protocol;
 243         s->cport = cp->cport;
 244         s->vport = cp->vport;
 245         s->dport = cp->dport;
 246         s->caddr = cp->caddr;
 247         s->vaddr = cp->vaddr;
 248         s->daddr = cp->daddr;
 249         s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
 250         s->state = htons(cp->state);
 251         if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
 252                 struct ip_vs_sync_conn_options *opt =
 253                         (struct ip_vs_sync_conn_options *)&s[1];
 254                 memcpy(opt, &cp->in_seq, sizeof(*opt));
 255         }
 256
 257         m->nr_conns++;
 258         m->size += len;
 259         curr_sb->head += len;
 260
 261         /* check if there is a space for next one */
 262         if (curr_sb->head+FULL_CONN_SIZE > curr_sb->end) {
 263                 sb_queue_tail(curr_sb);
 264                 curr_sb = NULL;
 265         }
 266         spin_unlock(&curr_sb_lock);
 267
 268         /* synchronize its controller if it has */
 269         if (cp->control)
 270                 ip_vs_sync_conn(cp->control);
 271 }
 272
 273
 274 /*
 275  *      Process received multicast message and create the corresponding
 276  *      ip_vs_conn entries.
 277  */
 278 static void ip_vs_process_message(const char *buffer, const size_t buflen)
 279 {
 280         struct ip_vs_sync_mesg *m = (struct ip_vs_sync_mesg *)buffer;
 281         struct ip_vs_sync_conn *s;
 282         struct ip_vs_sync_conn_options *opt;
 283         struct ip_vs_conn *cp;
 284         struct ip_vs_protocol *pp;
 285         struct ip_vs_dest *dest;
 286         char *p;
 287         int i;
 288
 289         if (buflen < sizeof(struct ip_vs_sync_mesg)) {
 290                 IP_VS_ERR_RL("sync message header too short\n");
 291                 return;
 292         }
 293
 294         /* Convert size back to host byte order */
 295         m->size = ntohs(m->size);
 296
 297         if (buflen != m->size) {
 298                 IP_VS_ERR_RL("bogus sync message size\n");
 299                 return;
 300         }
 301
 302         /* SyncID sanity check */
 303         if (ip_vs_backup_syncid != 0 && m->syncid != ip_vs_backup_syncid) {
 304                 IP_VS_DBG(7, "Ignoring incoming msg with syncid = %d\n",
 305                           m->syncid);
 306                 return;
 307         }
 308
 309         p = (char *)buffer + sizeof(struct ip_vs_sync_mesg);
 310         for (i=0; i<m->nr_conns; i++) {
 311                 unsigned flags, state;
 312
 313                 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
 314                         IP_VS_ERR_RL("bogus conn in sync message\n");
 315                         return;
 316                 }
 317                 s = (struct ip_vs_sync_conn *) p;
 318                 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
 319                 flags &= ~IP_VS_CONN_F_HASHED;
 320                 if (flags & IP_VS_CONN_F_SEQ_MASK) {
 321                         opt = (struct ip_vs_sync_conn_options *)&s[1];
 322                         p += FULL_CONN_SIZE;
 323                         if (p > buffer+buflen) {
 324                                 IP_VS_ERR_RL("bogus conn options in sync message\n");
 325                                 return;
 326                         }
 327                 } else {
 328                         opt = NULL;
 329                         p += SIMPLE_CONN_SIZE;
 330                 }
 331
 332                 state = ntohs(s->state);
 333                 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
 334                         pp = ip_vs_proto_get(s->protocol);
 335                         if (!pp) {
 336                                 IP_VS_ERR_RL("Unsupported protocol %u in sync msg\n",
 337                                         s->protocol);
 338                                 continue;
 339                         }
 340                         if (state >= pp->num_states) {
 341                                 IP_VS_DBG(2, "Invalid %s state %u in sync msg\n",
 342                                         pp->name, state);
 343                                 continue;
 344                         }
 345                 } else {
 346                         /* protocol in templates is not used for state/timeout */
 347                         pp = NULL;
 348                         if (state > 0) {
 349                                 IP_VS_DBG(2, "Invalid template state %u in sync msg\n",
 350                                         state);
 351                                 state = 0;
 352                         }
 353                 }
 354
 355                 if (!(flags & IP_VS_CONN_F_TEMPLATE))
 356                         cp = ip_vs_conn_in_get(s->protocol,
 357                                                s->caddr, s->cport,
 358                                                s->vaddr, s->vport);
 359                 else
 360                         cp = ip_vs_ct_in_get(s->protocol,
 361                                                s->caddr, s->cport,
 362                                                s->vaddr, s->vport);
 363                 if (!cp) {
 364                         /*
 365                          * Find the appropriate destination for the connection.
 366                          * If it is not found the connection will remain unbound
 367                          * but still handled.
 368                          */
 369                         dest = ip_vs_find_dest(s->daddr, s->dport,
 370                                                s->vaddr, s->vport,
 371                                                s->protocol);
 372                         /*  Set the approprite ativity flag */
 373                         if (s->protocol == IPPROTO_TCP) {
 374                                 if (state != IP_VS_TCP_S_ESTABLISHED)
 375                                         flags |= IP_VS_CONN_F_INACTIVE;
 376                                 else
 377                                         flags &= ~IP_VS_CONN_F_INACTIVE;
 378                         }
 379                         cp = ip_vs_conn_new(s->protocol,
 380                                             s->caddr, s->cport,
 381                                             s->vaddr, s->vport,
 382                                             s->daddr, s->dport,
 383                                             flags, dest);
 384                         if (dest)
 385                                 atomic_dec(&dest->refcnt);
 386                         if (!cp) {
 387                                 IP_VS_ERR("ip_vs_conn_new failed\n");
 388                                 return;
 389                         }
 390                 } else if (!cp->dest) {
 391                         dest = ip_vs_try_bind_dest(cp);
 392                         if (dest)
 393                                 atomic_dec(&dest->refcnt);
 394                 } else if ((cp->dest) && (cp->protocol == IPPROTO_TCP) &&
 395                            (cp->state != state)) {
 396                         /* update active/inactive flag for the connection */
 397                         dest = cp->dest;
 398                         if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
 399                                 (state != IP_VS_TCP_S_ESTABLISHED)) {
 400                                 atomic_dec(&dest->activeconns);
 401                                 atomic_inc(&dest->inactconns);
 402                                 cp->flags |= IP_VS_CONN_F_INACTIVE;
 403                         } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) &&
 404                                 (state == IP_VS_TCP_S_ESTABLISHED)) {
 405                                 atomic_inc(&dest->activeconns);
 406                                 atomic_dec(&dest->inactconns);
 407                                 cp->flags &= ~IP_VS_CONN_F_INACTIVE;
 408                         }
 409                 }
 410
 411                 if (opt)
 412                         memcpy(&cp->in_seq, opt, sizeof(*opt));
 413                 atomic_set(&cp->in_pkts, sysctl_ip_vs_sync_threshold[0]);
 414                 cp->state = state;
 415                 cp->old_state = cp->state;
 416                 /*
 417                  * We can not recover the right timeout for templates
 418                  * in all cases, we can not find the right fwmark
 419                  * virtual service. If needed, we can do it for
 420                  * non-fwmark persistent services.
 421                  */
 422                 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pp->timeout_table)
 423                         cp->timeout = pp->timeout_table[state];
 424                 else
 425                         cp->timeout = (3*60*HZ);
 426                 ip_vs_conn_put(cp);
 427         }
 428 }
 429
 430
 431 /*
 432  *      Setup loopback of outgoing multicasts on a sending socket
 433  */
 434 static void set_mcast_loop(struct sock *sk, u_char loop)
 435 {
 436         struct inet_sock *inet = inet_sk(sk);
 437
 438         /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
 439         lock_sock(sk);
 440         inet->mc_loop = loop ? 1 : 0;
 441         release_sock(sk);
 442 }
 443
 444 /*
 445  *      Specify TTL for outgoing multicasts on a sending socket
 446  */
 447 static void set_mcast_ttl(struct sock *sk, u_char ttl)
 448 {
 449         struct inet_sock *inet = inet_sk(sk);
 450
 451         /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
 452         lock_sock(sk);
 453         inet->mc_ttl = ttl;
 454         release_sock(sk);
 455 }
 456
 457 /*
 458  *      Specifiy default interface for outgoing multicasts
 459  */
 460 static int set_mcast_if(struct sock *sk, char *ifname)
 461 {
 462         struct net_device *dev;
 463         struct inet_sock *inet = inet_sk(sk);
 464
 465         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
 466                 return -ENODEV;
 467
 468         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
 469                 return -EINVAL;
 470
 471         lock_sock(sk);
 472         inet->mc_index = dev->ifindex;
 473         /*  inet->mc_addr  = 0; */
 474         release_sock(sk);
 475
 476         return 0;
 477 }
 478
 479
 480 /*
 481  *      Set the maximum length of sync message according to the
 482  *      specified interface's MTU.
 483  */
 484 static int set_sync_mesg_maxlen(int sync_state)
 485 {
 486         struct net_device *dev;
 487         int num;
 488
 489         if (sync_state == IP_VS_STATE_MASTER) {
 490                 if ((dev = __dev_get_by_name(&init_net, ip_vs_master_mcast_ifn)) == NULL)
 491                         return -ENODEV;
 492
 493                 num = (dev->mtu - sizeof(struct iphdr) -
 494                        sizeof(struct udphdr) -
 495                        SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE;
 496                 sync_send_mesg_maxlen =
 497                         SYNC_MESG_HEADER_LEN + SIMPLE_CONN_SIZE * num;
 498                 IP_VS_DBG(7, "setting the maximum length of sync sending "
 499                           "message %d.\n", sync_send_mesg_maxlen);
 500         } else if (sync_state == IP_VS_STATE_BACKUP) {
 501                 if ((dev = __dev_get_by_name(&init_net, ip_vs_backup_mcast_ifn)) == NULL)
 502                         return -ENODEV;
 503
 504                 sync_recv_mesg_maxlen = dev->mtu -
 505                         sizeof(struct iphdr) - sizeof(struct udphdr);
 506                 IP_VS_DBG(7, "setting the maximum length of sync receiving "
 507                           "message %d.\n", sync_recv_mesg_maxlen);
 508         }
 509
 510         return 0;
 511 }
 512
 513
 514 /*
 515  *      Join a multicast group.
 516  *      the group is specified by a class D multicast address 224.0.0.0/8
 517  *      in the in_addr structure passed in as a parameter.
 518  */
 519 static int
 520 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
 521 {
 522         struct ip_mreqn mreq;
 523         struct net_device *dev;
 524         int ret;
 525
 526         memset(&mreq, 0, sizeof(mreq));
 527         memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
 528
 529         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
 530                 return -ENODEV;
 531         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
 532                 return -EINVAL;
 533
 534         mreq.imr_ifindex = dev->ifindex;
 535
 536         lock_sock(sk);
 537         ret = ip_mc_join_group(sk, &mreq);
 538         release_sock(sk);
 539
 540         return ret;
 541 }
 542
 543
 544 static int bind_mcastif_addr(struct socket *sock, char *ifname)
 545 {
 546         struct net_device *dev;
 547         __be32 addr;
 548         struct sockaddr_in sin;
 549
 550         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
 551                 return -ENODEV;
 552
 553         addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
 554         if (!addr)
 555                 IP_VS_ERR("You probably need to specify IP address on "
 556                           "multicast interface.\n");
 557
 558         IP_VS_DBG(7, "binding socket with (%s) %u.%u.%u.%u\n",
 559                   ifname, NIPQUAD(addr));
 560
 561         /* Now bind the socket with the address of multicast interface */
 562         sin.sin_family       = AF_INET;
 563         sin.sin_addr.s_addr  = addr;
 564         sin.sin_port         = 0;
 565
 566         return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
 567 }
 568
 569 /*
 570  *      Set up sending multicast socket over UDP
 571  */
 572 static struct socket * make_send_sock(void)
 573 {
 574         struct socket *sock;
 575
 576         /* First create a socket */
 577         if (sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock) < 0) {
 578                 IP_VS_ERR("Error during creation of socket; terminating\n");
 579                 return NULL;
 580         }
 581
 582         if (set_mcast_if(sock->sk, ip_vs_master_mcast_ifn) < 0) {
 583                 IP_VS_ERR("Error setting outbound mcast interface\n");
 584                 goto error;
 585         }
 586
 587         set_mcast_loop(sock->sk, 0);
 588         set_mcast_ttl(sock->sk, 1);
 589
 590         if (bind_mcastif_addr(sock, ip_vs_master_mcast_ifn) < 0) {
 591                 IP_VS_ERR("Error binding address of the mcast interface\n");
 592                 goto error;
 593         }
 594
 595         if (sock->ops->connect(sock,
 596                                (struct sockaddr*)&mcast_addr,
 597                                sizeof(struct sockaddr), 0) < 0) {
 598                 IP_VS_ERR("Error connecting to the multicast addr\n");
 599                 goto error;
 600         }
 601
 602         return sock;
 603
 604   error:
 605         sock_release(sock);
 606         return NULL;
 607 }
 608
 609
 610 /*
 611  *      Set up receiving multicast socket over UDP
 612  */
 613 static struct socket * make_receive_sock(void)
 614 {
 615         struct socket *sock;
 616
 617         /* First create a socket */
 618         if (sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock) < 0) {
 619                 IP_VS_ERR("Error during creation of socket; terminating\n");
 620                 return NULL;
 621         }
 622
 623         /* it is equivalent to the REUSEADDR option in user-space */
 624         sock->sk->sk_reuse = 1;
 625
 626         if (sock->ops->bind(sock,
 627                             (struct sockaddr*)&mcast_addr,
 628                             sizeof(struct sockaddr)) < 0) {
 629                 IP_VS_ERR("Error binding to the multicast addr\n");
 630                 goto error;
 631         }
 632
 633         /* join the multicast group */
 634         if (join_mcast_group(sock->sk,
 635                              (struct in_addr*)&mcast_addr.sin_addr,
 636                              ip_vs_backup_mcast_ifn) < 0) {
 637                 IP_VS_ERR("Error joining to the multicast group\n");
 638                 goto error;
 639         }
 640
 641         return sock;
 642
 643   error:
 644         sock_release(sock);
 645         return NULL;
 646 }
 647
 648
 649 static int
 650 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
 651 {
 652         struct msghdr   msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
 653         struct kvec     iov;
 654         int             len;
 655
 656         EnterFunction(7);
 657         iov.iov_base     = (void *)buffer;
 658         iov.iov_len      = length;
 659
 660         len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
 661
 662         LeaveFunction(7);
 663         return len;
 664 }
 665
 666 static void
 667 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
 668 {
 669         int msize;
 670
 671         msize = msg->size;
 672
 673         /* Put size in network byte order */
 674         msg->size = htons(msg->size);
 675
 676         if (ip_vs_send_async(sock, (char *)msg, msize) != msize)
 677                 IP_VS_ERR("ip_vs_send_async error\n");
 678 }
 679
 680 static int
 681 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
 682 {
 683         struct msghdr           msg = {NULL,};
 684         struct kvec             iov;
 685         int                     len;
 686
 687         EnterFunction(7);
 688
 689         /* Receive a packet */
 690         iov.iov_base     = buffer;
 691         iov.iov_len      = (size_t)buflen;
 692
 693         len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, 0);
 694
 695         if (len < 0)
 696                 return -1;
 697
 698         LeaveFunction(7);
 699         return len;
 700 }
 701
 702
 703 static DECLARE_WAIT_QUEUE_HEAD(sync_wait);
 704 static pid_t sync_master_pid = 0;
 705 static pid_t sync_backup_pid = 0;
 706
 707 static DECLARE_WAIT_QUEUE_HEAD(stop_sync_wait);
 708 static int stop_master_sync = 0;
 709 static int stop_backup_sync = 0;
 710
 711 static void sync_master_loop(void)
 712 {
 713         struct socket *sock;
 714         struct ip_vs_sync_buff *sb;
 715
 716         /* create the sending multicast socket */
 717         sock = make_send_sock();
 718         if (!sock)
 719                 return;
 720
 721         IP_VS_INFO("sync thread started: state = MASTER, mcast_ifn = %s, "
 722                    "syncid = %d\n",
 723                    ip_vs_master_mcast_ifn, ip_vs_master_syncid);
 724
 725         for (;;) {
 726                 while ((sb=sb_dequeue())) {
 727                         ip_vs_send_sync_msg(sock, sb->mesg);
 728                         ip_vs_sync_buff_release(sb);
 729                 }
 730
 731                 /* check if entries stay in curr_sb for 2 seconds */
 732                 if ((sb = get_curr_sync_buff(2*HZ))) {
 733                         ip_vs_send_sync_msg(sock, sb->mesg);
 734                         ip_vs_sync_buff_release(sb);
 735                 }
 736
 737                 if (stop_master_sync)
 738                         break;
 739
 740                 msleep_interruptible(1000);
 741         }
 742
 743         /* clean up the sync_buff queue */
 744         while ((sb=sb_dequeue())) {
 745                 ip_vs_sync_buff_release(sb);
 746         }
 747
 748         /* clean up the current sync_buff */
 749         if ((sb = get_curr_sync_buff(0))) {
 750                 ip_vs_sync_buff_release(sb);
 751         }
 752
 753         /* release the sending multicast socket */
 754         sock_release(sock);
 755 }
 756
 757
 758 static void sync_backup_loop(void)
 759 {
 760         struct socket *sock;
 761         char *buf;
 762         int len;
 763
 764         if (!(buf = kmalloc(sync_recv_mesg_maxlen, GFP_ATOMIC))) {
 765                 IP_VS_ERR("sync_backup_loop: kmalloc error\n");
 766                 return;
 767         }
 768
 769         /* create the receiving multicast socket */
 770         sock = make_receive_sock();
 771         if (!sock)
 772                 goto out;
 773
 774         IP_VS_INFO("sync thread started: state = BACKUP, mcast_ifn = %s, "
 775                    "syncid = %d\n",
 776                    ip_vs_backup_mcast_ifn, ip_vs_backup_syncid);
 777
 778         for (;;) {
 779                 /* do you have data now? */
 780                 while (!skb_queue_empty(&(sock->sk->sk_receive_queue))) {
 781                         if ((len =
 782                              ip_vs_receive(sock, buf,
 783                                            sync_recv_mesg_maxlen)) <= 0) {
 784                                 IP_VS_ERR("receiving message error\n");
 785                                 break;
 786                         }
 787                         /* disable bottom half, because it accessed the data
 788                            shared by softirq while getting/creating conns */
 789                         local_bh_disable();
 790                         ip_vs_process_message(buf, len);
 791                         local_bh_enable();
 792                 }
 793
 794                 if (stop_backup_sync)
 795                         break;
 796
 797                 msleep_interruptible(1000);
 798         }
 799
 800         /* release the sending multicast socket */
 801         sock_release(sock);
 802
 803   out:
 804         kfree(buf);
 805 }
 806
 807
 808 static void set_sync_pid(int sync_state, pid_t sync_pid)
 809 {
 810         if (sync_state == IP_VS_STATE_MASTER)
 811                 sync_master_pid = sync_pid;
 812         else if (sync_state == IP_VS_STATE_BACKUP)
 813                 sync_backup_pid = sync_pid;
 814 }
 815
 816 static void set_stop_sync(int sync_state, int set)
 817 {
 818         if (sync_state == IP_VS_STATE_MASTER)
 819                 stop_master_sync = set;
 820         else if (sync_state == IP_VS_STATE_BACKUP)
 821                 stop_backup_sync = set;
 822         else {
 823                 stop_master_sync = set;
 824                 stop_backup_sync = set;
 825         }
 826 }
 827
 828 static int sync_thread(void *startup)
 829 {
 830         DECLARE_WAITQUEUE(wait, current);
 831         mm_segment_t oldmm;
 832         int state;
 833         const char *name;
 834         struct ip_vs_sync_thread_data *tinfo = startup;
 835
 836         /* increase the module use count */
 837         ip_vs_use_count_inc();
 838
 839         if (ip_vs_sync_state & IP_VS_STATE_MASTER && !sync_master_pid) {
 840                 state = IP_VS_STATE_MASTER;
 841                 name = "ipvs_syncmaster";
 842         } else if (ip_vs_sync_state & IP_VS_STATE_BACKUP && !sync_backup_pid) {
 843                 state = IP_VS_STATE_BACKUP;
 844                 name = "ipvs_syncbackup";
 845         } else {
 846                 IP_VS_BUG();
 847                 ip_vs_use_count_dec();
 848                 return -EINVAL;
 849         }
 850
 851         daemonize(name);
 852
 853         oldmm = get_fs();
 854         set_fs(KERNEL_DS);
 855
 856         /* Block all signals */
 857         spin_lock_irq(&current->sighand->siglock);
 858         siginitsetinv(&current->blocked, 0);
 859         recalc_sigpending();
 860         spin_unlock_irq(&current->sighand->siglock);
 861
 862         /* set the maximum length of sync message */
 863         set_sync_mesg_maxlen(state);
 864
 865         /* set up multicast address */
 866         mcast_addr.sin_family = AF_INET;
 867         mcast_addr.sin_port = htons(IP_VS_SYNC_PORT);
 868         mcast_addr.sin_addr.s_addr = htonl(IP_VS_SYNC_GROUP);
 869
 870         add_wait_queue(&sync_wait, &wait);
 871
 872         set_sync_pid(state, task_pid_nr(current));
 873         complete(tinfo->startup);
 874
 875         /*
 876          * once we call the completion queue above, we should
 877          * null out that reference, since its allocated on the
 878          * stack of the creating kernel thread
 879          */
 880         tinfo->startup = NULL;
 881
 882         /* processing master/backup loop here */
 883         if (state == IP_VS_STATE_MASTER)
 884                 sync_master_loop();
 885         else if (state == IP_VS_STATE_BACKUP)
 886                 sync_backup_loop();
 887         else IP_VS_BUG();
 888
 889         remove_wait_queue(&sync_wait, &wait);
 890
 891         /* thread exits */
 892
 893         /*
 894          * If we weren't explicitly stopped, then we
 895          * exited in error, and should undo our state
 896          */
 897         if ((!stop_master_sync) && (!stop_backup_sync))
 898                 ip_vs_sync_state -= tinfo->state;
 899
 900         set_sync_pid(state, 0);
 901         IP_VS_INFO("sync thread stopped!\n");
 902
 903         set_fs(oldmm);
 904
 905         /* decrease the module use count */
 906         ip_vs_use_count_dec();
 907
 908         set_stop_sync(state, 0);
 909         wake_up(&stop_sync_wait);
 910
 911         /*
 912          * we need to free the structure that was allocated
 913          * for us in start_sync_thread
 914          */
 915         kfree(tinfo);
 916         return 0;
 917 }
 918
 919
 920 static int fork_sync_thread(void *startup)
 921 {
 922         pid_t pid;
 923
 924         /* fork the sync thread here, then the parent process of the
 925            sync thread is the init process after this thread exits. */
 926   repeat:
 927         if ((pid = kernel_thread(sync_thread, startup, 0)) < 0) {
 928                 IP_VS_ERR("could not create sync_thread due to %d... "
 929                           "retrying.\n", pid);
 930                 msleep_interruptible(1000);
 931                 goto repeat;
 932         }
 933
 934         return 0;
 935 }
 936
 937
 938 int start_sync_thread(int state, char *mcast_ifn, __u8 syncid)
 939 {
 940         DECLARE_COMPLETION_ONSTACK(startup);
 941         pid_t pid;
 942         struct ip_vs_sync_thread_data *tinfo;
 943
 944         if ((state == IP_VS_STATE_MASTER && sync_master_pid) ||
 945             (state == IP_VS_STATE_BACKUP && sync_backup_pid))
 946                 return -EEXIST;
 947
 948         /*
 949          * Note that tinfo will be freed in sync_thread on exit
 950          */
 951         tinfo = kmalloc(sizeof(struct ip_vs_sync_thread_data), GFP_KERNEL);
 952         if (!tinfo)
 953                 return -ENOMEM;
 954
 955         IP_VS_DBG(7, "%s: pid %d\n", __func__, task_pid_nr(current));
 956         IP_VS_DBG(7, "Each ip_vs_sync_conn entry need %Zd bytes\n",
 957                   sizeof(struct ip_vs_sync_conn));
 958
 959         ip_vs_sync_state |= state;
 960         if (state == IP_VS_STATE_MASTER) {
 961                 strlcpy(ip_vs_master_mcast_ifn, mcast_ifn,
 962                         sizeof(ip_vs_master_mcast_ifn));
 963                 ip_vs_master_syncid = syncid;
 964         } else {
 965                 strlcpy(ip_vs_backup_mcast_ifn, mcast_ifn,
 966                         sizeof(ip_vs_backup_mcast_ifn));
 967                 ip_vs_backup_syncid = syncid;
 968         }
 969
 970         tinfo->state = state;
 971         tinfo->startup = &startup;
 972
 973   repeat:
 974         if ((pid = kernel_thread(fork_sync_thread, tinfo, 0)) < 0) {
 975                 IP_VS_ERR("could not create fork_sync_thread due to %d... "
 976                           "retrying.\n", pid);
 977                 msleep_interruptible(1000);
 978                 goto repeat;
 979         }
 980
 981         wait_for_completion(&startup);
 982
 983         return 0;
 984 }
 985
 986
 987 int stop_sync_thread(int state)
 988 {
 989         DECLARE_WAITQUEUE(wait, current);
 990
 991         if ((state == IP_VS_STATE_MASTER && !sync_master_pid) ||
 992             (state == IP_VS_STATE_BACKUP && !sync_backup_pid))
 993                 return -ESRCH;
 994
 995         IP_VS_DBG(7, "%s: pid %d\n", __func__, task_pid_nr(current));
 996         IP_VS_INFO("stopping sync thread %d ...\n",
 997                    (state == IP_VS_STATE_MASTER) ?
 998                    sync_master_pid : sync_backup_pid);
 999
1000         __set_current_state(TASK_UNINTERRUPTIBLE);
1001         add_wait_queue(&stop_sync_wait, &wait);
1002         set_stop_sync(state, 1);
1003         ip_vs_sync_state -= state;
1004         wake_up(&sync_wait);
1005         schedule();
1006         __set_current_state(TASK_RUNNING);
1007         remove_wait_queue(&stop_sync_wait, &wait);
1008
1009         /* Note: no need to reap the sync thread, because its parent
1010            process is the init process */
1011
1012         if ((state == IP_VS_STATE_MASTER && stop_master_sync) ||
1013             (state == IP_VS_STATE_BACKUP && stop_backup_sync))
1014                 IP_VS_BUG();
1015
1016         return 0;
1017 }