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 1.8 2002/08/17 14:06:02 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/net.h>
  25
  26 #include <linux/skbuff.h>
  27 #include <linux/in.h>
  28 #include <linux/igmp.h>                 /* for ip_mc_join_group */
  29
  30 #include <net/ip.h>
  31 #include <net/sock.h>
  32 #include <asm/uaccess.h>                /* for get_fs and set_fs */
  33
  34 #include <net/ip_vs.h>
  35
  36 #define IP_VS_SYNC_GROUP 0xe0000051    /* multicast addr - 224.0.0.81 */
  37 #define IP_VS_SYNC_PORT  8848          /* multicast port */
  38
  39
  40 /*
  41  *      IPVS sync connection entry
  42  */
  43 struct ip_vs_sync_conn {
  44         __u8                    reserved;
  45
  46         /* Protocol, addresses and port numbers */
  47         __u8                    protocol;       /* Which protocol (TCP/UDP) */
  48         __u16                   cport;
  49         __u16                   vport;
  50         __u16                   dport;
  51         __u32                   caddr;          /* client address */
  52         __u32                   vaddr;          /* virtual address */
  53         __u32                   daddr;          /* destination address */
  54
  55         /* Flags and state transition */
  56         __u16                   flags;          /* status flags */
  57         __u16                   state;          /* state info */
  58
  59         /* The sequence options start here */
  60 };
  61
  62 struct ip_vs_sync_conn_options {
  63         struct ip_vs_seq        in_seq;         /* incoming seq. struct */
  64         struct ip_vs_seq        out_seq;        /* outgoing seq. struct */
  65 };
  66
  67 #define IP_VS_SYNC_CONN_TIMEOUT (3*60*HZ)
  68 #define SIMPLE_CONN_SIZE  (sizeof(struct ip_vs_sync_conn))
  69 #define FULL_CONN_SIZE  \
  70 (sizeof(struct ip_vs_sync_conn) + sizeof(struct ip_vs_sync_conn_options))
  71
  72
  73 /*
  74   The master mulitcasts messages to the backup load balancers in the
  75   following format.
  76
  77        0                   1                   2                   3
  78        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
  79       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  80       |  Count Conns  |    Sync ID    |            Size               |
  81       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  82       |                                                               |
  83       |                    IPVS Sync Connection (1)                   |
  84       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  85       |                            .                                  |
  86       |                            .                                  |
  87       |                            .                                  |
  88       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  89       |                                                               |
  90       |                    IPVS Sync Connection (n)                   |
  91       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  92
  93    Count Conns : Number of IPVS sync Connection entries.
  94    Sync ID     : IPVS sync group we belong to.
  95    Size        : Size of packet.
  96
  97 */
  98 #define SYNC_MESG_MAX_SIZE      (24*50+4)
  99 struct ip_vs_sync_mesg {
 100         __u8                    nr_conns;
 101         __u8                    syncid;
 102         __u16                   size;
 103
 104         /* ip_vs_sync_conn entries start here */
 105 };
 106
 107
 108 struct ip_vs_sync_buff {
 109         struct list_head        list;
 110         unsigned long           firstuse;
 111
 112         /* pointers for the message data */
 113         struct ip_vs_sync_mesg  *mesg;
 114         unsigned char           *head;
 115         unsigned char           *end;
 116 };
 117
 118
 119 /* the sync_buff list head and the lock */
 120 static LIST_HEAD(ip_vs_sync_queue);
 121 static spinlock_t ip_vs_sync_lock = SPIN_LOCK_UNLOCKED;
 122
 123 /* current sync_buff for accepting new conn entries */
 124 static struct ip_vs_sync_buff   *curr_sb = NULL;
 125 static spinlock_t curr_sb_lock = SPIN_LOCK_UNLOCKED;
 126
 127 /* ipvs sync daemon state */
 128 volatile int ip_vs_sync_state = IP_VS_STATE_NONE;
 129 volatile int ip_vs_master_syncid = 0;
 130 volatile int ip_vs_backup_syncid = 0;
 131
 132 /* multicast interface name */
 133 char ip_vs_mcast_master_ifn[IP_VS_IFNAME_MAXLEN];
 134 char ip_vs_mcast_backup_ifn[IP_VS_IFNAME_MAXLEN];
 135
 136 /* multicast addr */
 137 static struct sockaddr_in mcast_addr;
 138
 139 static inline void sb_queue_tail(struct ip_vs_sync_buff *sb)
 140 {
 141         spin_lock(&ip_vs_sync_lock);
 142         list_add_tail(&sb->list, &ip_vs_sync_queue);
 143         spin_unlock(&ip_vs_sync_lock);
 144 }
 145
 146 static inline struct ip_vs_sync_buff * sb_dequeue(void)
 147 {
 148         struct ip_vs_sync_buff *sb;
 149
 150         spin_lock_bh(&ip_vs_sync_lock);
 151         if (list_empty(&ip_vs_sync_queue)) {
 152                 sb = NULL;
 153         } else {
 154                 sb = list_entry(ip_vs_sync_queue.next,
 155                                 struct ip_vs_sync_buff,
 156                                 list);
 157                 list_del(&sb->list);
 158         }
 159         spin_unlock_bh(&ip_vs_sync_lock);
 160
 161         return sb;
 162 }
 163
 164 static inline struct ip_vs_sync_buff * ip_vs_sync_buff_create(void)
 165 {
 166         struct ip_vs_sync_buff *sb;
 167
 168         if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
 169                 return NULL;
 170
 171         if (!(sb->mesg=kmalloc(SYNC_MESG_MAX_SIZE, GFP_ATOMIC))) {
 172                 kfree(sb);
 173                 return NULL;
 174         }
 175         sb->mesg->nr_conns = 0;
 176         sb->mesg->syncid = ip_vs_master_syncid;
 177         sb->mesg->size = 4;
 178         sb->head = (unsigned char *)sb->mesg + 4;
 179         sb->end = (unsigned char *)sb->mesg + SYNC_MESG_MAX_SIZE;
 180         sb->firstuse = jiffies;
 181         return sb;
 182 }
 183
 184 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
 185 {
 186         kfree(sb->mesg);
 187         kfree(sb);
 188 }
 189
 190 /*
 191  *      Get the current sync buffer if it has been created for more
 192  *      than the specified time or the specified time is zero.
 193  */
 194 static inline struct ip_vs_sync_buff *
 195 get_curr_sync_buff(unsigned long time)
 196 {
 197         struct ip_vs_sync_buff *sb;
 198
 199         spin_lock_bh(&curr_sb_lock);
 200         if (curr_sb &&
 201             (jiffies - curr_sb->firstuse > time || time == 0)) {
 202                 sb = curr_sb;
 203                 curr_sb = NULL;
 204         } else
 205                 sb = NULL;
 206         spin_unlock_bh(&curr_sb_lock);
 207         return sb;
 208 }
 209
 210
 211 /*
 212  *      Add an ip_vs_conn information into the current sync_buff.
 213  *      Called by ip_vs_in.
 214  */
 215 void ip_vs_sync_conn(struct ip_vs_conn *cp)
 216 {
 217         struct ip_vs_sync_mesg *m;
 218         struct ip_vs_sync_conn *s;
 219         int len;
 220
 221         spin_lock(&curr_sb_lock);
 222         if (!curr_sb) {
 223                 if (!(curr_sb=ip_vs_sync_buff_create())) {
 224                         spin_unlock(&curr_sb_lock);
 225                         IP_VS_ERR("ip_vs_sync_buff_create failed.\n");
 226                         return;
 227                 }
 228         }
 229
 230         len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
 231                 SIMPLE_CONN_SIZE;
 232         m = curr_sb->mesg;
 233         s = (struct ip_vs_sync_conn *)curr_sb->head;
 234
 235         /* copy members */
 236         s->protocol = cp->protocol;
 237         s->cport = cp->cport;
 238         s->vport = cp->vport;
 239         s->dport = cp->dport;
 240         s->caddr = cp->caddr;
 241         s->vaddr = cp->vaddr;
 242         s->daddr = cp->daddr;
 243         s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
 244         s->state = htons(cp->state);
 245         if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
 246                 struct ip_vs_sync_conn_options *opt =
 247                         (struct ip_vs_sync_conn_options *)&s[1];
 248                 memcpy(opt, &cp->in_seq, sizeof(*opt));
 249         }
 250
 251         m->nr_conns++;
 252         m->size += len;
 253         curr_sb->head += len;
 254
 255         /* check if there is a space for next one */
 256         if (curr_sb->head+FULL_CONN_SIZE > curr_sb->end) {
 257                 sb_queue_tail(curr_sb);
 258                 curr_sb = NULL;
 259         }
 260         spin_unlock(&curr_sb_lock);
 261
 262         /* synchronize its controller if it has */
 263         if (cp->control)
 264                 ip_vs_sync_conn(cp->control);
 265 }
 266
 267
 268 /*
 269  *      Process received multicast message and create the corresponding
 270  *      ip_vs_conn entries.
 271  */
 272 static void ip_vs_process_message(const char *buffer, const size_t buflen)
 273 {
 274         struct ip_vs_sync_mesg *m = (struct ip_vs_sync_mesg *)buffer;
 275         struct ip_vs_sync_conn *s;
 276         struct ip_vs_sync_conn_options *opt;
 277         struct ip_vs_conn *cp;
 278         char *p;
 279         int i;
 280
 281         /* Convert size back to host byte order */
 282         m->size = ntohs(m->size);
 283
 284         if (buflen != m->size) {
 285                 IP_VS_ERR("bogus message\n");
 286                 return;
 287         }
 288
 289         /* SyncID sanity check */
 290         if (ip_vs_backup_syncid != 0 && m->syncid != ip_vs_backup_syncid) {
 291                 IP_VS_DBG(7, "Ignoring incoming msg with syncid = %d\n",
 292                           m->syncid);
 293                 return;
 294         }
 295
 296         p = (char *)buffer + sizeof(struct ip_vs_sync_mesg);
 297         for (i=0; i<m->nr_conns; i++) {
 298                 unsigned flags;
 299
 300                 s = (struct ip_vs_sync_conn *)p;
 301                 flags = ntohs(s->flags);
 302                 if (!(flags & IP_VS_CONN_F_TEMPLATE))
 303                         cp = ip_vs_conn_in_get(s->protocol,
 304                                                s->caddr, s->cport,
 305                                                s->vaddr, s->vport);
 306                 else
 307                         cp = ip_vs_ct_in_get(s->protocol,
 308                                                s->caddr, s->cport,
 309                                                s->vaddr, s->vport);
 310                 if (!cp) {
 311                         cp = ip_vs_conn_new(s->protocol,
 312                                             s->caddr, s->cport,
 313                                             s->vaddr, s->vport,
 314                                             s->daddr, s->dport,
 315                                             flags, NULL);
 316                         if (!cp) {
 317                                 IP_VS_ERR("ip_vs_conn_new failed\n");
 318                                 return;
 319                         }
 320                         cp->state = ntohs(s->state);
 321                 } else if (!cp->dest) {
 322                         /* it is an entry created by the synchronization */
 323                         cp->state = ntohs(s->state);
 324                         cp->flags = flags | IP_VS_CONN_F_HASHED;
 325                 }       /* Note that we don't touch its state and flags
 326                            if it is a normal entry. */
 327
 328                 if (flags & IP_VS_CONN_F_SEQ_MASK) {
 329                         opt = (struct ip_vs_sync_conn_options *)&s[1];
 330                         memcpy(&cp->in_seq, opt, sizeof(*opt));
 331                         p += FULL_CONN_SIZE;
 332                 } else
 333                         p += SIMPLE_CONN_SIZE;
 334
 335                 atomic_set(&cp->in_pkts, sysctl_ip_vs_sync_threshold);
 336                 cp->timeout = IP_VS_SYNC_CONN_TIMEOUT;
 337                 ip_vs_conn_put(cp);
 338
 339                 if (p > buffer+buflen) {
 340                         IP_VS_ERR("bogus message\n");
 341                         return;
 342                 }
 343         }
 344 }
 345
 346
 347 /*
 348  *      Setup loopback of outgoing multicasts on a sending socket
 349  */
 350 static void set_mcast_loop(struct sock *sk, u_char loop)
 351 {
 352         /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
 353         lock_sock(sk);
 354         sk->protinfo.af_inet.mc_loop = loop ? 1 : 0;
 355         release_sock(sk);
 356 }
 357
 358 /*
 359  *      Specify TTL for outgoing multicasts on a sending socket
 360  */
 361 static void set_mcast_ttl(struct sock *sk, u_char ttl)
 362 {
 363         /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
 364         lock_sock(sk);
 365         sk->protinfo.af_inet.mc_ttl = ttl;
 366         release_sock(sk);
 367 }
 368
 369 /*
 370  *      Specifiy default interface for outgoing multicasts
 371  */
 372 static int set_mcast_if(struct sock *sk, char *ifname)
 373 {
 374         struct net_device *dev;
 375
 376         if ((dev = __dev_get_by_name(ifname)) == NULL)
 377                 return -ENODEV;
 378
 379         if (sk->bound_dev_if && dev->ifindex != sk->bound_dev_if)
 380                 return -EINVAL;
 381
 382         lock_sock(sk);
 383         sk->protinfo.af_inet.mc_index = dev->ifindex;
 384         /*  sk->protinfo.af_inet.mc_addr  = 0; */
 385         release_sock(sk);
 386
 387         return 0;
 388 }
 389
 390 /*
 391  *      Join a multicast group.
 392  *      the group is specified by a class D multicast address 224.0.0.0/8
 393  *      in the in_addr structure passed in as a parameter.
 394  */
 395 static int
 396 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
 397 {
 398         struct ip_mreqn mreq;
 399         struct net_device *dev;
 400         int ret;
 401
 402         memset(&mreq, 0, sizeof(mreq));
 403         memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
 404
 405         if ((dev = __dev_get_by_name(ifname)) == NULL)
 406                 return -ENODEV;
 407         if (sk->bound_dev_if && dev->ifindex != sk->bound_dev_if)
 408                 return -EINVAL;
 409
 410         mreq.imr_ifindex = dev->ifindex;
 411
 412         lock_sock(sk);
 413         ret = ip_mc_join_group(sk, &mreq);
 414         release_sock(sk);
 415
 416         return ret;
 417 }
 418
 419
 420 static int bind_mcastif_addr(struct socket *sock, char *ifname)
 421 {
 422         struct net_device *dev;
 423         u32 addr;
 424         struct sockaddr_in sin;
 425
 426         if ((dev = __dev_get_by_name(ifname)) == NULL)
 427                 return -ENODEV;
 428
 429         addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
 430         if (!addr)
 431                 IP_VS_ERR("You probably need to specify IP address on "
 432                           "multicast interface.\n");
 433
 434         IP_VS_DBG(7, "binding socket with (%s) %u.%u.%u.%u\n",
 435                   ifname, NIPQUAD(addr));
 436
 437         /* Now bind the socket with the address of multicast interface */
 438         sin.sin_family       = AF_INET;
 439         sin.sin_addr.s_addr  = addr;
 440         sin.sin_port         = 0;
 441
 442         return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
 443 }
 444
 445 /*
 446  *      Set up sending multicast socket over UDP
 447  */
 448 static struct socket * make_send_sock(void)
 449 {
 450         struct socket *sock;
 451
 452         /* First create a socket */
 453         if (sock_create(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock) < 0) {
 454                 IP_VS_ERR("Error during creation of socket; terminating\n");
 455                 return NULL;
 456         }
 457
 458         if (set_mcast_if(sock->sk, ip_vs_mcast_master_ifn) < 0) {
 459                 IP_VS_ERR("Error setting outbound mcast interface\n");
 460                 goto error;
 461         }
 462
 463         set_mcast_loop(sock->sk, 0);
 464         set_mcast_ttl(sock->sk, 1);
 465
 466         if (bind_mcastif_addr(sock, ip_vs_mcast_master_ifn) < 0) {
 467                 IP_VS_ERR("Error binding address of the mcast interface\n");
 468                 goto error;
 469         }
 470
 471         if (sock->ops->connect(sock,
 472                                (struct sockaddr*)&mcast_addr,
 473                                sizeof(struct sockaddr), 0) < 0) {
 474                 IP_VS_ERR("Error connecting to the multicast addr\n");
 475                 goto error;
 476         }
 477
 478         return sock;
 479
 480   error:
 481         sock_release(sock);
 482         return NULL;
 483 }
 484
 485
 486 /*
 487  *      Set up receiving multicast socket over UDP
 488  */
 489 static struct socket * make_receive_sock(void)
 490 {
 491         struct socket *sock;
 492
 493         /* First create a socket */
 494         if (sock_create(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock) < 0) {
 495                 IP_VS_ERR("Error during creation of socket; terminating\n");
 496                 return NULL;
 497         }
 498
 499         /* it is equivalent to the REUSEADDR option in user-space */
 500         sock->sk->reuse = 1;
 501
 502         if (sock->ops->bind(sock,
 503                             (struct sockaddr*)&mcast_addr,
 504                             sizeof(struct sockaddr)) < 0) {
 505                 IP_VS_ERR("Error binding to the multicast addr\n");
 506                 goto error;
 507         }
 508
 509         /* join the multicast group */
 510         if (join_mcast_group(sock->sk,
 511                              (struct in_addr*)&mcast_addr.sin_addr,
 512                              ip_vs_mcast_backup_ifn) < 0) {
 513                 IP_VS_ERR("Error joining to the multicast group\n");
 514                 goto error;
 515         }
 516
 517         return sock;
 518
 519   error:
 520         sock_release(sock);
 521         return NULL;
 522 }
 523
 524
 525 static int
 526 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
 527 {
 528         struct msghdr   msg;
 529         mm_segment_t    oldfs;
 530         struct iovec    iov;
 531         int             len;
 532
 533         EnterFunction(7);
 534         iov.iov_base     = (void *)buffer;
 535         iov.iov_len      = length;
 536         msg.msg_name     = 0;
 537         msg.msg_namelen  = 0;
 538         msg.msg_iov      = &iov;
 539         msg.msg_iovlen   = 1;
 540         msg.msg_control  = NULL;
 541         msg.msg_controllen = 0;
 542         msg.msg_flags    = MSG_DONTWAIT|MSG_NOSIGNAL;
 543
 544         oldfs = get_fs(); set_fs(KERNEL_DS);
 545         len = sock_sendmsg(sock, &msg, (size_t)(length));
 546         set_fs(oldfs);
 547
 548         LeaveFunction(7);
 549         return len;
 550 }
 551
 552 static void
 553 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
 554 {
 555         int msize;
 556
 557         msize = msg->size;
 558
 559         /* Put size in network byte order */
 560         msg->size = htons(msg->size);
 561
 562         if (ip_vs_send_async(sock, (char *)msg, msize) != msize)
 563                 IP_VS_ERR("ip_vs_send_async error\n");
 564 }
 565
 566 static int
 567 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
 568 {
 569         struct msghdr           msg;
 570         struct iovec            iov;
 571         int                     len;
 572         mm_segment_t            oldfs;
 573
 574         EnterFunction(7);
 575
 576         /* Receive a packet */
 577         iov.iov_base     = buffer;
 578         iov.iov_len      = (size_t)buflen;
 579         msg.msg_name     = 0;
 580         msg.msg_namelen  = 0;
 581         msg.msg_iov      = &iov;
 582         msg.msg_iovlen   = 1;
 583         msg.msg_control  = NULL;
 584         msg.msg_controllen = 0;
 585         msg.msg_flags    = 0;
 586
 587         oldfs = get_fs(); set_fs(KERNEL_DS);
 588         len = sock_recvmsg(sock, &msg, buflen, 0);
 589         set_fs(oldfs);
 590
 591         if (len < 0)
 592                 return -1;
 593
 594         LeaveFunction(7);
 595         return len;
 596 }
 597
 598
 599 static DECLARE_WAIT_QUEUE_HEAD(sync_wait);
 600 static pid_t sync_master_pid = 0;
 601 static pid_t sync_backup_pid = 0;
 602
 603 static DECLARE_WAIT_QUEUE_HEAD(stop_sync_wait);
 604 static int stop_master_sync = 0;
 605 static int stop_backup_sync = 0;
 606
 607 static void sync_master_loop(void)
 608 {
 609         struct socket *sock;
 610         struct ip_vs_sync_buff *sb;
 611
 612         /* create the sending multicast socket */
 613         sock = make_send_sock();
 614         if (!sock)
 615                 return;
 616
 617         IP_VS_INFO("sync thread started: state = MASTER, mcast_ifn = %s, "
 618                    "syncid = %d\n",
 619                    ip_vs_mcast_master_ifn, ip_vs_master_syncid);
 620
 621         for (;;) {
 622                 while ((sb=sb_dequeue())) {
 623                         ip_vs_send_sync_msg(sock, sb->mesg);
 624                         ip_vs_sync_buff_release(sb);
 625                 }
 626
 627                 /* check if entries stay in curr_sb for 2 seconds */
 628                 if ((sb = get_curr_sync_buff(2*HZ))) {
 629                         ip_vs_send_sync_msg(sock, sb->mesg);
 630                         ip_vs_sync_buff_release(sb);
 631                 }
 632
 633                 if (stop_master_sync)
 634                         break;
 635
 636                 __set_current_state(TASK_INTERRUPTIBLE);
 637                 schedule_timeout(HZ);
 638                 __set_current_state(TASK_RUNNING);
 639         }
 640
 641         /* clean up the sync_buff queue */
 642         while ((sb=sb_dequeue())) {
 643                 ip_vs_sync_buff_release(sb);
 644         }
 645
 646         /* clean up the current sync_buff */
 647         if ((sb = get_curr_sync_buff(0))) {
 648                 ip_vs_sync_buff_release(sb);
 649         }
 650
 651         /* release the sending multicast socket */
 652         sock_release(sock);
 653 }
 654
 655
 656 static void sync_backup_loop(void)
 657 {
 658         struct socket *sock;
 659         char *buf;
 660         int len;
 661
 662         if (!(buf=kmalloc(SYNC_MESG_MAX_SIZE, GFP_ATOMIC))) {
 663                 IP_VS_ERR("sync_backup_loop: kmalloc error\n");
 664                 return;
 665         }
 666
 667         /* create the receiving multicast socket */
 668         sock = make_receive_sock();
 669         if (!sock)
 670                 goto out;
 671
 672         IP_VS_INFO("sync thread started: state = BACKUP, mcast_ifn = %s, "
 673                    "syncid = %d\n",
 674                    ip_vs_mcast_backup_ifn, ip_vs_backup_syncid);
 675
 676         for (;;) {
 677                 /* do you have data now? */
 678                 while (!skb_queue_empty(&(sock->sk->receive_queue))) {
 679                         if ((len=ip_vs_receive(sock, buf,
 680                                                SYNC_MESG_MAX_SIZE))<=0) {
 681                                 IP_VS_ERR("receiving message error\n");
 682                                 break;
 683                         }
 684                         /* disable bottom half, because it accessed the data
 685                            shared by softirq while getting/creating conns */
 686                         local_bh_disable();
 687                         ip_vs_process_message(buf, len);
 688                         local_bh_enable();
 689                 }
 690
 691                 if (stop_backup_sync)
 692                         break;
 693
 694                 __set_current_state(TASK_INTERRUPTIBLE);
 695                 schedule_timeout(HZ);
 696                 __set_current_state(TASK_RUNNING);
 697         }
 698
 699         /* release the sending multicast socket */
 700         sock_release(sock);
 701
 702   out:
 703         kfree(buf);
 704 }
 705
 706 static void sync_pid_set(int sync_state, pid_t sync_pid)
 707 {
 708         if (sync_state == IP_VS_STATE_MASTER)
 709                 sync_master_pid = sync_pid;
 710         else if (sync_state == IP_VS_STATE_BACKUP)
 711                 sync_backup_pid = sync_pid;
 712 }
 713
 714 static void sync_stop_set(int sync_state, int set)
 715 {
 716         if (sync_state == IP_VS_STATE_MASTER)
 717                 stop_master_sync = set;
 718         else if (sync_state == IP_VS_STATE_BACKUP)
 719                 stop_backup_sync = set;
 720         else {
 721                 stop_master_sync = set;
 722                 stop_backup_sync = set;
 723         }
 724 }
 725
 726 static int sync_thread(void *startup)
 727 {
 728         DECLARE_WAITQUEUE(wait, current);
 729         mm_segment_t oldmm;
 730         int state = IP_VS_STATE_NONE;
 731
 732         MOD_INC_USE_COUNT;
 733         daemonize();
 734
 735         oldmm = get_fs();
 736         set_fs(KERNEL_DS);
 737
 738         if (ip_vs_sync_state & IP_VS_STATE_MASTER && !sync_master_pid) {
 739                 state = IP_VS_STATE_MASTER;
 740                 sprintf(current->comm, "ipvs_syncmaster");
 741         } else if (ip_vs_sync_state & IP_VS_STATE_BACKUP) {
 742                 state = IP_VS_STATE_BACKUP;
 743                 sprintf(current->comm, "ipvs_syncbackup");
 744         } else IP_VS_BUG();
 745
 746         /* Block all signals */
 747         spin_lock_irq(&current->sigmask_lock);
 748         siginitsetinv(&current->blocked, 0);
 749         recalc_sigpending(current);
 750         spin_unlock_irq(&current->sigmask_lock);
 751
 752         /* set up multicast address */
 753         mcast_addr.sin_family = AF_INET;
 754         mcast_addr.sin_port = htons(IP_VS_SYNC_PORT);
 755         mcast_addr.sin_addr.s_addr = htonl(IP_VS_SYNC_GROUP);
 756
 757         add_wait_queue(&sync_wait, &wait);
 758
 759         sync_pid_set(state, current->pid);
 760         complete((struct completion *)startup);
 761
 762         /* processing master/backup loop here */
 763         if (state == IP_VS_STATE_MASTER)
 764                 sync_master_loop();
 765         else if (state == IP_VS_STATE_BACKUP)
 766                 sync_backup_loop();
 767         else IP_VS_BUG();
 768
 769         remove_wait_queue(&sync_wait, &wait);
 770
 771         /* thread exits */
 772         sync_pid_set(state, 0);
 773         IP_VS_INFO("sync thread stopped!\n");
 774
 775         set_fs(oldmm);
 776         MOD_DEC_USE_COUNT;
 777
 778         sync_stop_set(state, 0);
 779         wake_up(&stop_sync_wait);
 780
 781         return 0;
 782 }
 783
 784
 785 static int fork_sync_thread(void *startup)
 786 {
 787         pid_t pid;
 788
 789         /* fork the sync thread here, then the parent process of the
 790            sync thread is the init process after this thread exits. */
 791   repeat:
 792         if ((pid = kernel_thread(sync_thread, startup, 0)) < 0) {
 793                 IP_VS_ERR("could not create sync_thread due to %d... "
 794                           "retrying.\n", pid);
 795                 current->state = TASK_UNINTERRUPTIBLE;
 796                 schedule_timeout(HZ);
 797                 goto repeat;
 798         }
 799
 800         return 0;
 801 }
 802
 803
 804 int start_sync_thread(int state, char *mcast_ifn, __u8 syncid)
 805 {
 806         DECLARE_COMPLETION(startup);
 807         pid_t pid;
 808
 809         if ((state == IP_VS_STATE_MASTER && sync_master_pid) ||
 810             (state == IP_VS_STATE_BACKUP && sync_backup_pid))
 811                 return -EEXIST;
 812
 813         IP_VS_DBG(7, "%s: pid %d\n", __FUNCTION__, current->pid);
 814         IP_VS_DBG(7, "Each ip_vs_sync_conn entry need %d bytes\n",
 815                   sizeof(struct ip_vs_sync_conn));
 816
 817         ip_vs_sync_state |= state;
 818         if (state == IP_VS_STATE_MASTER) {
 819                 strncpy(ip_vs_mcast_master_ifn, mcast_ifn, sizeof(ip_vs_mcast_master_ifn));
 820                 ip_vs_mcast_master_ifn[sizeof(ip_vs_mcast_master_ifn) - 1] = 0;
 821                 ip_vs_master_syncid = syncid;
 822         } else {
 823                 strncpy(ip_vs_mcast_backup_ifn, mcast_ifn, sizeof(ip_vs_mcast_backup_ifn));
 824                 ip_vs_mcast_backup_ifn[sizeof(ip_vs_mcast_backup_ifn) - 1] = 0;
 825                 ip_vs_backup_syncid = syncid;
 826         }
 827
 828   repeat:
 829         if ((pid = kernel_thread(fork_sync_thread, &startup, 0)) < 0) {
 830                 IP_VS_ERR("could not create fork_sync_thread due to %d... "
 831                           "retrying.\n", pid);
 832                 current->state = TASK_UNINTERRUPTIBLE;
 833                 schedule_timeout(HZ);
 834                 goto repeat;
 835         }
 836
 837         wait_for_completion(&startup);
 838
 839         return 0;
 840 }
 841
 842
 843 int stop_sync_thread(int state)
 844 {
 845         DECLARE_WAITQUEUE(wait, current);
 846
 847         if ((state == IP_VS_STATE_MASTER && !sync_master_pid) ||
 848             (state == IP_VS_STATE_BACKUP && !sync_backup_pid))
 849                 return -ESRCH;
 850
 851         IP_VS_DBG(7, "%s: pid %d\n", __FUNCTION__, current->pid);
 852         IP_VS_INFO("stopping sync thread %d ...\n",
 853                    (state == IP_VS_STATE_MASTER) ? sync_master_pid : sync_backup_pid);
 854
 855         __set_current_state(TASK_UNINTERRUPTIBLE);
 856         add_wait_queue(&stop_sync_wait, &wait);
 857         sync_stop_set(state, 1);
 858         ip_vs_sync_state -= state;
 859         wake_up(&sync_wait);
 860         schedule();
 861         __set_current_state(TASK_RUNNING);
 862         remove_wait_queue(&stop_sync_wait, &wait);
 863
 864         /* Note: no need to reap the sync thread, because its parent
 865            process is the init process */
 866
 867         if ((state == IP_VS_STATE_MASTER && stop_master_sync) ||
 868             (state == IP_VS_STATE_BACKUP && stop_backup_sync))
 869                 IP_VS_BUG();
 870
 871         return 0;
 872 }