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  */
  20
  21 #include <linux/module.h>
  22 #include <linux/slab.h>
  23 #include <linux/net.h>
  24 #include <linux/completion.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  |    SyncID     |            Size               |
  81       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  82       |                                                               |
  83       |                    IPVS Sync Connection (1)                   |
  84       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  85       |                            .                                  |
  86       |                            .                                  |
  87       |                            .                                  |
  88       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  89       |                                                               |
  90       |                    IPVS Sync Connection (n)                   |
  91       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  92 */
  93
  94 #define SYNC_MESG_HEADER_LEN    4
  95
  96 struct ip_vs_sync_mesg {
  97         __u8                    nr_conns;
  98         __u8                    syncid;
  99         __u16                   size;
 100
 101         /* ip_vs_sync_conn entries start here */
 102 };
 103
 104 /* the maximum length of sync (sending/receiving) message */
 105 static int sync_send_mesg_maxlen;
 106 static int sync_recv_mesg_maxlen;
 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_master_mcast_ifn[IP_VS_IFNAME_MAXLEN];
 134 char ip_vs_backup_mcast_ifn[IP_VS_IFNAME_MAXLEN];
 135
 136 /* multicast addr */
 137 static struct sockaddr_in mcast_addr;
 138
 139
 140 static inline void sb_queue_tail(struct ip_vs_sync_buff *sb)
 141 {
 142         spin_lock(&ip_vs_sync_lock);
 143         list_add_tail(&sb->list, &ip_vs_sync_queue);
 144         spin_unlock(&ip_vs_sync_lock);
 145 }
 146
 147 static inline struct ip_vs_sync_buff * sb_dequeue(void)
 148 {
 149         struct ip_vs_sync_buff *sb;
 150
 151         spin_lock_bh(&ip_vs_sync_lock);
 152         if (list_empty(&ip_vs_sync_queue)) {
 153                 sb = NULL;
 154         } else {
 155                 sb = list_entry(ip_vs_sync_queue.next,
 156                                 struct ip_vs_sync_buff,
 157                                 list);
 158                 list_del(&sb->list);
 159         }
 160         spin_unlock_bh(&ip_vs_sync_lock);
 161
 162         return sb;
 163 }
 164
 165 static inline struct ip_vs_sync_buff * ip_vs_sync_buff_create(void)
 166 {
 167         struct ip_vs_sync_buff *sb;
 168
 169         if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
 170                 return NULL;
 171
 172         if (!(sb->mesg=kmalloc(sync_send_mesg_maxlen, GFP_ATOMIC))) {
 173                 kfree(sb);
 174                 return NULL;
 175         }
 176         sb->mesg->nr_conns = 0;
 177         sb->mesg->syncid = ip_vs_master_syncid;
 178         sb->mesg->size = 4;
 179         sb->head = (unsigned char *)sb->mesg + 4;
 180         sb->end = (unsigned char *)sb->mesg + sync_send_mesg_maxlen;
 181         sb->firstuse = jiffies;
 182         return sb;
 183 }
 184
 185 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
 186 {
 187         kfree(sb->mesg);
 188         kfree(sb);
 189 }
 190
 191 /*
 192  *      Get the current sync buffer if it has been created for more
 193  *      than the specified time or the specified time is zero.
 194  */
 195 static inline struct ip_vs_sync_buff *
 196 get_curr_sync_buff(unsigned long time)
 197 {
 198         struct ip_vs_sync_buff *sb;
 199
 200         spin_lock_bh(&curr_sb_lock);
 201         if (curr_sb && (time == 0 ||
 202                         time_before(jiffies - curr_sb->firstuse, time))) {
 203                 sb = curr_sb;
 204                 curr_sb = NULL;
 205         } else
 206                 sb = NULL;
 207         spin_unlock_bh(&curr_sb_lock);
 208         return sb;
 209 }
 210
 211
 212 /*
 213  *      Add an ip_vs_conn information into the current sync_buff.
 214  *      Called by ip_vs_in.
 215  */
 216 void ip_vs_sync_conn(struct ip_vs_conn *cp)
 217 {
 218         struct ip_vs_sync_mesg *m;
 219         struct ip_vs_sync_conn *s;
 220         int len;
 221
 222         spin_lock(&curr_sb_lock);
 223         if (!curr_sb) {
 224                 if (!(curr_sb=ip_vs_sync_buff_create())) {
 225                         spin_unlock(&curr_sb_lock);
 226                         IP_VS_ERR("ip_vs_sync_buff_create failed.\n");
 227                         return;
 228                 }
 229         }
 230
 231         len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
 232                 SIMPLE_CONN_SIZE;
 233         m = curr_sb->mesg;
 234         s = (struct ip_vs_sync_conn *)curr_sb->head;
 235
 236         /* copy members */
 237         s->protocol = cp->protocol;
 238         s->cport = cp->cport;
 239         s->vport = cp->vport;
 240         s->dport = cp->dport;
 241         s->caddr = cp->caddr;
 242         s->vaddr = cp->vaddr;
 243         s->daddr = cp->daddr;
 244         s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
 245         s->state = htons(cp->state);
 246         if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
 247                 struct ip_vs_sync_conn_options *opt =
 248                         (struct ip_vs_sync_conn_options *)&s[1];
 249                 memcpy(opt, &cp->in_seq, sizeof(*opt));
 250         }
 251
 252         m->nr_conns++;
 253         m->size += len;
 254         curr_sb->head += len;
 255
 256         /* check if there is a space for next one */
 257         if (curr_sb->head+FULL_CONN_SIZE > curr_sb->end) {
 258                 sb_queue_tail(curr_sb);
 259                 curr_sb = NULL;
 260         }
 261         spin_unlock(&curr_sb_lock);
 262
 263         /* synchronize its controller if it has */
 264         if (cp->control)
 265                 ip_vs_sync_conn(cp->control);
 266 }
 267
 268
 269 /*
 270  *      Process received multicast message and create the corresponding
 271  *      ip_vs_conn entries.
 272  */
 273 static void ip_vs_process_message(const char *buffer, const size_t buflen)
 274 {
 275         struct ip_vs_sync_mesg *m = (struct ip_vs_sync_mesg *)buffer;
 276         struct ip_vs_sync_conn *s;
 277         struct ip_vs_sync_conn_options *opt;
 278         struct ip_vs_conn *cp;
 279         char *p;
 280         int i;
 281
 282         if (buflen != m->size) {
 283                 IP_VS_ERR("bogus message\n");
 284                 return;
 285         }
 286
 287         /* SyncID sanity check */
 288         if (ip_vs_backup_syncid != 0 && m->syncid != ip_vs_backup_syncid) {
 289                 IP_VS_DBG(7, "Ignoring incoming msg with syncid = %d\n",
 290                           m->syncid);
 291                 return;
 292         }
 293
 294         p = (char *)buffer + sizeof(struct ip_vs_sync_mesg);
 295         for (i=0; i<m->nr_conns; i++) {
 296                 s = (struct ip_vs_sync_conn *)p;
 297                 cp = ip_vs_conn_in_get(s->protocol,
 298                                        s->caddr, s->cport,
 299                                        s->vaddr, s->vport);
 300                 if (!cp) {
 301                         cp = ip_vs_conn_new(s->protocol,
 302                                             s->caddr, s->cport,
 303                                             s->vaddr, s->vport,
 304                                             s->daddr, s->dport,
 305                                             ntohs(s->flags), NULL);
 306                         if (!cp) {
 307                                 IP_VS_ERR("ip_vs_conn_new failed\n");
 308                                 return;
 309                         }
 310                         cp->state = ntohs(s->state);
 311                 } else if (!cp->dest) {
 312                         /* it is an entry created by the synchronization */
 313                         cp->state = ntohs(s->state);
 314                         cp->flags = ntohs(s->flags) | IP_VS_CONN_F_HASHED;
 315                 }       /* Note that we don't touch its state and flags
 316                            if it is a normal entry. */
 317
 318                 if (ntohs(s->flags) & IP_VS_CONN_F_SEQ_MASK) {
 319                         opt = (struct ip_vs_sync_conn_options *)&s[1];
 320                         memcpy(&cp->in_seq, opt, sizeof(*opt));
 321                         p += FULL_CONN_SIZE;
 322                 } else
 323                         p += SIMPLE_CONN_SIZE;
 324
 325                 atomic_set(&cp->in_pkts, sysctl_ip_vs_sync_threshold[0]);
 326                 cp->timeout = IP_VS_SYNC_CONN_TIMEOUT;
 327                 ip_vs_conn_put(cp);
 328
 329                 if (p > buffer+buflen) {
 330                         IP_VS_ERR("bogus message\n");
 331                         return;
 332                 }
 333         }
 334 }
 335
 336
 337 /*
 338  *      Setup loopback of outgoing multicasts on a sending socket
 339  */
 340 static void set_mcast_loop(struct sock *sk, u_char loop)
 341 {
 342         struct inet_opt *inet = inet_sk(sk);
 343
 344         /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
 345         lock_sock(sk);
 346         inet->mc_loop = loop ? 1 : 0;
 347         release_sock(sk);
 348 }
 349
 350 /*
 351  *      Specify TTL for outgoing multicasts on a sending socket
 352  */
 353 static void set_mcast_ttl(struct sock *sk, u_char ttl)
 354 {
 355         struct inet_opt *inet = inet_sk(sk);
 356
 357         /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
 358         lock_sock(sk);
 359         inet->mc_ttl = ttl;
 360         release_sock(sk);
 361 }
 362
 363 /*
 364  *      Specifiy default interface for outgoing multicasts
 365  */
 366 static int set_mcast_if(struct sock *sk, char *ifname)
 367 {
 368         struct net_device *dev;
 369         struct inet_opt *inet = inet_sk(sk);
 370
 371         if ((dev = __dev_get_by_name(ifname)) == NULL)
 372                 return -ENODEV;
 373
 374         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
 375                 return -EINVAL;
 376
 377         lock_sock(sk);
 378         inet->mc_index = dev->ifindex;
 379         /*  inet->mc_addr  = 0; */
 380         release_sock(sk);
 381
 382         return 0;
 383 }
 384
 385
 386 /*
 387  *      Set the maximum length of sync message according to the
 388  *      specified interface's MTU.
 389  */
 390 static int set_sync_mesg_maxlen(int sync_state)
 391 {
 392         struct net_device *dev;
 393         int num;
 394
 395         if (sync_state == IP_VS_STATE_MASTER) {
 396                 if ((dev = __dev_get_by_name(ip_vs_master_mcast_ifn)) == NULL)
 397                         return -ENODEV;
 398
 399                 num = (dev->mtu - sizeof(struct iphdr) -
 400                        sizeof(struct udphdr) -
 401                        SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE;
 402                 sync_send_mesg_maxlen =
 403                         SYNC_MESG_HEADER_LEN + SIMPLE_CONN_SIZE * num;
 404                 IP_VS_DBG(7, "setting the maximum length of sync sending "
 405                           "message %d.\n", sync_send_mesg_maxlen);
 406         } else if (sync_state == IP_VS_STATE_BACKUP) {
 407                 if ((dev = __dev_get_by_name(ip_vs_backup_mcast_ifn)) == NULL)
 408                         return -ENODEV;
 409
 410                 sync_recv_mesg_maxlen = dev->mtu -
 411                         sizeof(struct iphdr) - sizeof(struct udphdr);
 412                 IP_VS_DBG(7, "setting the maximum length of sync receiving "
 413                           "message %d.\n", sync_recv_mesg_maxlen);
 414         }
 415
 416         return 0;
 417 }
 418
 419
 420 /*
 421  *      Join a multicast group.
 422  *      the group is specified by a class D multicast address 224.0.0.0/8
 423  *      in the in_addr structure passed in as a parameter.
 424  */
 425 static int
 426 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
 427 {
 428         struct ip_mreqn mreq;
 429         struct net_device *dev;
 430         int ret;
 431
 432         memset(&mreq, 0, sizeof(mreq));
 433         memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
 434
 435         if ((dev = __dev_get_by_name(ifname)) == NULL)
 436                 return -ENODEV;
 437         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
 438                 return -EINVAL;
 439
 440         mreq.imr_ifindex = dev->ifindex;
 441
 442         lock_sock(sk);
 443         ret = ip_mc_join_group(sk, &mreq);
 444         release_sock(sk);
 445
 446         return ret;
 447 }
 448
 449
 450 static int bind_mcastif_addr(struct socket *sock, char *ifname)
 451 {
 452         struct net_device *dev;
 453         u32 addr;
 454         struct sockaddr_in sin;
 455
 456         if ((dev = __dev_get_by_name(ifname)) == NULL)
 457                 return -ENODEV;
 458
 459         addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
 460         if (!addr)
 461                 IP_VS_ERR("You probably need to specify IP address on "
 462                           "multicast interface.\n");
 463
 464         IP_VS_DBG(7, "binding socket with (%s) %u.%u.%u.%u\n",
 465                   ifname, NIPQUAD(addr));
 466
 467         /* Now bind the socket with the address of multicast interface */
 468         sin.sin_family       = AF_INET;
 469         sin.sin_addr.s_addr  = addr;
 470         sin.sin_port         = 0;
 471
 472         return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
 473 }
 474
 475 /*
 476  *      Set up sending multicast socket over UDP
 477  */
 478 static struct socket * make_send_sock(void)
 479 {
 480         struct socket *sock;
 481
 482         /* First create a socket */
 483         if (sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock) < 0) {
 484                 IP_VS_ERR("Error during creation of socket; terminating\n");
 485                 return NULL;
 486         }
 487
 488         if (set_mcast_if(sock->sk, ip_vs_master_mcast_ifn) < 0) {
 489                 IP_VS_ERR("Error setting outbound mcast interface\n");
 490                 goto error;
 491         }
 492
 493         set_mcast_loop(sock->sk, 0);
 494         set_mcast_ttl(sock->sk, 1);
 495
 496         if (bind_mcastif_addr(sock, ip_vs_master_mcast_ifn) < 0) {
 497                 IP_VS_ERR("Error binding address of the mcast interface\n");
 498                 goto error;
 499         }
 500
 501         if (sock->ops->connect(sock,
 502                                (struct sockaddr*)&mcast_addr,
 503                                sizeof(struct sockaddr), 0) < 0) {
 504                 IP_VS_ERR("Error connecting to the multicast addr\n");
 505                 goto error;
 506         }
 507
 508         return sock;
 509
 510   error:
 511         sock_release(sock);
 512         return NULL;
 513 }
 514
 515
 516 /*
 517  *      Set up receiving multicast socket over UDP
 518  */
 519 static struct socket * make_receive_sock(void)
 520 {
 521         struct socket *sock;
 522
 523         /* First create a socket */
 524         if (sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock) < 0) {
 525                 IP_VS_ERR("Error during creation of socket; terminating\n");
 526                 return NULL;
 527         }
 528
 529         /* it is equivalent to the REUSEADDR option in user-space */
 530         sock->sk->sk_reuse = 1;
 531
 532         if (sock->ops->bind(sock,
 533                             (struct sockaddr*)&mcast_addr,
 534                             sizeof(struct sockaddr)) < 0) {
 535                 IP_VS_ERR("Error binding to the multicast addr\n");
 536                 goto error;
 537         }
 538
 539         /* join the multicast group */
 540         if (join_mcast_group(sock->sk,
 541                              (struct in_addr*)&mcast_addr.sin_addr,
 542                              ip_vs_backup_mcast_ifn) < 0) {
 543                 IP_VS_ERR("Error joining to the multicast group\n");
 544                 goto error;
 545         }
 546
 547         return sock;
 548
 549   error:
 550         sock_release(sock);
 551         return NULL;
 552 }
 553
 554
 555 static int
 556 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
 557 {
 558         struct msghdr   msg;
 559         mm_segment_t    oldfs;
 560         struct iovec    iov;
 561         int             len;
 562
 563         EnterFunction(7);
 564         iov.iov_base     = (void *)buffer;
 565         iov.iov_len      = length;
 566         msg.msg_name     = 0;
 567         msg.msg_namelen  = 0;
 568         msg.msg_iov      = &iov;
 569         msg.msg_iovlen   = 1;
 570         msg.msg_control  = NULL;
 571         msg.msg_controllen = 0;
 572         msg.msg_flags    = MSG_DONTWAIT|MSG_NOSIGNAL;
 573
 574         oldfs = get_fs(); set_fs(KERNEL_DS);
 575         len = sock_sendmsg(sock, &msg, (size_t)(length));
 576         set_fs(oldfs);
 577
 578         LeaveFunction(7);
 579         return len;
 580 }
 581
 582
 583 static int
 584 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
 585 {
 586         struct msghdr           msg;
 587         struct iovec            iov;
 588         int                     len;
 589         mm_segment_t            oldfs;
 590
 591         EnterFunction(7);
 592
 593         /* Receive a packet */
 594         iov.iov_base     = buffer;
 595         iov.iov_len      = (size_t)buflen;
 596         msg.msg_name     = 0;
 597         msg.msg_namelen  = 0;
 598         msg.msg_iov      = &iov;
 599         msg.msg_iovlen   = 1;
 600         msg.msg_control  = NULL;
 601         msg.msg_controllen = 0;
 602         msg.msg_flags    = 0;
 603
 604         oldfs = get_fs(); set_fs(KERNEL_DS);
 605         len = sock_recvmsg(sock, &msg, buflen, 0);
 606         set_fs(oldfs);
 607
 608         if (len < 0)
 609                 return -1;
 610
 611         LeaveFunction(7);
 612         return len;
 613 }
 614
 615
 616 static DECLARE_WAIT_QUEUE_HEAD(sync_wait);
 617 static pid_t sync_master_pid = 0;
 618 static pid_t sync_backup_pid = 0;
 619
 620 static DECLARE_WAIT_QUEUE_HEAD(stop_sync_wait);
 621 static int stop_master_sync = 0;
 622 static int stop_backup_sync = 0;
 623
 624 static void sync_master_loop(void)
 625 {
 626         struct socket *sock;
 627         struct ip_vs_sync_buff *sb;
 628         struct ip_vs_sync_mesg *m;
 629
 630         /* create the sending multicast socket */
 631         sock = make_send_sock();
 632         if (!sock)
 633                 return;
 634
 635         IP_VS_INFO("sync thread started: state = MASTER, mcast_ifn = %s, "
 636                    "syncid = %d\n",
 637                    ip_vs_master_mcast_ifn, ip_vs_master_syncid);
 638
 639         for (;;) {
 640                 while ((sb=sb_dequeue())) {
 641                         m = sb->mesg;
 642                         if (ip_vs_send_async(sock, (char *)m,
 643                                              m->size) != m->size)
 644                                 IP_VS_ERR("ip_vs_send_async error\n");
 645                         ip_vs_sync_buff_release(sb);
 646                 }
 647
 648                 /* check if entries stay in curr_sb for 2 seconds */
 649                 if ((sb = get_curr_sync_buff(2*HZ))) {
 650                         m = sb->mesg;
 651                         if (ip_vs_send_async(sock, (char *)m,
 652                                              m->size) != m->size)
 653                                 IP_VS_ERR("ip_vs_send_async error\n");
 654                         ip_vs_sync_buff_release(sb);
 655                 }
 656
 657                 if (stop_master_sync)
 658                         break;
 659
 660                 __set_current_state(TASK_INTERRUPTIBLE);
 661                 schedule_timeout(HZ);
 662         }
 663
 664         /* clean up the sync_buff queue */
 665         while ((sb=sb_dequeue())) {
 666                 ip_vs_sync_buff_release(sb);
 667         }
 668
 669         /* clean up the current sync_buff */
 670         if ((sb = get_curr_sync_buff(0))) {
 671                 ip_vs_sync_buff_release(sb);
 672         }
 673
 674         /* release the sending multicast socket */
 675         sock_release(sock);
 676 }
 677
 678
 679 static void sync_backup_loop(void)
 680 {
 681         struct socket *sock;
 682         char *buf;
 683         int len;
 684
 685         if (!(buf = kmalloc(sync_recv_mesg_maxlen, GFP_ATOMIC))) {
 686                 IP_VS_ERR("sync_backup_loop: kmalloc error\n");
 687                 return;
 688         }
 689
 690         /* create the receiving multicast socket */
 691         sock = make_receive_sock();
 692         if (!sock)
 693                 goto out;
 694
 695         IP_VS_INFO("sync thread started: state = BACKUP, mcast_ifn = %s, "
 696                    "syncid = %d\n",
 697                    ip_vs_backup_mcast_ifn, ip_vs_backup_syncid);
 698
 699         for (;;) {
 700                 /* do you have data now? */
 701                 while (!skb_queue_empty(&(sock->sk->sk_receive_queue))) {
 702                         if ((len =
 703                              ip_vs_receive(sock, buf,
 704                                            sync_recv_mesg_maxlen)) <= 0) {
 705                                 IP_VS_ERR("receiving message error\n");
 706                                 break;
 707                         }
 708                         /* disable bottom half, because it accessed the data
 709                            shared by softirq while getting/creating conns */
 710                         local_bh_disable();
 711                         ip_vs_process_message(buf, len);
 712                         local_bh_enable();
 713                 }
 714
 715                 if (stop_backup_sync)
 716                         break;
 717
 718                 __set_current_state(TASK_INTERRUPTIBLE);
 719                 schedule_timeout(HZ);
 720         }
 721
 722         /* release the sending multicast socket */
 723         sock_release(sock);
 724
 725   out:
 726         kfree(buf);
 727 }
 728
 729
 730 static void set_sync_pid(int sync_state, pid_t sync_pid)
 731 {
 732         if (sync_state == IP_VS_STATE_MASTER)
 733                 sync_master_pid = sync_pid;
 734         else if (sync_state == IP_VS_STATE_BACKUP)
 735                 sync_backup_pid = sync_pid;
 736 }
 737
 738 static void set_stop_sync(int sync_state, int set)
 739 {
 740         if (sync_state == IP_VS_STATE_MASTER)
 741                 stop_master_sync = set;
 742         else if (sync_state == IP_VS_STATE_BACKUP)
 743                 stop_backup_sync = set;
 744         else {
 745                 stop_master_sync = set;
 746                 stop_backup_sync = set;
 747         }
 748 }
 749
 750 static int sync_thread(void *startup)
 751 {
 752         DECLARE_WAITQUEUE(wait, current);
 753         mm_segment_t oldmm;
 754         int state;
 755         const char *name;
 756
 757         /* increase the module use count */
 758         ip_vs_use_count_inc();
 759
 760         if (ip_vs_sync_state & IP_VS_STATE_MASTER && !sync_master_pid) {
 761                 state = IP_VS_STATE_MASTER;
 762                 name = "ipvs_syncmaster";
 763         } else if (ip_vs_sync_state & IP_VS_STATE_BACKUP && !sync_backup_pid) {
 764                 state = IP_VS_STATE_BACKUP;
 765                 name = "ipvs_syncbackup";
 766         } else {
 767                 IP_VS_BUG();
 768                 ip_vs_use_count_dec();
 769                 return -EINVAL;
 770         }
 771
 772         daemonize(name);
 773
 774         oldmm = get_fs();
 775         set_fs(KERNEL_DS);
 776
 777         /* Block all signals */
 778         spin_lock_irq(&current->sighand->siglock);
 779         siginitsetinv(&current->blocked, 0);
 780         recalc_sigpending();
 781         spin_unlock_irq(&current->sighand->siglock);
 782
 783         /* set the maximum length of sync message */
 784         set_sync_mesg_maxlen(state);
 785
 786         /* set up multicast address */
 787         mcast_addr.sin_family = AF_INET;
 788         mcast_addr.sin_port = htons(IP_VS_SYNC_PORT);
 789         mcast_addr.sin_addr.s_addr = htonl(IP_VS_SYNC_GROUP);
 790
 791         add_wait_queue(&sync_wait, &wait);
 792
 793         set_sync_pid(state, current->pid);
 794         complete((struct completion *)startup);
 795
 796         /* processing master/backup loop here */
 797         if (state == IP_VS_STATE_MASTER)
 798                 sync_master_loop();
 799         else if (state == IP_VS_STATE_BACKUP)
 800                 sync_backup_loop();
 801         else IP_VS_BUG();
 802
 803         remove_wait_queue(&sync_wait, &wait);
 804
 805         /* thread exits */
 806         set_sync_pid(state, 0);
 807         IP_VS_INFO("sync thread stopped!\n");
 808
 809         set_fs(oldmm);
 810
 811         /* decrease the module use count */
 812         ip_vs_use_count_dec();
 813
 814         set_stop_sync(state, 0);
 815         wake_up(&stop_sync_wait);
 816
 817         return 0;
 818 }
 819
 820
 821 static int fork_sync_thread(void *startup)
 822 {
 823         pid_t pid;
 824
 825         /* fork the sync thread here, then the parent process of the
 826            sync thread is the init process after this thread exits. */
 827   repeat:
 828         if ((pid = kernel_thread(sync_thread, startup, 0)) < 0) {
 829                 IP_VS_ERR("could not create sync_thread due to %d... "
 830                           "retrying.\n", pid);
 831                 current->state = TASK_UNINTERRUPTIBLE;
 832                 schedule_timeout(HZ);
 833                 goto repeat;
 834         }
 835
 836         return 0;
 837 }
 838
 839
 840 int start_sync_thread(int state, char *mcast_ifn, __u8 syncid)
 841 {
 842         DECLARE_COMPLETION(startup);
 843         pid_t pid;
 844
 845         if ((state == IP_VS_STATE_MASTER && sync_master_pid) ||
 846             (state == IP_VS_STATE_BACKUP && sync_backup_pid))
 847                 return -EEXIST;
 848
 849         IP_VS_DBG(7, "%s: pid %d\n", __FUNCTION__, current->pid);
 850         IP_VS_DBG(7, "Each ip_vs_sync_conn entry need %Zd bytes\n",
 851                   sizeof(struct ip_vs_sync_conn));
 852
 853         ip_vs_sync_state |= state;
 854         if (state == IP_VS_STATE_MASTER) {
 855                 strcpy(ip_vs_master_mcast_ifn, mcast_ifn);
 856                 ip_vs_master_syncid = syncid;
 857         } else {
 858                 strcpy(ip_vs_backup_mcast_ifn, mcast_ifn);
 859                 ip_vs_backup_syncid = syncid;
 860         }
 861
 862   repeat:
 863         if ((pid = kernel_thread(fork_sync_thread, &startup, 0)) < 0) {
 864                 IP_VS_ERR("could not create fork_sync_thread due to %d... "
 865                           "retrying.\n", pid);
 866                 current->state = TASK_UNINTERRUPTIBLE;
 867                 schedule_timeout(HZ);
 868                 goto repeat;
 869         }
 870
 871         wait_for_completion(&startup);
 872
 873         return 0;
 874 }
 875
 876
 877 int stop_sync_thread(int state)
 878 {
 879         DECLARE_WAITQUEUE(wait, current);
 880
 881         if ((state == IP_VS_STATE_MASTER && !sync_master_pid) ||
 882             (state == IP_VS_STATE_BACKUP && !sync_backup_pid))
 883                 return -ESRCH;
 884
 885         IP_VS_DBG(7, "%s: pid %d\n", __FUNCTION__, current->pid);
 886         IP_VS_INFO("stopping sync thread %d ...\n",
 887                    (state == IP_VS_STATE_MASTER) ? sync_master_pid : sync_backup_pid);
 888
 889         __set_current_state(TASK_UNINTERRUPTIBLE);
 890         add_wait_queue(&stop_sync_wait, &wait);
 891         set_stop_sync(state, 1);
 892         ip_vs_sync_state -= state;
 893         wake_up(&sync_wait);
 894         schedule();
 895         __set_current_state(TASK_RUNNING);
 896         remove_wait_queue(&stop_sync_wait, &wait);
 897
 898         /* Note: no need to reap the sync thread, because its parent
 899            process is the init process */
 900
 901         if ((state == IP_VS_STATE_MASTER && stop_master_sync) ||
 902             (state == IP_VS_STATE_BACKUP && stop_backup_sync))
 903                 IP_VS_BUG();
 904
 905         return 0;
 906 }