net/ipv4/ipvs/ip_vs_sync.c

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