net/ipv4/ipvs/ip_vs_sync.c

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