git.oblomov.eu Git - linux-2.6/blob - net/ipv4/ipvs/ip_vs_sync.c

   1 /*
   2  * IPVS         An implementation of the IP virtual server support for the
   3  *              LINUX operating system.  IPVS is now implemented as a module
   4  *              over the NetFilter framework. IPVS can be used to build a
   5  *              high-performance and highly available server based on a
   6  *              cluster of servers.
   7  *
   8  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
   9  *
  10  * ip_vs_sync:  sync connection info from master load balancer to backups
  11  *              through multicast
  12  *
  13  * Changes:
  14  *      Alexandre Cassen        :       Added master & backup support at a time.
  15  *      Alexandre Cassen        :       Added SyncID support for incoming sync
  16  *                                      messages filtering.
  17  *      Justin Ossevoort        :       Fix endian problem on sync message size.
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/slab.h>
  22 #include <linux/inetdevice.h>
  23 #include <linux/net.h>
  24 #include <linux/completion.h>
  25 #include <linux/delay.h>
  26 #include <linux/skbuff.h>
  27 #include <linux/in.h>
  28 #include <linux/igmp.h>                 /* for ip_mc_join_group */
  29 #include <linux/udp.h>
  30
  31 #include <net/ip.h>
  32 #include <net/sock.h>
  33 #include <asm/uaccess.h>                /* for get_fs and set_fs */
  34
  35 #include <net/ip_vs.h>
  36
  37 #define IP_VS_SYNC_GROUP 0xe0000051    /* multicast addr - 224.0.0.81 */
  38 #define IP_VS_SYNC_PORT  8848          /* multicast port */
  39
  40
  41 /*
  42  *      IPVS sync connection entry
  43  */
  44 struct ip_vs_sync_conn {
  45         __u8                    reserved;
  46
  47         /* Protocol, addresses and port numbers */
  48         __u8                    protocol;       /* Which protocol (TCP/UDP) */
  49         __be16                  cport;
  50         __be16                  vport;
  51         __be16                  dport;
  52         __be32                  caddr;          /* client address */
  53         __be32                  vaddr;          /* virtual address */
  54         __be32                  daddr;          /* destination address */
  55
  56         /* Flags and state transition */
  57         __be16                  flags;          /* status flags */
  58         __be16                  state;          /* state info */
  59
  60         /* The sequence options start here */
  61 };
  62
  63 struct ip_vs_sync_conn_options {
  64         struct ip_vs_seq        in_seq;         /* incoming seq. struct */
  65         struct ip_vs_seq        out_seq;        /* outgoing seq. struct */
  66 };
  67
  68 struct ip_vs_sync_thread_data {
  69         struct completion *startup;
  70         int state;
  71 };
  72
  73 #define SIMPLE_CONN_SIZE  (sizeof(struct ip_vs_sync_conn))
  74 #define FULL_CONN_SIZE  \
  75 (sizeof(struct ip_vs_sync_conn) + sizeof(struct ip_vs_sync_conn_options))
  76
  77
  78 /*
  79   The master mulitcasts messages to the backup load balancers in the
  80   following format.
  81
  82        0                   1                   2                   3
  83        0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
  84       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  85       |  Count Conns  |    SyncID     |            Size               |
  86       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  87       |                                                               |
  88       |                    IPVS Sync Connection (1)                   |
  89       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  90       |                            .                                  |
  91       |                            .                                  |
  92       |                            .                                  |
  93       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  94       |                                                               |
  95       |                    IPVS Sync Connection (n)                   |
  96       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  97 */
  98
  99 #define SYNC_MESG_HEADER_LEN    4
 100
 101 struct ip_vs_sync_mesg {
 102         __u8                    nr_conns;
 103         __u8                    syncid;
 104         __u16                   size;
 105
 106         /* ip_vs_sync_conn entries start here */
 107 };
 108
 109 /* the maximum length of sync (sending/receiving) message */
 110 static int sync_send_mesg_maxlen;
 111 static int sync_recv_mesg_maxlen;
 112
 113 struct ip_vs_sync_buff {
 114         struct list_head        list;
 115         unsigned long           firstuse;
 116
 117         /* pointers for the message data */
 118         struct ip_vs_sync_mesg  *mesg;
 119         unsigned char           *head;
 120         unsigned char           *end;
 121 };
 122
 123
 124 /* the sync_buff list head and the lock */
 125 static LIST_HEAD(ip_vs_sync_queue);
 126 static DEFINE_SPINLOCK(ip_vs_sync_lock);
 127
 128 /* current sync_buff for accepting new conn entries */
 129 static struct ip_vs_sync_buff   *curr_sb = NULL;
 130 static DEFINE_SPINLOCK(curr_sb_lock);
 131
 132 /* ipvs sync daemon state */
 133 volatile int ip_vs_sync_state = IP_VS_STATE_NONE;
 134 volatile int ip_vs_master_syncid = 0;
 135 volatile int ip_vs_backup_syncid = 0;
 136
 137 /* multicast interface name */
 138 char ip_vs_master_mcast_ifn[IP_VS_IFNAME_MAXLEN];
 139 char ip_vs_backup_mcast_ifn[IP_VS_IFNAME_MAXLEN];
 140
 141 /* multicast addr */
 142 static struct sockaddr_in mcast_addr = {
 143         .sin_family             = AF_INET,
 144         .sin_port               = __constant_htons(IP_VS_SYNC_PORT),
 145         .sin_addr.s_addr        = __constant_htonl(IP_VS_SYNC_GROUP),
 146 };
 147
 148
 149 static inline void sb_queue_tail(struct ip_vs_sync_buff *sb)
 150 {
 151         spin_lock(&ip_vs_sync_lock);
 152         list_add_tail(&sb->list, &ip_vs_sync_queue);
 153         spin_unlock(&ip_vs_sync_lock);
 154 }
 155
 156 static inline struct ip_vs_sync_buff * sb_dequeue(void)
 157 {
 158         struct ip_vs_sync_buff *sb;
 159
 160         spin_lock_bh(&ip_vs_sync_lock);
 161         if (list_empty(&ip_vs_sync_queue)) {
 162                 sb = NULL;
 163         } else {
 164                 sb = list_entry(ip_vs_sync_queue.next,
 165                                 struct ip_vs_sync_buff,
 166                                 list);
 167                 list_del(&sb->list);
 168         }
 169         spin_unlock_bh(&ip_vs_sync_lock);
 170
 171         return sb;
 172 }
 173
 174 static inline struct ip_vs_sync_buff * ip_vs_sync_buff_create(void)
 175 {
 176         struct ip_vs_sync_buff *sb;
 177
 178         if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
 179                 return NULL;
 180
 181         if (!(sb->mesg=kmalloc(sync_send_mesg_maxlen, GFP_ATOMIC))) {
 182                 kfree(sb);
 183                 return NULL;
 184         }
 185         sb->mesg->nr_conns = 0;
 186         sb->mesg->syncid = ip_vs_master_syncid;
 187         sb->mesg->size = 4;
 188         sb->head = (unsigned char *)sb->mesg + 4;
 189         sb->end = (unsigned char *)sb->mesg + sync_send_mesg_maxlen;
 190         sb->firstuse = jiffies;
 191         return sb;
 192 }
 193
 194 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
 195 {
 196         kfree(sb->mesg);
 197         kfree(sb);
 198 }
 199
 200 /*
 201  *      Get the current sync buffer if it has been created for more
 202  *      than the specified time or the specified time is zero.
 203  */
 204 static inline struct ip_vs_sync_buff *
 205 get_curr_sync_buff(unsigned long time)
 206 {
 207         struct ip_vs_sync_buff *sb;
 208
 209         spin_lock_bh(&curr_sb_lock);
 210         if (curr_sb && (time == 0 ||
 211                         time_before(jiffies - curr_sb->firstuse, time))) {
 212                 sb = curr_sb;
 213                 curr_sb = NULL;
 214         } else
 215                 sb = NULL;
 216         spin_unlock_bh(&curr_sb_lock);
 217         return sb;
 218 }
 219
 220
 221 /*
 222  *      Add an ip_vs_conn information into the current sync_buff.
 223  *      Called by ip_vs_in.
 224  */
 225 void ip_vs_sync_conn(struct ip_vs_conn *cp)
 226 {
 227         struct ip_vs_sync_mesg *m;
 228         struct ip_vs_sync_conn *s;
 229         int len;
 230
 231         spin_lock(&curr_sb_lock);
 232         if (!curr_sb) {
 233                 if (!(curr_sb=ip_vs_sync_buff_create())) {
 234                         spin_unlock(&curr_sb_lock);
 235                         IP_VS_ERR("ip_vs_sync_buff_create failed.\n");
 236                         return;
 237                 }
 238         }
 239
 240         len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
 241                 SIMPLE_CONN_SIZE;
 242         m = curr_sb->mesg;
 243         s = (struct ip_vs_sync_conn *)curr_sb->head;
 244
 245         /* copy members */
 246         s->protocol = cp->protocol;
 247         s->cport = cp->cport;
 248         s->vport = cp->vport;
 249         s->dport = cp->dport;
 250         s->caddr = cp->caddr;
 251         s->vaddr = cp->vaddr;
 252         s->daddr = cp->daddr;
 253         s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
 254         s->state = htons(cp->state);
 255         if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
 256                 struct ip_vs_sync_conn_options *opt =
 257                         (struct ip_vs_sync_conn_options *)&s[1];
 258                 memcpy(opt, &cp->in_seq, sizeof(*opt));
 259         }
 260
 261         m->nr_conns++;
 262         m->size += len;
 263         curr_sb->head += len;
 264
 265         /* check if there is a space for next one */
 266         if (curr_sb->head+FULL_CONN_SIZE > curr_sb->end) {
 267                 sb_queue_tail(curr_sb);
 268                 curr_sb = NULL;
 269         }
 270         spin_unlock(&curr_sb_lock);
 271
 272         /* synchronize its controller if it has */
 273         if (cp->control)
 274                 ip_vs_sync_conn(cp->control);
 275 }
 276
 277
 278 /*
 279  *      Process received multicast message and create the corresponding
 280  *      ip_vs_conn entries.
 281  */
 282 static void ip_vs_process_message(const char *buffer, const size_t buflen)
 283 {
 284         struct ip_vs_sync_mesg *m = (struct ip_vs_sync_mesg *)buffer;
 285         struct ip_vs_sync_conn *s;
 286         struct ip_vs_sync_conn_options *opt;
 287         struct ip_vs_conn *cp;
 288         struct ip_vs_protocol *pp;
 289         struct ip_vs_dest *dest;
 290         char *p;
 291         int i;
 292
 293         if (buflen < sizeof(struct ip_vs_sync_mesg)) {
 294                 IP_VS_ERR_RL("sync message header too short\n");
 295                 return;
 296         }
 297
 298         /* Convert size back to host byte order */
 299         m->size = ntohs(m->size);
 300
 301         if (buflen != m->size) {
 302                 IP_VS_ERR_RL("bogus sync message size\n");
 303                 return;
 304         }
 305
 306         /* SyncID sanity check */
 307         if (ip_vs_backup_syncid != 0 && m->syncid != ip_vs_backup_syncid) {
 308                 IP_VS_DBG(7, "Ignoring incoming msg with syncid = %d\n",
 309                           m->syncid);
 310                 return;
 311         }
 312
 313         p = (char *)buffer + sizeof(struct ip_vs_sync_mesg);
 314         for (i=0; i<m->nr_conns; i++) {
 315                 unsigned flags, state;
 316
 317                 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
 318                         IP_VS_ERR_RL("bogus conn in sync message\n");
 319                         return;
 320                 }
 321                 s = (struct ip_vs_sync_conn *) p;
 322                 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
 323                 flags &= ~IP_VS_CONN_F_HASHED;
 324                 if (flags & IP_VS_CONN_F_SEQ_MASK) {
 325                         opt = (struct ip_vs_sync_conn_options *)&s[1];
 326                         p += FULL_CONN_SIZE;
 327                         if (p > buffer+buflen) {
 328                                 IP_VS_ERR_RL("bogus conn options in sync message\n");
 329                                 return;
 330                         }
 331                 } else {
 332                         opt = NULL;
 333                         p += SIMPLE_CONN_SIZE;
 334                 }
 335
 336                 state = ntohs(s->state);
 337                 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
 338                         pp = ip_vs_proto_get(s->protocol);
 339                         if (!pp) {
 340                                 IP_VS_ERR_RL("Unsupported protocol %u in sync msg\n",
 341                                         s->protocol);
 342                                 continue;
 343                         }
 344                         if (state >= pp->num_states) {
 345                                 IP_VS_DBG(2, "Invalid %s state %u in sync msg\n",
 346                                         pp->name, state);
 347                                 continue;
 348                         }
 349                 } else {
 350                         /* protocol in templates is not used for state/timeout */
 351                         pp = NULL;
 352                         if (state > 0) {
 353                                 IP_VS_DBG(2, "Invalid template state %u in sync msg\n",
 354                                         state);
 355                                 state = 0;
 356                         }
 357                 }
 358
 359                 if (!(flags & IP_VS_CONN_F_TEMPLATE))
 360                         cp = ip_vs_conn_in_get(s->protocol,
 361                                                s->caddr, s->cport,
 362                                                s->vaddr, s->vport);
 363                 else
 364                         cp = ip_vs_ct_in_get(s->protocol,
 365                                                s->caddr, s->cport,
 366                                                s->vaddr, s->vport);
 367                 if (!cp) {
 368                         /*
 369                          * Find the appropriate destination for the connection.
 370                          * If it is not found the connection will remain unbound
 371                          * but still handled.
 372                          */
 373                         dest = ip_vs_find_dest(s->daddr, s->dport,
 374                                                s->vaddr, s->vport,
 375                                                s->protocol);
 376                         /*  Set the approprite ativity flag */
 377                         if (s->protocol == IPPROTO_TCP) {
 378                                 if (state != IP_VS_TCP_S_ESTABLISHED)
 379                                         flags |= IP_VS_CONN_F_INACTIVE;
 380                                 else
 381                                         flags &= ~IP_VS_CONN_F_INACTIVE;
 382                         }
 383                         cp = ip_vs_conn_new(s->protocol,
 384                                             s->caddr, s->cport,
 385                                             s->vaddr, s->vport,
 386                                             s->daddr, s->dport,
 387                                             flags, dest);
 388                         if (dest)
 389                                 atomic_dec(&dest->refcnt);
 390                         if (!cp) {
 391                                 IP_VS_ERR("ip_vs_conn_new failed\n");
 392                                 return;
 393                         }
 394                 } else if (!cp->dest) {
 395                         dest = ip_vs_try_bind_dest(cp);
 396                         if (dest)
 397                                 atomic_dec(&dest->refcnt);
 398                 } else if ((cp->dest) && (cp->protocol == IPPROTO_TCP) &&
 399                            (cp->state != state)) {
 400                         /* update active/inactive flag for the connection */
 401                         dest = cp->dest;
 402                         if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
 403                                 (state != IP_VS_TCP_S_ESTABLISHED)) {
 404                                 atomic_dec(&dest->activeconns);
 405                                 atomic_inc(&dest->inactconns);
 406                                 cp->flags |= IP_VS_CONN_F_INACTIVE;
 407                         } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) &&
 408                                 (state == IP_VS_TCP_S_ESTABLISHED)) {
 409                                 atomic_inc(&dest->activeconns);
 410                                 atomic_dec(&dest->inactconns);
 411                                 cp->flags &= ~IP_VS_CONN_F_INACTIVE;
 412                         }
 413                 }
 414
 415                 if (opt)
 416                         memcpy(&cp->in_seq, opt, sizeof(*opt));
 417                 atomic_set(&cp->in_pkts, sysctl_ip_vs_sync_threshold[0]);
 418                 cp->state = state;
 419                 cp->old_state = cp->state;
 420                 /*
 421                  * We can not recover the right timeout for templates
 422                  * in all cases, we can not find the right fwmark
 423                  * virtual service. If needed, we can do it for
 424                  * non-fwmark persistent services.
 425                  */
 426                 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pp->timeout_table)
 427                         cp->timeout = pp->timeout_table[state];
 428                 else
 429                         cp->timeout = (3*60*HZ);
 430                 ip_vs_conn_put(cp);
 431         }
 432 }
 433
 434
 435 /*
 436  *      Setup loopback of outgoing multicasts on a sending socket
 437  */
 438 static void set_mcast_loop(struct sock *sk, u_char loop)
 439 {
 440         struct inet_sock *inet = inet_sk(sk);
 441
 442         /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
 443         lock_sock(sk);
 444         inet->mc_loop = loop ? 1 : 0;
 445         release_sock(sk);
 446 }
 447
 448 /*
 449  *      Specify TTL for outgoing multicasts on a sending socket
 450  */
 451 static void set_mcast_ttl(struct sock *sk, u_char ttl)
 452 {
 453         struct inet_sock *inet = inet_sk(sk);
 454
 455         /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
 456         lock_sock(sk);
 457         inet->mc_ttl = ttl;
 458         release_sock(sk);
 459 }
 460
 461 /*
 462  *      Specifiy default interface for outgoing multicasts
 463  */
 464 static int set_mcast_if(struct sock *sk, char *ifname)
 465 {
 466         struct net_device *dev;
 467         struct inet_sock *inet = inet_sk(sk);
 468
 469         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
 470                 return -ENODEV;
 471
 472         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
 473                 return -EINVAL;
 474
 475         lock_sock(sk);
 476         inet->mc_index = dev->ifindex;
 477         /*  inet->mc_addr  = 0; */
 478         release_sock(sk);
 479
 480         return 0;
 481 }
 482
 483
 484 /*
 485  *      Set the maximum length of sync message according to the
 486  *      specified interface's MTU.
 487  */
 488 static int set_sync_mesg_maxlen(int sync_state)
 489 {
 490         struct net_device *dev;
 491         int num;
 492
 493         if (sync_state == IP_VS_STATE_MASTER) {
 494                 if ((dev = __dev_get_by_name(&init_net, ip_vs_master_mcast_ifn)) == NULL)
 495                         return -ENODEV;
 496
 497                 num = (dev->mtu - sizeof(struct iphdr) -
 498                        sizeof(struct udphdr) -
 499                        SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE;
 500                 sync_send_mesg_maxlen =
 501                         SYNC_MESG_HEADER_LEN + SIMPLE_CONN_SIZE * num;
 502                 IP_VS_DBG(7, "setting the maximum length of sync sending "
 503                           "message %d.\n", sync_send_mesg_maxlen);
 504         } else if (sync_state == IP_VS_STATE_BACKUP) {
 505                 if ((dev = __dev_get_by_name(&init_net, ip_vs_backup_mcast_ifn)) == NULL)
 506                         return -ENODEV;
 507
 508                 sync_recv_mesg_maxlen = dev->mtu -
 509                         sizeof(struct iphdr) - sizeof(struct udphdr);
 510                 IP_VS_DBG(7, "setting the maximum length of sync receiving "
 511                           "message %d.\n", sync_recv_mesg_maxlen);
 512         }
 513
 514         return 0;
 515 }
 516
 517
 518 /*
 519  *      Join a multicast group.
 520  *      the group is specified by a class D multicast address 224.0.0.0/8
 521  *      in the in_addr structure passed in as a parameter.
 522  */
 523 static int
 524 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
 525 {
 526         struct ip_mreqn mreq;
 527         struct net_device *dev;
 528         int ret;
 529
 530         memset(&mreq, 0, sizeof(mreq));
 531         memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
 532
 533         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
 534                 return -ENODEV;
 535         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
 536                 return -EINVAL;
 537
 538         mreq.imr_ifindex = dev->ifindex;
 539
 540         lock_sock(sk);
 541         ret = ip_mc_join_group(sk, &mreq);
 542         release_sock(sk);
 543
 544         return ret;
 545 }
 546
 547
 548 static int bind_mcastif_addr(struct socket *sock, char *ifname)
 549 {
 550         struct net_device *dev;
 551         __be32 addr;
 552         struct sockaddr_in sin;
 553
 554         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
 555                 return -ENODEV;
 556
 557         addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
 558         if (!addr)
 559                 IP_VS_ERR("You probably need to specify IP address on "
 560                           "multicast interface.\n");
 561
 562         IP_VS_DBG(7, "binding socket with (%s) %u.%u.%u.%u\n",
 563                   ifname, NIPQUAD(addr));
 564
 565         /* Now bind the socket with the address of multicast interface */
 566         sin.sin_family       = AF_INET;
 567         sin.sin_addr.s_addr  = addr;
 568         sin.sin_port         = 0;
 569
 570         return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
 571 }
 572
 573 /*
 574  *      Set up sending multicast socket over UDP
 575  */
 576 static struct socket * make_send_sock(void)
 577 {
 578         struct socket *sock;
 579
 580         /* First create a socket */
 581         if (sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock) < 0) {
 582                 IP_VS_ERR("Error during creation of socket; terminating\n");
 583                 return NULL;
 584         }
 585
 586         if (set_mcast_if(sock->sk, ip_vs_master_mcast_ifn) < 0) {
 587                 IP_VS_ERR("Error setting outbound mcast interface\n");
 588                 goto error;
 589         }
 590
 591         set_mcast_loop(sock->sk, 0);
 592         set_mcast_ttl(sock->sk, 1);
 593
 594         if (bind_mcastif_addr(sock, ip_vs_master_mcast_ifn) < 0) {
 595                 IP_VS_ERR("Error binding address of the mcast interface\n");
 596                 goto error;
 597         }
 598
 599         if (sock->ops->connect(sock,
 600                                (struct sockaddr*)&mcast_addr,
 601                                sizeof(struct sockaddr), 0) < 0) {
 602                 IP_VS_ERR("Error connecting to the multicast addr\n");
 603                 goto error;
 604         }
 605
 606         return sock;
 607
 608   error:
 609         sock_release(sock);
 610         return NULL;
 611 }
 612
 613
 614 /*
 615  *      Set up receiving multicast socket over UDP
 616  */
 617 static struct socket * make_receive_sock(void)
 618 {
 619         struct socket *sock;
 620
 621         /* First create a socket */
 622         if (sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock) < 0) {
 623                 IP_VS_ERR("Error during creation of socket; terminating\n");
 624                 return NULL;
 625         }
 626
 627         /* it is equivalent to the REUSEADDR option in user-space */
 628         sock->sk->sk_reuse = 1;
 629
 630         if (sock->ops->bind(sock,
 631                             (struct sockaddr*)&mcast_addr,
 632                             sizeof(struct sockaddr)) < 0) {
 633                 IP_VS_ERR("Error binding to the multicast addr\n");
 634                 goto error;
 635         }
 636
 637         /* join the multicast group */
 638         if (join_mcast_group(sock->sk,
 639                              (struct in_addr*)&mcast_addr.sin_addr,
 640                              ip_vs_backup_mcast_ifn) < 0) {
 641                 IP_VS_ERR("Error joining to the multicast group\n");
 642                 goto error;
 643         }
 644
 645         return sock;
 646
 647   error:
 648         sock_release(sock);
 649         return NULL;
 650 }
 651
 652
 653 static int
 654 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
 655 {
 656         struct msghdr   msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
 657         struct kvec     iov;
 658         int             len;
 659
 660         EnterFunction(7);
 661         iov.iov_base     = (void *)buffer;
 662         iov.iov_len      = length;
 663
 664         len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
 665
 666         LeaveFunction(7);
 667         return len;
 668 }
 669
 670 static void
 671 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
 672 {
 673         int msize;
 674
 675         msize = msg->size;
 676
 677         /* Put size in network byte order */
 678         msg->size = htons(msg->size);
 679
 680         if (ip_vs_send_async(sock, (char *)msg, msize) != msize)
 681                 IP_VS_ERR("ip_vs_send_async error\n");
 682 }
 683
 684 static int
 685 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
 686 {
 687         struct msghdr           msg = {NULL,};
 688         struct kvec             iov;
 689         int                     len;
 690
 691         EnterFunction(7);
 692
 693         /* Receive a packet */
 694         iov.iov_base     = buffer;
 695         iov.iov_len      = (size_t)buflen;
 696
 697         len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, 0);
 698
 699         if (len < 0)
 700                 return -1;
 701
 702         LeaveFunction(7);
 703         return len;
 704 }
 705
 706
 707 static DECLARE_WAIT_QUEUE_HEAD(sync_wait);
 708 static pid_t sync_master_pid = 0;
 709 static pid_t sync_backup_pid = 0;
 710
 711 static DECLARE_WAIT_QUEUE_HEAD(stop_sync_wait);
 712 static int stop_master_sync = 0;
 713 static int stop_backup_sync = 0;
 714
 715 static void sync_master_loop(void)
 716 {
 717         struct socket *sock;
 718         struct ip_vs_sync_buff *sb;
 719
 720         /* create the sending multicast socket */
 721         sock = make_send_sock();
 722         if (!sock)
 723                 return;
 724
 725         IP_VS_INFO("sync thread started: state = MASTER, mcast_ifn = %s, "
 726                    "syncid = %d\n",
 727                    ip_vs_master_mcast_ifn, ip_vs_master_syncid);
 728
 729         for (;;) {
 730                 while ((sb=sb_dequeue())) {
 731                         ip_vs_send_sync_msg(sock, sb->mesg);
 732                         ip_vs_sync_buff_release(sb);
 733                 }
 734
 735                 /* check if entries stay in curr_sb for 2 seconds */
 736                 if ((sb = get_curr_sync_buff(2*HZ))) {
 737                         ip_vs_send_sync_msg(sock, sb->mesg);
 738                         ip_vs_sync_buff_release(sb);
 739                 }
 740
 741                 if (stop_master_sync)
 742                         break;
 743
 744                 msleep_interruptible(1000);
 745         }
 746
 747         /* clean up the sync_buff queue */
 748         while ((sb=sb_dequeue())) {
 749                 ip_vs_sync_buff_release(sb);
 750         }
 751
 752         /* clean up the current sync_buff */
 753         if ((sb = get_curr_sync_buff(0))) {
 754                 ip_vs_sync_buff_release(sb);
 755         }
 756
 757         /* release the sending multicast socket */
 758         sock_release(sock);
 759 }
 760
 761
 762 static void sync_backup_loop(void)
 763 {
 764         struct socket *sock;
 765         char *buf;
 766         int len;
 767
 768         if (!(buf = kmalloc(sync_recv_mesg_maxlen, GFP_ATOMIC))) {
 769                 IP_VS_ERR("sync_backup_loop: kmalloc error\n");
 770                 return;
 771         }
 772
 773         /* create the receiving multicast socket */
 774         sock = make_receive_sock();
 775         if (!sock)
 776                 goto out;
 777
 778         IP_VS_INFO("sync thread started: state = BACKUP, mcast_ifn = %s, "
 779                    "syncid = %d\n",
 780                    ip_vs_backup_mcast_ifn, ip_vs_backup_syncid);
 781
 782         for (;;) {
 783                 /* do you have data now? */
 784                 while (!skb_queue_empty(&(sock->sk->sk_receive_queue))) {
 785                         if ((len =
 786                              ip_vs_receive(sock, buf,
 787                                            sync_recv_mesg_maxlen)) <= 0) {
 788                                 IP_VS_ERR("receiving message error\n");
 789                                 break;
 790                         }
 791                         /* disable bottom half, because it accessed the data
 792                            shared by softirq while getting/creating conns */
 793                         local_bh_disable();
 794                         ip_vs_process_message(buf, len);
 795                         local_bh_enable();
 796                 }
 797
 798                 if (stop_backup_sync)
 799                         break;
 800
 801                 msleep_interruptible(1000);
 802         }
 803
 804         /* release the sending multicast socket */
 805         sock_release(sock);
 806
 807   out:
 808         kfree(buf);
 809 }
 810
 811
 812 static void set_sync_pid(int sync_state, pid_t sync_pid)
 813 {
 814         if (sync_state == IP_VS_STATE_MASTER)
 815                 sync_master_pid = sync_pid;
 816         else if (sync_state == IP_VS_STATE_BACKUP)
 817                 sync_backup_pid = sync_pid;
 818 }
 819
 820 static void set_stop_sync(int sync_state, int set)
 821 {
 822         if (sync_state == IP_VS_STATE_MASTER)
 823                 stop_master_sync = set;
 824         else if (sync_state == IP_VS_STATE_BACKUP)
 825                 stop_backup_sync = set;
 826         else {
 827                 stop_master_sync = set;
 828                 stop_backup_sync = set;
 829         }
 830 }
 831
 832 static int sync_thread(void *startup)
 833 {
 834         DECLARE_WAITQUEUE(wait, current);
 835         mm_segment_t oldmm;
 836         int state;
 837         const char *name;
 838         struct ip_vs_sync_thread_data *tinfo = startup;
 839
 840         /* increase the module use count */
 841         ip_vs_use_count_inc();
 842
 843         if (ip_vs_sync_state & IP_VS_STATE_MASTER && !sync_master_pid) {
 844                 state = IP_VS_STATE_MASTER;
 845                 name = "ipvs_syncmaster";
 846         } else if (ip_vs_sync_state & IP_VS_STATE_BACKUP && !sync_backup_pid) {
 847                 state = IP_VS_STATE_BACKUP;
 848                 name = "ipvs_syncbackup";
 849         } else {
 850                 IP_VS_BUG();
 851                 ip_vs_use_count_dec();
 852                 return -EINVAL;
 853         }
 854
 855         daemonize(name);
 856
 857         oldmm = get_fs();
 858         set_fs(KERNEL_DS);
 859
 860         /* Block all signals */
 861         spin_lock_irq(&current->sighand->siglock);
 862         siginitsetinv(&current->blocked, 0);
 863         recalc_sigpending();
 864         spin_unlock_irq(&current->sighand->siglock);
 865
 866         /* set the maximum length of sync message */
 867         set_sync_mesg_maxlen(state);
 868
 869         add_wait_queue(&sync_wait, &wait);
 870
 871         set_sync_pid(state, task_pid_nr(current));
 872         complete(tinfo->startup);
 873
 874         /*
 875          * once we call the completion queue above, we should
 876          * null out that reference, since its allocated on the
 877          * stack of the creating kernel thread
 878          */
 879         tinfo->startup = NULL;
 880
 881         /* processing master/backup loop here */
 882         if (state == IP_VS_STATE_MASTER)
 883                 sync_master_loop();
 884         else if (state == IP_VS_STATE_BACKUP)
 885                 sync_backup_loop();
 886         else IP_VS_BUG();
 887
 888         remove_wait_queue(&sync_wait, &wait);
 889
 890         /* thread exits */
 891
 892         /*
 893          * If we weren't explicitly stopped, then we
 894          * exited in error, and should undo our state
 895          */
 896         if ((!stop_master_sync) && (!stop_backup_sync))
 897                 ip_vs_sync_state -= tinfo->state;
 898
 899         set_sync_pid(state, 0);
 900         IP_VS_INFO("sync thread stopped!\n");
 901
 902         set_fs(oldmm);
 903
 904         /* decrease the module use count */
 905         ip_vs_use_count_dec();
 906
 907         set_stop_sync(state, 0);
 908         wake_up(&stop_sync_wait);
 909
 910         /*
 911          * we need to free the structure that was allocated
 912          * for us in start_sync_thread
 913          */
 914         kfree(tinfo);
 915         return 0;
 916 }
 917
 918
 919 static int fork_sync_thread(void *startup)
 920 {
 921         pid_t pid;
 922
 923         /* fork the sync thread here, then the parent process of the
 924            sync thread is the init process after this thread exits. */
 925   repeat:
 926         if ((pid = kernel_thread(sync_thread, startup, 0)) < 0) {
 927                 IP_VS_ERR("could not create sync_thread due to %d... "
 928                           "retrying.\n", pid);
 929                 msleep_interruptible(1000);
 930                 goto repeat;
 931         }
 932
 933         return 0;
 934 }
 935
 936
 937 int start_sync_thread(int state, char *mcast_ifn, __u8 syncid)
 938 {
 939         DECLARE_COMPLETION_ONSTACK(startup);
 940         pid_t pid;
 941         struct ip_vs_sync_thread_data *tinfo;
 942
 943         if ((state == IP_VS_STATE_MASTER && sync_master_pid) ||
 944             (state == IP_VS_STATE_BACKUP && sync_backup_pid))
 945                 return -EEXIST;
 946
 947         /*
 948          * Note that tinfo will be freed in sync_thread on exit
 949          */
 950         tinfo = kmalloc(sizeof(struct ip_vs_sync_thread_data), GFP_KERNEL);
 951         if (!tinfo)
 952                 return -ENOMEM;
 953
 954         IP_VS_DBG(7, "%s: pid %d\n", __func__, task_pid_nr(current));
 955         IP_VS_DBG(7, "Each ip_vs_sync_conn entry need %Zd bytes\n",
 956                   sizeof(struct ip_vs_sync_conn));
 957
 958         ip_vs_sync_state |= state;
 959         if (state == IP_VS_STATE_MASTER) {
 960                 strlcpy(ip_vs_master_mcast_ifn, mcast_ifn,
 961                         sizeof(ip_vs_master_mcast_ifn));
 962                 ip_vs_master_syncid = syncid;
 963         } else {
 964                 strlcpy(ip_vs_backup_mcast_ifn, mcast_ifn,
 965                         sizeof(ip_vs_backup_mcast_ifn));
 966                 ip_vs_backup_syncid = syncid;
 967         }
 968
 969         tinfo->state = state;
 970         tinfo->startup = &startup;
 971
 972   repeat:
 973         if ((pid = kernel_thread(fork_sync_thread, tinfo, 0)) < 0) {
 974                 IP_VS_ERR("could not create fork_sync_thread due to %d... "
 975                           "retrying.\n", pid);
 976                 msleep_interruptible(1000);
 977                 goto repeat;
 978         }
 979
 980         wait_for_completion(&startup);
 981
 982         return 0;
 983 }
 984
 985
 986 int stop_sync_thread(int state)
 987 {
 988         DECLARE_WAITQUEUE(wait, current);
 989
 990         if ((state == IP_VS_STATE_MASTER && !sync_master_pid) ||
 991             (state == IP_VS_STATE_BACKUP && !sync_backup_pid))
 992                 return -ESRCH;
 993
 994         IP_VS_DBG(7, "%s: pid %d\n", __func__, task_pid_nr(current));
 995         IP_VS_INFO("stopping sync thread %d ...\n",
 996                    (state == IP_VS_STATE_MASTER) ?
 997                    sync_master_pid : sync_backup_pid);
 998
 999         __set_current_state(TASK_UNINTERRUPTIBLE);
1000         add_wait_queue(&stop_sync_wait, &wait);
1001         set_stop_sync(state, 1);
1002         ip_vs_sync_state -= state;
1003         wake_up(&sync_wait);
1004         schedule();
1005         __set_current_state(TASK_RUNNING);
1006         remove_wait_queue(&stop_sync_wait, &wait);
1007
1008         /* Note: no need to reap the sync thread, because its parent
1009            process is the init process */
1010
1011         if ((state == IP_VS_STATE_MASTER && stop_master_sync) ||
1012             (state == IP_VS_STATE_BACKUP && stop_backup_sync))
1013                 IP_VS_BUG();
1014
1015         return 0;
1016 }