SAFE public projects git trees. - safe/jmp/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 #include <linux/err.h>
  31
  32 #include <net/ip.h>
  33 #include <net/sock.h>
  34 #include <asm/uaccess.h>                /* for get_fs and set_fs */
  35
  36 #include <net/ip_vs.h>
  37
  38 #define IP_VS_SYNC_GROUP 0xe0000051    /* multicast addr - 224.0.0.81 */
  39 #define IP_VS_SYNC_PORT  8848          /* multicast port */
  40
  41
  42 /*
  43  *      IPVS sync connection entry
  44  */
  45 struct ip_vs_sync_conn {
  46         __u8                    reserved;
  47
  48         /* Protocol, addresses and port numbers */
  49         __u8                    protocol;       /* Which protocol (TCP/UDP) */
  50         __be16                  cport;
  51         __be16                  vport;
  52         __be16                  dport;
  53         __be32                  caddr;          /* client address */
  54         __be32                  vaddr;          /* virtual address */
  55         __be32                  daddr;          /* destination address */
  56
  57         /* Flags and state transition */
  58         __be16                  flags;          /* status flags */
  59         __be16                  state;          /* state info */
  60
  61         /* The sequence options start here */
  62 };
  63
  64 struct ip_vs_sync_conn_options {
  65         struct ip_vs_seq        in_seq;         /* incoming seq. struct */
  66         struct ip_vs_seq        out_seq;        /* outgoing seq. struct */
  67 };
  68
  69 struct ip_vs_sync_thread_data {
  70         struct completion *startup;
  71         int state;
  72 };
  73
  74 #define SIMPLE_CONN_SIZE  (sizeof(struct ip_vs_sync_conn))
  75 #define FULL_CONN_SIZE  \
  76 (sizeof(struct ip_vs_sync_conn) + sizeof(struct ip_vs_sync_conn_options))
  77
  78
  79 /*
  80   The master mulitcasts messages to the backup load balancers in the
  81   following format.
  82
  83        0                   1                   2                   3
  84        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
  85       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  86       |  Count Conns  |    SyncID     |            Size               |
  87       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  88       |                                                               |
  89       |                    IPVS Sync Connection (1)                   |
  90       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  91       |                            .                                  |
  92       |                            .                                  |
  93       |                            .                                  |
  94       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  95       |                                                               |
  96       |                    IPVS Sync Connection (n)                   |
  97       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  98 */
  99
 100 #define SYNC_MESG_HEADER_LEN    4
 101
 102 struct ip_vs_sync_mesg {
 103         __u8                    nr_conns;
 104         __u8                    syncid;
 105         __u16                   size;
 106
 107         /* ip_vs_sync_conn entries start here */
 108 };
 109
 110 /* the maximum length of sync (sending/receiving) message */
 111 static int sync_send_mesg_maxlen;
 112 static int sync_recv_mesg_maxlen;
 113
 114 struct ip_vs_sync_buff {
 115         struct list_head        list;
 116         unsigned long           firstuse;
 117
 118         /* pointers for the message data */
 119         struct ip_vs_sync_mesg  *mesg;
 120         unsigned char           *head;
 121         unsigned char           *end;
 122 };
 123
 124
 125 /* the sync_buff list head and the lock */
 126 static LIST_HEAD(ip_vs_sync_queue);
 127 static DEFINE_SPINLOCK(ip_vs_sync_lock);
 128
 129 /* current sync_buff for accepting new conn entries */
 130 static struct ip_vs_sync_buff   *curr_sb = NULL;
 131 static DEFINE_SPINLOCK(curr_sb_lock);
 132
 133 /* ipvs sync daemon state */
 134 volatile int ip_vs_sync_state = IP_VS_STATE_NONE;
 135 volatile int ip_vs_master_syncid = 0;
 136 volatile int ip_vs_backup_syncid = 0;
 137
 138 /* multicast interface name */
 139 char ip_vs_master_mcast_ifn[IP_VS_IFNAME_MAXLEN];
 140 char ip_vs_backup_mcast_ifn[IP_VS_IFNAME_MAXLEN];
 141
 142 /* multicast addr */
 143 static struct sockaddr_in mcast_addr = {
 144         .sin_family             = AF_INET,
 145         .sin_port               = __constant_htons(IP_VS_SYNC_PORT),
 146         .sin_addr.s_addr        = __constant_htonl(IP_VS_SYNC_GROUP),
 147 };
 148
 149
 150 static inline void sb_queue_tail(struct ip_vs_sync_buff *sb)
 151 {
 152         spin_lock(&ip_vs_sync_lock);
 153         list_add_tail(&sb->list, &ip_vs_sync_queue);
 154         spin_unlock(&ip_vs_sync_lock);
 155 }
 156
 157 static inline struct ip_vs_sync_buff * sb_dequeue(void)
 158 {
 159         struct ip_vs_sync_buff *sb;
 160
 161         spin_lock_bh(&ip_vs_sync_lock);
 162         if (list_empty(&ip_vs_sync_queue)) {
 163                 sb = NULL;
 164         } else {
 165                 sb = list_entry(ip_vs_sync_queue.next,
 166                                 struct ip_vs_sync_buff,
 167                                 list);
 168                 list_del(&sb->list);
 169         }
 170         spin_unlock_bh(&ip_vs_sync_lock);
 171
 172         return sb;
 173 }
 174
 175 static inline struct ip_vs_sync_buff * ip_vs_sync_buff_create(void)
 176 {
 177         struct ip_vs_sync_buff *sb;
 178
 179         if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
 180                 return NULL;
 181
 182         if (!(sb->mesg=kmalloc(sync_send_mesg_maxlen, GFP_ATOMIC))) {
 183                 kfree(sb);
 184                 return NULL;
 185         }
 186         sb->mesg->nr_conns = 0;
 187         sb->mesg->syncid = ip_vs_master_syncid;
 188         sb->mesg->size = 4;
 189         sb->head = (unsigned char *)sb->mesg + 4;
 190         sb->end = (unsigned char *)sb->mesg + sync_send_mesg_maxlen;
 191         sb->firstuse = jiffies;
 192         return sb;
 193 }
 194
 195 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
 196 {
 197         kfree(sb->mesg);
 198         kfree(sb);
 199 }
 200
 201 /*
 202  *      Get the current sync buffer if it has been created for more
 203  *      than the specified time or the specified time is zero.
 204  */
 205 static inline struct ip_vs_sync_buff *
 206 get_curr_sync_buff(unsigned long time)
 207 {
 208         struct ip_vs_sync_buff *sb;
 209
 210         spin_lock_bh(&curr_sb_lock);
 211         if (curr_sb && (time == 0 ||
 212                         time_before(jiffies - curr_sb->firstuse, time))) {
 213                 sb = curr_sb;
 214                 curr_sb = NULL;
 215         } else
 216                 sb = NULL;
 217         spin_unlock_bh(&curr_sb_lock);
 218         return sb;
 219 }
 220
 221
 222 /*
 223  *      Add an ip_vs_conn information into the current sync_buff.
 224  *      Called by ip_vs_in.
 225  */
 226 void ip_vs_sync_conn(struct ip_vs_conn *cp)
 227 {
 228         struct ip_vs_sync_mesg *m;
 229         struct ip_vs_sync_conn *s;
 230         int len;
 231
 232         spin_lock(&curr_sb_lock);
 233         if (!curr_sb) {
 234                 if (!(curr_sb=ip_vs_sync_buff_create())) {
 235                         spin_unlock(&curr_sb_lock);
 236                         IP_VS_ERR("ip_vs_sync_buff_create failed.\n");
 237                         return;
 238                 }
 239         }
 240
 241         len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
 242                 SIMPLE_CONN_SIZE;
 243         m = curr_sb->mesg;
 244         s = (struct ip_vs_sync_conn *)curr_sb->head;
 245
 246         /* copy members */
 247         s->protocol = cp->protocol;
 248         s->cport = cp->cport;
 249         s->vport = cp->vport;
 250         s->dport = cp->dport;
 251         s->caddr = cp->caddr;
 252         s->vaddr = cp->vaddr;
 253         s->daddr = cp->daddr;
 254         s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
 255         s->state = htons(cp->state);
 256         if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
 257                 struct ip_vs_sync_conn_options *opt =
 258                         (struct ip_vs_sync_conn_options *)&s[1];
 259                 memcpy(opt, &cp->in_seq, sizeof(*opt));
 260         }
 261
 262         m->nr_conns++;
 263         m->size += len;
 264         curr_sb->head += len;
 265
 266         /* check if there is a space for next one */
 267         if (curr_sb->head+FULL_CONN_SIZE > curr_sb->end) {
 268                 sb_queue_tail(curr_sb);
 269                 curr_sb = NULL;
 270         }
 271         spin_unlock(&curr_sb_lock);
 272
 273         /* synchronize its controller if it has */
 274         if (cp->control)
 275                 ip_vs_sync_conn(cp->control);
 276 }
 277
 278
 279 /*
 280  *      Process received multicast message and create the corresponding
 281  *      ip_vs_conn entries.
 282  */
 283 static void ip_vs_process_message(const char *buffer, const size_t buflen)
 284 {
 285         struct ip_vs_sync_mesg *m = (struct ip_vs_sync_mesg *)buffer;
 286         struct ip_vs_sync_conn *s;
 287         struct ip_vs_sync_conn_options *opt;
 288         struct ip_vs_conn *cp;
 289         struct ip_vs_protocol *pp;
 290         struct ip_vs_dest *dest;
 291         char *p;
 292         int i;
 293
 294         if (buflen < sizeof(struct ip_vs_sync_mesg)) {
 295                 IP_VS_ERR_RL("sync message header too short\n");
 296                 return;
 297         }
 298
 299         /* Convert size back to host byte order */
 300         m->size = ntohs(m->size);
 301
 302         if (buflen != m->size) {
 303                 IP_VS_ERR_RL("bogus sync message size\n");
 304                 return;
 305         }
 306
 307         /* SyncID sanity check */
 308         if (ip_vs_backup_syncid != 0 && m->syncid != ip_vs_backup_syncid) {
 309                 IP_VS_DBG(7, "Ignoring incoming msg with syncid = %d\n",
 310                           m->syncid);
 311                 return;
 312         }
 313
 314         p = (char *)buffer + sizeof(struct ip_vs_sync_mesg);
 315         for (i=0; i<m->nr_conns; i++) {
 316                 unsigned flags, state;
 317
 318                 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
 319                         IP_VS_ERR_RL("bogus conn in sync message\n");
 320                         return;
 321                 }
 322                 s = (struct ip_vs_sync_conn *) p;
 323                 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
 324                 flags &= ~IP_VS_CONN_F_HASHED;
 325                 if (flags & IP_VS_CONN_F_SEQ_MASK) {
 326                         opt = (struct ip_vs_sync_conn_options *)&s[1];
 327                         p += FULL_CONN_SIZE;
 328                         if (p > buffer+buflen) {
 329                                 IP_VS_ERR_RL("bogus conn options in sync message\n");
 330                                 return;
 331                         }
 332                 } else {
 333                         opt = NULL;
 334                         p += SIMPLE_CONN_SIZE;
 335                 }
 336
 337                 state = ntohs(s->state);
 338                 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
 339                         pp = ip_vs_proto_get(s->protocol);
 340                         if (!pp) {
 341                                 IP_VS_ERR_RL("Unsupported protocol %u in sync msg\n",
 342                                         s->protocol);
 343                                 continue;
 344                         }
 345                         if (state >= pp->num_states) {
 346                                 IP_VS_DBG(2, "Invalid %s state %u in sync msg\n",
 347                                         pp->name, state);
 348                                 continue;
 349                         }
 350                 } else {
 351                         /* protocol in templates is not used for state/timeout */
 352                         pp = NULL;
 353                         if (state > 0) {
 354                                 IP_VS_DBG(2, "Invalid template state %u in sync msg\n",
 355                                         state);
 356                                 state = 0;
 357                         }
 358                 }
 359
 360                 if (!(flags & IP_VS_CONN_F_TEMPLATE))
 361                         cp = ip_vs_conn_in_get(s->protocol,
 362                                                s->caddr, s->cport,
 363                                                s->vaddr, s->vport);
 364                 else
 365                         cp = ip_vs_ct_in_get(s->protocol,
 366                                                s->caddr, s->cport,
 367                                                s->vaddr, s->vport);
 368                 if (!cp) {
 369                         /*
 370                          * Find the appropriate destination for the connection.
 371                          * If it is not found the connection will remain unbound
 372                          * but still handled.
 373                          */
 374                         dest = ip_vs_find_dest(s->daddr, s->dport,
 375                                                s->vaddr, s->vport,
 376                                                s->protocol);
 377                         /*  Set the approprite ativity flag */
 378                         if (s->protocol == IPPROTO_TCP) {
 379                                 if (state != IP_VS_TCP_S_ESTABLISHED)
 380                                         flags |= IP_VS_CONN_F_INACTIVE;
 381                                 else
 382                                         flags &= ~IP_VS_CONN_F_INACTIVE;
 383                         }
 384                         cp = ip_vs_conn_new(s->protocol,
 385                                             s->caddr, s->cport,
 386                                             s->vaddr, s->vport,
 387                                             s->daddr, s->dport,
 388                                             flags, dest);
 389                         if (dest)
 390                                 atomic_dec(&dest->refcnt);
 391                         if (!cp) {
 392                                 IP_VS_ERR("ip_vs_conn_new failed\n");
 393                                 return;
 394                         }
 395                 } else if (!cp->dest) {
 396                         dest = ip_vs_try_bind_dest(cp);
 397                         if (dest)
 398                                 atomic_dec(&dest->refcnt);
 399                 } else if ((cp->dest) && (cp->protocol == IPPROTO_TCP) &&
 400                            (cp->state != state)) {
 401                         /* update active/inactive flag for the connection */
 402                         dest = cp->dest;
 403                         if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
 404                                 (state != IP_VS_TCP_S_ESTABLISHED)) {
 405                                 atomic_dec(&dest->activeconns);
 406                                 atomic_inc(&dest->inactconns);
 407                                 cp->flags |= IP_VS_CONN_F_INACTIVE;
 408                         } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) &&
 409                                 (state == IP_VS_TCP_S_ESTABLISHED)) {
 410                                 atomic_inc(&dest->activeconns);
 411                                 atomic_dec(&dest->inactconns);
 412                                 cp->flags &= ~IP_VS_CONN_F_INACTIVE;
 413                         }
 414                 }
 415
 416                 if (opt)
 417                         memcpy(&cp->in_seq, opt, sizeof(*opt));
 418                 atomic_set(&cp->in_pkts, sysctl_ip_vs_sync_threshold[0]);
 419                 cp->state = state;
 420                 cp->old_state = cp->state;
 421                 /*
 422                  * We can not recover the right timeout for templates
 423                  * in all cases, we can not find the right fwmark
 424                  * virtual service. If needed, we can do it for
 425                  * non-fwmark persistent services.
 426                  */
 427                 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pp->timeout_table)
 428                         cp->timeout = pp->timeout_table[state];
 429                 else
 430                         cp->timeout = (3*60*HZ);
 431                 ip_vs_conn_put(cp);
 432         }
 433 }
 434
 435
 436 /*
 437  *      Setup loopback of outgoing multicasts on a sending socket
 438  */
 439 static void set_mcast_loop(struct sock *sk, u_char loop)
 440 {
 441         struct inet_sock *inet = inet_sk(sk);
 442
 443         /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
 444         lock_sock(sk);
 445         inet->mc_loop = loop ? 1 : 0;
 446         release_sock(sk);
 447 }
 448
 449 /*
 450  *      Specify TTL for outgoing multicasts on a sending socket
 451  */
 452 static void set_mcast_ttl(struct sock *sk, u_char ttl)
 453 {
 454         struct inet_sock *inet = inet_sk(sk);
 455
 456         /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
 457         lock_sock(sk);
 458         inet->mc_ttl = ttl;
 459         release_sock(sk);
 460 }
 461
 462 /*
 463  *      Specifiy default interface for outgoing multicasts
 464  */
 465 static int set_mcast_if(struct sock *sk, char *ifname)
 466 {
 467         struct net_device *dev;
 468         struct inet_sock *inet = inet_sk(sk);
 469
 470         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
 471                 return -ENODEV;
 472
 473         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
 474                 return -EINVAL;
 475
 476         lock_sock(sk);
 477         inet->mc_index = dev->ifindex;
 478         /*  inet->mc_addr  = 0; */
 479         release_sock(sk);
 480
 481         return 0;
 482 }
 483
 484
 485 /*
 486  *      Set the maximum length of sync message according to the
 487  *      specified interface's MTU.
 488  */
 489 static int set_sync_mesg_maxlen(int sync_state)
 490 {
 491         struct net_device *dev;
 492         int num;
 493
 494         if (sync_state == IP_VS_STATE_MASTER) {
 495                 if ((dev = __dev_get_by_name(&init_net, ip_vs_master_mcast_ifn)) == NULL)
 496                         return -ENODEV;
 497
 498                 num = (dev->mtu - sizeof(struct iphdr) -
 499                        sizeof(struct udphdr) -
 500                        SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE;
 501                 sync_send_mesg_maxlen =
 502                         SYNC_MESG_HEADER_LEN + SIMPLE_CONN_SIZE * num;
 503                 IP_VS_DBG(7, "setting the maximum length of sync sending "
 504                           "message %d.\n", sync_send_mesg_maxlen);
 505         } else if (sync_state == IP_VS_STATE_BACKUP) {
 506                 if ((dev = __dev_get_by_name(&init_net, ip_vs_backup_mcast_ifn)) == NULL)
 507                         return -ENODEV;
 508
 509                 sync_recv_mesg_maxlen = dev->mtu -
 510                         sizeof(struct iphdr) - sizeof(struct udphdr);
 511                 IP_VS_DBG(7, "setting the maximum length of sync receiving "
 512                           "message %d.\n", sync_recv_mesg_maxlen);
 513         }
 514
 515         return 0;
 516 }
 517
 518
 519 /*
 520  *      Join a multicast group.
 521  *      the group is specified by a class D multicast address 224.0.0.0/8
 522  *      in the in_addr structure passed in as a parameter.
 523  */
 524 static int
 525 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
 526 {
 527         struct ip_mreqn mreq;
 528         struct net_device *dev;
 529         int ret;
 530
 531         memset(&mreq, 0, sizeof(mreq));
 532         memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
 533
 534         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
 535                 return -ENODEV;
 536         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
 537                 return -EINVAL;
 538
 539         mreq.imr_ifindex = dev->ifindex;
 540
 541         lock_sock(sk);
 542         ret = ip_mc_join_group(sk, &mreq);
 543         release_sock(sk);
 544
 545         return ret;
 546 }
 547
 548
 549 static int bind_mcastif_addr(struct socket *sock, char *ifname)
 550 {
 551         struct net_device *dev;
 552         __be32 addr;
 553         struct sockaddr_in sin;
 554
 555         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
 556                 return -ENODEV;
 557
 558         addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
 559         if (!addr)
 560                 IP_VS_ERR("You probably need to specify IP address on "
 561                           "multicast interface.\n");
 562
 563         IP_VS_DBG(7, "binding socket with (%s) %u.%u.%u.%u\n",
 564                   ifname, NIPQUAD(addr));
 565
 566         /* Now bind the socket with the address of multicast interface */
 567         sin.sin_family       = AF_INET;
 568         sin.sin_addr.s_addr  = addr;
 569         sin.sin_port         = 0;
 570
 571         return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
 572 }
 573
 574 /*
 575  *      Set up sending multicast socket over UDP
 576  */
 577 static struct socket * make_send_sock(void)
 578 {
 579         struct socket *sock;
 580         int result;
 581
 582         /* First create a socket */
 583         result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
 584         if (result < 0) {
 585                 IP_VS_ERR("Error during creation of socket; terminating\n");
 586                 return ERR_PTR(result);
 587         }
 588
 589         result = set_mcast_if(sock->sk, ip_vs_master_mcast_ifn);
 590         if (result < 0) {
 591                 IP_VS_ERR("Error setting outbound mcast interface\n");
 592                 goto error;
 593         }
 594
 595         set_mcast_loop(sock->sk, 0);
 596         set_mcast_ttl(sock->sk, 1);
 597
 598         result = bind_mcastif_addr(sock, ip_vs_master_mcast_ifn);
 599         if (result < 0) {
 600                 IP_VS_ERR("Error binding address of the mcast interface\n");
 601                 goto error;
 602         }
 603
 604         result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr,
 605                         sizeof(struct sockaddr), 0);
 606         if (result < 0) {
 607                 IP_VS_ERR("Error connecting to the multicast addr\n");
 608                 goto error;
 609         }
 610
 611         return sock;
 612
 613   error:
 614         sock_release(sock);
 615         return ERR_PTR(result);
 616 }
 617
 618
 619 /*
 620  *      Set up receiving multicast socket over UDP
 621  */
 622 static struct socket * make_receive_sock(void)
 623 {
 624         struct socket *sock;
 625         int result;
 626
 627         /* First create a socket */
 628         result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
 629         if (result < 0) {
 630                 IP_VS_ERR("Error during creation of socket; terminating\n");
 631                 return ERR_PTR(result);
 632         }
 633
 634         /* it is equivalent to the REUSEADDR option in user-space */
 635         sock->sk->sk_reuse = 1;
 636
 637         result = sock->ops->bind(sock, (struct sockaddr *) &mcast_addr,
 638                         sizeof(struct sockaddr));
 639         if (result < 0) {
 640                 IP_VS_ERR("Error binding to the multicast addr\n");
 641                 goto error;
 642         }
 643
 644         /* join the multicast group */
 645         result = join_mcast_group(sock->sk,
 646                         (struct in_addr *) &mcast_addr.sin_addr,
 647                         ip_vs_backup_mcast_ifn);
 648         if (result < 0) {
 649                 IP_VS_ERR("Error joining to the multicast group\n");
 650                 goto error;
 651         }
 652
 653         return sock;
 654
 655   error:
 656         sock_release(sock);
 657         return ERR_PTR(result);
 658 }
 659
 660
 661 static int
 662 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
 663 {
 664         struct msghdr   msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
 665         struct kvec     iov;
 666         int             len;
 667
 668         EnterFunction(7);
 669         iov.iov_base     = (void *)buffer;
 670         iov.iov_len      = length;
 671
 672         len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
 673
 674         LeaveFunction(7);
 675         return len;
 676 }
 677
 678 static void
 679 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
 680 {
 681         int msize;
 682
 683         msize = msg->size;
 684
 685         /* Put size in network byte order */
 686         msg->size = htons(msg->size);
 687
 688         if (ip_vs_send_async(sock, (char *)msg, msize) != msize)
 689                 IP_VS_ERR("ip_vs_send_async error\n");
 690 }
 691
 692 static int
 693 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
 694 {
 695         struct msghdr           msg = {NULL,};
 696         struct kvec             iov;
 697         int                     len;
 698
 699         EnterFunction(7);
 700
 701         /* Receive a packet */
 702         iov.iov_base     = buffer;
 703         iov.iov_len      = (size_t)buflen;
 704
 705         len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, 0);
 706
 707         if (len < 0)
 708                 return -1;
 709
 710         LeaveFunction(7);
 711         return len;
 712 }
 713
 714
 715 static DECLARE_WAIT_QUEUE_HEAD(sync_wait);
 716 static pid_t sync_master_pid = 0;
 717 static pid_t sync_backup_pid = 0;
 718
 719 static DECLARE_WAIT_QUEUE_HEAD(stop_sync_wait);
 720 static int stop_master_sync = 0;
 721 static int stop_backup_sync = 0;
 722
 723 static void sync_master_loop(void)
 724 {
 725         struct socket *sock;
 726         struct ip_vs_sync_buff *sb;
 727
 728         /* create the sending multicast socket */
 729         sock = make_send_sock();
 730         if (IS_ERR(sock))
 731                 return;
 732
 733         IP_VS_INFO("sync thread started: state = MASTER, mcast_ifn = %s, "
 734                    "syncid = %d\n",
 735                    ip_vs_master_mcast_ifn, ip_vs_master_syncid);
 736
 737         for (;;) {
 738                 while ((sb=sb_dequeue())) {
 739                         ip_vs_send_sync_msg(sock, sb->mesg);
 740                         ip_vs_sync_buff_release(sb);
 741                 }
 742
 743                 /* check if entries stay in curr_sb for 2 seconds */
 744                 if ((sb = get_curr_sync_buff(2*HZ))) {
 745                         ip_vs_send_sync_msg(sock, sb->mesg);
 746                         ip_vs_sync_buff_release(sb);
 747                 }
 748
 749                 if (stop_master_sync)
 750                         break;
 751
 752                 msleep_interruptible(1000);
 753         }
 754
 755         /* clean up the sync_buff queue */
 756         while ((sb=sb_dequeue())) {
 757                 ip_vs_sync_buff_release(sb);
 758         }
 759
 760         /* clean up the current sync_buff */
 761         if ((sb = get_curr_sync_buff(0))) {
 762                 ip_vs_sync_buff_release(sb);
 763         }
 764
 765         /* release the sending multicast socket */
 766         sock_release(sock);
 767 }
 768
 769
 770 static void sync_backup_loop(void)
 771 {
 772         struct socket *sock;
 773         char *buf;
 774         int len;
 775
 776         if (!(buf = kmalloc(sync_recv_mesg_maxlen, GFP_ATOMIC))) {
 777                 IP_VS_ERR("sync_backup_loop: kmalloc error\n");
 778                 return;
 779         }
 780
 781         /* create the receiving multicast socket */
 782         sock = make_receive_sock();
 783         if (IS_ERR(sock))
 784                 goto out;
 785
 786         IP_VS_INFO("sync thread started: state = BACKUP, mcast_ifn = %s, "
 787                    "syncid = %d\n",
 788                    ip_vs_backup_mcast_ifn, ip_vs_backup_syncid);
 789
 790         for (;;) {
 791                 /* do you have data now? */
 792                 while (!skb_queue_empty(&(sock->sk->sk_receive_queue))) {
 793                         if ((len =
 794                              ip_vs_receive(sock, buf,
 795                                            sync_recv_mesg_maxlen)) <= 0) {
 796                                 IP_VS_ERR("receiving message error\n");
 797                                 break;
 798                         }
 799                         /* disable bottom half, because it accessed the data
 800                            shared by softirq while getting/creating conns */
 801                         local_bh_disable();
 802                         ip_vs_process_message(buf, len);
 803                         local_bh_enable();
 804                 }
 805
 806                 if (stop_backup_sync)
 807                         break;
 808
 809                 msleep_interruptible(1000);
 810         }
 811
 812         /* release the sending multicast socket */
 813         sock_release(sock);
 814
 815   out:
 816         kfree(buf);
 817 }
 818
 819
 820 static void set_sync_pid(int sync_state, pid_t sync_pid)
 821 {
 822         if (sync_state == IP_VS_STATE_MASTER)
 823                 sync_master_pid = sync_pid;
 824         else if (sync_state == IP_VS_STATE_BACKUP)
 825                 sync_backup_pid = sync_pid;
 826 }
 827
 828 static void set_stop_sync(int sync_state, int set)
 829 {
 830         if (sync_state == IP_VS_STATE_MASTER)
 831                 stop_master_sync = set;
 832         else if (sync_state == IP_VS_STATE_BACKUP)
 833                 stop_backup_sync = set;
 834         else {
 835                 stop_master_sync = set;
 836                 stop_backup_sync = set;
 837         }
 838 }
 839
 840 static int sync_thread(void *startup)
 841 {
 842         DECLARE_WAITQUEUE(wait, current);
 843         mm_segment_t oldmm;
 844         int state;
 845         const char *name;
 846         struct ip_vs_sync_thread_data *tinfo = startup;
 847
 848         /* increase the module use count */
 849         ip_vs_use_count_inc();
 850
 851         if (ip_vs_sync_state & IP_VS_STATE_MASTER && !sync_master_pid) {
 852                 state = IP_VS_STATE_MASTER;
 853                 name = "ipvs_syncmaster";
 854         } else if (ip_vs_sync_state & IP_VS_STATE_BACKUP && !sync_backup_pid) {
 855                 state = IP_VS_STATE_BACKUP;
 856                 name = "ipvs_syncbackup";
 857         } else {
 858                 IP_VS_BUG();
 859                 ip_vs_use_count_dec();
 860                 return -EINVAL;
 861         }
 862
 863         daemonize(name);
 864
 865         oldmm = get_fs();
 866         set_fs(KERNEL_DS);
 867
 868         /* Block all signals */
 869         spin_lock_irq(&current->sighand->siglock);
 870         siginitsetinv(&current->blocked, 0);
 871         recalc_sigpending();
 872         spin_unlock_irq(&current->sighand->siglock);
 873
 874         /* set the maximum length of sync message */
 875         set_sync_mesg_maxlen(state);
 876
 877         add_wait_queue(&sync_wait, &wait);
 878
 879         set_sync_pid(state, task_pid_nr(current));
 880         complete(tinfo->startup);
 881
 882         /*
 883          * once we call the completion queue above, we should
 884          * null out that reference, since its allocated on the
 885          * stack of the creating kernel thread
 886          */
 887         tinfo->startup = NULL;
 888
 889         /* processing master/backup loop here */
 890         if (state == IP_VS_STATE_MASTER)
 891                 sync_master_loop();
 892         else if (state == IP_VS_STATE_BACKUP)
 893                 sync_backup_loop();
 894         else IP_VS_BUG();
 895
 896         remove_wait_queue(&sync_wait, &wait);
 897
 898         /* thread exits */
 899
 900         /*
 901          * If we weren't explicitly stopped, then we
 902          * exited in error, and should undo our state
 903          */
 904         if ((!stop_master_sync) && (!stop_backup_sync))
 905                 ip_vs_sync_state -= tinfo->state;
 906
 907         set_sync_pid(state, 0);
 908         IP_VS_INFO("sync thread stopped!\n");
 909
 910         set_fs(oldmm);
 911
 912         /* decrease the module use count */
 913         ip_vs_use_count_dec();
 914
 915         set_stop_sync(state, 0);
 916         wake_up(&stop_sync_wait);
 917
 918         /*
 919          * we need to free the structure that was allocated
 920          * for us in start_sync_thread
 921          */
 922         kfree(tinfo);
 923         return 0;
 924 }
 925
 926
 927 static int fork_sync_thread(void *startup)
 928 {
 929         pid_t pid;
 930
 931         /* fork the sync thread here, then the parent process of the
 932            sync thread is the init process after this thread exits. */
 933   repeat:
 934         if ((pid = kernel_thread(sync_thread, startup, 0)) < 0) {
 935                 IP_VS_ERR("could not create sync_thread due to %d... "
 936                           "retrying.\n", pid);
 937                 msleep_interruptible(1000);
 938                 goto repeat;
 939         }
 940
 941         return 0;
 942 }
 943
 944
 945 int start_sync_thread(int state, char *mcast_ifn, __u8 syncid)
 946 {
 947         DECLARE_COMPLETION_ONSTACK(startup);
 948         pid_t pid;
 949         struct ip_vs_sync_thread_data *tinfo;
 950
 951         if ((state == IP_VS_STATE_MASTER && sync_master_pid) ||
 952             (state == IP_VS_STATE_BACKUP && sync_backup_pid))
 953                 return -EEXIST;
 954
 955         /*
 956          * Note that tinfo will be freed in sync_thread on exit
 957          */
 958         tinfo = kmalloc(sizeof(struct ip_vs_sync_thread_data), GFP_KERNEL);
 959         if (!tinfo)
 960                 return -ENOMEM;
 961
 962         IP_VS_DBG(7, "%s: pid %d\n", __func__, task_pid_nr(current));
 963         IP_VS_DBG(7, "Each ip_vs_sync_conn entry need %Zd bytes\n",
 964                   sizeof(struct ip_vs_sync_conn));
 965
 966         ip_vs_sync_state |= state;
 967         if (state == IP_VS_STATE_MASTER) {
 968                 strlcpy(ip_vs_master_mcast_ifn, mcast_ifn,
 969                         sizeof(ip_vs_master_mcast_ifn));
 970                 ip_vs_master_syncid = syncid;
 971         } else {
 972                 strlcpy(ip_vs_backup_mcast_ifn, mcast_ifn,
 973                         sizeof(ip_vs_backup_mcast_ifn));
 974                 ip_vs_backup_syncid = syncid;
 975         }
 976
 977         tinfo->state = state;
 978         tinfo->startup = &startup;
 979
 980   repeat:
 981         if ((pid = kernel_thread(fork_sync_thread, tinfo, 0)) < 0) {
 982                 IP_VS_ERR("could not create fork_sync_thread due to %d... "
 983                           "retrying.\n", pid);
 984                 msleep_interruptible(1000);
 985                 goto repeat;
 986         }
 987
 988         wait_for_completion(&startup);
 989
 990         return 0;
 991 }
 992
 993
 994 int stop_sync_thread(int state)
 995 {
 996         DECLARE_WAITQUEUE(wait, current);
 997
 998         if ((state == IP_VS_STATE_MASTER && !sync_master_pid) ||
 999             (state == IP_VS_STATE_BACKUP && !sync_backup_pid))
1000                 return -ESRCH;
1001
1002         IP_VS_DBG(7, "%s: pid %d\n", __func__, task_pid_nr(current));
1003         IP_VS_INFO("stopping sync thread %d ...\n",
1004                    (state == IP_VS_STATE_MASTER) ?
1005                    sync_master_pid : sync_backup_pid);
1006
1007         __set_current_state(TASK_UNINTERRUPTIBLE);
1008         add_wait_queue(&stop_sync_wait, &wait);
1009         set_stop_sync(state, 1);
1010         ip_vs_sync_state -= state;
1011         wake_up(&sync_wait);
1012         schedule();
1013         __set_current_state(TASK_RUNNING);
1014         remove_wait_queue(&stop_sync_wait, &wait);
1015
1016         /* Note: no need to reap the sync thread, because its parent
1017            process is the init process */
1018
1019         if ((state == IP_VS_STATE_MASTER && stop_master_sync) ||
1020             (state == IP_VS_STATE_BACKUP && stop_backup_sync))
1021                 IP_VS_BUG();
1022
1023         return 0;
1024 }