net/netfilter/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 #define KMSG_COMPONENT "IPVS"
  21 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  22
  23 #include <linux/module.h>
  24 #include <linux/slab.h>
  25 #include <linux/inetdevice.h>
  26 #include <linux/net.h>
  27 #include <linux/completion.h>
  28 #include <linux/delay.h>
  29 #include <linux/skbuff.h>
  30 #include <linux/in.h>
  31 #include <linux/igmp.h>                 /* for ip_mc_join_group */
  32 #include <linux/udp.h>
  33 #include <linux/err.h>
  34 #include <linux/kthread.h>
  35 #include <linux/wait.h>
  36 #include <linux/kernel.h>
  37
  38 #include <net/ip.h>
  39 #include <net/sock.h>
  40
  41 #include <net/ip_vs.h>
  42
  43 #define IP_VS_SYNC_GROUP 0xe0000051    /* multicast addr - 224.0.0.81 */
  44 #define IP_VS_SYNC_PORT  8848          /* multicast port */
  45
  46
  47 /*
  48  *      IPVS sync connection entry
  49  */
  50 struct ip_vs_sync_conn {
  51         __u8                    reserved;
  52
  53         /* Protocol, addresses and port numbers */
  54         __u8                    protocol;       /* Which protocol (TCP/UDP) */
  55         __be16                  cport;
  56         __be16                  vport;
  57         __be16                  dport;
  58         __be32                  caddr;          /* client address */
  59         __be32                  vaddr;          /* virtual address */
  60         __be32                  daddr;          /* destination address */
  61
  62         /* Flags and state transition */
  63         __be16                  flags;          /* status flags */
  64         __be16                  state;          /* state info */
  65
  66         /* The sequence options start here */
  67 };
  68
  69 struct ip_vs_sync_conn_options {
  70         struct ip_vs_seq        in_seq;         /* incoming seq. struct */
  71         struct ip_vs_seq        out_seq;        /* outgoing seq. struct */
  72 };
  73
  74 struct ip_vs_sync_thread_data {
  75         struct socket *sock;
  76         char *buf;
  77 };
  78
  79 #define SIMPLE_CONN_SIZE  (sizeof(struct ip_vs_sync_conn))
  80 #define FULL_CONN_SIZE  \
  81 (sizeof(struct ip_vs_sync_conn) + sizeof(struct ip_vs_sync_conn_options))
  82
  83
  84 /*
  85   The master mulitcasts messages to the backup load balancers in the
  86   following format.
  87
  88        0                   1                   2                   3
  89        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
  90       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  91       |  Count Conns  |    SyncID     |            Size               |
  92       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  93       |                                                               |
  94       |                    IPVS Sync Connection (1)                   |
  95       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  96       |                            .                                  |
  97       |                            .                                  |
  98       |                            .                                  |
  99       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 100       |                                                               |
 101       |                    IPVS Sync Connection (n)                   |
 102       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 103 */
 104
 105 #define SYNC_MESG_HEADER_LEN    4
 106 #define MAX_CONNS_PER_SYNCBUFF  255 /* nr_conns in ip_vs_sync_mesg is 8 bit */
 107
 108 struct ip_vs_sync_mesg {
 109         __u8                    nr_conns;
 110         __u8                    syncid;
 111         __u16                   size;
 112
 113         /* ip_vs_sync_conn entries start here */
 114 };
 115
 116 /* the maximum length of sync (sending/receiving) message */
 117 static int sync_send_mesg_maxlen;
 118 static int sync_recv_mesg_maxlen;
 119
 120 struct ip_vs_sync_buff {
 121         struct list_head        list;
 122         unsigned long           firstuse;
 123
 124         /* pointers for the message data */
 125         struct ip_vs_sync_mesg  *mesg;
 126         unsigned char           *head;
 127         unsigned char           *end;
 128 };
 129
 130
 131 /* the sync_buff list head and the lock */
 132 static LIST_HEAD(ip_vs_sync_queue);
 133 static DEFINE_SPINLOCK(ip_vs_sync_lock);
 134
 135 /* current sync_buff for accepting new conn entries */
 136 static struct ip_vs_sync_buff   *curr_sb = NULL;
 137 static DEFINE_SPINLOCK(curr_sb_lock);
 138
 139 /* ipvs sync daemon state */
 140 volatile int ip_vs_sync_state = IP_VS_STATE_NONE;
 141 volatile int ip_vs_master_syncid = 0;
 142 volatile int ip_vs_backup_syncid = 0;
 143
 144 /* multicast interface name */
 145 char ip_vs_master_mcast_ifn[IP_VS_IFNAME_MAXLEN];
 146 char ip_vs_backup_mcast_ifn[IP_VS_IFNAME_MAXLEN];
 147
 148 /* sync daemon tasks */
 149 static struct task_struct *sync_master_thread;
 150 static struct task_struct *sync_backup_thread;
 151
 152 /* multicast addr */
 153 static struct sockaddr_in mcast_addr = {
 154         .sin_family             = AF_INET,
 155         .sin_port               = cpu_to_be16(IP_VS_SYNC_PORT),
 156         .sin_addr.s_addr        = cpu_to_be32(IP_VS_SYNC_GROUP),
 157 };
 158
 159
 160 static inline struct ip_vs_sync_buff *sb_dequeue(void)
 161 {
 162         struct ip_vs_sync_buff *sb;
 163
 164         spin_lock_bh(&ip_vs_sync_lock);
 165         if (list_empty(&ip_vs_sync_queue)) {
 166                 sb = NULL;
 167         } else {
 168                 sb = list_entry(ip_vs_sync_queue.next,
 169                                 struct ip_vs_sync_buff,
 170                                 list);
 171                 list_del(&sb->list);
 172         }
 173         spin_unlock_bh(&ip_vs_sync_lock);
 174
 175         return sb;
 176 }
 177
 178 static inline struct ip_vs_sync_buff * ip_vs_sync_buff_create(void)
 179 {
 180         struct ip_vs_sync_buff *sb;
 181
 182         if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
 183                 return NULL;
 184
 185         if (!(sb->mesg=kmalloc(sync_send_mesg_maxlen, GFP_ATOMIC))) {
 186                 kfree(sb);
 187                 return NULL;
 188         }
 189         sb->mesg->nr_conns = 0;
 190         sb->mesg->syncid = ip_vs_master_syncid;
 191         sb->mesg->size = 4;
 192         sb->head = (unsigned char *)sb->mesg + 4;
 193         sb->end = (unsigned char *)sb->mesg + sync_send_mesg_maxlen;
 194         sb->firstuse = jiffies;
 195         return sb;
 196 }
 197
 198 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
 199 {
 200         kfree(sb->mesg);
 201         kfree(sb);
 202 }
 203
 204 static inline void sb_queue_tail(struct ip_vs_sync_buff *sb)
 205 {
 206         spin_lock(&ip_vs_sync_lock);
 207         if (ip_vs_sync_state & IP_VS_STATE_MASTER)
 208                 list_add_tail(&sb->list, &ip_vs_sync_queue);
 209         else
 210                 ip_vs_sync_buff_release(sb);
 211         spin_unlock(&ip_vs_sync_lock);
 212 }
 213
 214 /*
 215  *      Get the current sync buffer if it has been created for more
 216  *      than the specified time or the specified time is zero.
 217  */
 218 static inline struct ip_vs_sync_buff *
 219 get_curr_sync_buff(unsigned long time)
 220 {
 221         struct ip_vs_sync_buff *sb;
 222
 223         spin_lock_bh(&curr_sb_lock);
 224         if (curr_sb && (time == 0 ||
 225                         time_before(jiffies - curr_sb->firstuse, time))) {
 226                 sb = curr_sb;
 227                 curr_sb = NULL;
 228         } else
 229                 sb = NULL;
 230         spin_unlock_bh(&curr_sb_lock);
 231         return sb;
 232 }
 233
 234
 235 /*
 236  *      Add an ip_vs_conn information into the current sync_buff.
 237  *      Called by ip_vs_in.
 238  */
 239 void ip_vs_sync_conn(struct ip_vs_conn *cp)
 240 {
 241         struct ip_vs_sync_mesg *m;
 242         struct ip_vs_sync_conn *s;
 243         int len;
 244
 245         spin_lock(&curr_sb_lock);
 246         if (!curr_sb) {
 247                 if (!(curr_sb=ip_vs_sync_buff_create())) {
 248                         spin_unlock(&curr_sb_lock);
 249                         pr_err("ip_vs_sync_buff_create failed.\n");
 250                         return;
 251                 }
 252         }
 253
 254         len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
 255                 SIMPLE_CONN_SIZE;
 256         m = curr_sb->mesg;
 257         s = (struct ip_vs_sync_conn *)curr_sb->head;
 258
 259         /* copy members */
 260         s->protocol = cp->protocol;
 261         s->cport = cp->cport;
 262         s->vport = cp->vport;
 263         s->dport = cp->dport;
 264         s->caddr = cp->caddr.ip;
 265         s->vaddr = cp->vaddr.ip;
 266         s->daddr = cp->daddr.ip;
 267         s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
 268         s->state = htons(cp->state);
 269         if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
 270                 struct ip_vs_sync_conn_options *opt =
 271                         (struct ip_vs_sync_conn_options *)&s[1];
 272                 memcpy(opt, &cp->in_seq, sizeof(*opt));
 273         }
 274
 275         m->nr_conns++;
 276         m->size += len;
 277         curr_sb->head += len;
 278
 279         /* check if there is a space for next one */
 280         if (curr_sb->head+FULL_CONN_SIZE > curr_sb->end) {
 281                 sb_queue_tail(curr_sb);
 282                 curr_sb = NULL;
 283         }
 284         spin_unlock(&curr_sb_lock);
 285
 286         /* synchronize its controller if it has */
 287         if (cp->control)
 288                 ip_vs_sync_conn(cp->control);
 289 }
 290
 291
 292 /*
 293  *      Process received multicast message and create the corresponding
 294  *      ip_vs_conn entries.
 295  */
 296 static void ip_vs_process_message(const char *buffer, const size_t buflen)
 297 {
 298         struct ip_vs_sync_mesg *m = (struct ip_vs_sync_mesg *)buffer;
 299         struct ip_vs_sync_conn *s;
 300         struct ip_vs_sync_conn_options *opt;
 301         struct ip_vs_conn *cp;
 302         struct ip_vs_protocol *pp;
 303         struct ip_vs_dest *dest;
 304         char *p;
 305         int i;
 306
 307         if (buflen < sizeof(struct ip_vs_sync_mesg)) {
 308                 IP_VS_ERR_RL("sync message header too short\n");
 309                 return;
 310         }
 311
 312         /* Convert size back to host byte order */
 313         m->size = ntohs(m->size);
 314
 315         if (buflen != m->size) {
 316                 IP_VS_ERR_RL("bogus sync message size\n");
 317                 return;
 318         }
 319
 320         /* SyncID sanity check */
 321         if (ip_vs_backup_syncid != 0 && m->syncid != ip_vs_backup_syncid) {
 322                 IP_VS_DBG(7, "Ignoring incoming msg with syncid = %d\n",
 323                           m->syncid);
 324                 return;
 325         }
 326
 327         p = (char *)buffer + sizeof(struct ip_vs_sync_mesg);
 328         for (i=0; i<m->nr_conns; i++) {
 329                 unsigned flags, state;
 330
 331                 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
 332                         IP_VS_ERR_RL("bogus conn in sync message\n");
 333                         return;
 334                 }
 335                 s = (struct ip_vs_sync_conn *) p;
 336                 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
 337                 flags &= ~IP_VS_CONN_F_HASHED;
 338                 if (flags & IP_VS_CONN_F_SEQ_MASK) {
 339                         opt = (struct ip_vs_sync_conn_options *)&s[1];
 340                         p += FULL_CONN_SIZE;
 341                         if (p > buffer+buflen) {
 342                                 IP_VS_ERR_RL("bogus conn options in sync message\n");
 343                                 return;
 344                         }
 345                 } else {
 346                         opt = NULL;
 347                         p += SIMPLE_CONN_SIZE;
 348                 }
 349
 350                 state = ntohs(s->state);
 351                 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
 352                         pp = ip_vs_proto_get(s->protocol);
 353                         if (!pp) {
 354                                 IP_VS_ERR_RL("Unsupported protocol %u in sync msg\n",
 355                                         s->protocol);
 356                                 continue;
 357                         }
 358                         if (state >= pp->num_states) {
 359                                 IP_VS_DBG(2, "Invalid %s state %u in sync msg\n",
 360                                         pp->name, state);
 361                                 continue;
 362                         }
 363                 } else {
 364                         /* protocol in templates is not used for state/timeout */
 365                         pp = NULL;
 366                         if (state > 0) {
 367                                 IP_VS_DBG(2, "Invalid template state %u in sync msg\n",
 368                                         state);
 369                                 state = 0;
 370                         }
 371                 }
 372
 373                 if (!(flags & IP_VS_CONN_F_TEMPLATE))
 374                         cp = ip_vs_conn_in_get(AF_INET, s->protocol,
 375                                                (union nf_inet_addr *)&s->caddr,
 376                                                s->cport,
 377                                                (union nf_inet_addr *)&s->vaddr,
 378                                                s->vport);
 379                 else
 380                         cp = ip_vs_ct_in_get(AF_INET, s->protocol,
 381                                              (union nf_inet_addr *)&s->caddr,
 382                                              s->cport,
 383                                              (union nf_inet_addr *)&s->vaddr,
 384                                              s->vport);
 385                 if (!cp) {
 386                         /*
 387                          * Find the appropriate destination for the connection.
 388                          * If it is not found the connection will remain unbound
 389                          * but still handled.
 390                          */
 391                         dest = ip_vs_find_dest(AF_INET,
 392                                                (union nf_inet_addr *)&s->daddr,
 393                                                s->dport,
 394                                                (union nf_inet_addr *)&s->vaddr,
 395                                                s->vport,
 396                                                s->protocol);
 397                         /*  Set the approprite ativity flag */
 398                         if (s->protocol == IPPROTO_TCP) {
 399                                 if (state != IP_VS_TCP_S_ESTABLISHED)
 400                                         flags |= IP_VS_CONN_F_INACTIVE;
 401                                 else
 402                                         flags &= ~IP_VS_CONN_F_INACTIVE;
 403                         } else if (s->protocol == IPPROTO_SCTP) {
 404                                 if (state != IP_VS_SCTP_S_ESTABLISHED)
 405                                         flags |= IP_VS_CONN_F_INACTIVE;
 406                                 else
 407                                         flags &= ~IP_VS_CONN_F_INACTIVE;
 408                         }
 409                         cp = ip_vs_conn_new(AF_INET, s->protocol,
 410                                             (union nf_inet_addr *)&s->caddr,
 411                                             s->cport,
 412                                             (union nf_inet_addr *)&s->vaddr,
 413                                             s->vport,
 414                                             (union nf_inet_addr *)&s->daddr,
 415                                             s->dport,
 416                                             flags, dest);
 417                         if (dest)
 418                                 atomic_dec(&dest->refcnt);
 419                         if (!cp) {
 420                                 pr_err("ip_vs_conn_new failed\n");
 421                                 return;
 422                         }
 423                 } else if (!cp->dest) {
 424                         dest = ip_vs_try_bind_dest(cp);
 425                         if (dest)
 426                                 atomic_dec(&dest->refcnt);
 427                 } else if ((cp->dest) && (cp->protocol == IPPROTO_TCP) &&
 428                            (cp->state != state)) {
 429                         /* update active/inactive flag for the connection */
 430                         dest = cp->dest;
 431                         if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
 432                                 (state != IP_VS_TCP_S_ESTABLISHED)) {
 433                                 atomic_dec(&dest->activeconns);
 434                                 atomic_inc(&dest->inactconns);
 435                                 cp->flags |= IP_VS_CONN_F_INACTIVE;
 436                         } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) &&
 437                                 (state == IP_VS_TCP_S_ESTABLISHED)) {
 438                                 atomic_inc(&dest->activeconns);
 439                                 atomic_dec(&dest->inactconns);
 440                                 cp->flags &= ~IP_VS_CONN_F_INACTIVE;
 441                         }
 442                 } else if ((cp->dest) && (cp->protocol == IPPROTO_SCTP) &&
 443                            (cp->state != state)) {
 444                         dest = cp->dest;
 445                         if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
 446                              (state != IP_VS_SCTP_S_ESTABLISHED)) {
 447                             atomic_dec(&dest->activeconns);
 448                             atomic_inc(&dest->inactconns);
 449                             cp->flags &= ~IP_VS_CONN_F_INACTIVE;
 450                         }
 451                 }
 452
 453                 if (opt)
 454                         memcpy(&cp->in_seq, opt, sizeof(*opt));
 455                 atomic_set(&cp->in_pkts, sysctl_ip_vs_sync_threshold[0]);
 456                 cp->state = state;
 457                 cp->old_state = cp->state;
 458                 /*
 459                  * We can not recover the right timeout for templates
 460                  * in all cases, we can not find the right fwmark
 461                  * virtual service. If needed, we can do it for
 462                  * non-fwmark persistent services.
 463                  */
 464                 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pp->timeout_table)
 465                         cp->timeout = pp->timeout_table[state];
 466                 else
 467                         cp->timeout = (3*60*HZ);
 468                 ip_vs_conn_put(cp);
 469         }
 470 }
 471
 472
 473 /*
 474  *      Setup loopback of outgoing multicasts on a sending socket
 475  */
 476 static void set_mcast_loop(struct sock *sk, u_char loop)
 477 {
 478         struct inet_sock *inet = inet_sk(sk);
 479
 480         /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
 481         lock_sock(sk);
 482         inet->mc_loop = loop ? 1 : 0;
 483         release_sock(sk);
 484 }
 485
 486 /*
 487  *      Specify TTL for outgoing multicasts on a sending socket
 488  */
 489 static void set_mcast_ttl(struct sock *sk, u_char ttl)
 490 {
 491         struct inet_sock *inet = inet_sk(sk);
 492
 493         /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
 494         lock_sock(sk);
 495         inet->mc_ttl = ttl;
 496         release_sock(sk);
 497 }
 498
 499 /*
 500  *      Specifiy default interface for outgoing multicasts
 501  */
 502 static int set_mcast_if(struct sock *sk, char *ifname)
 503 {
 504         struct net_device *dev;
 505         struct inet_sock *inet = inet_sk(sk);
 506
 507         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
 508                 return -ENODEV;
 509
 510         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
 511                 return -EINVAL;
 512
 513         lock_sock(sk);
 514         inet->mc_index = dev->ifindex;
 515         /*  inet->mc_addr  = 0; */
 516         release_sock(sk);
 517
 518         return 0;
 519 }
 520
 521
 522 /*
 523  *      Set the maximum length of sync message according to the
 524  *      specified interface's MTU.
 525  */
 526 static int set_sync_mesg_maxlen(int sync_state)
 527 {
 528         struct net_device *dev;
 529         int num;
 530
 531         if (sync_state == IP_VS_STATE_MASTER) {
 532                 if ((dev = __dev_get_by_name(&init_net, ip_vs_master_mcast_ifn)) == NULL)
 533                         return -ENODEV;
 534
 535                 num = (dev->mtu - sizeof(struct iphdr) -
 536                        sizeof(struct udphdr) -
 537                        SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE;
 538                 sync_send_mesg_maxlen = SYNC_MESG_HEADER_LEN +
 539                         SIMPLE_CONN_SIZE * min(num, MAX_CONNS_PER_SYNCBUFF);
 540                 IP_VS_DBG(7, "setting the maximum length of sync sending "
 541                           "message %d.\n", sync_send_mesg_maxlen);
 542         } else if (sync_state == IP_VS_STATE_BACKUP) {
 543                 if ((dev = __dev_get_by_name(&init_net, ip_vs_backup_mcast_ifn)) == NULL)
 544                         return -ENODEV;
 545
 546                 sync_recv_mesg_maxlen = dev->mtu -
 547                         sizeof(struct iphdr) - sizeof(struct udphdr);
 548                 IP_VS_DBG(7, "setting the maximum length of sync receiving "
 549                           "message %d.\n", sync_recv_mesg_maxlen);
 550         }
 551
 552         return 0;
 553 }
 554
 555
 556 /*
 557  *      Join a multicast group.
 558  *      the group is specified by a class D multicast address 224.0.0.0/8
 559  *      in the in_addr structure passed in as a parameter.
 560  */
 561 static int
 562 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
 563 {
 564         struct ip_mreqn mreq;
 565         struct net_device *dev;
 566         int ret;
 567
 568         memset(&mreq, 0, sizeof(mreq));
 569         memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
 570
 571         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
 572                 return -ENODEV;
 573         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
 574                 return -EINVAL;
 575
 576         mreq.imr_ifindex = dev->ifindex;
 577
 578         lock_sock(sk);
 579         ret = ip_mc_join_group(sk, &mreq);
 580         release_sock(sk);
 581
 582         return ret;
 583 }
 584
 585
 586 static int bind_mcastif_addr(struct socket *sock, char *ifname)
 587 {
 588         struct net_device *dev;
 589         __be32 addr;
 590         struct sockaddr_in sin;
 591
 592         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
 593                 return -ENODEV;
 594
 595         addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
 596         if (!addr)
 597                 pr_err("You probably need to specify IP address on "
 598                        "multicast interface.\n");
 599
 600         IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
 601                   ifname, &addr);
 602
 603         /* Now bind the socket with the address of multicast interface */
 604         sin.sin_family       = AF_INET;
 605         sin.sin_addr.s_addr  = addr;
 606         sin.sin_port         = 0;
 607
 608         return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
 609 }
 610
 611 /*
 612  *      Set up sending multicast socket over UDP
 613  */
 614 static struct socket * make_send_sock(void)
 615 {
 616         struct socket *sock;
 617         int result;
 618
 619         /* First create a socket */
 620         result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
 621         if (result < 0) {
 622                 pr_err("Error during creation of socket; terminating\n");
 623                 return ERR_PTR(result);
 624         }
 625
 626         result = set_mcast_if(sock->sk, ip_vs_master_mcast_ifn);
 627         if (result < 0) {
 628                 pr_err("Error setting outbound mcast interface\n");
 629                 goto error;
 630         }
 631
 632         set_mcast_loop(sock->sk, 0);
 633         set_mcast_ttl(sock->sk, 1);
 634
 635         result = bind_mcastif_addr(sock, ip_vs_master_mcast_ifn);
 636         if (result < 0) {
 637                 pr_err("Error binding address of the mcast interface\n");
 638                 goto error;
 639         }
 640
 641         result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr,
 642                         sizeof(struct sockaddr), 0);
 643         if (result < 0) {
 644                 pr_err("Error connecting to the multicast addr\n");
 645                 goto error;
 646         }
 647
 648         return sock;
 649
 650   error:
 651         sock_release(sock);
 652         return ERR_PTR(result);
 653 }
 654
 655
 656 /*
 657  *      Set up receiving multicast socket over UDP
 658  */
 659 static struct socket * make_receive_sock(void)
 660 {
 661         struct socket *sock;
 662         int result;
 663
 664         /* First create a socket */
 665         result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
 666         if (result < 0) {
 667                 pr_err("Error during creation of socket; terminating\n");
 668                 return ERR_PTR(result);
 669         }
 670
 671         /* it is equivalent to the REUSEADDR option in user-space */
 672         sock->sk->sk_reuse = 1;
 673
 674         result = sock->ops->bind(sock, (struct sockaddr *) &mcast_addr,
 675                         sizeof(struct sockaddr));
 676         if (result < 0) {
 677                 pr_err("Error binding to the multicast addr\n");
 678                 goto error;
 679         }
 680
 681         /* join the multicast group */
 682         result = join_mcast_group(sock->sk,
 683                         (struct in_addr *) &mcast_addr.sin_addr,
 684                         ip_vs_backup_mcast_ifn);
 685         if (result < 0) {
 686                 pr_err("Error joining to the multicast group\n");
 687                 goto error;
 688         }
 689
 690         return sock;
 691
 692   error:
 693         sock_release(sock);
 694         return ERR_PTR(result);
 695 }
 696
 697
 698 static int
 699 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
 700 {
 701         struct msghdr   msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
 702         struct kvec     iov;
 703         int             len;
 704
 705         EnterFunction(7);
 706         iov.iov_base     = (void *)buffer;
 707         iov.iov_len      = length;
 708
 709         len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
 710
 711         LeaveFunction(7);
 712         return len;
 713 }
 714
 715 static void
 716 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
 717 {
 718         int msize;
 719
 720         msize = msg->size;
 721
 722         /* Put size in network byte order */
 723         msg->size = htons(msg->size);
 724
 725         if (ip_vs_send_async(sock, (char *)msg, msize) != msize)
 726                 pr_err("ip_vs_send_async error\n");
 727 }
 728
 729 static int
 730 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
 731 {
 732         struct msghdr           msg = {NULL,};
 733         struct kvec             iov;
 734         int                     len;
 735
 736         EnterFunction(7);
 737
 738         /* Receive a packet */
 739         iov.iov_base     = buffer;
 740         iov.iov_len      = (size_t)buflen;
 741
 742         len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, 0);
 743
 744         if (len < 0)
 745                 return -1;
 746
 747         LeaveFunction(7);
 748         return len;
 749 }
 750
 751
 752 static int sync_thread_master(void *data)
 753 {
 754         struct ip_vs_sync_thread_data *tinfo = data;
 755         struct ip_vs_sync_buff *sb;
 756
 757         pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
 758                 "syncid = %d\n",
 759                 ip_vs_master_mcast_ifn, ip_vs_master_syncid);
 760
 761         while (!kthread_should_stop()) {
 762                 while ((sb = sb_dequeue())) {
 763                         ip_vs_send_sync_msg(tinfo->sock, sb->mesg);
 764                         ip_vs_sync_buff_release(sb);
 765                 }
 766
 767                 /* check if entries stay in curr_sb for 2 seconds */
 768                 sb = get_curr_sync_buff(2 * HZ);
 769                 if (sb) {
 770                         ip_vs_send_sync_msg(tinfo->sock, sb->mesg);
 771                         ip_vs_sync_buff_release(sb);
 772                 }
 773
 774                 schedule_timeout_interruptible(HZ);
 775         }
 776
 777         /* clean up the sync_buff queue */
 778         while ((sb=sb_dequeue())) {
 779                 ip_vs_sync_buff_release(sb);
 780         }
 781
 782         /* clean up the current sync_buff */
 783         if ((sb = get_curr_sync_buff(0))) {
 784                 ip_vs_sync_buff_release(sb);
 785         }
 786
 787         /* release the sending multicast socket */
 788         sock_release(tinfo->sock);
 789         kfree(tinfo);
 790
 791         return 0;
 792 }
 793
 794
 795 static int sync_thread_backup(void *data)
 796 {
 797         struct ip_vs_sync_thread_data *tinfo = data;
 798         int len;
 799
 800         pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
 801                 "syncid = %d\n",
 802                 ip_vs_backup_mcast_ifn, ip_vs_backup_syncid);
 803
 804         while (!kthread_should_stop()) {
 805                 wait_event_interruptible(*sk_sleep(tinfo->sock->sk),
 806                          !skb_queue_empty(&tinfo->sock->sk->sk_receive_queue)
 807                          || kthread_should_stop());
 808
 809                 /* do we have data now? */
 810                 while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) {
 811                         len = ip_vs_receive(tinfo->sock, tinfo->buf,
 812                                         sync_recv_mesg_maxlen);
 813                         if (len <= 0) {
 814                                 pr_err("receiving message error\n");
 815                                 break;
 816                         }
 817
 818                         /* disable bottom half, because it accesses the data
 819                            shared by softirq while getting/creating conns */
 820                         local_bh_disable();
 821                         ip_vs_process_message(tinfo->buf, len);
 822                         local_bh_enable();
 823                 }
 824         }
 825
 826         /* release the sending multicast socket */
 827         sock_release(tinfo->sock);
 828         kfree(tinfo->buf);
 829         kfree(tinfo);
 830
 831         return 0;
 832 }
 833
 834
 835 int start_sync_thread(int state, char *mcast_ifn, __u8 syncid)
 836 {
 837         struct ip_vs_sync_thread_data *tinfo;
 838         struct task_struct **realtask, *task;
 839         struct socket *sock;
 840         char *name, *buf = NULL;
 841         int (*threadfn)(void *data);
 842         int result = -ENOMEM;
 843
 844         IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
 845         IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n",
 846                   sizeof(struct ip_vs_sync_conn));
 847
 848         if (state == IP_VS_STATE_MASTER) {
 849                 if (sync_master_thread)
 850                         return -EEXIST;
 851
 852                 strlcpy(ip_vs_master_mcast_ifn, mcast_ifn,
 853                         sizeof(ip_vs_master_mcast_ifn));
 854                 ip_vs_master_syncid = syncid;
 855                 realtask = &sync_master_thread;
 856                 name = "ipvs_syncmaster";
 857                 threadfn = sync_thread_master;
 858                 sock = make_send_sock();
 859         } else if (state == IP_VS_STATE_BACKUP) {
 860                 if (sync_backup_thread)
 861                         return -EEXIST;
 862
 863                 strlcpy(ip_vs_backup_mcast_ifn, mcast_ifn,
 864                         sizeof(ip_vs_backup_mcast_ifn));
 865                 ip_vs_backup_syncid = syncid;
 866                 realtask = &sync_backup_thread;
 867                 name = "ipvs_syncbackup";
 868                 threadfn = sync_thread_backup;
 869                 sock = make_receive_sock();
 870         } else {
 871                 return -EINVAL;
 872         }
 873
 874         if (IS_ERR(sock)) {
 875                 result = PTR_ERR(sock);
 876                 goto out;
 877         }
 878
 879         set_sync_mesg_maxlen(state);
 880         if (state == IP_VS_STATE_BACKUP) {
 881                 buf = kmalloc(sync_recv_mesg_maxlen, GFP_KERNEL);
 882                 if (!buf)
 883                         goto outsocket;
 884         }
 885
 886         tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL);
 887         if (!tinfo)
 888                 goto outbuf;
 889
 890         tinfo->sock = sock;
 891         tinfo->buf = buf;
 892
 893         task = kthread_run(threadfn, tinfo, name);
 894         if (IS_ERR(task)) {
 895                 result = PTR_ERR(task);
 896                 goto outtinfo;
 897         }
 898
 899         /* mark as active */
 900         *realtask = task;
 901         ip_vs_sync_state |= state;
 902
 903         /* increase the module use count */
 904         ip_vs_use_count_inc();
 905
 906         return 0;
 907
 908 outtinfo:
 909         kfree(tinfo);
 910 outbuf:
 911         kfree(buf);
 912 outsocket:
 913         sock_release(sock);
 914 out:
 915         return result;
 916 }
 917
 918
 919 int stop_sync_thread(int state)
 920 {
 921         IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
 922
 923         if (state == IP_VS_STATE_MASTER) {
 924                 if (!sync_master_thread)
 925                         return -ESRCH;
 926
 927                 pr_info("stopping master sync thread %d ...\n",
 928                         task_pid_nr(sync_master_thread));
 929
 930                 /*
 931                  * The lock synchronizes with sb_queue_tail(), so that we don't
 932                  * add sync buffers to the queue, when we are already in
 933                  * progress of stopping the master sync daemon.
 934                  */
 935
 936                 spin_lock_bh(&ip_vs_sync_lock);
 937                 ip_vs_sync_state &= ~IP_VS_STATE_MASTER;
 938                 spin_unlock_bh(&ip_vs_sync_lock);
 939                 kthread_stop(sync_master_thread);
 940                 sync_master_thread = NULL;
 941         } else if (state == IP_VS_STATE_BACKUP) {
 942                 if (!sync_backup_thread)
 943                         return -ESRCH;
 944
 945                 pr_info("stopping backup sync thread %d ...\n",
 946                         task_pid_nr(sync_backup_thread));
 947
 948                 ip_vs_sync_state &= ~IP_VS_STATE_BACKUP;
 949                 kthread_stop(sync_backup_thread);
 950                 sync_backup_thread = NULL;
 951         } else {
 952                 return -EINVAL;
 953         }
 954
 955         /* decrease the module use count */
 956         ip_vs_use_count_dec();
 957
 958         return 0;
 959 }