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 static inline int
 292 ip_vs_conn_fill_param_sync(int af, int protocol,
 293                            const union nf_inet_addr *caddr, __be16 cport,
 294                            const union nf_inet_addr *vaddr, __be16 vport,
 295                            struct ip_vs_conn_param *p)
 296 {
 297         /* XXX: Need to take into account persistence engine */
 298         ip_vs_conn_fill_param(af, protocol, caddr, cport, vaddr, vport, p);
 299         return 0;
 300 }
 301
 302 /*
 303  *      Process received multicast message and create the corresponding
 304  *      ip_vs_conn entries.
 305  */
 306 static void ip_vs_process_message(const char *buffer, const size_t buflen)
 307 {
 308         struct ip_vs_sync_mesg *m = (struct ip_vs_sync_mesg *)buffer;
 309         struct ip_vs_sync_conn *s;
 310         struct ip_vs_sync_conn_options *opt;
 311         struct ip_vs_conn *cp;
 312         struct ip_vs_protocol *pp;
 313         struct ip_vs_dest *dest;
 314         struct ip_vs_conn_param param;
 315         char *p;
 316         int i;
 317
 318         if (buflen < sizeof(struct ip_vs_sync_mesg)) {
 319                 IP_VS_ERR_RL("sync message header too short\n");
 320                 return;
 321         }
 322
 323         /* Convert size back to host byte order */
 324         m->size = ntohs(m->size);
 325
 326         if (buflen != m->size) {
 327                 IP_VS_ERR_RL("bogus sync message size\n");
 328                 return;
 329         }
 330
 331         /* SyncID sanity check */
 332         if (ip_vs_backup_syncid != 0 && m->syncid != ip_vs_backup_syncid) {
 333                 IP_VS_DBG(7, "Ignoring incoming msg with syncid = %d\n",
 334                           m->syncid);
 335                 return;
 336         }
 337
 338         p = (char *)buffer + sizeof(struct ip_vs_sync_mesg);
 339         for (i=0; i<m->nr_conns; i++) {
 340                 unsigned flags, state;
 341
 342                 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
 343                         IP_VS_ERR_RL("bogus conn in sync message\n");
 344                         return;
 345                 }
 346                 s = (struct ip_vs_sync_conn *) p;
 347                 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
 348                 flags &= ~IP_VS_CONN_F_HASHED;
 349                 if (flags & IP_VS_CONN_F_SEQ_MASK) {
 350                         opt = (struct ip_vs_sync_conn_options *)&s[1];
 351                         p += FULL_CONN_SIZE;
 352                         if (p > buffer+buflen) {
 353                                 IP_VS_ERR_RL("bogus conn options in sync message\n");
 354                                 return;
 355                         }
 356                 } else {
 357                         opt = NULL;
 358                         p += SIMPLE_CONN_SIZE;
 359                 }
 360
 361                 state = ntohs(s->state);
 362                 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
 363                         pp = ip_vs_proto_get(s->protocol);
 364                         if (!pp) {
 365                                 IP_VS_ERR_RL("Unsupported protocol %u in sync msg\n",
 366                                         s->protocol);
 367                                 continue;
 368                         }
 369                         if (state >= pp->num_states) {
 370                                 IP_VS_DBG(2, "Invalid %s state %u in sync msg\n",
 371                                         pp->name, state);
 372                                 continue;
 373                         }
 374                 } else {
 375                         /* protocol in templates is not used for state/timeout */
 376                         pp = NULL;
 377                         if (state > 0) {
 378                                 IP_VS_DBG(2, "Invalid template state %u in sync msg\n",
 379                                         state);
 380                                 state = 0;
 381                         }
 382                 }
 383
 384                 {
 385                         if (ip_vs_conn_fill_param_sync(AF_INET, s->protocol,
 386                                               (union nf_inet_addr *)&s->caddr,
 387                                               s->cport,
 388                                               (union nf_inet_addr *)&s->vaddr,
 389                                               s->vport, &param)) {
 390                                 pr_err("ip_vs_conn_fill_param_sync failed");
 391                                 return;
 392                         }
 393                         if (!(flags & IP_VS_CONN_F_TEMPLATE))
 394                                 cp = ip_vs_conn_in_get(&param);
 395                         else
 396                                 cp = ip_vs_ct_in_get(&param);
 397                 }
 398                 if (!cp) {
 399                         /*
 400                          * Find the appropriate destination for the connection.
 401                          * If it is not found the connection will remain unbound
 402                          * but still handled.
 403                          */
 404                         dest = ip_vs_find_dest(AF_INET,
 405                                                (union nf_inet_addr *)&s->daddr,
 406                                                s->dport,
 407                                                (union nf_inet_addr *)&s->vaddr,
 408                                                s->vport,
 409                                                s->protocol);
 410                         /*  Set the approprite ativity flag */
 411                         if (s->protocol == IPPROTO_TCP) {
 412                                 if (state != IP_VS_TCP_S_ESTABLISHED)
 413                                         flags |= IP_VS_CONN_F_INACTIVE;
 414                                 else
 415                                         flags &= ~IP_VS_CONN_F_INACTIVE;
 416                         } else if (s->protocol == IPPROTO_SCTP) {
 417                                 if (state != IP_VS_SCTP_S_ESTABLISHED)
 418                                         flags |= IP_VS_CONN_F_INACTIVE;
 419                                 else
 420                                         flags &= ~IP_VS_CONN_F_INACTIVE;
 421                         }
 422                         cp = ip_vs_conn_new(&param,
 423                                             (union nf_inet_addr *)&s->daddr,
 424                                             s->dport, flags, dest);
 425                         if (dest)
 426                                 atomic_dec(&dest->refcnt);
 427                         if (!cp) {
 428                                 pr_err("ip_vs_conn_new failed\n");
 429                                 return;
 430                         }
 431                 } else if (!cp->dest) {
 432                         dest = ip_vs_try_bind_dest(cp);
 433                         if (dest)
 434                                 atomic_dec(&dest->refcnt);
 435                 } else if ((cp->dest) && (cp->protocol == IPPROTO_TCP) &&
 436                            (cp->state != state)) {
 437                         /* update active/inactive flag for the connection */
 438                         dest = cp->dest;
 439                         if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
 440                                 (state != IP_VS_TCP_S_ESTABLISHED)) {
 441                                 atomic_dec(&dest->activeconns);
 442                                 atomic_inc(&dest->inactconns);
 443                                 cp->flags |= IP_VS_CONN_F_INACTIVE;
 444                         } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) &&
 445                                 (state == IP_VS_TCP_S_ESTABLISHED)) {
 446                                 atomic_inc(&dest->activeconns);
 447                                 atomic_dec(&dest->inactconns);
 448                                 cp->flags &= ~IP_VS_CONN_F_INACTIVE;
 449                         }
 450                 } else if ((cp->dest) && (cp->protocol == IPPROTO_SCTP) &&
 451                            (cp->state != state)) {
 452                         dest = cp->dest;
 453                         if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
 454                              (state != IP_VS_SCTP_S_ESTABLISHED)) {
 455                             atomic_dec(&dest->activeconns);
 456                             atomic_inc(&dest->inactconns);
 457                             cp->flags &= ~IP_VS_CONN_F_INACTIVE;
 458                         }
 459                 }
 460
 461                 if (opt)
 462                         memcpy(&cp->in_seq, opt, sizeof(*opt));
 463                 atomic_set(&cp->in_pkts, sysctl_ip_vs_sync_threshold[0]);
 464                 cp->state = state;
 465                 cp->old_state = cp->state;
 466                 /*
 467                  * We can not recover the right timeout for templates
 468                  * in all cases, we can not find the right fwmark
 469                  * virtual service. If needed, we can do it for
 470                  * non-fwmark persistent services.
 471                  */
 472                 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pp->timeout_table)
 473                         cp->timeout = pp->timeout_table[state];
 474                 else
 475                         cp->timeout = (3*60*HZ);
 476                 ip_vs_conn_put(cp);
 477         }
 478 }
 479
 480
 481 /*
 482  *      Setup loopback of outgoing multicasts on a sending socket
 483  */
 484 static void set_mcast_loop(struct sock *sk, u_char loop)
 485 {
 486         struct inet_sock *inet = inet_sk(sk);
 487
 488         /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
 489         lock_sock(sk);
 490         inet->mc_loop = loop ? 1 : 0;
 491         release_sock(sk);
 492 }
 493
 494 /*
 495  *      Specify TTL for outgoing multicasts on a sending socket
 496  */
 497 static void set_mcast_ttl(struct sock *sk, u_char ttl)
 498 {
 499         struct inet_sock *inet = inet_sk(sk);
 500
 501         /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
 502         lock_sock(sk);
 503         inet->mc_ttl = ttl;
 504         release_sock(sk);
 505 }
 506
 507 /*
 508  *      Specifiy default interface for outgoing multicasts
 509  */
 510 static int set_mcast_if(struct sock *sk, char *ifname)
 511 {
 512         struct net_device *dev;
 513         struct inet_sock *inet = inet_sk(sk);
 514
 515         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
 516                 return -ENODEV;
 517
 518         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
 519                 return -EINVAL;
 520
 521         lock_sock(sk);
 522         inet->mc_index = dev->ifindex;
 523         /*  inet->mc_addr  = 0; */
 524         release_sock(sk);
 525
 526         return 0;
 527 }
 528
 529
 530 /*
 531  *      Set the maximum length of sync message according to the
 532  *      specified interface's MTU.
 533  */
 534 static int set_sync_mesg_maxlen(int sync_state)
 535 {
 536         struct net_device *dev;
 537         int num;
 538
 539         if (sync_state == IP_VS_STATE_MASTER) {
 540                 if ((dev = __dev_get_by_name(&init_net, ip_vs_master_mcast_ifn)) == NULL)
 541                         return -ENODEV;
 542
 543                 num = (dev->mtu - sizeof(struct iphdr) -
 544                        sizeof(struct udphdr) -
 545                        SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE;
 546                 sync_send_mesg_maxlen = SYNC_MESG_HEADER_LEN +
 547                         SIMPLE_CONN_SIZE * min(num, MAX_CONNS_PER_SYNCBUFF);
 548                 IP_VS_DBG(7, "setting the maximum length of sync sending "
 549                           "message %d.\n", sync_send_mesg_maxlen);
 550         } else if (sync_state == IP_VS_STATE_BACKUP) {
 551                 if ((dev = __dev_get_by_name(&init_net, ip_vs_backup_mcast_ifn)) == NULL)
 552                         return -ENODEV;
 553
 554                 sync_recv_mesg_maxlen = dev->mtu -
 555                         sizeof(struct iphdr) - sizeof(struct udphdr);
 556                 IP_VS_DBG(7, "setting the maximum length of sync receiving "
 557                           "message %d.\n", sync_recv_mesg_maxlen);
 558         }
 559
 560         return 0;
 561 }
 562
 563
 564 /*
 565  *      Join a multicast group.
 566  *      the group is specified by a class D multicast address 224.0.0.0/8
 567  *      in the in_addr structure passed in as a parameter.
 568  */
 569 static int
 570 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
 571 {
 572         struct ip_mreqn mreq;
 573         struct net_device *dev;
 574         int ret;
 575
 576         memset(&mreq, 0, sizeof(mreq));
 577         memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
 578
 579         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
 580                 return -ENODEV;
 581         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
 582                 return -EINVAL;
 583
 584         mreq.imr_ifindex = dev->ifindex;
 585
 586         lock_sock(sk);
 587         ret = ip_mc_join_group(sk, &mreq);
 588         release_sock(sk);
 589
 590         return ret;
 591 }
 592
 593
 594 static int bind_mcastif_addr(struct socket *sock, char *ifname)
 595 {
 596         struct net_device *dev;
 597         __be32 addr;
 598         struct sockaddr_in sin;
 599
 600         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
 601                 return -ENODEV;
 602
 603         addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
 604         if (!addr)
 605                 pr_err("You probably need to specify IP address on "
 606                        "multicast interface.\n");
 607
 608         IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
 609                   ifname, &addr);
 610
 611         /* Now bind the socket with the address of multicast interface */
 612         sin.sin_family       = AF_INET;
 613         sin.sin_addr.s_addr  = addr;
 614         sin.sin_port         = 0;
 615
 616         return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
 617 }
 618
 619 /*
 620  *      Set up sending multicast socket over UDP
 621  */
 622 static struct socket * make_send_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                 pr_err("Error during creation of socket; terminating\n");
 631                 return ERR_PTR(result);
 632         }
 633
 634         result = set_mcast_if(sock->sk, ip_vs_master_mcast_ifn);
 635         if (result < 0) {
 636                 pr_err("Error setting outbound mcast interface\n");
 637                 goto error;
 638         }
 639
 640         set_mcast_loop(sock->sk, 0);
 641         set_mcast_ttl(sock->sk, 1);
 642
 643         result = bind_mcastif_addr(sock, ip_vs_master_mcast_ifn);
 644         if (result < 0) {
 645                 pr_err("Error binding address of the mcast interface\n");
 646                 goto error;
 647         }
 648
 649         result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr,
 650                         sizeof(struct sockaddr), 0);
 651         if (result < 0) {
 652                 pr_err("Error connecting to the multicast addr\n");
 653                 goto error;
 654         }
 655
 656         return sock;
 657
 658   error:
 659         sock_release(sock);
 660         return ERR_PTR(result);
 661 }
 662
 663
 664 /*
 665  *      Set up receiving multicast socket over UDP
 666  */
 667 static struct socket * make_receive_sock(void)
 668 {
 669         struct socket *sock;
 670         int result;
 671
 672         /* First create a socket */
 673         result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
 674         if (result < 0) {
 675                 pr_err("Error during creation of socket; terminating\n");
 676                 return ERR_PTR(result);
 677         }
 678
 679         /* it is equivalent to the REUSEADDR option in user-space */
 680         sock->sk->sk_reuse = 1;
 681
 682         result = sock->ops->bind(sock, (struct sockaddr *) &mcast_addr,
 683                         sizeof(struct sockaddr));
 684         if (result < 0) {
 685                 pr_err("Error binding to the multicast addr\n");
 686                 goto error;
 687         }
 688
 689         /* join the multicast group */
 690         result = join_mcast_group(sock->sk,
 691                         (struct in_addr *) &mcast_addr.sin_addr,
 692                         ip_vs_backup_mcast_ifn);
 693         if (result < 0) {
 694                 pr_err("Error joining to the multicast group\n");
 695                 goto error;
 696         }
 697
 698         return sock;
 699
 700   error:
 701         sock_release(sock);
 702         return ERR_PTR(result);
 703 }
 704
 705
 706 static int
 707 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
 708 {
 709         struct msghdr   msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
 710         struct kvec     iov;
 711         int             len;
 712
 713         EnterFunction(7);
 714         iov.iov_base     = (void *)buffer;
 715         iov.iov_len      = length;
 716
 717         len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
 718
 719         LeaveFunction(7);
 720         return len;
 721 }
 722
 723 static void
 724 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
 725 {
 726         int msize;
 727
 728         msize = msg->size;
 729
 730         /* Put size in network byte order */
 731         msg->size = htons(msg->size);
 732
 733         if (ip_vs_send_async(sock, (char *)msg, msize) != msize)
 734                 pr_err("ip_vs_send_async error\n");
 735 }
 736
 737 static int
 738 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
 739 {
 740         struct msghdr           msg = {NULL,};
 741         struct kvec             iov;
 742         int                     len;
 743
 744         EnterFunction(7);
 745
 746         /* Receive a packet */
 747         iov.iov_base     = buffer;
 748         iov.iov_len      = (size_t)buflen;
 749
 750         len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, 0);
 751
 752         if (len < 0)
 753                 return -1;
 754
 755         LeaveFunction(7);
 756         return len;
 757 }
 758
 759
 760 static int sync_thread_master(void *data)
 761 {
 762         struct ip_vs_sync_thread_data *tinfo = data;
 763         struct ip_vs_sync_buff *sb;
 764
 765         pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
 766                 "syncid = %d\n",
 767                 ip_vs_master_mcast_ifn, ip_vs_master_syncid);
 768
 769         while (!kthread_should_stop()) {
 770                 while ((sb = sb_dequeue())) {
 771                         ip_vs_send_sync_msg(tinfo->sock, sb->mesg);
 772                         ip_vs_sync_buff_release(sb);
 773                 }
 774
 775                 /* check if entries stay in curr_sb for 2 seconds */
 776                 sb = get_curr_sync_buff(2 * HZ);
 777                 if (sb) {
 778                         ip_vs_send_sync_msg(tinfo->sock, sb->mesg);
 779                         ip_vs_sync_buff_release(sb);
 780                 }
 781
 782                 schedule_timeout_interruptible(HZ);
 783         }
 784
 785         /* clean up the sync_buff queue */
 786         while ((sb=sb_dequeue())) {
 787                 ip_vs_sync_buff_release(sb);
 788         }
 789
 790         /* clean up the current sync_buff */
 791         if ((sb = get_curr_sync_buff(0))) {
 792                 ip_vs_sync_buff_release(sb);
 793         }
 794
 795         /* release the sending multicast socket */
 796         sock_release(tinfo->sock);
 797         kfree(tinfo);
 798
 799         return 0;
 800 }
 801
 802
 803 static int sync_thread_backup(void *data)
 804 {
 805         struct ip_vs_sync_thread_data *tinfo = data;
 806         int len;
 807
 808         pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
 809                 "syncid = %d\n",
 810                 ip_vs_backup_mcast_ifn, ip_vs_backup_syncid);
 811
 812         while (!kthread_should_stop()) {
 813                 wait_event_interruptible(*sk_sleep(tinfo->sock->sk),
 814                          !skb_queue_empty(&tinfo->sock->sk->sk_receive_queue)
 815                          || kthread_should_stop());
 816
 817                 /* do we have data now? */
 818                 while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) {
 819                         len = ip_vs_receive(tinfo->sock, tinfo->buf,
 820                                         sync_recv_mesg_maxlen);
 821                         if (len <= 0) {
 822                                 pr_err("receiving message error\n");
 823                                 break;
 824                         }
 825
 826                         /* disable bottom half, because it accesses the data
 827                            shared by softirq while getting/creating conns */
 828                         local_bh_disable();
 829                         ip_vs_process_message(tinfo->buf, len);
 830                         local_bh_enable();
 831                 }
 832         }
 833
 834         /* release the sending multicast socket */
 835         sock_release(tinfo->sock);
 836         kfree(tinfo->buf);
 837         kfree(tinfo);
 838
 839         return 0;
 840 }
 841
 842
 843 int start_sync_thread(int state, char *mcast_ifn, __u8 syncid)
 844 {
 845         struct ip_vs_sync_thread_data *tinfo;
 846         struct task_struct **realtask, *task;
 847         struct socket *sock;
 848         char *name, *buf = NULL;
 849         int (*threadfn)(void *data);
 850         int result = -ENOMEM;
 851
 852         IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
 853         IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n",
 854                   sizeof(struct ip_vs_sync_conn));
 855
 856         if (state == IP_VS_STATE_MASTER) {
 857                 if (sync_master_thread)
 858                         return -EEXIST;
 859
 860                 strlcpy(ip_vs_master_mcast_ifn, mcast_ifn,
 861                         sizeof(ip_vs_master_mcast_ifn));
 862                 ip_vs_master_syncid = syncid;
 863                 realtask = &sync_master_thread;
 864                 name = "ipvs_syncmaster";
 865                 threadfn = sync_thread_master;
 866                 sock = make_send_sock();
 867         } else if (state == IP_VS_STATE_BACKUP) {
 868                 if (sync_backup_thread)
 869                         return -EEXIST;
 870
 871                 strlcpy(ip_vs_backup_mcast_ifn, mcast_ifn,
 872                         sizeof(ip_vs_backup_mcast_ifn));
 873                 ip_vs_backup_syncid = syncid;
 874                 realtask = &sync_backup_thread;
 875                 name = "ipvs_syncbackup";
 876                 threadfn = sync_thread_backup;
 877                 sock = make_receive_sock();
 878         } else {
 879                 return -EINVAL;
 880         }
 881
 882         if (IS_ERR(sock)) {
 883                 result = PTR_ERR(sock);
 884                 goto out;
 885         }
 886
 887         set_sync_mesg_maxlen(state);
 888         if (state == IP_VS_STATE_BACKUP) {
 889                 buf = kmalloc(sync_recv_mesg_maxlen, GFP_KERNEL);
 890                 if (!buf)
 891                         goto outsocket;
 892         }
 893
 894         tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL);
 895         if (!tinfo)
 896                 goto outbuf;
 897
 898         tinfo->sock = sock;
 899         tinfo->buf = buf;
 900
 901         task = kthread_run(threadfn, tinfo, name);
 902         if (IS_ERR(task)) {
 903                 result = PTR_ERR(task);
 904                 goto outtinfo;
 905         }
 906
 907         /* mark as active */
 908         *realtask = task;
 909         ip_vs_sync_state |= state;
 910
 911         /* increase the module use count */
 912         ip_vs_use_count_inc();
 913
 914         return 0;
 915
 916 outtinfo:
 917         kfree(tinfo);
 918 outbuf:
 919         kfree(buf);
 920 outsocket:
 921         sock_release(sock);
 922 out:
 923         return result;
 924 }
 925
 926
 927 int stop_sync_thread(int state)
 928 {
 929         IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
 930
 931         if (state == IP_VS_STATE_MASTER) {
 932                 if (!sync_master_thread)
 933                         return -ESRCH;
 934
 935                 pr_info("stopping master sync thread %d ...\n",
 936                         task_pid_nr(sync_master_thread));
 937
 938                 /*
 939                  * The lock synchronizes with sb_queue_tail(), so that we don't
 940                  * add sync buffers to the queue, when we are already in
 941                  * progress of stopping the master sync daemon.
 942                  */
 943
 944                 spin_lock_bh(&ip_vs_sync_lock);
 945                 ip_vs_sync_state &= ~IP_VS_STATE_MASTER;
 946                 spin_unlock_bh(&ip_vs_sync_lock);
 947                 kthread_stop(sync_master_thread);
 948                 sync_master_thread = NULL;
 949         } else if (state == IP_VS_STATE_BACKUP) {
 950                 if (!sync_backup_thread)
 951                         return -ESRCH;
 952
 953                 pr_info("stopping backup sync thread %d ...\n",
 954                         task_pid_nr(sync_backup_thread));
 955
 956                 ip_vs_sync_state &= ~IP_VS_STATE_BACKUP;
 957                 kthread_stop(sync_backup_thread);
 958                 sync_backup_thread = NULL;
 959         } else {
 960                 return -EINVAL;
 961         }
 962
 963         /* decrease the module use count */
 964         ip_vs_use_count_dec();
 965
 966         return 0;
 967 }