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  * Version 1,   is capable of handling both version 0 and 1 messages.
   9  *              Version 0 is the plain old format.
  10  *              Note Version 0 receivers will just drop Ver 1 messages.
  11  *              Version 1 is capable of handle IPv6, Persistence data,
  12  *              time-outs, and firewall marks.
  13  *              In ver.1 "ip_vs_sync_conn_options" will be sent in netw. order.
  14  *              Ver. 0 can be turned on by sysctl -w net.ipv4.vs.sync_version=0
  15  *
  16  * Definitions  Message: is a complete datagram
  17  *              Sync_conn: is a part of a Message
  18  *              Param Data is an option to a Sync_conn.
  19  *
  20  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
  21  *
  22  * ip_vs_sync:  sync connection info from master load balancer to backups
  23  *              through multicast
  24  *
  25  * Changes:
  26  *      Alexandre Cassen        :       Added master & backup support at a time.
  27  *      Alexandre Cassen        :       Added SyncID support for incoming sync
  28  *                                      messages filtering.
  29  *      Justin Ossevoort        :       Fix endian problem on sync message size.
  30  *      Hans Schillstrom        :       Added Version 1: i.e. IPv6,
  31  *                                      Persistence support, fwmark and time-out.
  32  */
  33
  34 #define KMSG_COMPONENT "IPVS"
  35 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  36
  37 #include <linux/module.h>
  38 #include <linux/slab.h>
  39 #include <linux/inetdevice.h>
  40 #include <linux/net.h>
  41 #include <linux/completion.h>
  42 #include <linux/delay.h>
  43 #include <linux/skbuff.h>
  44 #include <linux/in.h>
  45 #include <linux/igmp.h>                 /* for ip_mc_join_group */
  46 #include <linux/udp.h>
  47 #include <linux/err.h>
  48 #include <linux/kthread.h>
  49 #include <linux/wait.h>
  50 #include <linux/kernel.h>
  51
  52 #include <asm/unaligned.h>              /* Used for ntoh_seq and hton_seq */
  53
  54 #include <net/ip.h>
  55 #include <net/sock.h>
  56
  57 #include <net/ip_vs.h>
  58
  59 #define IP_VS_SYNC_GROUP 0xe0000051    /* multicast addr - 224.0.0.81 */
  60 #define IP_VS_SYNC_PORT  8848          /* multicast port */
  61
  62 #define SYNC_PROTO_VER  1               /* Protocol version in header */
  63
  64 static struct lock_class_key __ipvs_sync_key;
  65 /*
  66  *      IPVS sync connection entry
  67  *      Version 0, i.e. original version.
  68  */
  69 struct ip_vs_sync_conn_v0 {
  70         __u8                    reserved;
  71
  72         /* Protocol, addresses and port numbers */
  73         __u8                    protocol;       /* Which protocol (TCP/UDP) */
  74         __be16                  cport;
  75         __be16                  vport;
  76         __be16                  dport;
  77         __be32                  caddr;          /* client address */
  78         __be32                  vaddr;          /* virtual address */
  79         __be32                  daddr;          /* destination address */
  80
  81         /* Flags and state transition */
  82         __be16                  flags;          /* status flags */
  83         __be16                  state;          /* state info */
  84
  85         /* The sequence options start here */
  86 };
  87
  88 struct ip_vs_sync_conn_options {
  89         struct ip_vs_seq        in_seq;         /* incoming seq. struct */
  90         struct ip_vs_seq        out_seq;        /* outgoing seq. struct */
  91 };
  92
  93 /*
  94      Sync Connection format (sync_conn)
  95
  96        0                   1                   2                   3
  97        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
  98       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  99       |    Type       |    Protocol   | Ver.  |        Size           |
 100       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 101       |                             Flags                             |
 102       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 103       |            State              |         cport                 |
 104       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 105       |            vport              |         dport                 |
 106       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 107       |                             fwmark                            |
 108       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 109       |                             timeout  (in sec.)                |
 110       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 111       |                              ...                              |
 112       |                        IP-Addresses  (v4 or v6)               |
 113       |                              ...                              |
 114       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 115   Optional Parameters.
 116       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 117       | Param. Type    | Param. Length |   Param. data                |
 118       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+                               |
 119       |                              ...                              |
 120       |                               +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 121       |                               | Param Type    | Param. Length |
 122       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 123       |                           Param  data                         |
 124       |         Last Param data should be padded for 32 bit alignment |
 125       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 126 */
 127
 128 /*
 129  *  Type 0, IPv4 sync connection format
 130  */
 131 struct ip_vs_sync_v4 {
 132         __u8                    type;
 133         __u8                    protocol;       /* Which protocol (TCP/UDP) */
 134         __be16                  ver_size;       /* Version msb 4 bits */
 135         /* Flags and state transition */
 136         __be32                  flags;          /* status flags */
 137         __be16                  state;          /* state info   */
 138         /* Protocol, addresses and port numbers */
 139         __be16                  cport;
 140         __be16                  vport;
 141         __be16                  dport;
 142         __be32                  fwmark;         /* Firewall mark from skb */
 143         __be32                  timeout;        /* cp timeout */
 144         __be32                  caddr;          /* client address */
 145         __be32                  vaddr;          /* virtual address */
 146         __be32                  daddr;          /* destination address */
 147         /* The sequence options start here */
 148         /* PE data padded to 32bit alignment after seq. options */
 149 };
 150 /*
 151  * Type 2 messages IPv6
 152  */
 153 struct ip_vs_sync_v6 {
 154         __u8                    type;
 155         __u8                    protocol;       /* Which protocol (TCP/UDP) */
 156         __be16                  ver_size;       /* Version msb 4 bits */
 157         /* Flags and state transition */
 158         __be32                  flags;          /* status flags */
 159         __be16                  state;          /* state info   */
 160         /* Protocol, addresses and port numbers */
 161         __be16                  cport;
 162         __be16                  vport;
 163         __be16                  dport;
 164         __be32                  fwmark;         /* Firewall mark from skb */
 165         __be32                  timeout;        /* cp timeout */
 166         struct in6_addr         caddr;          /* client address */
 167         struct in6_addr         vaddr;          /* virtual address */
 168         struct in6_addr         daddr;          /* destination address */
 169         /* The sequence options start here */
 170         /* PE data padded to 32bit alignment after seq. options */
 171 };
 172
 173 union ip_vs_sync_conn {
 174         struct ip_vs_sync_v4    v4;
 175         struct ip_vs_sync_v6    v6;
 176 };
 177
 178 /* Bits in Type field in above */
 179 #define STYPE_INET6             0
 180 #define STYPE_F_INET6           (1 << STYPE_INET6)
 181
 182 #define SVER_SHIFT              12              /* Shift to get version */
 183 #define SVER_MASK               0x0fff          /* Mask to strip version */
 184
 185 #define IPVS_OPT_SEQ_DATA       1
 186 #define IPVS_OPT_PE_DATA        2
 187 #define IPVS_OPT_PE_NAME        3
 188 #define IPVS_OPT_PARAM          7
 189
 190 #define IPVS_OPT_F_SEQ_DATA     (1 << (IPVS_OPT_SEQ_DATA-1))
 191 #define IPVS_OPT_F_PE_DATA      (1 << (IPVS_OPT_PE_DATA-1))
 192 #define IPVS_OPT_F_PE_NAME      (1 << (IPVS_OPT_PE_NAME-1))
 193 #define IPVS_OPT_F_PARAM        (1 << (IPVS_OPT_PARAM-1))
 194
 195 struct ip_vs_sync_thread_data {
 196         struct net *net;
 197         struct socket *sock;
 198         char *buf;
 199         int id;
 200 };
 201
 202 /* Version 0 definition of packet sizes */
 203 #define SIMPLE_CONN_SIZE  (sizeof(struct ip_vs_sync_conn_v0))
 204 #define FULL_CONN_SIZE  \
 205 (sizeof(struct ip_vs_sync_conn_v0) + sizeof(struct ip_vs_sync_conn_options))
 206
 207
 208 /*
 209   The master mulitcasts messages (Datagrams) to the backup load balancers
 210   in the following format.
 211
 212  Version 1:
 213   Note, first byte should be Zero, so ver 0 receivers will drop the packet.
 214
 215        0                   1                   2                   3
 216        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
 217       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 218       |      0        |    SyncID     |            Size               |
 219       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 220       |  Count Conns  |    Version    |    Reserved, set to Zero      |
 221       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 222       |                                                               |
 223       |                    IPVS Sync Connection (1)                   |
 224       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 225       |                            .                                  |
 226       ~                            .                                  ~
 227       |                            .                                  |
 228       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 229       |                                                               |
 230       |                    IPVS Sync Connection (n)                   |
 231       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 232
 233  Version 0 Header
 234        0                   1                   2                   3
 235        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
 236       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 237       |  Count Conns  |    SyncID     |            Size               |
 238       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 239       |                    IPVS Sync Connection (1)                   |
 240 */
 241
 242 #define SYNC_MESG_HEADER_LEN    4
 243 #define MAX_CONNS_PER_SYNCBUFF  255 /* nr_conns in ip_vs_sync_mesg is 8 bit */
 244
 245 /* Version 0 header */
 246 struct ip_vs_sync_mesg_v0 {
 247         __u8                    nr_conns;
 248         __u8                    syncid;
 249         __be16                  size;
 250
 251         /* ip_vs_sync_conn entries start here */
 252 };
 253
 254 /* Version 1 header */
 255 struct ip_vs_sync_mesg {
 256         __u8                    reserved;       /* must be zero */
 257         __u8                    syncid;
 258         __be16                  size;
 259         __u8                    nr_conns;
 260         __s8                    version;        /* SYNC_PROTO_VER  */
 261         __u16                   spare;
 262         /* ip_vs_sync_conn entries start here */
 263 };
 264
 265 struct ip_vs_sync_buff {
 266         struct list_head        list;
 267         unsigned long           firstuse;
 268
 269         /* pointers for the message data */
 270         struct ip_vs_sync_mesg  *mesg;
 271         unsigned char           *head;
 272         unsigned char           *end;
 273 };
 274
 275 /*
 276  * Copy of struct ip_vs_seq
 277  * From unaligned network order to aligned host order
 278  */
 279 static void ntoh_seq(struct ip_vs_seq *no, struct ip_vs_seq *ho)
 280 {
 281         ho->init_seq       = get_unaligned_be32(&no->init_seq);
 282         ho->delta          = get_unaligned_be32(&no->delta);
 283         ho->previous_delta = get_unaligned_be32(&no->previous_delta);
 284 }
 285
 286 /*
 287  * Copy of struct ip_vs_seq
 288  * From Aligned host order to unaligned network order
 289  */
 290 static void hton_seq(struct ip_vs_seq *ho, struct ip_vs_seq *no)
 291 {
 292         put_unaligned_be32(ho->init_seq, &no->init_seq);
 293         put_unaligned_be32(ho->delta, &no->delta);
 294         put_unaligned_be32(ho->previous_delta, &no->previous_delta);
 295 }
 296
 297 static inline struct ip_vs_sync_buff *
 298 sb_dequeue(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
 299 {
 300         struct ip_vs_sync_buff *sb;
 301
 302         spin_lock_bh(&ipvs->sync_lock);
 303         if (list_empty(&ms->sync_queue)) {
 304                 sb = NULL;
 305                 __set_current_state(TASK_INTERRUPTIBLE);
 306         } else {
 307                 sb = list_entry(ms->sync_queue.next, struct ip_vs_sync_buff,
 308                                 list);
 309                 list_del(&sb->list);
 310                 ms->sync_queue_len--;
 311                 if (!ms->sync_queue_len)
 312                         ms->sync_queue_delay = 0;
 313         }
 314         spin_unlock_bh(&ipvs->sync_lock);
 315
 316         return sb;
 317 }
 318
 319 /*
 320  * Create a new sync buffer for Version 1 proto.
 321  */
 322 static inline struct ip_vs_sync_buff *
 323 ip_vs_sync_buff_create(struct netns_ipvs *ipvs)
 324 {
 325         struct ip_vs_sync_buff *sb;
 326
 327         if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
 328                 return NULL;
 329
 330         sb->mesg = kmalloc(ipvs->send_mesg_maxlen, GFP_ATOMIC);
 331         if (!sb->mesg) {
 332                 kfree(sb);
 333                 return NULL;
 334         }
 335         sb->mesg->reserved = 0;  /* old nr_conns i.e. must be zero now */
 336         sb->mesg->version = SYNC_PROTO_VER;
 337         sb->mesg->syncid = ipvs->master_syncid;
 338         sb->mesg->size = htons(sizeof(struct ip_vs_sync_mesg));
 339         sb->mesg->nr_conns = 0;
 340         sb->mesg->spare = 0;
 341         sb->head = (unsigned char *)sb->mesg + sizeof(struct ip_vs_sync_mesg);
 342         sb->end = (unsigned char *)sb->mesg + ipvs->send_mesg_maxlen;
 343
 344         sb->firstuse = jiffies;
 345         return sb;
 346 }
 347
 348 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
 349 {
 350         kfree(sb->mesg);
 351         kfree(sb);
 352 }
 353
 354 static inline void sb_queue_tail(struct netns_ipvs *ipvs,
 355                                  struct ipvs_master_sync_state *ms)
 356 {
 357         struct ip_vs_sync_buff *sb = ms->sync_buff;
 358
 359         spin_lock(&ipvs->sync_lock);
 360         if (ipvs->sync_state & IP_VS_STATE_MASTER &&
 361             ms->sync_queue_len < sysctl_sync_qlen_max(ipvs)) {
 362                 if (!ms->sync_queue_len)
 363                         schedule_delayed_work(&ms->master_wakeup_work,
 364                                               max(IPVS_SYNC_SEND_DELAY, 1));
 365                 ms->sync_queue_len++;
 366                 list_add_tail(&sb->list, &ms->sync_queue);
 367                 if ((++ms->sync_queue_delay) == IPVS_SYNC_WAKEUP_RATE)
 368                         wake_up_process(ms->master_thread);
 369         } else
 370                 ip_vs_sync_buff_release(sb);
 371         spin_unlock(&ipvs->sync_lock);
 372 }
 373
 374 /*
 375  *      Get the current sync buffer if it has been created for more
 376  *      than the specified time or the specified time is zero.
 377  */
 378 static inline struct ip_vs_sync_buff *
 379 get_curr_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms,
 380                    unsigned long time)
 381 {
 382         struct ip_vs_sync_buff *sb;
 383
 384         spin_lock_bh(&ipvs->sync_buff_lock);
 385         sb = ms->sync_buff;
 386         if (sb && time_after_eq(jiffies - sb->firstuse, time)) {
 387                 ms->sync_buff = NULL;
 388                 __set_current_state(TASK_RUNNING);
 389         } else
 390                 sb = NULL;
 391         spin_unlock_bh(&ipvs->sync_buff_lock);
 392         return sb;
 393 }
 394
 395 static inline int
 396 select_master_thread_id(struct netns_ipvs *ipvs, struct ip_vs_conn *cp)
 397 {
 398         return ((long) cp >> (1 + ilog2(sizeof(*cp)))) & ipvs->threads_mask;
 399 }
 400
 401 /*
 402  * Create a new sync buffer for Version 0 proto.
 403  */
 404 static inline struct ip_vs_sync_buff *
 405 ip_vs_sync_buff_create_v0(struct netns_ipvs *ipvs)
 406 {
 407         struct ip_vs_sync_buff *sb;
 408         struct ip_vs_sync_mesg_v0 *mesg;
 409
 410         if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
 411                 return NULL;
 412
 413         sb->mesg = kmalloc(ipvs->send_mesg_maxlen, GFP_ATOMIC);
 414         if (!sb->mesg) {
 415                 kfree(sb);
 416                 return NULL;
 417         }
 418         mesg = (struct ip_vs_sync_mesg_v0 *)sb->mesg;
 419         mesg->nr_conns = 0;
 420         mesg->syncid = ipvs->master_syncid;
 421         mesg->size = htons(sizeof(struct ip_vs_sync_mesg_v0));
 422         sb->head = (unsigned char *)mesg + sizeof(struct ip_vs_sync_mesg_v0);
 423         sb->end = (unsigned char *)mesg + ipvs->send_mesg_maxlen;
 424         sb->firstuse = jiffies;
 425         return sb;
 426 }
 427
 428 /* Check if conn should be synced.
 429  * pkts: conn packets, use sysctl_sync_threshold to avoid packet check
 430  * - (1) sync_refresh_period: reduce sync rate. Additionally, retry
 431  *      sync_retries times with period of sync_refresh_period/8
 432  * - (2) if both sync_refresh_period and sync_period are 0 send sync only
 433  *      for state changes or only once when pkts matches sync_threshold
 434  * - (3) templates: rate can be reduced only with sync_refresh_period or
 435  *      with (2)
 436  */
 437 static int ip_vs_sync_conn_needed(struct netns_ipvs *ipvs,
 438                                   struct ip_vs_conn *cp, int pkts)
 439 {
 440         unsigned long orig = ACCESS_ONCE(cp->sync_endtime);
 441         unsigned long now = jiffies;
 442         unsigned long n = (now + cp->timeout) & ~3UL;
 443         unsigned int sync_refresh_period;
 444         int sync_period;
 445         int force;
 446
 447         /* Check if we sync in current state */
 448         if (unlikely(cp->flags & IP_VS_CONN_F_TEMPLATE))
 449                 force = 0;
 450         else if (likely(cp->protocol == IPPROTO_TCP)) {
 451                 if (!((1 << cp->state) &
 452                       ((1 << IP_VS_TCP_S_ESTABLISHED) |
 453                        (1 << IP_VS_TCP_S_FIN_WAIT) |
 454                        (1 << IP_VS_TCP_S_CLOSE) |
 455                        (1 << IP_VS_TCP_S_CLOSE_WAIT) |
 456                        (1 << IP_VS_TCP_S_TIME_WAIT))))
 457                         return 0;
 458                 force = cp->state != cp->old_state;
 459                 if (force && cp->state != IP_VS_TCP_S_ESTABLISHED)
 460                         goto set;
 461         } else if (unlikely(cp->protocol == IPPROTO_SCTP)) {
 462                 if (!((1 << cp->state) &
 463                       ((1 << IP_VS_SCTP_S_ESTABLISHED) |
 464                        (1 << IP_VS_SCTP_S_SHUTDOWN_SENT) |
 465                        (1 << IP_VS_SCTP_S_SHUTDOWN_RECEIVED) |
 466                        (1 << IP_VS_SCTP_S_SHUTDOWN_ACK_SENT) |
 467                        (1 << IP_VS_SCTP_S_CLOSED))))
 468                         return 0;
 469                 force = cp->state != cp->old_state;
 470                 if (force && cp->state != IP_VS_SCTP_S_ESTABLISHED)
 471                         goto set;
 472         } else {
 473                 /* UDP or another protocol with single state */
 474                 force = 0;
 475         }
 476
 477         sync_refresh_period = sysctl_sync_refresh_period(ipvs);
 478         if (sync_refresh_period > 0) {
 479                 long diff = n - orig;
 480                 long min_diff = max(cp->timeout >> 1, 10UL * HZ);
 481
 482                 /* Avoid sync if difference is below sync_refresh_period
 483                  * and below the half timeout.
 484                  */
 485                 if (abs(diff) < min_t(long, sync_refresh_period, min_diff)) {
 486                         int retries = orig & 3;
 487
 488                         if (retries >= sysctl_sync_retries(ipvs))
 489                                 return 0;
 490                         if (time_before(now, orig - cp->timeout +
 491                                         (sync_refresh_period >> 3)))
 492                                 return 0;
 493                         n |= retries + 1;
 494                 }
 495         }
 496         sync_period = sysctl_sync_period(ipvs);
 497         if (sync_period > 0) {
 498                 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE) &&
 499                     pkts % sync_period != sysctl_sync_threshold(ipvs))
 500                         return 0;
 501         } else if (sync_refresh_period <= 0 &&
 502                    pkts != sysctl_sync_threshold(ipvs))
 503                 return 0;
 504
 505 set:
 506         cp->old_state = cp->state;
 507         n = cmpxchg(&cp->sync_endtime, orig, n);
 508         return n == orig || force;
 509 }
 510
 511 /*
 512  *      Version 0 , could be switched in by sys_ctl.
 513  *      Add an ip_vs_conn information into the current sync_buff.
 514  */
 515 static void ip_vs_sync_conn_v0(struct net *net, struct ip_vs_conn *cp,
 516                                int pkts)
 517 {
 518         struct netns_ipvs *ipvs = net_ipvs(net);
 519         struct ip_vs_sync_mesg_v0 *m;
 520         struct ip_vs_sync_conn_v0 *s;
 521         struct ip_vs_sync_buff *buff;
 522         struct ipvs_master_sync_state *ms;
 523         int id;
 524         int len;
 525
 526         if (unlikely(cp->af != AF_INET))
 527                 return;
 528         /* Do not sync ONE PACKET */
 529         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
 530                 return;
 531
 532         if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
 533                 return;
 534
 535         spin_lock_bh(&ipvs->sync_buff_lock);
 536         if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
 537                 spin_unlock_bh(&ipvs->sync_buff_lock);
 538                 return;
 539         }
 540
 541         id = select_master_thread_id(ipvs, cp);
 542         ms = &ipvs->ms[id];
 543         buff = ms->sync_buff;
 544         if (buff) {
 545                 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
 546                 /* Send buffer if it is for v1 */
 547                 if (!m->nr_conns) {
 548                         sb_queue_tail(ipvs, ms);
 549                         ms->sync_buff = NULL;
 550                         buff = NULL;
 551                 }
 552         }
 553         if (!buff) {
 554                 buff = ip_vs_sync_buff_create_v0(ipvs);
 555                 if (!buff) {
 556                         spin_unlock_bh(&ipvs->sync_buff_lock);
 557                         pr_err("ip_vs_sync_buff_create failed.\n");
 558                         return;
 559                 }
 560                 ms->sync_buff = buff;
 561         }
 562
 563         len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
 564                 SIMPLE_CONN_SIZE;
 565         m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
 566         s = (struct ip_vs_sync_conn_v0 *) buff->head;
 567
 568         /* copy members */
 569         s->reserved = 0;
 570         s->protocol = cp->protocol;
 571         s->cport = cp->cport;
 572         s->vport = cp->vport;
 573         s->dport = cp->dport;
 574         s->caddr = cp->caddr.ip;
 575         s->vaddr = cp->vaddr.ip;
 576         s->daddr = cp->daddr.ip;
 577         s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
 578         s->state = htons(cp->state);
 579         if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
 580                 struct ip_vs_sync_conn_options *opt =
 581                         (struct ip_vs_sync_conn_options *)&s[1];
 582                 memcpy(opt, &cp->in_seq, sizeof(*opt));
 583         }
 584
 585         m->nr_conns++;
 586         m->size = htons(ntohs(m->size) + len);
 587         buff->head += len;
 588
 589         /* check if there is a space for next one */
 590         if (buff->head + FULL_CONN_SIZE > buff->end) {
 591                 sb_queue_tail(ipvs, ms);
 592                 ms->sync_buff = NULL;
 593         }
 594         spin_unlock_bh(&ipvs->sync_buff_lock);
 595
 596         /* synchronize its controller if it has */
 597         cp = cp->control;
 598         if (cp) {
 599                 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
 600                         pkts = atomic_add_return(1, &cp->in_pkts);
 601                 else
 602                         pkts = sysctl_sync_threshold(ipvs);
 603                 ip_vs_sync_conn(net, cp->control, pkts);
 604         }
 605 }
 606
 607 /*
 608  *      Add an ip_vs_conn information into the current sync_buff.
 609  *      Called by ip_vs_in.
 610  *      Sending Version 1 messages
 611  */
 612 void ip_vs_sync_conn(struct net *net, struct ip_vs_conn *cp, int pkts)
 613 {
 614         struct netns_ipvs *ipvs = net_ipvs(net);
 615         struct ip_vs_sync_mesg *m;
 616         union ip_vs_sync_conn *s;
 617         struct ip_vs_sync_buff *buff;
 618         struct ipvs_master_sync_state *ms;
 619         int id;
 620         __u8 *p;
 621         unsigned int len, pe_name_len, pad;
 622
 623         /* Handle old version of the protocol */
 624         if (sysctl_sync_ver(ipvs) == 0) {
 625                 ip_vs_sync_conn_v0(net, cp, pkts);
 626                 return;
 627         }
 628         /* Do not sync ONE PACKET */
 629         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
 630                 goto control;
 631 sloop:
 632         if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
 633                 goto control;
 634
 635         /* Sanity checks */
 636         pe_name_len = 0;
 637         if (cp->pe_data_len) {
 638                 if (!cp->pe_data || !cp->dest) {
 639                         IP_VS_ERR_RL("SYNC, connection pe_data invalid\n");
 640                         return;
 641                 }
 642                 pe_name_len = strnlen(cp->pe->name, IP_VS_PENAME_MAXLEN);
 643         }
 644
 645         spin_lock_bh(&ipvs->sync_buff_lock);
 646         if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
 647                 spin_unlock_bh(&ipvs->sync_buff_lock);
 648                 return;
 649         }
 650
 651         id = select_master_thread_id(ipvs, cp);
 652         ms = &ipvs->ms[id];
 653
 654 #ifdef CONFIG_IP_VS_IPV6
 655         if (cp->af == AF_INET6)
 656                 len = sizeof(struct ip_vs_sync_v6);
 657         else
 658 #endif
 659                 len = sizeof(struct ip_vs_sync_v4);
 660
 661         if (cp->flags & IP_VS_CONN_F_SEQ_MASK)
 662                 len += sizeof(struct ip_vs_sync_conn_options) + 2;
 663
 664         if (cp->pe_data_len)
 665                 len += cp->pe_data_len + 2;     /* + Param hdr field */
 666         if (pe_name_len)
 667                 len += pe_name_len + 2;
 668
 669         /* check if there is a space for this one  */
 670         pad = 0;
 671         buff = ms->sync_buff;
 672         if (buff) {
 673                 m = buff->mesg;
 674                 pad = (4 - (size_t) buff->head) & 3;
 675                 /* Send buffer if it is for v0 */
 676                 if (buff->head + len + pad > buff->end || m->reserved) {
 677                         sb_queue_tail(ipvs, ms);
 678                         ms->sync_buff = NULL;
 679                         buff = NULL;
 680                         pad = 0;
 681                 }
 682         }
 683
 684         if (!buff) {
 685                 buff = ip_vs_sync_buff_create(ipvs);
 686                 if (!buff) {
 687                         spin_unlock_bh(&ipvs->sync_buff_lock);
 688                         pr_err("ip_vs_sync_buff_create failed.\n");
 689                         return;
 690                 }
 691                 ms->sync_buff = buff;
 692                 m = buff->mesg;
 693         }
 694
 695         p = buff->head;
 696         buff->head += pad + len;
 697         m->size = htons(ntohs(m->size) + pad + len);
 698         /* Add ev. padding from prev. sync_conn */
 699         while (pad--)
 700                 *(p++) = 0;
 701
 702         s = (union ip_vs_sync_conn *)p;
 703
 704         /* Set message type  & copy members */
 705         s->v4.type = (cp->af == AF_INET6 ? STYPE_F_INET6 : 0);
 706         s->v4.ver_size = htons(len & SVER_MASK);        /* Version 0 */
 707         s->v4.flags = htonl(cp->flags & ~IP_VS_CONN_F_HASHED);
 708         s->v4.state = htons(cp->state);
 709         s->v4.protocol = cp->protocol;
 710         s->v4.cport = cp->cport;
 711         s->v4.vport = cp->vport;
 712         s->v4.dport = cp->dport;
 713         s->v4.fwmark = htonl(cp->fwmark);
 714         s->v4.timeout = htonl(cp->timeout / HZ);
 715         m->nr_conns++;
 716
 717 #ifdef CONFIG_IP_VS_IPV6
 718         if (cp->af == AF_INET6) {
 719                 p += sizeof(struct ip_vs_sync_v6);
 720                 s->v6.caddr = cp->caddr.in6;
 721                 s->v6.vaddr = cp->vaddr.in6;
 722                 s->v6.daddr = cp->daddr.in6;
 723         } else
 724 #endif
 725         {
 726                 p += sizeof(struct ip_vs_sync_v4);      /* options ptr */
 727                 s->v4.caddr = cp->caddr.ip;
 728                 s->v4.vaddr = cp->vaddr.ip;
 729                 s->v4.daddr = cp->daddr.ip;
 730         }
 731         if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
 732                 *(p++) = IPVS_OPT_SEQ_DATA;
 733                 *(p++) = sizeof(struct ip_vs_sync_conn_options);
 734                 hton_seq((struct ip_vs_seq *)p, &cp->in_seq);
 735                 p += sizeof(struct ip_vs_seq);
 736                 hton_seq((struct ip_vs_seq *)p, &cp->out_seq);
 737                 p += sizeof(struct ip_vs_seq);
 738         }
 739         /* Handle pe data */
 740         if (cp->pe_data_len && cp->pe_data) {
 741                 *(p++) = IPVS_OPT_PE_DATA;
 742                 *(p++) = cp->pe_data_len;
 743                 memcpy(p, cp->pe_data, cp->pe_data_len);
 744                 p += cp->pe_data_len;
 745                 if (pe_name_len) {
 746                         /* Add PE_NAME */
 747                         *(p++) = IPVS_OPT_PE_NAME;
 748                         *(p++) = pe_name_len;
 749                         memcpy(p, cp->pe->name, pe_name_len);
 750                         p += pe_name_len;
 751                 }
 752         }
 753
 754         spin_unlock_bh(&ipvs->sync_buff_lock);
 755
 756 control:
 757         /* synchronize its controller if it has */
 758         cp = cp->control;
 759         if (!cp)
 760                 return;
 761         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
 762                 pkts = atomic_add_return(1, &cp->in_pkts);
 763         else
 764                 pkts = sysctl_sync_threshold(ipvs);
 765         goto sloop;
 766 }
 767
 768 /*
 769  *  fill_param used by version 1
 770  */
 771 static inline int
 772 ip_vs_conn_fill_param_sync(struct net *net, int af, union ip_vs_sync_conn *sc,
 773                            struct ip_vs_conn_param *p,
 774                            __u8 *pe_data, unsigned int pe_data_len,
 775                            __u8 *pe_name, unsigned int pe_name_len)
 776 {
 777 #ifdef CONFIG_IP_VS_IPV6
 778         if (af == AF_INET6)
 779                 ip_vs_conn_fill_param(net, af, sc->v6.protocol,
 780                                       (const union nf_inet_addr *)&sc->v6.caddr,
 781                                       sc->v6.cport,
 782                                       (const union nf_inet_addr *)&sc->v6.vaddr,
 783                                       sc->v6.vport, p);
 784         else
 785 #endif
 786                 ip_vs_conn_fill_param(net, af, sc->v4.protocol,
 787                                       (const union nf_inet_addr *)&sc->v4.caddr,
 788                                       sc->v4.cport,
 789                                       (const union nf_inet_addr *)&sc->v4.vaddr,
 790                                       sc->v4.vport, p);
 791         /* Handle pe data */
 792         if (pe_data_len) {
 793                 if (pe_name_len) {
 794                         char buff[IP_VS_PENAME_MAXLEN+1];
 795
 796                         memcpy(buff, pe_name, pe_name_len);
 797                         buff[pe_name_len]=0;
 798                         p->pe = __ip_vs_pe_getbyname(buff);
 799                         if (!p->pe) {
 800                                 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n",
 801                                              buff);
 802                                 return 1;
 803                         }
 804                 } else {
 805                         IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n");
 806                         return 1;
 807                 }
 808
 809                 p->pe_data = kmemdup(pe_data, pe_data_len, GFP_ATOMIC);
 810                 if (!p->pe_data) {
 811                         if (p->pe->module)
 812                                 module_put(p->pe->module);
 813                         return -ENOMEM;
 814                 }
 815                 p->pe_data_len = pe_data_len;
 816         }
 817         return 0;
 818 }
 819
 820 /*
 821  *  Connection Add / Update.
 822  *  Common for version 0 and 1 reception of backup sync_conns.
 823  *  Param: ...
 824  *         timeout is in sec.
 825  */
 826 static void ip_vs_proc_conn(struct net *net, struct ip_vs_conn_param *param,
 827                             unsigned int flags, unsigned int state,
 828                             unsigned int protocol, unsigned int type,
 829                             const union nf_inet_addr *daddr, __be16 dport,
 830                             unsigned long timeout, __u32 fwmark,
 831                             struct ip_vs_sync_conn_options *opt)
 832 {
 833         struct ip_vs_dest *dest;
 834         struct ip_vs_conn *cp;
 835         struct netns_ipvs *ipvs = net_ipvs(net);
 836
 837         if (!(flags & IP_VS_CONN_F_TEMPLATE))
 838                 cp = ip_vs_conn_in_get(param);
 839         else
 840                 cp = ip_vs_ct_in_get(param);
 841
 842         if (cp) {
 843                 /* Free pe_data */
 844                 kfree(param->pe_data);
 845
 846                 dest = cp->dest;
 847                 spin_lock_bh(&cp->lock);
 848                 if ((cp->flags ^ flags) & IP_VS_CONN_F_INACTIVE &&
 849                     !(flags & IP_VS_CONN_F_TEMPLATE) && dest) {
 850                         if (flags & IP_VS_CONN_F_INACTIVE) {
 851                                 atomic_dec(&dest->activeconns);
 852                                 atomic_inc(&dest->inactconns);
 853                         } else {
 854                                 atomic_inc(&dest->activeconns);
 855                                 atomic_dec(&dest->inactconns);
 856                         }
 857                 }
 858                 flags &= IP_VS_CONN_F_BACKUP_UPD_MASK;
 859                 flags |= cp->flags & ~IP_VS_CONN_F_BACKUP_UPD_MASK;
 860                 cp->flags = flags;
 861                 spin_unlock_bh(&cp->lock);
 862                 if (!dest)
 863                         ip_vs_try_bind_dest(cp);
 864         } else {
 865                 /*
 866                  * Find the appropriate destination for the connection.
 867                  * If it is not found the connection will remain unbound
 868                  * but still handled.
 869                  */
 870                 rcu_read_lock();
 871                 dest = ip_vs_find_dest(net, type, daddr, dport, param->vaddr,
 872                                        param->vport, protocol, fwmark, flags);
 873
 874                 cp = ip_vs_conn_new(param, daddr, dport, flags, dest, fwmark);
 875                 rcu_read_unlock();
 876                 if (!cp) {
 877                         if (param->pe_data)
 878                                 kfree(param->pe_data);
 879                         IP_VS_DBG(2, "BACKUP, add new conn. failed\n");
 880                         return;
 881                 }
 882         }
 883
 884         if (opt)
 885                 memcpy(&cp->in_seq, opt, sizeof(*opt));
 886         atomic_set(&cp->in_pkts, sysctl_sync_threshold(ipvs));
 887         cp->state = state;
 888         cp->old_state = cp->state;
 889         /*
 890          * For Ver 0 messages style
 891          *  - Not possible to recover the right timeout for templates
 892          *  - can not find the right fwmark
 893          *    virtual service. If needed, we can do it for
 894          *    non-fwmark persistent services.
 895          * Ver 1 messages style.
 896          *  - No problem.
 897          */
 898         if (timeout) {
 899                 if (timeout > MAX_SCHEDULE_TIMEOUT / HZ)
 900                         timeout = MAX_SCHEDULE_TIMEOUT / HZ;
 901                 cp->timeout = timeout*HZ;
 902         } else {
 903                 struct ip_vs_proto_data *pd;
 904
 905                 pd = ip_vs_proto_data_get(net, protocol);
 906                 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pd && pd->timeout_table)
 907                         cp->timeout = pd->timeout_table[state];
 908                 else
 909                         cp->timeout = (3*60*HZ);
 910         }
 911         ip_vs_conn_put(cp);
 912 }
 913
 914 /*
 915  *  Process received multicast message for Version 0
 916  */
 917 static void ip_vs_process_message_v0(struct net *net, const char *buffer,
 918                                      const size_t buflen)
 919 {
 920         struct ip_vs_sync_mesg_v0 *m = (struct ip_vs_sync_mesg_v0 *)buffer;
 921         struct ip_vs_sync_conn_v0 *s;
 922         struct ip_vs_sync_conn_options *opt;
 923         struct ip_vs_protocol *pp;
 924         struct ip_vs_conn_param param;
 925         char *p;
 926         int i;
 927
 928         p = (char *)buffer + sizeof(struct ip_vs_sync_mesg_v0);
 929         for (i=0; i<m->nr_conns; i++) {
 930                 unsigned int flags, state;
 931
 932                 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
 933                         IP_VS_ERR_RL("BACKUP v0, bogus conn\n");
 934                         return;
 935                 }
 936                 s = (struct ip_vs_sync_conn_v0 *) p;
 937                 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
 938                 flags &= ~IP_VS_CONN_F_HASHED;
 939                 if (flags & IP_VS_CONN_F_SEQ_MASK) {
 940                         opt = (struct ip_vs_sync_conn_options *)&s[1];
 941                         p += FULL_CONN_SIZE;
 942                         if (p > buffer+buflen) {
 943                                 IP_VS_ERR_RL("BACKUP v0, Dropping buffer bogus conn options\n");
 944                                 return;
 945                         }
 946                 } else {
 947                         opt = NULL;
 948                         p += SIMPLE_CONN_SIZE;
 949                 }
 950
 951                 state = ntohs(s->state);
 952                 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
 953                         pp = ip_vs_proto_get(s->protocol);
 954                         if (!pp) {
 955                                 IP_VS_DBG(2, "BACKUP v0, Unsupported protocol %u\n",
 956                                         s->protocol);
 957                                 continue;
 958                         }
 959                         if (state >= pp->num_states) {
 960                                 IP_VS_DBG(2, "BACKUP v0, Invalid %s state %u\n",
 961                                         pp->name, state);
 962                                 continue;
 963                         }
 964                 } else {
 965                         /* protocol in templates is not used for state/timeout */
 966                         if (state > 0) {
 967                                 IP_VS_DBG(2, "BACKUP v0, Invalid template state %u\n",
 968                                         state);
 969                                 state = 0;
 970                         }
 971                 }
 972
 973                 ip_vs_conn_fill_param(net, AF_INET, s->protocol,
 974                                       (const union nf_inet_addr *)&s->caddr,
 975                                       s->cport,
 976                                       (const union nf_inet_addr *)&s->vaddr,
 977                                       s->vport, &param);
 978
 979                 /* Send timeout as Zero */
 980                 ip_vs_proc_conn(net, &param, flags, state, s->protocol, AF_INET,
 981                                 (union nf_inet_addr *)&s->daddr, s->dport,
 982                                 0, 0, opt);
 983         }
 984 }
 985
 986 /*
 987  * Handle options
 988  */
 989 static inline int ip_vs_proc_seqopt(__u8 *p, unsigned int plen,
 990                                     __u32 *opt_flags,
 991                                     struct ip_vs_sync_conn_options *opt)
 992 {
 993         struct ip_vs_sync_conn_options *topt;
 994
 995         topt = (struct ip_vs_sync_conn_options *)p;
 996
 997         if (plen != sizeof(struct ip_vs_sync_conn_options)) {
 998                 IP_VS_DBG(2, "BACKUP, bogus conn options length\n");
 999                 return -EINVAL;
1000         }
1001         if (*opt_flags & IPVS_OPT_F_SEQ_DATA) {
1002                 IP_VS_DBG(2, "BACKUP, conn options found twice\n");
1003                 return -EINVAL;
1004         }
1005         ntoh_seq(&topt->in_seq, &opt->in_seq);
1006         ntoh_seq(&topt->out_seq, &opt->out_seq);
1007         *opt_flags |= IPVS_OPT_F_SEQ_DATA;
1008         return 0;
1009 }
1010
1011 static int ip_vs_proc_str(__u8 *p, unsigned int plen, unsigned int *data_len,
1012                           __u8 **data, unsigned int maxlen,
1013                           __u32 *opt_flags, __u32 flag)
1014 {
1015         if (plen > maxlen) {
1016                 IP_VS_DBG(2, "BACKUP, bogus par.data len > %d\n", maxlen);
1017                 return -EINVAL;
1018         }
1019         if (*opt_flags & flag) {
1020                 IP_VS_DBG(2, "BACKUP, Par.data found twice 0x%x\n", flag);
1021                 return -EINVAL;
1022         }
1023         *data_len = plen;
1024         *data = p;
1025         *opt_flags |= flag;
1026         return 0;
1027 }
1028 /*
1029  *   Process a Version 1 sync. connection
1030  */
1031 static inline int ip_vs_proc_sync_conn(struct net *net, __u8 *p, __u8 *msg_end)
1032 {
1033         struct ip_vs_sync_conn_options opt;
1034         union  ip_vs_sync_conn *s;
1035         struct ip_vs_protocol *pp;
1036         struct ip_vs_conn_param param;
1037         __u32 flags;
1038         unsigned int af, state, pe_data_len=0, pe_name_len=0;
1039         __u8 *pe_data=NULL, *pe_name=NULL;
1040         __u32 opt_flags=0;
1041         int retc=0;
1042
1043         s = (union ip_vs_sync_conn *) p;
1044
1045         if (s->v6.type & STYPE_F_INET6) {
1046 #ifdef CONFIG_IP_VS_IPV6
1047                 af = AF_INET6;
1048                 p += sizeof(struct ip_vs_sync_v6);
1049 #else
1050                 IP_VS_DBG(3,"BACKUP, IPv6 msg received, and IPVS is not compiled for IPv6\n");
1051                 retc = 10;
1052                 goto out;
1053 #endif
1054         } else if (!s->v4.type) {
1055                 af = AF_INET;
1056                 p += sizeof(struct ip_vs_sync_v4);
1057         } else {
1058                 return -10;
1059         }
1060         if (p > msg_end)
1061                 return -20;
1062
1063         /* Process optional params check Type & Len. */
1064         while (p < msg_end) {
1065                 int ptype;
1066                 int plen;
1067
1068                 if (p+2 > msg_end)
1069                         return -30;
1070                 ptype = *(p++);
1071                 plen  = *(p++);
1072
1073                 if (!plen || ((p + plen) > msg_end))
1074                         return -40;
1075                 /* Handle seq option  p = param data */
1076                 switch (ptype & ~IPVS_OPT_F_PARAM) {
1077                 case IPVS_OPT_SEQ_DATA:
1078                         if (ip_vs_proc_seqopt(p, plen, &opt_flags, &opt))
1079                                 return -50;
1080                         break;
1081
1082                 case IPVS_OPT_PE_DATA:
1083                         if (ip_vs_proc_str(p, plen, &pe_data_len, &pe_data,
1084                                            IP_VS_PEDATA_MAXLEN, &opt_flags,
1085                                            IPVS_OPT_F_PE_DATA))
1086                                 return -60;
1087                         break;
1088
1089                 case IPVS_OPT_PE_NAME:
1090                         if (ip_vs_proc_str(p, plen,&pe_name_len, &pe_name,
1091                                            IP_VS_PENAME_MAXLEN, &opt_flags,
1092                                            IPVS_OPT_F_PE_NAME))
1093                                 return -70;
1094                         break;
1095
1096                 default:
1097                         /* Param data mandatory ? */
1098                         if (!(ptype & IPVS_OPT_F_PARAM)) {
1099                                 IP_VS_DBG(3, "BACKUP, Unknown mandatory param %d found\n",
1100                                           ptype & ~IPVS_OPT_F_PARAM);
1101                                 retc = 20;
1102                                 goto out;
1103                         }
1104                 }
1105                 p += plen;  /* Next option */
1106         }
1107
1108         /* Get flags and Mask off unsupported */
1109         flags  = ntohl(s->v4.flags) & IP_VS_CONN_F_BACKUP_MASK;
1110         flags |= IP_VS_CONN_F_SYNC;
1111         state = ntohs(s->v4.state);
1112
1113         if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
1114                 pp = ip_vs_proto_get(s->v4.protocol);
1115                 if (!pp) {
1116                         IP_VS_DBG(3,"BACKUP, Unsupported protocol %u\n",
1117                                 s->v4.protocol);
1118                         retc = 30;
1119                         goto out;
1120                 }
1121                 if (state >= pp->num_states) {
1122                         IP_VS_DBG(3, "BACKUP, Invalid %s state %u\n",
1123                                 pp->name, state);
1124                         retc = 40;
1125                         goto out;
1126                 }
1127         } else {
1128                 /* protocol in templates is not used for state/timeout */
1129                 if (state > 0) {
1130                         IP_VS_DBG(3, "BACKUP, Invalid template state %u\n",
1131                                 state);
1132                         state = 0;
1133                 }
1134         }
1135         if (ip_vs_conn_fill_param_sync(net, af, s, &param, pe_data,
1136                                        pe_data_len, pe_name, pe_name_len)) {
1137                 retc = 50;
1138                 goto out;
1139         }
1140         /* If only IPv4, just silent skip IPv6 */
1141         if (af == AF_INET)
1142                 ip_vs_proc_conn(net, &param, flags, state, s->v4.protocol, af,
1143                                 (union nf_inet_addr *)&s->v4.daddr, s->v4.dport,
1144                                 ntohl(s->v4.timeout), ntohl(s->v4.fwmark),
1145                                 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1146                                 );
1147 #ifdef CONFIG_IP_VS_IPV6
1148         else
1149                 ip_vs_proc_conn(net, &param, flags, state, s->v6.protocol, af,
1150                                 (union nf_inet_addr *)&s->v6.daddr, s->v6.dport,
1151                                 ntohl(s->v6.timeout), ntohl(s->v6.fwmark),
1152                                 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1153                                 );
1154 #endif
1155         return 0;
1156         /* Error exit */
1157 out:
1158         IP_VS_DBG(2, "BACKUP, Single msg dropped err:%d\n", retc);
1159         return retc;
1160
1161 }
1162 /*
1163  *      Process received multicast message and create the corresponding
1164  *      ip_vs_conn entries.
1165  *      Handles Version 0 & 1
1166  */
1167 static void ip_vs_process_message(struct net *net, __u8 *buffer,
1168                                   const size_t buflen)
1169 {
1170         struct netns_ipvs *ipvs = net_ipvs(net);
1171         struct ip_vs_sync_mesg *m2 = (struct ip_vs_sync_mesg *)buffer;
1172         __u8 *p, *msg_end;
1173         int i, nr_conns;
1174
1175         if (buflen < sizeof(struct ip_vs_sync_mesg_v0)) {
1176                 IP_VS_DBG(2, "BACKUP, message header too short\n");
1177                 return;
1178         }
1179
1180         if (buflen != ntohs(m2->size)) {
1181                 IP_VS_DBG(2, "BACKUP, bogus message size\n");
1182                 return;
1183         }
1184         /* SyncID sanity check */
1185         if (ipvs->backup_syncid != 0 && m2->syncid != ipvs->backup_syncid) {
1186                 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2->syncid);
1187                 return;
1188         }
1189         /* Handle version 1  message */
1190         if ((m2->version == SYNC_PROTO_VER) && (m2->reserved == 0)
1191             && (m2->spare == 0)) {
1192
1193                 msg_end = buffer + sizeof(struct ip_vs_sync_mesg);
1194                 nr_conns = m2->nr_conns;
1195
1196                 for (i=0; i<nr_conns; i++) {
1197                         union ip_vs_sync_conn *s;
1198                         unsigned int size;
1199                         int retc;
1200
1201                         p = msg_end;
1202                         if (p + sizeof(s->v4) > buffer+buflen) {
1203                                 IP_VS_ERR_RL("BACKUP, Dropping buffer, to small\n");
1204                                 return;
1205                         }
1206                         s = (union ip_vs_sync_conn *)p;
1207                         size = ntohs(s->v4.ver_size) & SVER_MASK;
1208                         msg_end = p + size;
1209                         /* Basic sanity checks */
1210                         if (msg_end  > buffer+buflen) {
1211                                 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n");
1212                                 return;
1213                         }
1214                         if (ntohs(s->v4.ver_size) >> SVER_SHIFT) {
1215                                 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n",
1216                                               ntohs(s->v4.ver_size) >> SVER_SHIFT);
1217                                 return;
1218                         }
1219                         /* Process a single sync_conn */
1220                         retc = ip_vs_proc_sync_conn(net, p, msg_end);
1221                         if (retc < 0) {
1222                                 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n",
1223                                              retc);
1224                                 return;
1225                         }
1226                         /* Make sure we have 32 bit alignment */
1227                         msg_end = p + ((size + 3) & ~3);
1228                 }
1229         } else {
1230                 /* Old type of message */
1231                 ip_vs_process_message_v0(net, buffer, buflen);
1232                 return;
1233         }
1234 }
1235
1236
1237 /*
1238  *      Setup sndbuf (mode=1) or rcvbuf (mode=0)
1239  */
1240 static void set_sock_size(struct sock *sk, int mode, int val)
1241 {
1242         /* setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &val, sizeof(val)); */
1243         /* setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)); */
1244         lock_sock(sk);
1245         if (mode) {
1246                 val = clamp_t(int, val, (SOCK_MIN_SNDBUF + 1) / 2,
1247                               sysctl_wmem_max);
1248                 sk->sk_sndbuf = val * 2;
1249                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1250         } else {
1251                 val = clamp_t(int, val, (SOCK_MIN_RCVBUF + 1) / 2,
1252                               sysctl_rmem_max);
1253                 sk->sk_rcvbuf = val * 2;
1254                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1255         }
1256         release_sock(sk);
1257 }
1258
1259 /*
1260  *      Setup loopback of outgoing multicasts on a sending socket
1261  */
1262 static void set_mcast_loop(struct sock *sk, u_char loop)
1263 {
1264         struct inet_sock *inet = inet_sk(sk);
1265
1266         /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
1267         lock_sock(sk);
1268         inet->mc_loop = loop ? 1 : 0;
1269         release_sock(sk);
1270 }
1271
1272 /*
1273  *      Specify TTL for outgoing multicasts on a sending socket
1274  */
1275 static void set_mcast_ttl(struct sock *sk, u_char ttl)
1276 {
1277         struct inet_sock *inet = inet_sk(sk);
1278
1279         /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
1280         lock_sock(sk);
1281         inet->mc_ttl = ttl;
1282         release_sock(sk);
1283 }
1284
1285 /*
1286  *      Specifiy default interface for outgoing multicasts
1287  */
1288 static int set_mcast_if(struct sock *sk, char *ifname)
1289 {
1290         struct net_device *dev;
1291         struct inet_sock *inet = inet_sk(sk);
1292         struct net *net = sock_net(sk);
1293
1294         dev = __dev_get_by_name(net, ifname);
1295         if (!dev)
1296                 return -ENODEV;
1297
1298         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1299                 return -EINVAL;
1300
1301         lock_sock(sk);
1302         inet->mc_index = dev->ifindex;
1303         /*  inet->mc_addr  = 0; */
1304         release_sock(sk);
1305
1306         return 0;
1307 }
1308
1309
1310 /*
1311  *      Set the maximum length of sync message according to the
1312  *      specified interface's MTU.
1313  */
1314 static int set_sync_mesg_maxlen(struct net *net, int sync_state)
1315 {
1316         struct netns_ipvs *ipvs = net_ipvs(net);
1317         struct net_device *dev;
1318         int num;
1319
1320         if (sync_state == IP_VS_STATE_MASTER) {
1321                 dev = __dev_get_by_name(net, ipvs->master_mcast_ifn);
1322                 if (!dev)
1323                         return -ENODEV;
1324
1325                 num = (dev->mtu - sizeof(struct iphdr) -
1326                        sizeof(struct udphdr) -
1327                        SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE;
1328                 ipvs->send_mesg_maxlen = SYNC_MESG_HEADER_LEN +
1329                         SIMPLE_CONN_SIZE * min(num, MAX_CONNS_PER_SYNCBUFF);
1330                 IP_VS_DBG(7, "setting the maximum length of sync sending "
1331                           "message %d.\n", ipvs->send_mesg_maxlen);
1332         } else if (sync_state == IP_VS_STATE_BACKUP) {
1333                 dev = __dev_get_by_name(net, ipvs->backup_mcast_ifn);
1334                 if (!dev)
1335                         return -ENODEV;
1336
1337                 ipvs->recv_mesg_maxlen = dev->mtu -
1338                         sizeof(struct iphdr) - sizeof(struct udphdr);
1339                 IP_VS_DBG(7, "setting the maximum length of sync receiving "
1340                           "message %d.\n", ipvs->recv_mesg_maxlen);
1341         }
1342
1343         return 0;
1344 }
1345
1346
1347 /*
1348  *      Join a multicast group.
1349  *      the group is specified by a class D multicast address 224.0.0.0/8
1350  *      in the in_addr structure passed in as a parameter.
1351  */
1352 static int
1353 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
1354 {
1355         struct net *net = sock_net(sk);
1356         struct ip_mreqn mreq;
1357         struct net_device *dev;
1358         int ret;
1359
1360         memset(&mreq, 0, sizeof(mreq));
1361         memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
1362
1363         dev = __dev_get_by_name(net, ifname);
1364         if (!dev)
1365                 return -ENODEV;
1366         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1367                 return -EINVAL;
1368
1369         mreq.imr_ifindex = dev->ifindex;
1370
1371         lock_sock(sk);
1372         ret = ip_mc_join_group(sk, &mreq);
1373         release_sock(sk);
1374
1375         return ret;
1376 }
1377
1378
1379 static int bind_mcastif_addr(struct socket *sock, char *ifname)
1380 {
1381         struct net *net = sock_net(sock->sk);
1382         struct net_device *dev;
1383         __be32 addr;
1384         struct sockaddr_in sin;
1385
1386         dev = __dev_get_by_name(net, ifname);
1387         if (!dev)
1388                 return -ENODEV;
1389
1390         addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
1391         if (!addr)
1392                 pr_err("You probably need to specify IP address on "
1393                        "multicast interface.\n");
1394
1395         IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
1396                   ifname, &addr);
1397
1398         /* Now bind the socket with the address of multicast interface */
1399         sin.sin_family       = AF_INET;
1400         sin.sin_addr.s_addr  = addr;
1401         sin.sin_port         = 0;
1402
1403         return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
1404 }
1405
1406 /*
1407  *      Set up sending multicast socket over UDP
1408  */
1409 static struct socket *make_send_sock(struct net *net, int id)
1410 {
1411         struct netns_ipvs *ipvs = net_ipvs(net);
1412         /* multicast addr */
1413         struct sockaddr_in mcast_addr = {
1414                 .sin_family             = AF_INET,
1415                 .sin_port               = cpu_to_be16(IP_VS_SYNC_PORT + id),
1416                 .sin_addr.s_addr        = cpu_to_be32(IP_VS_SYNC_GROUP),
1417         };
1418         struct socket *sock;
1419         int result;
1420
1421         /* First create a socket move it to right name space later */
1422         result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1423         if (result < 0) {
1424                 pr_err("Error during creation of socket; terminating\n");
1425                 return ERR_PTR(result);
1426         }
1427         /*
1428          * Kernel sockets that are a part of a namespace, should not
1429          * hold a reference to a namespace in order to allow to stop it.
1430          * After sk_change_net should be released using sk_release_kernel.
1431          */
1432         sk_change_net(sock->sk, net);
1433         result = set_mcast_if(sock->sk, ipvs->master_mcast_ifn);
1434         if (result < 0) {
1435                 pr_err("Error setting outbound mcast interface\n");
1436                 goto error;
1437         }
1438
1439         set_mcast_loop(sock->sk, 0);
1440         set_mcast_ttl(sock->sk, 1);
1441         result = sysctl_sync_sock_size(ipvs);
1442         if (result > 0)
1443                 set_sock_size(sock->sk, 1, result);
1444
1445         result = bind_mcastif_addr(sock, ipvs->master_mcast_ifn);
1446         if (result < 0) {
1447                 pr_err("Error binding address of the mcast interface\n");
1448                 goto error;
1449         }
1450
1451         result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr,
1452                         sizeof(struct sockaddr), 0);
1453         if (result < 0) {
1454                 pr_err("Error connecting to the multicast addr\n");
1455                 goto error;
1456         }
1457
1458         return sock;
1459
1460 error:
1461         sk_release_kernel(sock->sk);
1462         return ERR_PTR(result);
1463 }
1464
1465
1466 /*
1467  *      Set up receiving multicast socket over UDP
1468  */
1469 static struct socket *make_receive_sock(struct net *net, int id)
1470 {
1471         struct netns_ipvs *ipvs = net_ipvs(net);
1472         /* multicast addr */
1473         struct sockaddr_in mcast_addr = {
1474                 .sin_family             = AF_INET,
1475                 .sin_port               = cpu_to_be16(IP_VS_SYNC_PORT + id),
1476                 .sin_addr.s_addr        = cpu_to_be32(IP_VS_SYNC_GROUP),
1477         };
1478         struct socket *sock;
1479         int result;
1480
1481         /* First create a socket */
1482         result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1483         if (result < 0) {
1484                 pr_err("Error during creation of socket; terminating\n");
1485                 return ERR_PTR(result);
1486         }
1487         /*
1488          * Kernel sockets that are a part of a namespace, should not
1489          * hold a reference to a namespace in order to allow to stop it.
1490          * After sk_change_net should be released using sk_release_kernel.
1491          */
1492         sk_change_net(sock->sk, net);
1493         /* it is equivalent to the REUSEADDR option in user-space */
1494         sock->sk->sk_reuse = SK_CAN_REUSE;
1495         result = sysctl_sync_sock_size(ipvs);
1496         if (result > 0)
1497                 set_sock_size(sock->sk, 0, result);
1498
1499         result = sock->ops->bind(sock, (struct sockaddr *) &mcast_addr,
1500                         sizeof(struct sockaddr));
1501         if (result < 0) {
1502                 pr_err("Error binding to the multicast addr\n");
1503                 goto error;
1504         }
1505
1506         /* join the multicast group */
1507         result = join_mcast_group(sock->sk,
1508                         (struct in_addr *) &mcast_addr.sin_addr,
1509                         ipvs->backup_mcast_ifn);
1510         if (result < 0) {
1511                 pr_err("Error joining to the multicast group\n");
1512                 goto error;
1513         }
1514
1515         return sock;
1516
1517 error:
1518         sk_release_kernel(sock->sk);
1519         return ERR_PTR(result);
1520 }
1521
1522
1523 static int
1524 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
1525 {
1526         struct msghdr   msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
1527         struct kvec     iov;
1528         int             len;
1529
1530         EnterFunction(7);
1531         iov.iov_base     = (void *)buffer;
1532         iov.iov_len      = length;
1533
1534         len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
1535
1536         LeaveFunction(7);
1537         return len;
1538 }
1539
1540 static int
1541 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
1542 {
1543         int msize;
1544         int ret;
1545
1546         msize = ntohs(msg->size);
1547
1548         ret = ip_vs_send_async(sock, (char *)msg, msize);
1549         if (ret >= 0 || ret == -EAGAIN)
1550                 return ret;
1551         pr_err("ip_vs_send_async error %d\n", ret);
1552         return 0;
1553 }
1554
1555 static int
1556 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
1557 {
1558         struct msghdr           msg = {NULL,};
1559         struct kvec             iov;
1560         int                     len;
1561
1562         EnterFunction(7);
1563
1564         /* Receive a packet */
1565         iov.iov_base     = buffer;
1566         iov.iov_len      = (size_t)buflen;
1567
1568         len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, MSG_DONTWAIT);
1569
1570         if (len < 0)
1571                 return len;
1572
1573         LeaveFunction(7);
1574         return len;
1575 }
1576
1577 /* Wakeup the master thread for sending */
1578 static void master_wakeup_work_handler(struct work_struct *work)
1579 {
1580         struct ipvs_master_sync_state *ms =
1581                 container_of(work, struct ipvs_master_sync_state,
1582                              master_wakeup_work.work);
1583         struct netns_ipvs *ipvs = ms->ipvs;
1584
1585         spin_lock_bh(&ipvs->sync_lock);
1586         if (ms->sync_queue_len &&
1587             ms->sync_queue_delay < IPVS_SYNC_WAKEUP_RATE) {
1588                 ms->sync_queue_delay = IPVS_SYNC_WAKEUP_RATE;
1589                 wake_up_process(ms->master_thread);
1590         }
1591         spin_unlock_bh(&ipvs->sync_lock);
1592 }
1593
1594 /* Get next buffer to send */
1595 static inline struct ip_vs_sync_buff *
1596 next_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
1597 {
1598         struct ip_vs_sync_buff *sb;
1599
1600         sb = sb_dequeue(ipvs, ms);
1601         if (sb)
1602                 return sb;
1603         /* Do not delay entries in buffer for more than 2 seconds */
1604         return get_curr_sync_buff(ipvs, ms, IPVS_SYNC_FLUSH_TIME);
1605 }
1606
1607 static int sync_thread_master(void *data)
1608 {
1609         struct ip_vs_sync_thread_data *tinfo = data;
1610         struct netns_ipvs *ipvs = net_ipvs(tinfo->net);
1611         struct ipvs_master_sync_state *ms = &ipvs->ms[tinfo->id];
1612         struct sock *sk = tinfo->sock->sk;
1613         struct ip_vs_sync_buff *sb;
1614
1615         pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
1616                 "syncid = %d, id = %d\n",
1617                 ipvs->master_mcast_ifn, ipvs->master_syncid, tinfo->id);
1618
1619         for (;;) {
1620                 sb = next_sync_buff(ipvs, ms);
1621                 if (unlikely(kthread_should_stop()))
1622                         break;
1623                 if (!sb) {
1624                         schedule_timeout(IPVS_SYNC_CHECK_PERIOD);
1625                         continue;
1626                 }
1627                 while (ip_vs_send_sync_msg(tinfo->sock, sb->mesg) < 0) {
1628                         int ret = 0;
1629
1630                         __wait_event_interruptible(*sk_sleep(sk),
1631                                                    sock_writeable(sk) ||
1632                                                    kthread_should_stop(),
1633                                                    ret);
1634                         if (unlikely(kthread_should_stop()))
1635                                 goto done;
1636                 }
1637                 ip_vs_sync_buff_release(sb);
1638         }
1639
1640 done:
1641         __set_current_state(TASK_RUNNING);
1642         if (sb)
1643                 ip_vs_sync_buff_release(sb);
1644
1645         /* clean up the sync_buff queue */
1646         while ((sb = sb_dequeue(ipvs, ms)))
1647                 ip_vs_sync_buff_release(sb);
1648         __set_current_state(TASK_RUNNING);
1649
1650         /* clean up the current sync_buff */
1651         sb = get_curr_sync_buff(ipvs, ms, 0);
1652         if (sb)
1653                 ip_vs_sync_buff_release(sb);
1654
1655         /* release the sending multicast socket */
1656         sk_release_kernel(tinfo->sock->sk);
1657         kfree(tinfo);
1658
1659         return 0;
1660 }
1661
1662
1663 static int sync_thread_backup(void *data)
1664 {
1665         struct ip_vs_sync_thread_data *tinfo = data;
1666         struct netns_ipvs *ipvs = net_ipvs(tinfo->net);
1667         int len;
1668
1669         pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
1670                 "syncid = %d, id = %d\n",
1671                 ipvs->backup_mcast_ifn, ipvs->backup_syncid, tinfo->id);
1672
1673         while (!kthread_should_stop()) {
1674                 wait_event_interruptible(*sk_sleep(tinfo->sock->sk),
1675                          !skb_queue_empty(&tinfo->sock->sk->sk_receive_queue)
1676                          || kthread_should_stop());
1677
1678                 /* do we have data now? */
1679                 while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) {
1680                         len = ip_vs_receive(tinfo->sock, tinfo->buf,
1681                                         ipvs->recv_mesg_maxlen);
1682                         if (len <= 0) {
1683                                 if (len != -EAGAIN)
1684                                         pr_err("receiving message error\n");
1685                                 break;
1686                         }
1687
1688                         ip_vs_process_message(tinfo->net, tinfo->buf, len);
1689                 }
1690         }
1691
1692         /* release the sending multicast socket */
1693         sk_release_kernel(tinfo->sock->sk);
1694         kfree(tinfo->buf);
1695         kfree(tinfo);
1696
1697         return 0;
1698 }
1699
1700
1701 int start_sync_thread(struct net *net, int state, char *mcast_ifn, __u8 syncid)
1702 {
1703         struct ip_vs_sync_thread_data *tinfo;
1704         struct task_struct **array = NULL, *task;
1705         struct socket *sock;
1706         struct netns_ipvs *ipvs = net_ipvs(net);
1707         char *name;
1708         int (*threadfn)(void *data);
1709         int id, count;
1710         int result = -ENOMEM;
1711
1712         IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1713         IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n",
1714                   sizeof(struct ip_vs_sync_conn_v0));
1715
1716         if (!ipvs->sync_state) {
1717                 count = clamp(sysctl_sync_ports(ipvs), 1, IPVS_SYNC_PORTS_MAX);
1718                 ipvs->threads_mask = count - 1;
1719         } else
1720                 count = ipvs->threads_mask + 1;
1721
1722         if (state == IP_VS_STATE_MASTER) {
1723                 if (ipvs->ms)
1724                         return -EEXIST;
1725
1726                 strlcpy(ipvs->master_mcast_ifn, mcast_ifn,
1727                         sizeof(ipvs->master_mcast_ifn));
1728                 ipvs->master_syncid = syncid;
1729                 name = "ipvs-m:%d:%d";
1730                 threadfn = sync_thread_master;
1731         } else if (state == IP_VS_STATE_BACKUP) {
1732                 if (ipvs->backup_threads)
1733                         return -EEXIST;
1734
1735                 strlcpy(ipvs->backup_mcast_ifn, mcast_ifn,
1736                         sizeof(ipvs->backup_mcast_ifn));
1737                 ipvs->backup_syncid = syncid;
1738                 name = "ipvs-b:%d:%d";
1739                 threadfn = sync_thread_backup;
1740         } else {
1741                 return -EINVAL;
1742         }
1743
1744         if (state == IP_VS_STATE_MASTER) {
1745                 struct ipvs_master_sync_state *ms;
1746
1747                 ipvs->ms = kzalloc(count * sizeof(ipvs->ms[0]), GFP_KERNEL);
1748                 if (!ipvs->ms)
1749                         goto out;
1750                 ms = ipvs->ms;
1751                 for (id = 0; id < count; id++, ms++) {
1752                         INIT_LIST_HEAD(&ms->sync_queue);
1753                         ms->sync_queue_len = 0;
1754                         ms->sync_queue_delay = 0;
1755                         INIT_DELAYED_WORK(&ms->master_wakeup_work,
1756                                           master_wakeup_work_handler);
1757                         ms->ipvs = ipvs;
1758                 }
1759         } else {
1760                 array = kzalloc(count * sizeof(struct task_struct *),
1761                                 GFP_KERNEL);
1762                 if (!array)
1763                         goto out;
1764         }
1765         set_sync_mesg_maxlen(net, state);
1766
1767         tinfo = NULL;
1768         for (id = 0; id < count; id++) {
1769                 if (state == IP_VS_STATE_MASTER)
1770                         sock = make_send_sock(net, id);
1771                 else
1772                         sock = make_receive_sock(net, id);
1773                 if (IS_ERR(sock)) {
1774                         result = PTR_ERR(sock);
1775                         goto outtinfo;
1776                 }
1777                 tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL);
1778                 if (!tinfo)
1779                         goto outsocket;
1780                 tinfo->net = net;
1781                 tinfo->sock = sock;
1782                 if (state == IP_VS_STATE_BACKUP) {
1783                         tinfo->buf = kmalloc(ipvs->recv_mesg_maxlen,
1784                                              GFP_KERNEL);
1785                         if (!tinfo->buf)
1786                                 goto outtinfo;
1787                 } else {
1788                         tinfo->buf = NULL;
1789                 }
1790                 tinfo->id = id;
1791
1792                 task = kthread_run(threadfn, tinfo, name, ipvs->gen, id);
1793                 if (IS_ERR(task)) {
1794                         result = PTR_ERR(task);
1795                         goto outtinfo;
1796                 }
1797                 tinfo = NULL;
1798                 if (state == IP_VS_STATE_MASTER)
1799                         ipvs->ms[id].master_thread = task;
1800                 else
1801                         array[id] = task;
1802         }
1803
1804         /* mark as active */
1805
1806         if (state == IP_VS_STATE_BACKUP)
1807                 ipvs->backup_threads = array;
1808         spin_lock_bh(&ipvs->sync_buff_lock);
1809         ipvs->sync_state |= state;
1810         spin_unlock_bh(&ipvs->sync_buff_lock);
1811
1812         /* increase the module use count */
1813         ip_vs_use_count_inc();
1814
1815         return 0;
1816
1817 outsocket:
1818         sk_release_kernel(sock->sk);
1819
1820 outtinfo:
1821         if (tinfo) {
1822                 sk_release_kernel(tinfo->sock->sk);
1823                 kfree(tinfo->buf);
1824                 kfree(tinfo);
1825         }
1826         count = id;
1827         while (count-- > 0) {
1828                 if (state == IP_VS_STATE_MASTER)
1829                         kthread_stop(ipvs->ms[count].master_thread);
1830                 else
1831                         kthread_stop(array[count]);
1832         }
1833         kfree(array);
1834
1835 out:
1836         if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
1837                 kfree(ipvs->ms);
1838                 ipvs->ms = NULL;
1839         }
1840         return result;
1841 }
1842
1843
1844 int stop_sync_thread(struct net *net, int state)
1845 {
1846         struct netns_ipvs *ipvs = net_ipvs(net);
1847         struct task_struct **array;
1848         int id;
1849         int retc = -EINVAL;
1850
1851         IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1852
1853         if (state == IP_VS_STATE_MASTER) {
1854                 if (!ipvs->ms)
1855                         return -ESRCH;
1856
1857                 /*
1858                  * The lock synchronizes with sb_queue_tail(), so that we don't
1859                  * add sync buffers to the queue, when we are already in
1860                  * progress of stopping the master sync daemon.
1861                  */
1862
1863                 spin_lock_bh(&ipvs->sync_buff_lock);
1864                 spin_lock(&ipvs->sync_lock);
1865                 ipvs->sync_state &= ~IP_VS_STATE_MASTER;
1866                 spin_unlock(&ipvs->sync_lock);
1867                 spin_unlock_bh(&ipvs->sync_buff_lock);
1868
1869                 retc = 0;
1870                 for (id = ipvs->threads_mask; id >= 0; id--) {
1871                         struct ipvs_master_sync_state *ms = &ipvs->ms[id];
1872                         int ret;
1873
1874                         pr_info("stopping master sync thread %d ...\n",
1875                                 task_pid_nr(ms->master_thread));
1876                         cancel_delayed_work_sync(&ms->master_wakeup_work);
1877                         ret = kthread_stop(ms->master_thread);
1878                         if (retc >= 0)
1879                                 retc = ret;
1880                 }
1881                 kfree(ipvs->ms);
1882                 ipvs->ms = NULL;
1883         } else if (state == IP_VS_STATE_BACKUP) {
1884                 if (!ipvs->backup_threads)
1885                         return -ESRCH;
1886
1887                 ipvs->sync_state &= ~IP_VS_STATE_BACKUP;
1888                 array = ipvs->backup_threads;
1889                 retc = 0;
1890                 for (id = ipvs->threads_mask; id >= 0; id--) {
1891                         int ret;
1892
1893                         pr_info("stopping backup sync thread %d ...\n",
1894                                 task_pid_nr(array[id]));
1895                         ret = kthread_stop(array[id]);
1896                         if (retc >= 0)
1897                                 retc = ret;
1898                 }
1899                 kfree(array);
1900                 ipvs->backup_threads = NULL;
1901         }
1902
1903         /* decrease the module use count */
1904         ip_vs_use_count_dec();
1905
1906         return retc;
1907 }
1908
1909 /*
1910  * Initialize data struct for each netns
1911  */
1912 int __net_init ip_vs_sync_net_init(struct net *net)
1913 {
1914         struct netns_ipvs *ipvs = net_ipvs(net);
1915
1916         __mutex_init(&ipvs->sync_mutex, "ipvs->sync_mutex", &__ipvs_sync_key);
1917         spin_lock_init(&ipvs->sync_lock);
1918         spin_lock_init(&ipvs->sync_buff_lock);
1919         return 0;
1920 }
1921
1922 void ip_vs_sync_net_cleanup(struct net *net)
1923 {
1924         int retc;
1925         struct netns_ipvs *ipvs = net_ipvs(net);
1926
1927         mutex_lock(&ipvs->sync_mutex);
1928         retc = stop_sync_thread(net, IP_VS_STATE_MASTER);
1929         if (retc && retc != -ESRCH)
1930                 pr_err("Failed to stop Master Daemon\n");
1931
1932         retc = stop_sync_thread(net, IP_VS_STATE_BACKUP);
1933         if (retc && retc != -ESRCH)
1934                 pr_err("Failed to stop Backup Daemon\n");
1935         mutex_unlock(&ipvs->sync_mutex);
1936 }