kernel/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 connection is controlled by persistence */
 429 static inline bool in_persistence(struct ip_vs_conn *cp)
 430 {
 431         for (cp = cp->control; cp; cp = cp->control) {
 432                 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
 433                         return true;
 434         }
 435         return false;
 436 }
 437
 438 /* Check if conn should be synced.
 439  * pkts: conn packets, use sysctl_sync_threshold to avoid packet check
 440  * - (1) sync_refresh_period: reduce sync rate. Additionally, retry
 441  *      sync_retries times with period of sync_refresh_period/8
 442  * - (2) if both sync_refresh_period and sync_period are 0 send sync only
 443  *      for state changes or only once when pkts matches sync_threshold
 444  * - (3) templates: rate can be reduced only with sync_refresh_period or
 445  *      with (2)
 446  */
 447 static int ip_vs_sync_conn_needed(struct netns_ipvs *ipvs,
 448                                   struct ip_vs_conn *cp, int pkts)
 449 {
 450         unsigned long orig = ACCESS_ONCE(cp->sync_endtime);
 451         unsigned long now = jiffies;
 452         unsigned long n = (now + cp->timeout) & ~3UL;
 453         unsigned int sync_refresh_period;
 454         int sync_period;
 455         int force;
 456
 457         /* Check if we sync in current state */
 458         if (unlikely(cp->flags & IP_VS_CONN_F_TEMPLATE))
 459                 force = 0;
 460         else if (unlikely(sysctl_sync_persist_mode(ipvs) && in_persistence(cp)))
 461                 return 0;
 462         else if (likely(cp->protocol == IPPROTO_TCP)) {
 463                 if (!((1 << cp->state) &
 464                       ((1 << IP_VS_TCP_S_ESTABLISHED) |
 465                        (1 << IP_VS_TCP_S_FIN_WAIT) |
 466                        (1 << IP_VS_TCP_S_CLOSE) |
 467                        (1 << IP_VS_TCP_S_CLOSE_WAIT) |
 468                        (1 << IP_VS_TCP_S_TIME_WAIT))))
 469                         return 0;
 470                 force = cp->state != cp->old_state;
 471                 if (force && cp->state != IP_VS_TCP_S_ESTABLISHED)
 472                         goto set;
 473         } else if (unlikely(cp->protocol == IPPROTO_SCTP)) {
 474                 if (!((1 << cp->state) &
 475                       ((1 << IP_VS_SCTP_S_ESTABLISHED) |
 476                        (1 << IP_VS_SCTP_S_SHUTDOWN_SENT) |
 477                        (1 << IP_VS_SCTP_S_SHUTDOWN_RECEIVED) |
 478                        (1 << IP_VS_SCTP_S_SHUTDOWN_ACK_SENT) |
 479                        (1 << IP_VS_SCTP_S_CLOSED))))
 480                         return 0;
 481                 force = cp->state != cp->old_state;
 482                 if (force && cp->state != IP_VS_SCTP_S_ESTABLISHED)
 483                         goto set;
 484         } else {
 485                 /* UDP or another protocol with single state */
 486                 force = 0;
 487         }
 488
 489         sync_refresh_period = sysctl_sync_refresh_period(ipvs);
 490         if (sync_refresh_period > 0) {
 491                 long diff = n - orig;
 492                 long min_diff = max(cp->timeout >> 1, 10UL * HZ);
 493
 494                 /* Avoid sync if difference is below sync_refresh_period
 495                  * and below the half timeout.
 496                  */
 497                 if (abs(diff) < min_t(long, sync_refresh_period, min_diff)) {
 498                         int retries = orig & 3;
 499
 500                         if (retries >= sysctl_sync_retries(ipvs))
 501                                 return 0;
 502                         if (time_before(now, orig - cp->timeout +
 503                                         (sync_refresh_period >> 3)))
 504                                 return 0;
 505                         n |= retries + 1;
 506                 }
 507         }
 508         sync_period = sysctl_sync_period(ipvs);
 509         if (sync_period > 0) {
 510                 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE) &&
 511                     pkts % sync_period != sysctl_sync_threshold(ipvs))
 512                         return 0;
 513         } else if (sync_refresh_period <= 0 &&
 514                    pkts != sysctl_sync_threshold(ipvs))
 515                 return 0;
 516
 517 set:
 518         cp->old_state = cp->state;
 519         n = cmpxchg(&cp->sync_endtime, orig, n);
 520         return n == orig || force;
 521 }
 522
 523 /*
 524  *      Version 0 , could be switched in by sys_ctl.
 525  *      Add an ip_vs_conn information into the current sync_buff.
 526  */
 527 static void ip_vs_sync_conn_v0(struct net *net, struct ip_vs_conn *cp,
 528                                int pkts)
 529 {
 530         struct netns_ipvs *ipvs = net_ipvs(net);
 531         struct ip_vs_sync_mesg_v0 *m;
 532         struct ip_vs_sync_conn_v0 *s;
 533         struct ip_vs_sync_buff *buff;
 534         struct ipvs_master_sync_state *ms;
 535         int id;
 536         int len;
 537
 538         if (unlikely(cp->af != AF_INET))
 539                 return;
 540         /* Do not sync ONE PACKET */
 541         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
 542                 return;
 543
 544         if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
 545                 return;
 546
 547         spin_lock_bh(&ipvs->sync_buff_lock);
 548         if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
 549                 spin_unlock_bh(&ipvs->sync_buff_lock);
 550                 return;
 551         }
 552
 553         id = select_master_thread_id(ipvs, cp);
 554         ms = &ipvs->ms[id];
 555         buff = ms->sync_buff;
 556         if (buff) {
 557                 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
 558                 /* Send buffer if it is for v1 */
 559                 if (!m->nr_conns) {
 560                         sb_queue_tail(ipvs, ms);
 561                         ms->sync_buff = NULL;
 562                         buff = NULL;
 563                 }
 564         }
 565         if (!buff) {
 566                 buff = ip_vs_sync_buff_create_v0(ipvs);
 567                 if (!buff) {
 568                         spin_unlock_bh(&ipvs->sync_buff_lock);
 569                         pr_err("ip_vs_sync_buff_create failed.\n");
 570                         return;
 571                 }
 572                 ms->sync_buff = buff;
 573         }
 574
 575         len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
 576                 SIMPLE_CONN_SIZE;
 577         m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
 578         s = (struct ip_vs_sync_conn_v0 *) buff->head;
 579
 580         /* copy members */
 581         s->reserved = 0;
 582         s->protocol = cp->protocol;
 583         s->cport = cp->cport;
 584         s->vport = cp->vport;
 585         s->dport = cp->dport;
 586         s->caddr = cp->caddr.ip;
 587         s->vaddr = cp->vaddr.ip;
 588         s->daddr = cp->daddr.ip;
 589         s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
 590         s->state = htons(cp->state);
 591         if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
 592                 struct ip_vs_sync_conn_options *opt =
 593                         (struct ip_vs_sync_conn_options *)&s[1];
 594                 memcpy(opt, &cp->in_seq, sizeof(*opt));
 595         }
 596
 597         m->nr_conns++;
 598         m->size = htons(ntohs(m->size) + len);
 599         buff->head += len;
 600
 601         /* check if there is a space for next one */
 602         if (buff->head + FULL_CONN_SIZE > buff->end) {
 603                 sb_queue_tail(ipvs, ms);
 604                 ms->sync_buff = NULL;
 605         }
 606         spin_unlock_bh(&ipvs->sync_buff_lock);
 607
 608         /* synchronize its controller if it has */
 609         cp = cp->control;
 610         if (cp) {
 611                 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
 612                         pkts = atomic_add_return(1, &cp->in_pkts);
 613                 else
 614                         pkts = sysctl_sync_threshold(ipvs);
 615                 ip_vs_sync_conn(net, cp->control, pkts);
 616         }
 617 }
 618
 619 /*
 620  *      Add an ip_vs_conn information into the current sync_buff.
 621  *      Called by ip_vs_in.
 622  *      Sending Version 1 messages
 623  */
 624 void ip_vs_sync_conn(struct net *net, struct ip_vs_conn *cp, int pkts)
 625 {
 626         struct netns_ipvs *ipvs = net_ipvs(net);
 627         struct ip_vs_sync_mesg *m;
 628         union ip_vs_sync_conn *s;
 629         struct ip_vs_sync_buff *buff;
 630         struct ipvs_master_sync_state *ms;
 631         int id;
 632         __u8 *p;
 633         unsigned int len, pe_name_len, pad;
 634
 635         /* Handle old version of the protocol */
 636         if (sysctl_sync_ver(ipvs) == 0) {
 637                 ip_vs_sync_conn_v0(net, cp, pkts);
 638                 return;
 639         }
 640         /* Do not sync ONE PACKET */
 641         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
 642                 goto control;
 643 sloop:
 644         if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
 645                 goto control;
 646
 647         /* Sanity checks */
 648         pe_name_len = 0;
 649         if (cp->pe_data_len) {
 650                 if (!cp->pe_data || !cp->dest) {
 651                         IP_VS_ERR_RL("SYNC, connection pe_data invalid\n");
 652                         return;
 653                 }
 654                 pe_name_len = strnlen(cp->pe->name, IP_VS_PENAME_MAXLEN);
 655         }
 656
 657         spin_lock_bh(&ipvs->sync_buff_lock);
 658         if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
 659                 spin_unlock_bh(&ipvs->sync_buff_lock);
 660                 return;
 661         }
 662
 663         id = select_master_thread_id(ipvs, cp);
 664         ms = &ipvs->ms[id];
 665
 666 #ifdef CONFIG_IP_VS_IPV6
 667         if (cp->af == AF_INET6)
 668                 len = sizeof(struct ip_vs_sync_v6);
 669         else
 670 #endif
 671                 len = sizeof(struct ip_vs_sync_v4);
 672
 673         if (cp->flags & IP_VS_CONN_F_SEQ_MASK)
 674                 len += sizeof(struct ip_vs_sync_conn_options) + 2;
 675
 676         if (cp->pe_data_len)
 677                 len += cp->pe_data_len + 2;     /* + Param hdr field */
 678         if (pe_name_len)
 679                 len += pe_name_len + 2;
 680
 681         /* check if there is a space for this one  */
 682         pad = 0;
 683         buff = ms->sync_buff;
 684         if (buff) {
 685                 m = buff->mesg;
 686                 pad = (4 - (size_t) buff->head) & 3;
 687                 /* Send buffer if it is for v0 */
 688                 if (buff->head + len + pad > buff->end || m->reserved) {
 689                         sb_queue_tail(ipvs, ms);
 690                         ms->sync_buff = NULL;
 691                         buff = NULL;
 692                         pad = 0;
 693                 }
 694         }
 695
 696         if (!buff) {
 697                 buff = ip_vs_sync_buff_create(ipvs);
 698                 if (!buff) {
 699                         spin_unlock_bh(&ipvs->sync_buff_lock);
 700                         pr_err("ip_vs_sync_buff_create failed.\n");
 701                         return;
 702                 }
 703                 ms->sync_buff = buff;
 704                 m = buff->mesg;
 705         }
 706
 707         p = buff->head;
 708         buff->head += pad + len;
 709         m->size = htons(ntohs(m->size) + pad + len);
 710         /* Add ev. padding from prev. sync_conn */
 711         while (pad--)
 712                 *(p++) = 0;
 713
 714         s = (union ip_vs_sync_conn *)p;
 715
 716         /* Set message type  & copy members */
 717         s->v4.type = (cp->af == AF_INET6 ? STYPE_F_INET6 : 0);
 718         s->v4.ver_size = htons(len & SVER_MASK);        /* Version 0 */
 719         s->v4.flags = htonl(cp->flags & ~IP_VS_CONN_F_HASHED);
 720         s->v4.state = htons(cp->state);
 721         s->v4.protocol = cp->protocol;
 722         s->v4.cport = cp->cport;
 723         s->v4.vport = cp->vport;
 724         s->v4.dport = cp->dport;
 725         s->v4.fwmark = htonl(cp->fwmark);
 726         s->v4.timeout = htonl(cp->timeout / HZ);
 727         m->nr_conns++;
 728
 729 #ifdef CONFIG_IP_VS_IPV6
 730         if (cp->af == AF_INET6) {
 731                 p += sizeof(struct ip_vs_sync_v6);
 732                 s->v6.caddr = cp->caddr.in6;
 733                 s->v6.vaddr = cp->vaddr.in6;
 734                 s->v6.daddr = cp->daddr.in6;
 735         } else
 736 #endif
 737         {
 738                 p += sizeof(struct ip_vs_sync_v4);      /* options ptr */
 739                 s->v4.caddr = cp->caddr.ip;
 740                 s->v4.vaddr = cp->vaddr.ip;
 741                 s->v4.daddr = cp->daddr.ip;
 742         }
 743         if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
 744                 *(p++) = IPVS_OPT_SEQ_DATA;
 745                 *(p++) = sizeof(struct ip_vs_sync_conn_options);
 746                 hton_seq((struct ip_vs_seq *)p, &cp->in_seq);
 747                 p += sizeof(struct ip_vs_seq);
 748                 hton_seq((struct ip_vs_seq *)p, &cp->out_seq);
 749                 p += sizeof(struct ip_vs_seq);
 750         }
 751         /* Handle pe data */
 752         if (cp->pe_data_len && cp->pe_data) {
 753                 *(p++) = IPVS_OPT_PE_DATA;
 754                 *(p++) = cp->pe_data_len;
 755                 memcpy(p, cp->pe_data, cp->pe_data_len);
 756                 p += cp->pe_data_len;
 757                 if (pe_name_len) {
 758                         /* Add PE_NAME */
 759                         *(p++) = IPVS_OPT_PE_NAME;
 760                         *(p++) = pe_name_len;
 761                         memcpy(p, cp->pe->name, pe_name_len);
 762                         p += pe_name_len;
 763                 }
 764         }
 765
 766         spin_unlock_bh(&ipvs->sync_buff_lock);
 767
 768 control:
 769         /* synchronize its controller if it has */
 770         cp = cp->control;
 771         if (!cp)
 772                 return;
 773         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
 774                 pkts = atomic_add_return(1, &cp->in_pkts);
 775         else
 776                 pkts = sysctl_sync_threshold(ipvs);
 777         goto sloop;
 778 }
 779
 780 /*
 781  *  fill_param used by version 1
 782  */
 783 static inline int
 784 ip_vs_conn_fill_param_sync(struct net *net, int af, union ip_vs_sync_conn *sc,
 785                            struct ip_vs_conn_param *p,
 786                            __u8 *pe_data, unsigned int pe_data_len,
 787                            __u8 *pe_name, unsigned int pe_name_len)
 788 {
 789 #ifdef CONFIG_IP_VS_IPV6
 790         if (af == AF_INET6)
 791                 ip_vs_conn_fill_param(net, af, sc->v6.protocol,
 792                                       (const union nf_inet_addr *)&sc->v6.caddr,
 793                                       sc->v6.cport,
 794                                       (const union nf_inet_addr *)&sc->v6.vaddr,
 795                                       sc->v6.vport, p);
 796         else
 797 #endif
 798                 ip_vs_conn_fill_param(net, af, sc->v4.protocol,
 799                                       (const union nf_inet_addr *)&sc->v4.caddr,
 800                                       sc->v4.cport,
 801                                       (const union nf_inet_addr *)&sc->v4.vaddr,
 802                                       sc->v4.vport, p);
 803         /* Handle pe data */
 804         if (pe_data_len) {
 805                 if (pe_name_len) {
 806                         char buff[IP_VS_PENAME_MAXLEN+1];
 807
 808                         memcpy(buff, pe_name, pe_name_len);
 809                         buff[pe_name_len]=0;
 810                         p->pe = __ip_vs_pe_getbyname(buff);
 811                         if (!p->pe) {
 812                                 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n",
 813                                              buff);
 814                                 return 1;
 815                         }
 816                 } else {
 817                         IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n");
 818                         return 1;
 819                 }
 820
 821                 p->pe_data = kmemdup(pe_data, pe_data_len, GFP_ATOMIC);
 822                 if (!p->pe_data) {
 823                         module_put(p->pe->module);
 824                         return -ENOMEM;
 825                 }
 826                 p->pe_data_len = pe_data_len;
 827         }
 828         return 0;
 829 }
 830
 831 /*
 832  *  Connection Add / Update.
 833  *  Common for version 0 and 1 reception of backup sync_conns.
 834  *  Param: ...
 835  *         timeout is in sec.
 836  */
 837 static void ip_vs_proc_conn(struct net *net, struct ip_vs_conn_param *param,
 838                             unsigned int flags, unsigned int state,
 839                             unsigned int protocol, unsigned int type,
 840                             const union nf_inet_addr *daddr, __be16 dport,
 841                             unsigned long timeout, __u32 fwmark,
 842                             struct ip_vs_sync_conn_options *opt)
 843 {
 844         struct ip_vs_dest *dest;
 845         struct ip_vs_conn *cp;
 846         struct netns_ipvs *ipvs = net_ipvs(net);
 847
 848         if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
 849                 cp = ip_vs_conn_in_get(param);
 850                 if (cp && ((cp->dport != dport) ||
 851                            !ip_vs_addr_equal(cp->daf, &cp->daddr, daddr))) {
 852                         if (!(flags & IP_VS_CONN_F_INACTIVE)) {
 853                                 ip_vs_conn_expire_now(cp);
 854                                 __ip_vs_conn_put(cp);
 855                                 cp = NULL;
 856                         } else {
 857                                 /* This is the expiration message for the
 858                                  * connection that was already replaced, so we
 859                                  * just ignore it.
 860                                  */
 861                                 __ip_vs_conn_put(cp);
 862                                 kfree(param->pe_data);
 863                                 return;
 864                         }
 865                 }
 866         } else {
 867                 cp = ip_vs_ct_in_get(param);
 868         }
 869
 870         if (cp) {
 871                 /* Free pe_data */
 872                 kfree(param->pe_data);
 873
 874                 dest = cp->dest;
 875                 spin_lock_bh(&cp->lock);
 876                 if ((cp->flags ^ flags) & IP_VS_CONN_F_INACTIVE &&
 877                     !(flags & IP_VS_CONN_F_TEMPLATE) && dest) {
 878                         if (flags & IP_VS_CONN_F_INACTIVE) {
 879                                 atomic_dec(&dest->activeconns);
 880                                 atomic_inc(&dest->inactconns);
 881                         } else {
 882                                 atomic_inc(&dest->activeconns);
 883                                 atomic_dec(&dest->inactconns);
 884                         }
 885                 }
 886                 flags &= IP_VS_CONN_F_BACKUP_UPD_MASK;
 887                 flags |= cp->flags & ~IP_VS_CONN_F_BACKUP_UPD_MASK;
 888                 cp->flags = flags;
 889                 spin_unlock_bh(&cp->lock);
 890                 if (!dest)
 891                         ip_vs_try_bind_dest(cp);
 892         } else {
 893                 /*
 894                  * Find the appropriate destination for the connection.
 895                  * If it is not found the connection will remain unbound
 896                  * but still handled.
 897                  */
 898                 rcu_read_lock();
 899                 /* This function is only invoked by the synchronization
 900                  * code. We do not currently support heterogeneous pools
 901                  * with synchronization, so we can make the assumption that
 902                  * the svc_af is the same as the dest_af
 903                  */
 904                 dest = ip_vs_find_dest(net, type, type, daddr, dport,
 905                                        param->vaddr, param->vport, protocol,
 906                                        fwmark, flags);
 907
 908                 cp = ip_vs_conn_new(param, type, daddr, dport, flags, dest,
 909                                     fwmark);
 910                 rcu_read_unlock();
 911                 if (!cp) {
 912                         kfree(param->pe_data);
 913                         IP_VS_DBG(2, "BACKUP, add new conn. failed\n");
 914                         return;
 915                 }
 916                 if (!(flags & IP_VS_CONN_F_TEMPLATE))
 917                         kfree(param->pe_data);
 918         }
 919
 920         if (opt)
 921                 memcpy(&cp->in_seq, opt, sizeof(*opt));
 922         atomic_set(&cp->in_pkts, sysctl_sync_threshold(ipvs));
 923         cp->state = state;
 924         cp->old_state = cp->state;
 925         /*
 926          * For Ver 0 messages style
 927          *  - Not possible to recover the right timeout for templates
 928          *  - can not find the right fwmark
 929          *    virtual service. If needed, we can do it for
 930          *    non-fwmark persistent services.
 931          * Ver 1 messages style.
 932          *  - No problem.
 933          */
 934         if (timeout) {
 935                 if (timeout > MAX_SCHEDULE_TIMEOUT / HZ)
 936                         timeout = MAX_SCHEDULE_TIMEOUT / HZ;
 937                 cp->timeout = timeout*HZ;
 938         } else {
 939                 struct ip_vs_proto_data *pd;
 940
 941                 pd = ip_vs_proto_data_get(net, protocol);
 942                 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pd && pd->timeout_table)
 943                         cp->timeout = pd->timeout_table[state];
 944                 else
 945                         cp->timeout = (3*60*HZ);
 946         }
 947         ip_vs_conn_put(cp);
 948 }
 949
 950 /*
 951  *  Process received multicast message for Version 0
 952  */
 953 static void ip_vs_process_message_v0(struct net *net, const char *buffer,
 954                                      const size_t buflen)
 955 {
 956         struct ip_vs_sync_mesg_v0 *m = (struct ip_vs_sync_mesg_v0 *)buffer;
 957         struct ip_vs_sync_conn_v0 *s;
 958         struct ip_vs_sync_conn_options *opt;
 959         struct ip_vs_protocol *pp;
 960         struct ip_vs_conn_param param;
 961         char *p;
 962         int i;
 963
 964         p = (char *)buffer + sizeof(struct ip_vs_sync_mesg_v0);
 965         for (i=0; i<m->nr_conns; i++) {
 966                 unsigned int flags, state;
 967
 968                 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
 969                         IP_VS_ERR_RL("BACKUP v0, bogus conn\n");
 970                         return;
 971                 }
 972                 s = (struct ip_vs_sync_conn_v0 *) p;
 973                 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
 974                 flags &= ~IP_VS_CONN_F_HASHED;
 975                 if (flags & IP_VS_CONN_F_SEQ_MASK) {
 976                         opt = (struct ip_vs_sync_conn_options *)&s[1];
 977                         p += FULL_CONN_SIZE;
 978                         if (p > buffer+buflen) {
 979                                 IP_VS_ERR_RL("BACKUP v0, Dropping buffer bogus conn options\n");
 980                                 return;
 981                         }
 982                 } else {
 983                         opt = NULL;
 984                         p += SIMPLE_CONN_SIZE;
 985                 }
 986
 987                 state = ntohs(s->state);
 988                 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
 989                         pp = ip_vs_proto_get(s->protocol);
 990                         if (!pp) {
 991                                 IP_VS_DBG(2, "BACKUP v0, Unsupported protocol %u\n",
 992                                         s->protocol);
 993                                 continue;
 994                         }
 995                         if (state >= pp->num_states) {
 996                                 IP_VS_DBG(2, "BACKUP v0, Invalid %s state %u\n",
 997                                         pp->name, state);
 998                                 continue;
 999                         }
1000                 } else {
1001                         /* protocol in templates is not used for state/timeout */
1002                         if (state > 0) {
1003                                 IP_VS_DBG(2, "BACKUP v0, Invalid template state %u\n",
1004                                         state);
1005                                 state = 0;
1006                         }
1007                 }
1008
1009                 ip_vs_conn_fill_param(net, AF_INET, s->protocol,
1010                                       (const union nf_inet_addr *)&s->caddr,
1011                                       s->cport,
1012                                       (const union nf_inet_addr *)&s->vaddr,
1013                                       s->vport, &param);
1014
1015                 /* Send timeout as Zero */
1016                 ip_vs_proc_conn(net, &param, flags, state, s->protocol, AF_INET,
1017                                 (union nf_inet_addr *)&s->daddr, s->dport,
1018                                 0, 0, opt);
1019         }
1020 }
1021
1022 /*
1023  * Handle options
1024  */
1025 static inline int ip_vs_proc_seqopt(__u8 *p, unsigned int plen,
1026                                     __u32 *opt_flags,
1027                                     struct ip_vs_sync_conn_options *opt)
1028 {
1029         struct ip_vs_sync_conn_options *topt;
1030
1031         topt = (struct ip_vs_sync_conn_options *)p;
1032
1033         if (plen != sizeof(struct ip_vs_sync_conn_options)) {
1034                 IP_VS_DBG(2, "BACKUP, bogus conn options length\n");
1035                 return -EINVAL;
1036         }
1037         if (*opt_flags & IPVS_OPT_F_SEQ_DATA) {
1038                 IP_VS_DBG(2, "BACKUP, conn options found twice\n");
1039                 return -EINVAL;
1040         }
1041         ntoh_seq(&topt->in_seq, &opt->in_seq);
1042         ntoh_seq(&topt->out_seq, &opt->out_seq);
1043         *opt_flags |= IPVS_OPT_F_SEQ_DATA;
1044         return 0;
1045 }
1046
1047 static int ip_vs_proc_str(__u8 *p, unsigned int plen, unsigned int *data_len,
1048                           __u8 **data, unsigned int maxlen,
1049                           __u32 *opt_flags, __u32 flag)
1050 {
1051         if (plen > maxlen) {
1052                 IP_VS_DBG(2, "BACKUP, bogus par.data len > %d\n", maxlen);
1053                 return -EINVAL;
1054         }
1055         if (*opt_flags & flag) {
1056                 IP_VS_DBG(2, "BACKUP, Par.data found twice 0x%x\n", flag);
1057                 return -EINVAL;
1058         }
1059         *data_len = plen;
1060         *data = p;
1061         *opt_flags |= flag;
1062         return 0;
1063 }
1064 /*
1065  *   Process a Version 1 sync. connection
1066  */
1067 static inline int ip_vs_proc_sync_conn(struct net *net, __u8 *p, __u8 *msg_end)
1068 {
1069         struct ip_vs_sync_conn_options opt;
1070         union  ip_vs_sync_conn *s;
1071         struct ip_vs_protocol *pp;
1072         struct ip_vs_conn_param param;
1073         __u32 flags;
1074         unsigned int af, state, pe_data_len=0, pe_name_len=0;
1075         __u8 *pe_data=NULL, *pe_name=NULL;
1076         __u32 opt_flags=0;
1077         int retc=0;
1078
1079         s = (union ip_vs_sync_conn *) p;
1080
1081         if (s->v6.type & STYPE_F_INET6) {
1082 #ifdef CONFIG_IP_VS_IPV6
1083                 af = AF_INET6;
1084                 p += sizeof(struct ip_vs_sync_v6);
1085 #else
1086                 IP_VS_DBG(3,"BACKUP, IPv6 msg received, and IPVS is not compiled for IPv6\n");
1087                 retc = 10;
1088                 goto out;
1089 #endif
1090         } else if (!s->v4.type) {
1091                 af = AF_INET;
1092                 p += sizeof(struct ip_vs_sync_v4);
1093         } else {
1094                 return -10;
1095         }
1096         if (p > msg_end)
1097                 return -20;
1098
1099         /* Process optional params check Type & Len. */
1100         while (p < msg_end) {
1101                 int ptype;
1102                 int plen;
1103
1104                 if (p+2 > msg_end)
1105                         return -30;
1106                 ptype = *(p++);
1107                 plen  = *(p++);
1108
1109                 if (!plen || ((p + plen) > msg_end))
1110                         return -40;
1111                 /* Handle seq option  p = param data */
1112                 switch (ptype & ~IPVS_OPT_F_PARAM) {
1113                 case IPVS_OPT_SEQ_DATA:
1114                         if (ip_vs_proc_seqopt(p, plen, &opt_flags, &opt))
1115                                 return -50;
1116                         break;
1117
1118                 case IPVS_OPT_PE_DATA:
1119                         if (ip_vs_proc_str(p, plen, &pe_data_len, &pe_data,
1120                                            IP_VS_PEDATA_MAXLEN, &opt_flags,
1121                                            IPVS_OPT_F_PE_DATA))
1122                                 return -60;
1123                         break;
1124
1125                 case IPVS_OPT_PE_NAME:
1126                         if (ip_vs_proc_str(p, plen,&pe_name_len, &pe_name,
1127                                            IP_VS_PENAME_MAXLEN, &opt_flags,
1128                                            IPVS_OPT_F_PE_NAME))
1129                                 return -70;
1130                         break;
1131
1132                 default:
1133                         /* Param data mandatory ? */
1134                         if (!(ptype & IPVS_OPT_F_PARAM)) {
1135                                 IP_VS_DBG(3, "BACKUP, Unknown mandatory param %d found\n",
1136                                           ptype & ~IPVS_OPT_F_PARAM);
1137                                 retc = 20;
1138                                 goto out;
1139                         }
1140                 }
1141                 p += plen;  /* Next option */
1142         }
1143
1144         /* Get flags and Mask off unsupported */
1145         flags  = ntohl(s->v4.flags) & IP_VS_CONN_F_BACKUP_MASK;
1146         flags |= IP_VS_CONN_F_SYNC;
1147         state = ntohs(s->v4.state);
1148
1149         if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
1150                 pp = ip_vs_proto_get(s->v4.protocol);
1151                 if (!pp) {
1152                         IP_VS_DBG(3,"BACKUP, Unsupported protocol %u\n",
1153                                 s->v4.protocol);
1154                         retc = 30;
1155                         goto out;
1156                 }
1157                 if (state >= pp->num_states) {
1158                         IP_VS_DBG(3, "BACKUP, Invalid %s state %u\n",
1159                                 pp->name, state);
1160                         retc = 40;
1161                         goto out;
1162                 }
1163         } else {
1164                 /* protocol in templates is not used for state/timeout */
1165                 if (state > 0) {
1166                         IP_VS_DBG(3, "BACKUP, Invalid template state %u\n",
1167                                 state);
1168                         state = 0;
1169                 }
1170         }
1171         if (ip_vs_conn_fill_param_sync(net, af, s, &param, pe_data,
1172                                        pe_data_len, pe_name, pe_name_len)) {
1173                 retc = 50;
1174                 goto out;
1175         }
1176         /* If only IPv4, just silent skip IPv6 */
1177         if (af == AF_INET)
1178                 ip_vs_proc_conn(net, &param, flags, state, s->v4.protocol, af,
1179                                 (union nf_inet_addr *)&s->v4.daddr, s->v4.dport,
1180                                 ntohl(s->v4.timeout), ntohl(s->v4.fwmark),
1181                                 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1182                                 );
1183 #ifdef CONFIG_IP_VS_IPV6
1184         else
1185                 ip_vs_proc_conn(net, &param, flags, state, s->v6.protocol, af,
1186                                 (union nf_inet_addr *)&s->v6.daddr, s->v6.dport,
1187                                 ntohl(s->v6.timeout), ntohl(s->v6.fwmark),
1188                                 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1189                                 );
1190 #endif
1191         ip_vs_pe_put(param.pe);
1192         return 0;
1193         /* Error exit */
1194 out:
1195         IP_VS_DBG(2, "BACKUP, Single msg dropped err:%d\n", retc);
1196         return retc;
1197
1198 }
1199 /*
1200  *      Process received multicast message and create the corresponding
1201  *      ip_vs_conn entries.
1202  *      Handles Version 0 & 1
1203  */
1204 static void ip_vs_process_message(struct net *net, __u8 *buffer,
1205                                   const size_t buflen)
1206 {
1207         struct netns_ipvs *ipvs = net_ipvs(net);
1208         struct ip_vs_sync_mesg *m2 = (struct ip_vs_sync_mesg *)buffer;
1209         __u8 *p, *msg_end;
1210         int i, nr_conns;
1211
1212         if (buflen < sizeof(struct ip_vs_sync_mesg_v0)) {
1213                 IP_VS_DBG(2, "BACKUP, message header too short\n");
1214                 return;
1215         }
1216
1217         if (buflen != ntohs(m2->size)) {
1218                 IP_VS_DBG(2, "BACKUP, bogus message size\n");
1219                 return;
1220         }
1221         /* SyncID sanity check */
1222         if (ipvs->backup_syncid != 0 && m2->syncid != ipvs->backup_syncid) {
1223                 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2->syncid);
1224                 return;
1225         }
1226         /* Handle version 1  message */
1227         if ((m2->version == SYNC_PROTO_VER) && (m2->reserved == 0)
1228             && (m2->spare == 0)) {
1229
1230                 msg_end = buffer + sizeof(struct ip_vs_sync_mesg);
1231                 nr_conns = m2->nr_conns;
1232
1233                 for (i=0; i<nr_conns; i++) {
1234                         union ip_vs_sync_conn *s;
1235                         unsigned int size;
1236                         int retc;
1237
1238                         p = msg_end;
1239                         if (p + sizeof(s->v4) > buffer+buflen) {
1240                                 IP_VS_ERR_RL("BACKUP, Dropping buffer, to small\n");
1241                                 return;
1242                         }
1243                         s = (union ip_vs_sync_conn *)p;
1244                         size = ntohs(s->v4.ver_size) & SVER_MASK;
1245                         msg_end = p + size;
1246                         /* Basic sanity checks */
1247                         if (msg_end  > buffer+buflen) {
1248                                 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n");
1249                                 return;
1250                         }
1251                         if (ntohs(s->v4.ver_size) >> SVER_SHIFT) {
1252                                 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n",
1253                                               ntohs(s->v4.ver_size) >> SVER_SHIFT);
1254                                 return;
1255                         }
1256                         /* Process a single sync_conn */
1257                         retc = ip_vs_proc_sync_conn(net, p, msg_end);
1258                         if (retc < 0) {
1259                                 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n",
1260                                              retc);
1261                                 return;
1262                         }
1263                         /* Make sure we have 32 bit alignment */
1264                         msg_end = p + ((size + 3) & ~3);
1265                 }
1266         } else {
1267                 /* Old type of message */
1268                 ip_vs_process_message_v0(net, buffer, buflen);
1269                 return;
1270         }
1271 }
1272
1273
1274 /*
1275  *      Setup sndbuf (mode=1) or rcvbuf (mode=0)
1276  */
1277 static void set_sock_size(struct sock *sk, int mode, int val)
1278 {
1279         /* setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &val, sizeof(val)); */
1280         /* setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)); */
1281         lock_sock(sk);
1282         if (mode) {
1283                 val = clamp_t(int, val, (SOCK_MIN_SNDBUF + 1) / 2,
1284                               sysctl_wmem_max);
1285                 sk->sk_sndbuf = val * 2;
1286                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1287         } else {
1288                 val = clamp_t(int, val, (SOCK_MIN_RCVBUF + 1) / 2,
1289                               sysctl_rmem_max);
1290                 sk->sk_rcvbuf = val * 2;
1291                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1292         }
1293         release_sock(sk);
1294 }
1295
1296 /*
1297  *      Setup loopback of outgoing multicasts on a sending socket
1298  */
1299 static void set_mcast_loop(struct sock *sk, u_char loop)
1300 {
1301         struct inet_sock *inet = inet_sk(sk);
1302
1303         /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
1304         lock_sock(sk);
1305         inet->mc_loop = loop ? 1 : 0;
1306         release_sock(sk);
1307 }
1308
1309 /*
1310  *      Specify TTL for outgoing multicasts on a sending socket
1311  */
1312 static void set_mcast_ttl(struct sock *sk, u_char ttl)
1313 {
1314         struct inet_sock *inet = inet_sk(sk);
1315
1316         /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
1317         lock_sock(sk);
1318         inet->mc_ttl = ttl;
1319         release_sock(sk);
1320 }
1321
1322 /*
1323  *      Specifiy default interface for outgoing multicasts
1324  */
1325 static int set_mcast_if(struct sock *sk, char *ifname)
1326 {
1327         struct net_device *dev;
1328         struct inet_sock *inet = inet_sk(sk);
1329         struct net *net = sock_net(sk);
1330
1331         dev = __dev_get_by_name(net, ifname);
1332         if (!dev)
1333                 return -ENODEV;
1334
1335         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1336                 return -EINVAL;
1337
1338         lock_sock(sk);
1339         inet->mc_index = dev->ifindex;
1340         /*  inet->mc_addr  = 0; */
1341         release_sock(sk);
1342
1343         return 0;
1344 }
1345
1346
1347 /*
1348  *      Set the maximum length of sync message according to the
1349  *      specified interface's MTU.
1350  */
1351 static int set_sync_mesg_maxlen(struct net *net, int sync_state)
1352 {
1353         struct netns_ipvs *ipvs = net_ipvs(net);
1354         struct net_device *dev;
1355         int num;
1356
1357         if (sync_state == IP_VS_STATE_MASTER) {
1358                 dev = __dev_get_by_name(net, ipvs->master_mcast_ifn);
1359                 if (!dev)
1360                         return -ENODEV;
1361
1362                 num = (dev->mtu - sizeof(struct iphdr) -
1363                        sizeof(struct udphdr) -
1364                        SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE;
1365                 ipvs->send_mesg_maxlen = SYNC_MESG_HEADER_LEN +
1366                         SIMPLE_CONN_SIZE * min(num, MAX_CONNS_PER_SYNCBUFF);
1367                 IP_VS_DBG(7, "setting the maximum length of sync sending "
1368                           "message %d.\n", ipvs->send_mesg_maxlen);
1369         } else if (sync_state == IP_VS_STATE_BACKUP) {
1370                 dev = __dev_get_by_name(net, ipvs->backup_mcast_ifn);
1371                 if (!dev)
1372                         return -ENODEV;
1373
1374                 ipvs->recv_mesg_maxlen = dev->mtu -
1375                         sizeof(struct iphdr) - sizeof(struct udphdr);
1376                 IP_VS_DBG(7, "setting the maximum length of sync receiving "
1377                           "message %d.\n", ipvs->recv_mesg_maxlen);
1378         }
1379
1380         return 0;
1381 }
1382
1383
1384 /*
1385  *      Join a multicast group.
1386  *      the group is specified by a class D multicast address 224.0.0.0/8
1387  *      in the in_addr structure passed in as a parameter.
1388  */
1389 static int
1390 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
1391 {
1392         struct net *net = sock_net(sk);
1393         struct ip_mreqn mreq;
1394         struct net_device *dev;
1395         int ret;
1396
1397         memset(&mreq, 0, sizeof(mreq));
1398         memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
1399
1400         dev = __dev_get_by_name(net, ifname);
1401         if (!dev)
1402                 return -ENODEV;
1403         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1404                 return -EINVAL;
1405
1406         mreq.imr_ifindex = dev->ifindex;
1407
1408         rtnl_lock();
1409         lock_sock(sk);
1410         ret = ip_mc_join_group(sk, &mreq);
1411         release_sock(sk);
1412         rtnl_unlock();
1413
1414         return ret;
1415 }
1416
1417
1418 static int bind_mcastif_addr(struct socket *sock, char *ifname)
1419 {
1420         struct net *net = sock_net(sock->sk);
1421         struct net_device *dev;
1422         __be32 addr;
1423         struct sockaddr_in sin;
1424
1425         dev = __dev_get_by_name(net, ifname);
1426         if (!dev)
1427                 return -ENODEV;
1428
1429         addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
1430         if (!addr)
1431                 pr_err("You probably need to specify IP address on "
1432                        "multicast interface.\n");
1433
1434         IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
1435                   ifname, &addr);
1436
1437         /* Now bind the socket with the address of multicast interface */
1438         sin.sin_family       = AF_INET;
1439         sin.sin_addr.s_addr  = addr;
1440         sin.sin_port         = 0;
1441
1442         return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
1443 }
1444
1445 /*
1446  *      Set up sending multicast socket over UDP
1447  */
1448 static struct socket *make_send_sock(struct net *net, int id)
1449 {
1450         struct netns_ipvs *ipvs = net_ipvs(net);
1451         /* multicast addr */
1452         struct sockaddr_in mcast_addr = {
1453                 .sin_family             = AF_INET,
1454                 .sin_port               = cpu_to_be16(IP_VS_SYNC_PORT + id),
1455                 .sin_addr.s_addr        = cpu_to_be32(IP_VS_SYNC_GROUP),
1456         };
1457         struct socket *sock;
1458         int result;
1459
1460         /* First create a socket move it to right name space later */
1461         result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1462         if (result < 0) {
1463                 pr_err("Error during creation of socket; terminating\n");
1464                 return ERR_PTR(result);
1465         }
1466         /*
1467          * Kernel sockets that are a part of a namespace, should not
1468          * hold a reference to a namespace in order to allow to stop it.
1469          * After sk_change_net should be released using sk_release_kernel.
1470          */
1471         sk_change_net(sock->sk, net);
1472         result = set_mcast_if(sock->sk, ipvs->master_mcast_ifn);
1473         if (result < 0) {
1474                 pr_err("Error setting outbound mcast interface\n");
1475                 goto error;
1476         }
1477
1478         set_mcast_loop(sock->sk, 0);
1479         set_mcast_ttl(sock->sk, 1);
1480         result = sysctl_sync_sock_size(ipvs);
1481         if (result > 0)
1482                 set_sock_size(sock->sk, 1, result);
1483
1484         result = bind_mcastif_addr(sock, ipvs->master_mcast_ifn);
1485         if (result < 0) {
1486                 pr_err("Error binding address of the mcast interface\n");
1487                 goto error;
1488         }
1489
1490         result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr,
1491                         sizeof(struct sockaddr), 0);
1492         if (result < 0) {
1493                 pr_err("Error connecting to the multicast addr\n");
1494                 goto error;
1495         }
1496
1497         return sock;
1498
1499 error:
1500         sk_release_kernel(sock->sk);
1501         return ERR_PTR(result);
1502 }
1503
1504
1505 /*
1506  *      Set up receiving multicast socket over UDP
1507  */
1508 static struct socket *make_receive_sock(struct net *net, int id)
1509 {
1510         struct netns_ipvs *ipvs = net_ipvs(net);
1511         /* multicast addr */
1512         struct sockaddr_in mcast_addr = {
1513                 .sin_family             = AF_INET,
1514                 .sin_port               = cpu_to_be16(IP_VS_SYNC_PORT + id),
1515                 .sin_addr.s_addr        = cpu_to_be32(IP_VS_SYNC_GROUP),
1516         };
1517         struct socket *sock;
1518         int result;
1519
1520         /* First create a socket */
1521         result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1522         if (result < 0) {
1523                 pr_err("Error during creation of socket; terminating\n");
1524                 return ERR_PTR(result);
1525         }
1526         /*
1527          * Kernel sockets that are a part of a namespace, should not
1528          * hold a reference to a namespace in order to allow to stop it.
1529          * After sk_change_net should be released using sk_release_kernel.
1530          */
1531         sk_change_net(sock->sk, net);
1532         /* it is equivalent to the REUSEADDR option in user-space */
1533         sock->sk->sk_reuse = SK_CAN_REUSE;
1534         result = sysctl_sync_sock_size(ipvs);
1535         if (result > 0)
1536                 set_sock_size(sock->sk, 0, result);
1537
1538         result = sock->ops->bind(sock, (struct sockaddr *) &mcast_addr,
1539                         sizeof(struct sockaddr));
1540         if (result < 0) {
1541                 pr_err("Error binding to the multicast addr\n");
1542                 goto error;
1543         }
1544
1545         /* join the multicast group */
1546         result = join_mcast_group(sock->sk,
1547                         (struct in_addr *) &mcast_addr.sin_addr,
1548                         ipvs->backup_mcast_ifn);
1549         if (result < 0) {
1550                 pr_err("Error joining to the multicast group\n");
1551                 goto error;
1552         }
1553
1554         return sock;
1555
1556 error:
1557         sk_release_kernel(sock->sk);
1558         return ERR_PTR(result);
1559 }
1560
1561
1562 static int
1563 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
1564 {
1565         struct msghdr   msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
1566         struct kvec     iov;
1567         int             len;
1568
1569         EnterFunction(7);
1570         iov.iov_base     = (void *)buffer;
1571         iov.iov_len      = length;
1572
1573         len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
1574
1575         LeaveFunction(7);
1576         return len;
1577 }
1578
1579 static int
1580 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
1581 {
1582         int msize;
1583         int ret;
1584
1585         msize = ntohs(msg->size);
1586
1587         ret = ip_vs_send_async(sock, (char *)msg, msize);
1588         if (ret >= 0 || ret == -EAGAIN)
1589                 return ret;
1590         pr_err("ip_vs_send_async error %d\n", ret);
1591         return 0;
1592 }
1593
1594 static int
1595 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
1596 {
1597         struct msghdr           msg = {NULL,};
1598         struct kvec             iov;
1599         int                     len;
1600
1601         EnterFunction(7);
1602
1603         /* Receive a packet */
1604         iov.iov_base     = buffer;
1605         iov.iov_len      = (size_t)buflen;
1606
1607         len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, MSG_DONTWAIT);
1608
1609         if (len < 0)
1610                 return len;
1611
1612         LeaveFunction(7);
1613         return len;
1614 }
1615
1616 /* Wakeup the master thread for sending */
1617 static void master_wakeup_work_handler(struct work_struct *work)
1618 {
1619         struct ipvs_master_sync_state *ms =
1620                 container_of(work, struct ipvs_master_sync_state,
1621                              master_wakeup_work.work);
1622         struct netns_ipvs *ipvs = ms->ipvs;
1623
1624         spin_lock_bh(&ipvs->sync_lock);
1625         if (ms->sync_queue_len &&
1626             ms->sync_queue_delay < IPVS_SYNC_WAKEUP_RATE) {
1627                 ms->sync_queue_delay = IPVS_SYNC_WAKEUP_RATE;
1628                 wake_up_process(ms->master_thread);
1629         }
1630         spin_unlock_bh(&ipvs->sync_lock);
1631 }
1632
1633 /* Get next buffer to send */
1634 static inline struct ip_vs_sync_buff *
1635 next_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
1636 {
1637         struct ip_vs_sync_buff *sb;
1638
1639         sb = sb_dequeue(ipvs, ms);
1640         if (sb)
1641                 return sb;
1642         /* Do not delay entries in buffer for more than 2 seconds */
1643         return get_curr_sync_buff(ipvs, ms, IPVS_SYNC_FLUSH_TIME);
1644 }
1645
1646 static int sync_thread_master(void *data)
1647 {
1648         struct ip_vs_sync_thread_data *tinfo = data;
1649         struct netns_ipvs *ipvs = net_ipvs(tinfo->net);
1650         struct ipvs_master_sync_state *ms = &ipvs->ms[tinfo->id];
1651         struct sock *sk = tinfo->sock->sk;
1652         struct ip_vs_sync_buff *sb;
1653
1654         pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
1655                 "syncid = %d, id = %d\n",
1656                 ipvs->master_mcast_ifn, ipvs->master_syncid, tinfo->id);
1657
1658         for (;;) {
1659                 sb = next_sync_buff(ipvs, ms);
1660                 if (unlikely(kthread_should_stop()))
1661                         break;
1662                 if (!sb) {
1663                         schedule_timeout(IPVS_SYNC_CHECK_PERIOD);
1664                         continue;
1665                 }
1666                 while (ip_vs_send_sync_msg(tinfo->sock, sb->mesg) < 0) {
1667                         /* (Ab)use interruptible sleep to avoid increasing
1668                          * the load avg.
1669                          */
1670                         __wait_event_interruptible(*sk_sleep(sk),
1671                                                    sock_writeable(sk) ||
1672                                                    kthread_should_stop());
1673                         if (unlikely(kthread_should_stop()))
1674                                 goto done;
1675                 }
1676                 ip_vs_sync_buff_release(sb);
1677         }
1678
1679 done:
1680         __set_current_state(TASK_RUNNING);
1681         if (sb)
1682                 ip_vs_sync_buff_release(sb);
1683
1684         /* clean up the sync_buff queue */
1685         while ((sb = sb_dequeue(ipvs, ms)))
1686                 ip_vs_sync_buff_release(sb);
1687         __set_current_state(TASK_RUNNING);
1688
1689         /* clean up the current sync_buff */
1690         sb = get_curr_sync_buff(ipvs, ms, 0);
1691         if (sb)
1692                 ip_vs_sync_buff_release(sb);
1693
1694         /* release the sending multicast socket */
1695         sk_release_kernel(tinfo->sock->sk);
1696         kfree(tinfo);
1697
1698         return 0;
1699 }
1700
1701
1702 static int sync_thread_backup(void *data)
1703 {
1704         struct ip_vs_sync_thread_data *tinfo = data;
1705         struct netns_ipvs *ipvs = net_ipvs(tinfo->net);
1706         int len;
1707
1708         pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
1709                 "syncid = %d, id = %d\n",
1710                 ipvs->backup_mcast_ifn, ipvs->backup_syncid, tinfo->id);
1711
1712         while (!kthread_should_stop()) {
1713                 wait_event_interruptible(*sk_sleep(tinfo->sock->sk),
1714                          !skb_queue_empty(&tinfo->sock->sk->sk_receive_queue)
1715                          || kthread_should_stop());
1716
1717                 /* do we have data now? */
1718                 while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) {
1719                         len = ip_vs_receive(tinfo->sock, tinfo->buf,
1720                                         ipvs->recv_mesg_maxlen);
1721                         if (len <= 0) {
1722                                 if (len != -EAGAIN)
1723                                         pr_err("receiving message error\n");
1724                                 break;
1725                         }
1726
1727                         ip_vs_process_message(tinfo->net, tinfo->buf, len);
1728                 }
1729         }
1730
1731         /* release the sending multicast socket */
1732         sk_release_kernel(tinfo->sock->sk);
1733         kfree(tinfo->buf);
1734         kfree(tinfo);
1735
1736         return 0;
1737 }
1738
1739
1740 int start_sync_thread(struct net *net, int state, char *mcast_ifn, __u8 syncid)
1741 {
1742         struct ip_vs_sync_thread_data *tinfo;
1743         struct task_struct **array = NULL, *task;
1744         struct socket *sock;
1745         struct netns_ipvs *ipvs = net_ipvs(net);
1746         char *name;
1747         int (*threadfn)(void *data);
1748         int id, count;
1749         int result = -ENOMEM;
1750
1751         IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1752         IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n",
1753                   sizeof(struct ip_vs_sync_conn_v0));
1754
1755         if (!ipvs->sync_state) {
1756                 count = clamp(sysctl_sync_ports(ipvs), 1, IPVS_SYNC_PORTS_MAX);
1757                 ipvs->threads_mask = count - 1;
1758         } else
1759                 count = ipvs->threads_mask + 1;
1760
1761         if (state == IP_VS_STATE_MASTER) {
1762                 if (ipvs->ms)
1763                         return -EEXIST;
1764
1765                 strlcpy(ipvs->master_mcast_ifn, mcast_ifn,
1766                         sizeof(ipvs->master_mcast_ifn));
1767                 ipvs->master_syncid = syncid;
1768                 name = "ipvs-m:%d:%d";
1769                 threadfn = sync_thread_master;
1770         } else if (state == IP_VS_STATE_BACKUP) {
1771                 if (ipvs->backup_threads)
1772                         return -EEXIST;
1773
1774                 strlcpy(ipvs->backup_mcast_ifn, mcast_ifn,
1775                         sizeof(ipvs->backup_mcast_ifn));
1776                 ipvs->backup_syncid = syncid;
1777                 name = "ipvs-b:%d:%d";
1778                 threadfn = sync_thread_backup;
1779         } else {
1780                 return -EINVAL;
1781         }
1782
1783         if (state == IP_VS_STATE_MASTER) {
1784                 struct ipvs_master_sync_state *ms;
1785
1786                 ipvs->ms = kzalloc(count * sizeof(ipvs->ms[0]), GFP_KERNEL);
1787                 if (!ipvs->ms)
1788                         goto out;
1789                 ms = ipvs->ms;
1790                 for (id = 0; id < count; id++, ms++) {
1791                         INIT_LIST_HEAD(&ms->sync_queue);
1792                         ms->sync_queue_len = 0;
1793                         ms->sync_queue_delay = 0;
1794                         INIT_DELAYED_WORK(&ms->master_wakeup_work,
1795                                           master_wakeup_work_handler);
1796                         ms->ipvs = ipvs;
1797                 }
1798         } else {
1799                 array = kzalloc(count * sizeof(struct task_struct *),
1800                                 GFP_KERNEL);
1801                 if (!array)
1802                         goto out;
1803         }
1804         set_sync_mesg_maxlen(net, state);
1805
1806         tinfo = NULL;
1807         for (id = 0; id < count; id++) {
1808                 if (state == IP_VS_STATE_MASTER)
1809                         sock = make_send_sock(net, id);
1810                 else
1811                         sock = make_receive_sock(net, id);
1812                 if (IS_ERR(sock)) {
1813                         result = PTR_ERR(sock);
1814                         goto outtinfo;
1815                 }
1816                 tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL);
1817                 if (!tinfo)
1818                         goto outsocket;
1819                 tinfo->net = net;
1820                 tinfo->sock = sock;
1821                 if (state == IP_VS_STATE_BACKUP) {
1822                         tinfo->buf = kmalloc(ipvs->recv_mesg_maxlen,
1823                                              GFP_KERNEL);
1824                         if (!tinfo->buf)
1825                                 goto outtinfo;
1826                 } else {
1827                         tinfo->buf = NULL;
1828                 }
1829                 tinfo->id = id;
1830
1831                 task = kthread_run(threadfn, tinfo, name, ipvs->gen, id);
1832                 if (IS_ERR(task)) {
1833                         result = PTR_ERR(task);
1834                         goto outtinfo;
1835                 }
1836                 tinfo = NULL;
1837                 if (state == IP_VS_STATE_MASTER)
1838                         ipvs->ms[id].master_thread = task;
1839                 else
1840                         array[id] = task;
1841         }
1842
1843         /* mark as active */
1844
1845         if (state == IP_VS_STATE_BACKUP)
1846                 ipvs->backup_threads = array;
1847         spin_lock_bh(&ipvs->sync_buff_lock);
1848         ipvs->sync_state |= state;
1849         spin_unlock_bh(&ipvs->sync_buff_lock);
1850
1851         /* increase the module use count */
1852         ip_vs_use_count_inc();
1853
1854         return 0;
1855
1856 outsocket:
1857         sk_release_kernel(sock->sk);
1858
1859 outtinfo:
1860         if (tinfo) {
1861                 sk_release_kernel(tinfo->sock->sk);
1862                 kfree(tinfo->buf);
1863                 kfree(tinfo);
1864         }
1865         count = id;
1866         while (count-- > 0) {
1867                 if (state == IP_VS_STATE_MASTER)
1868                         kthread_stop(ipvs->ms[count].master_thread);
1869                 else
1870                         kthread_stop(array[count]);
1871         }
1872         kfree(array);
1873
1874 out:
1875         if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
1876                 kfree(ipvs->ms);
1877                 ipvs->ms = NULL;
1878         }
1879         return result;
1880 }
1881
1882
1883 int stop_sync_thread(struct net *net, int state)
1884 {
1885         struct netns_ipvs *ipvs = net_ipvs(net);
1886         struct task_struct **array;
1887         int id;
1888         int retc = -EINVAL;
1889
1890         IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1891
1892         if (state == IP_VS_STATE_MASTER) {
1893                 if (!ipvs->ms)
1894                         return -ESRCH;
1895
1896                 /*
1897                  * The lock synchronizes with sb_queue_tail(), so that we don't
1898                  * add sync buffers to the queue, when we are already in
1899                  * progress of stopping the master sync daemon.
1900                  */
1901
1902                 spin_lock_bh(&ipvs->sync_buff_lock);
1903                 spin_lock(&ipvs->sync_lock);
1904                 ipvs->sync_state &= ~IP_VS_STATE_MASTER;
1905                 spin_unlock(&ipvs->sync_lock);
1906                 spin_unlock_bh(&ipvs->sync_buff_lock);
1907
1908                 retc = 0;
1909                 for (id = ipvs->threads_mask; id >= 0; id--) {
1910                         struct ipvs_master_sync_state *ms = &ipvs->ms[id];
1911                         int ret;
1912
1913                         pr_info("stopping master sync thread %d ...\n",
1914                                 task_pid_nr(ms->master_thread));
1915                         cancel_delayed_work_sync(&ms->master_wakeup_work);
1916                         ret = kthread_stop(ms->master_thread);
1917                         if (retc >= 0)
1918                                 retc = ret;
1919                 }
1920                 kfree(ipvs->ms);
1921                 ipvs->ms = NULL;
1922         } else if (state == IP_VS_STATE_BACKUP) {
1923                 if (!ipvs->backup_threads)
1924                         return -ESRCH;
1925
1926                 ipvs->sync_state &= ~IP_VS_STATE_BACKUP;
1927                 array = ipvs->backup_threads;
1928                 retc = 0;
1929                 for (id = ipvs->threads_mask; id >= 0; id--) {
1930                         int ret;
1931
1932                         pr_info("stopping backup sync thread %d ...\n",
1933                                 task_pid_nr(array[id]));
1934                         ret = kthread_stop(array[id]);
1935                         if (retc >= 0)
1936                                 retc = ret;
1937                 }
1938                 kfree(array);
1939                 ipvs->backup_threads = NULL;
1940         }
1941
1942         /* decrease the module use count */
1943         ip_vs_use_count_dec();
1944
1945         return retc;
1946 }
1947
1948 /*
1949  * Initialize data struct for each netns
1950  */
1951 int __net_init ip_vs_sync_net_init(struct net *net)
1952 {
1953         struct netns_ipvs *ipvs = net_ipvs(net);
1954
1955         __mutex_init(&ipvs->sync_mutex, "ipvs->sync_mutex", &__ipvs_sync_key);
1956         spin_lock_init(&ipvs->sync_lock);
1957         spin_lock_init(&ipvs->sync_buff_lock);
1958         return 0;
1959 }
1960
1961 void ip_vs_sync_net_cleanup(struct net *net)
1962 {
1963         int retc;
1964         struct netns_ipvs *ipvs = net_ipvs(net);
1965
1966         mutex_lock(&ipvs->sync_mutex);
1967         retc = stop_sync_thread(net, IP_VS_STATE_MASTER);
1968         if (retc && retc != -ESRCH)
1969                 pr_err("Failed to stop Master Daemon\n");
1970
1971         retc = stop_sync_thread(net, IP_VS_STATE_BACKUP);
1972         if (retc && retc != -ESRCH)
1973                 pr_err("Failed to stop Backup Daemon\n");
1974         mutex_unlock(&ipvs->sync_mutex);
1975 }