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