net: replace NIPQUAD() in net/netfilter/
[linux-2.6.git] / net / netfilter / ipvs / ip_vs_sync.c
1 /*
2  * IPVS         An implementation of the IP virtual server support for the
3  *              LINUX operating system.  IPVS is now implemented as a module
4  *              over the NetFilter framework. IPVS can be used to build a
5  *              high-performance and highly available server based on a
6  *              cluster of servers.
7  *
8  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
9  *
10  * ip_vs_sync:  sync connection info from master load balancer to backups
11  *              through multicast
12  *
13  * Changes:
14  *      Alexandre Cassen        :       Added master & backup support at a time.
15  *      Alexandre Cassen        :       Added SyncID support for incoming sync
16  *                                      messages filtering.
17  *      Justin Ossevoort        :       Fix endian problem on sync message size.
18  */
19
20 #include <linux/module.h>
21 #include <linux/slab.h>
22 #include <linux/inetdevice.h>
23 #include <linux/net.h>
24 #include <linux/completion.h>
25 #include <linux/delay.h>
26 #include <linux/skbuff.h>
27 #include <linux/in.h>
28 #include <linux/igmp.h>                 /* for ip_mc_join_group */
29 #include <linux/udp.h>
30 #include <linux/err.h>
31 #include <linux/kthread.h>
32 #include <linux/wait.h>
33 #include <linux/kernel.h>
34
35 #include <net/ip.h>
36 #include <net/sock.h>
37
38 #include <net/ip_vs.h>
39
40 #define IP_VS_SYNC_GROUP 0xe0000051    /* multicast addr - 224.0.0.81 */
41 #define IP_VS_SYNC_PORT  8848          /* multicast port */
42
43
44 /*
45  *      IPVS sync connection entry
46  */
47 struct ip_vs_sync_conn {
48         __u8                    reserved;
49
50         /* Protocol, addresses and port numbers */
51         __u8                    protocol;       /* Which protocol (TCP/UDP) */
52         __be16                  cport;
53         __be16                  vport;
54         __be16                  dport;
55         __be32                  caddr;          /* client address */
56         __be32                  vaddr;          /* virtual address */
57         __be32                  daddr;          /* destination address */
58
59         /* Flags and state transition */
60         __be16                  flags;          /* status flags */
61         __be16                  state;          /* state info */
62
63         /* The sequence options start here */
64 };
65
66 struct ip_vs_sync_conn_options {
67         struct ip_vs_seq        in_seq;         /* incoming seq. struct */
68         struct ip_vs_seq        out_seq;        /* outgoing seq. struct */
69 };
70
71 struct ip_vs_sync_thread_data {
72         struct socket *sock;
73         char *buf;
74 };
75
76 #define SIMPLE_CONN_SIZE  (sizeof(struct ip_vs_sync_conn))
77 #define FULL_CONN_SIZE  \
78 (sizeof(struct ip_vs_sync_conn) + sizeof(struct ip_vs_sync_conn_options))
79
80
81 /*
82   The master mulitcasts messages to the backup load balancers in the
83   following format.
84
85        0                   1                   2                   3
86        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
87       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
88       |  Count Conns  |    SyncID     |            Size               |
89       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
90       |                                                               |
91       |                    IPVS Sync Connection (1)                   |
92       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
93       |                            .                                  |
94       |                            .                                  |
95       |                            .                                  |
96       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
97       |                                                               |
98       |                    IPVS Sync Connection (n)                   |
99       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
100 */
101
102 #define SYNC_MESG_HEADER_LEN    4
103 #define MAX_CONNS_PER_SYNCBUFF  255 /* nr_conns in ip_vs_sync_mesg is 8 bit */
104
105 struct ip_vs_sync_mesg {
106         __u8                    nr_conns;
107         __u8                    syncid;
108         __u16                   size;
109
110         /* ip_vs_sync_conn entries start here */
111 };
112
113 /* the maximum length of sync (sending/receiving) message */
114 static int sync_send_mesg_maxlen;
115 static int sync_recv_mesg_maxlen;
116
117 struct ip_vs_sync_buff {
118         struct list_head        list;
119         unsigned long           firstuse;
120
121         /* pointers for the message data */
122         struct ip_vs_sync_mesg  *mesg;
123         unsigned char           *head;
124         unsigned char           *end;
125 };
126
127
128 /* the sync_buff list head and the lock */
129 static LIST_HEAD(ip_vs_sync_queue);
130 static DEFINE_SPINLOCK(ip_vs_sync_lock);
131
132 /* current sync_buff for accepting new conn entries */
133 static struct ip_vs_sync_buff   *curr_sb = NULL;
134 static DEFINE_SPINLOCK(curr_sb_lock);
135
136 /* ipvs sync daemon state */
137 volatile int ip_vs_sync_state = IP_VS_STATE_NONE;
138 volatile int ip_vs_master_syncid = 0;
139 volatile int ip_vs_backup_syncid = 0;
140
141 /* multicast interface name */
142 char ip_vs_master_mcast_ifn[IP_VS_IFNAME_MAXLEN];
143 char ip_vs_backup_mcast_ifn[IP_VS_IFNAME_MAXLEN];
144
145 /* sync daemon tasks */
146 static struct task_struct *sync_master_thread;
147 static struct task_struct *sync_backup_thread;
148
149 /* multicast addr */
150 static struct sockaddr_in mcast_addr = {
151         .sin_family             = AF_INET,
152         .sin_port               = __constant_htons(IP_VS_SYNC_PORT),
153         .sin_addr.s_addr        = __constant_htonl(IP_VS_SYNC_GROUP),
154 };
155
156
157 static inline struct ip_vs_sync_buff *sb_dequeue(void)
158 {
159         struct ip_vs_sync_buff *sb;
160
161         spin_lock_bh(&ip_vs_sync_lock);
162         if (list_empty(&ip_vs_sync_queue)) {
163                 sb = NULL;
164         } else {
165                 sb = list_entry(ip_vs_sync_queue.next,
166                                 struct ip_vs_sync_buff,
167                                 list);
168                 list_del(&sb->list);
169         }
170         spin_unlock_bh(&ip_vs_sync_lock);
171
172         return sb;
173 }
174
175 static inline struct ip_vs_sync_buff * ip_vs_sync_buff_create(void)
176 {
177         struct ip_vs_sync_buff *sb;
178
179         if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
180                 return NULL;
181
182         if (!(sb->mesg=kmalloc(sync_send_mesg_maxlen, GFP_ATOMIC))) {
183                 kfree(sb);
184                 return NULL;
185         }
186         sb->mesg->nr_conns = 0;
187         sb->mesg->syncid = ip_vs_master_syncid;
188         sb->mesg->size = 4;
189         sb->head = (unsigned char *)sb->mesg + 4;
190         sb->end = (unsigned char *)sb->mesg + sync_send_mesg_maxlen;
191         sb->firstuse = jiffies;
192         return sb;
193 }
194
195 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
196 {
197         kfree(sb->mesg);
198         kfree(sb);
199 }
200
201 static inline void sb_queue_tail(struct ip_vs_sync_buff *sb)
202 {
203         spin_lock(&ip_vs_sync_lock);
204         if (ip_vs_sync_state & IP_VS_STATE_MASTER)
205                 list_add_tail(&sb->list, &ip_vs_sync_queue);
206         else
207                 ip_vs_sync_buff_release(sb);
208         spin_unlock(&ip_vs_sync_lock);
209 }
210
211 /*
212  *      Get the current sync buffer if it has been created for more
213  *      than the specified time or the specified time is zero.
214  */
215 static inline struct ip_vs_sync_buff *
216 get_curr_sync_buff(unsigned long time)
217 {
218         struct ip_vs_sync_buff *sb;
219
220         spin_lock_bh(&curr_sb_lock);
221         if (curr_sb && (time == 0 ||
222                         time_before(jiffies - curr_sb->firstuse, time))) {
223                 sb = curr_sb;
224                 curr_sb = NULL;
225         } else
226                 sb = NULL;
227         spin_unlock_bh(&curr_sb_lock);
228         return sb;
229 }
230
231
232 /*
233  *      Add an ip_vs_conn information into the current sync_buff.
234  *      Called by ip_vs_in.
235  */
236 void ip_vs_sync_conn(struct ip_vs_conn *cp)
237 {
238         struct ip_vs_sync_mesg *m;
239         struct ip_vs_sync_conn *s;
240         int len;
241
242         spin_lock(&curr_sb_lock);
243         if (!curr_sb) {
244                 if (!(curr_sb=ip_vs_sync_buff_create())) {
245                         spin_unlock(&curr_sb_lock);
246                         IP_VS_ERR("ip_vs_sync_buff_create failed.\n");
247                         return;
248                 }
249         }
250
251         len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
252                 SIMPLE_CONN_SIZE;
253         m = curr_sb->mesg;
254         s = (struct ip_vs_sync_conn *)curr_sb->head;
255
256         /* copy members */
257         s->protocol = cp->protocol;
258         s->cport = cp->cport;
259         s->vport = cp->vport;
260         s->dport = cp->dport;
261         s->caddr = cp->caddr.ip;
262         s->vaddr = cp->vaddr.ip;
263         s->daddr = cp->daddr.ip;
264         s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
265         s->state = htons(cp->state);
266         if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
267                 struct ip_vs_sync_conn_options *opt =
268                         (struct ip_vs_sync_conn_options *)&s[1];
269                 memcpy(opt, &cp->in_seq, sizeof(*opt));
270         }
271
272         m->nr_conns++;
273         m->size += len;
274         curr_sb->head += len;
275
276         /* check if there is a space for next one */
277         if (curr_sb->head+FULL_CONN_SIZE > curr_sb->end) {
278                 sb_queue_tail(curr_sb);
279                 curr_sb = NULL;
280         }
281         spin_unlock(&curr_sb_lock);
282
283         /* synchronize its controller if it has */
284         if (cp->control)
285                 ip_vs_sync_conn(cp->control);
286 }
287
288
289 /*
290  *      Process received multicast message and create the corresponding
291  *      ip_vs_conn entries.
292  */
293 static void ip_vs_process_message(const char *buffer, const size_t buflen)
294 {
295         struct ip_vs_sync_mesg *m = (struct ip_vs_sync_mesg *)buffer;
296         struct ip_vs_sync_conn *s;
297         struct ip_vs_sync_conn_options *opt;
298         struct ip_vs_conn *cp;
299         struct ip_vs_protocol *pp;
300         struct ip_vs_dest *dest;
301         char *p;
302         int i;
303
304         if (buflen < sizeof(struct ip_vs_sync_mesg)) {
305                 IP_VS_ERR_RL("sync message header too short\n");
306                 return;
307         }
308
309         /* Convert size back to host byte order */
310         m->size = ntohs(m->size);
311
312         if (buflen != m->size) {
313                 IP_VS_ERR_RL("bogus sync message size\n");
314                 return;
315         }
316
317         /* SyncID sanity check */
318         if (ip_vs_backup_syncid != 0 && m->syncid != ip_vs_backup_syncid) {
319                 IP_VS_DBG(7, "Ignoring incoming msg with syncid = %d\n",
320                           m->syncid);
321                 return;
322         }
323
324         p = (char *)buffer + sizeof(struct ip_vs_sync_mesg);
325         for (i=0; i<m->nr_conns; i++) {
326                 unsigned flags, state;
327
328                 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
329                         IP_VS_ERR_RL("bogus conn in sync message\n");
330                         return;
331                 }
332                 s = (struct ip_vs_sync_conn *) p;
333                 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
334                 flags &= ~IP_VS_CONN_F_HASHED;
335                 if (flags & IP_VS_CONN_F_SEQ_MASK) {
336                         opt = (struct ip_vs_sync_conn_options *)&s[1];
337                         p += FULL_CONN_SIZE;
338                         if (p > buffer+buflen) {
339                                 IP_VS_ERR_RL("bogus conn options in sync message\n");
340                                 return;
341                         }
342                 } else {
343                         opt = NULL;
344                         p += SIMPLE_CONN_SIZE;
345                 }
346
347                 state = ntohs(s->state);
348                 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
349                         pp = ip_vs_proto_get(s->protocol);
350                         if (!pp) {
351                                 IP_VS_ERR_RL("Unsupported protocol %u in sync msg\n",
352                                         s->protocol);
353                                 continue;
354                         }
355                         if (state >= pp->num_states) {
356                                 IP_VS_DBG(2, "Invalid %s state %u in sync msg\n",
357                                         pp->name, state);
358                                 continue;
359                         }
360                 } else {
361                         /* protocol in templates is not used for state/timeout */
362                         pp = NULL;
363                         if (state > 0) {
364                                 IP_VS_DBG(2, "Invalid template state %u in sync msg\n",
365                                         state);
366                                 state = 0;
367                         }
368                 }
369
370                 if (!(flags & IP_VS_CONN_F_TEMPLATE))
371                         cp = ip_vs_conn_in_get(AF_INET, s->protocol,
372                                                (union nf_inet_addr *)&s->caddr,
373                                                s->cport,
374                                                (union nf_inet_addr *)&s->vaddr,
375                                                s->vport);
376                 else
377                         cp = ip_vs_ct_in_get(AF_INET, s->protocol,
378                                              (union nf_inet_addr *)&s->caddr,
379                                              s->cport,
380                                              (union nf_inet_addr *)&s->vaddr,
381                                              s->vport);
382                 if (!cp) {
383                         /*
384                          * Find the appropriate destination for the connection.
385                          * If it is not found the connection will remain unbound
386                          * but still handled.
387                          */
388                         dest = ip_vs_find_dest(AF_INET,
389                                                (union nf_inet_addr *)&s->daddr,
390                                                s->dport,
391                                                (union nf_inet_addr *)&s->vaddr,
392                                                s->vport,
393                                                s->protocol);
394                         /*  Set the approprite ativity flag */
395                         if (s->protocol == IPPROTO_TCP) {
396                                 if (state != IP_VS_TCP_S_ESTABLISHED)
397                                         flags |= IP_VS_CONN_F_INACTIVE;
398                                 else
399                                         flags &= ~IP_VS_CONN_F_INACTIVE;
400                         }
401                         cp = ip_vs_conn_new(AF_INET, s->protocol,
402                                             (union nf_inet_addr *)&s->caddr,
403                                             s->cport,
404                                             (union nf_inet_addr *)&s->vaddr,
405                                             s->vport,
406                                             (union nf_inet_addr *)&s->daddr,
407                                             s->dport,
408                                             flags, dest);
409                         if (dest)
410                                 atomic_dec(&dest->refcnt);
411                         if (!cp) {
412                                 IP_VS_ERR("ip_vs_conn_new failed\n");
413                                 return;
414                         }
415                 } else if (!cp->dest) {
416                         dest = ip_vs_try_bind_dest(cp);
417                         if (dest)
418                                 atomic_dec(&dest->refcnt);
419                 } else if ((cp->dest) && (cp->protocol == IPPROTO_TCP) &&
420                            (cp->state != state)) {
421                         /* update active/inactive flag for the connection */
422                         dest = cp->dest;
423                         if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
424                                 (state != IP_VS_TCP_S_ESTABLISHED)) {
425                                 atomic_dec(&dest->activeconns);
426                                 atomic_inc(&dest->inactconns);
427                                 cp->flags |= IP_VS_CONN_F_INACTIVE;
428                         } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) &&
429                                 (state == IP_VS_TCP_S_ESTABLISHED)) {
430                                 atomic_inc(&dest->activeconns);
431                                 atomic_dec(&dest->inactconns);
432                                 cp->flags &= ~IP_VS_CONN_F_INACTIVE;
433                         }
434                 }
435
436                 if (opt)
437                         memcpy(&cp->in_seq, opt, sizeof(*opt));
438                 atomic_set(&cp->in_pkts, sysctl_ip_vs_sync_threshold[0]);
439                 cp->state = state;
440                 cp->old_state = cp->state;
441                 /*
442                  * We can not recover the right timeout for templates
443                  * in all cases, we can not find the right fwmark
444                  * virtual service. If needed, we can do it for
445                  * non-fwmark persistent services.
446                  */
447                 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pp->timeout_table)
448                         cp->timeout = pp->timeout_table[state];
449                 else
450                         cp->timeout = (3*60*HZ);
451                 ip_vs_conn_put(cp);
452         }
453 }
454
455
456 /*
457  *      Setup loopback of outgoing multicasts on a sending socket
458  */
459 static void set_mcast_loop(struct sock *sk, u_char loop)
460 {
461         struct inet_sock *inet = inet_sk(sk);
462
463         /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
464         lock_sock(sk);
465         inet->mc_loop = loop ? 1 : 0;
466         release_sock(sk);
467 }
468
469 /*
470  *      Specify TTL for outgoing multicasts on a sending socket
471  */
472 static void set_mcast_ttl(struct sock *sk, u_char ttl)
473 {
474         struct inet_sock *inet = inet_sk(sk);
475
476         /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
477         lock_sock(sk);
478         inet->mc_ttl = ttl;
479         release_sock(sk);
480 }
481
482 /*
483  *      Specifiy default interface for outgoing multicasts
484  */
485 static int set_mcast_if(struct sock *sk, char *ifname)
486 {
487         struct net_device *dev;
488         struct inet_sock *inet = inet_sk(sk);
489
490         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
491                 return -ENODEV;
492
493         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
494                 return -EINVAL;
495
496         lock_sock(sk);
497         inet->mc_index = dev->ifindex;
498         /*  inet->mc_addr  = 0; */
499         release_sock(sk);
500
501         return 0;
502 }
503
504
505 /*
506  *      Set the maximum length of sync message according to the
507  *      specified interface's MTU.
508  */
509 static int set_sync_mesg_maxlen(int sync_state)
510 {
511         struct net_device *dev;
512         int num;
513
514         if (sync_state == IP_VS_STATE_MASTER) {
515                 if ((dev = __dev_get_by_name(&init_net, ip_vs_master_mcast_ifn)) == NULL)
516                         return -ENODEV;
517
518                 num = (dev->mtu - sizeof(struct iphdr) -
519                        sizeof(struct udphdr) -
520                        SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE;
521                 sync_send_mesg_maxlen = SYNC_MESG_HEADER_LEN +
522                         SIMPLE_CONN_SIZE * min(num, MAX_CONNS_PER_SYNCBUFF);
523                 IP_VS_DBG(7, "setting the maximum length of sync sending "
524                           "message %d.\n", sync_send_mesg_maxlen);
525         } else if (sync_state == IP_VS_STATE_BACKUP) {
526                 if ((dev = __dev_get_by_name(&init_net, ip_vs_backup_mcast_ifn)) == NULL)
527                         return -ENODEV;
528
529                 sync_recv_mesg_maxlen = dev->mtu -
530                         sizeof(struct iphdr) - sizeof(struct udphdr);
531                 IP_VS_DBG(7, "setting the maximum length of sync receiving "
532                           "message %d.\n", sync_recv_mesg_maxlen);
533         }
534
535         return 0;
536 }
537
538
539 /*
540  *      Join a multicast group.
541  *      the group is specified by a class D multicast address 224.0.0.0/8
542  *      in the in_addr structure passed in as a parameter.
543  */
544 static int
545 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
546 {
547         struct ip_mreqn mreq;
548         struct net_device *dev;
549         int ret;
550
551         memset(&mreq, 0, sizeof(mreq));
552         memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
553
554         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
555                 return -ENODEV;
556         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
557                 return -EINVAL;
558
559         mreq.imr_ifindex = dev->ifindex;
560
561         lock_sock(sk);
562         ret = ip_mc_join_group(sk, &mreq);
563         release_sock(sk);
564
565         return ret;
566 }
567
568
569 static int bind_mcastif_addr(struct socket *sock, char *ifname)
570 {
571         struct net_device *dev;
572         __be32 addr;
573         struct sockaddr_in sin;
574
575         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
576                 return -ENODEV;
577
578         addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
579         if (!addr)
580                 IP_VS_ERR("You probably need to specify IP address on "
581                           "multicast interface.\n");
582
583         IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
584                   ifname, &addr);
585
586         /* Now bind the socket with the address of multicast interface */
587         sin.sin_family       = AF_INET;
588         sin.sin_addr.s_addr  = addr;
589         sin.sin_port         = 0;
590
591         return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
592 }
593
594 /*
595  *      Set up sending multicast socket over UDP
596  */
597 static struct socket * make_send_sock(void)
598 {
599         struct socket *sock;
600         int result;
601
602         /* First create a socket */
603         result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
604         if (result < 0) {
605                 IP_VS_ERR("Error during creation of socket; terminating\n");
606                 return ERR_PTR(result);
607         }
608
609         result = set_mcast_if(sock->sk, ip_vs_master_mcast_ifn);
610         if (result < 0) {
611                 IP_VS_ERR("Error setting outbound mcast interface\n");
612                 goto error;
613         }
614
615         set_mcast_loop(sock->sk, 0);
616         set_mcast_ttl(sock->sk, 1);
617
618         result = bind_mcastif_addr(sock, ip_vs_master_mcast_ifn);
619         if (result < 0) {
620                 IP_VS_ERR("Error binding address of the mcast interface\n");
621                 goto error;
622         }
623
624         result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr,
625                         sizeof(struct sockaddr), 0);
626         if (result < 0) {
627                 IP_VS_ERR("Error connecting to the multicast addr\n");
628                 goto error;
629         }
630
631         return sock;
632
633   error:
634         sock_release(sock);
635         return ERR_PTR(result);
636 }
637
638
639 /*
640  *      Set up receiving multicast socket over UDP
641  */
642 static struct socket * make_receive_sock(void)
643 {
644         struct socket *sock;
645         int result;
646
647         /* First create a socket */
648         result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
649         if (result < 0) {
650                 IP_VS_ERR("Error during creation of socket; terminating\n");
651                 return ERR_PTR(result);
652         }
653
654         /* it is equivalent to the REUSEADDR option in user-space */
655         sock->sk->sk_reuse = 1;
656
657         result = sock->ops->bind(sock, (struct sockaddr *) &mcast_addr,
658                         sizeof(struct sockaddr));
659         if (result < 0) {
660                 IP_VS_ERR("Error binding to the multicast addr\n");
661                 goto error;
662         }
663
664         /* join the multicast group */
665         result = join_mcast_group(sock->sk,
666                         (struct in_addr *) &mcast_addr.sin_addr,
667                         ip_vs_backup_mcast_ifn);
668         if (result < 0) {
669                 IP_VS_ERR("Error joining to the multicast group\n");
670                 goto error;
671         }
672
673         return sock;
674
675   error:
676         sock_release(sock);
677         return ERR_PTR(result);
678 }
679
680
681 static int
682 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
683 {
684         struct msghdr   msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
685         struct kvec     iov;
686         int             len;
687
688         EnterFunction(7);
689         iov.iov_base     = (void *)buffer;
690         iov.iov_len      = length;
691
692         len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
693
694         LeaveFunction(7);
695         return len;
696 }
697
698 static void
699 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
700 {
701         int msize;
702
703         msize = msg->size;
704
705         /* Put size in network byte order */
706         msg->size = htons(msg->size);
707
708         if (ip_vs_send_async(sock, (char *)msg, msize) != msize)
709                 IP_VS_ERR("ip_vs_send_async error\n");
710 }
711
712 static int
713 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
714 {
715         struct msghdr           msg = {NULL,};
716         struct kvec             iov;
717         int                     len;
718
719         EnterFunction(7);
720
721         /* Receive a packet */
722         iov.iov_base     = buffer;
723         iov.iov_len      = (size_t)buflen;
724
725         len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, 0);
726
727         if (len < 0)
728                 return -1;
729
730         LeaveFunction(7);
731         return len;
732 }
733
734
735 static int sync_thread_master(void *data)
736 {
737         struct ip_vs_sync_thread_data *tinfo = data;
738         struct ip_vs_sync_buff *sb;
739
740         IP_VS_INFO("sync thread started: state = MASTER, mcast_ifn = %s, "
741                    "syncid = %d\n",
742                    ip_vs_master_mcast_ifn, ip_vs_master_syncid);
743
744         while (!kthread_should_stop()) {
745                 while ((sb = sb_dequeue())) {
746                         ip_vs_send_sync_msg(tinfo->sock, sb->mesg);
747                         ip_vs_sync_buff_release(sb);
748                 }
749
750                 /* check if entries stay in curr_sb for 2 seconds */
751                 sb = get_curr_sync_buff(2 * HZ);
752                 if (sb) {
753                         ip_vs_send_sync_msg(tinfo->sock, sb->mesg);
754                         ip_vs_sync_buff_release(sb);
755                 }
756
757                 schedule_timeout_interruptible(HZ);
758         }
759
760         /* clean up the sync_buff queue */
761         while ((sb=sb_dequeue())) {
762                 ip_vs_sync_buff_release(sb);
763         }
764
765         /* clean up the current sync_buff */
766         if ((sb = get_curr_sync_buff(0))) {
767                 ip_vs_sync_buff_release(sb);
768         }
769
770         /* release the sending multicast socket */
771         sock_release(tinfo->sock);
772         kfree(tinfo);
773
774         return 0;
775 }
776
777
778 static int sync_thread_backup(void *data)
779 {
780         struct ip_vs_sync_thread_data *tinfo = data;
781         int len;
782
783         IP_VS_INFO("sync thread started: state = BACKUP, mcast_ifn = %s, "
784                    "syncid = %d\n",
785                    ip_vs_backup_mcast_ifn, ip_vs_backup_syncid);
786
787         while (!kthread_should_stop()) {
788                 wait_event_interruptible(*tinfo->sock->sk->sk_sleep,
789                          !skb_queue_empty(&tinfo->sock->sk->sk_receive_queue)
790                          || kthread_should_stop());
791
792                 /* do we have data now? */
793                 while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) {
794                         len = ip_vs_receive(tinfo->sock, tinfo->buf,
795                                         sync_recv_mesg_maxlen);
796                         if (len <= 0) {
797                                 IP_VS_ERR("receiving message error\n");
798                                 break;
799                         }
800
801                         /* disable bottom half, because it accesses the data
802                            shared by softirq while getting/creating conns */
803                         local_bh_disable();
804                         ip_vs_process_message(tinfo->buf, len);
805                         local_bh_enable();
806                 }
807         }
808
809         /* release the sending multicast socket */
810         sock_release(tinfo->sock);
811         kfree(tinfo->buf);
812         kfree(tinfo);
813
814         return 0;
815 }
816
817
818 int start_sync_thread(int state, char *mcast_ifn, __u8 syncid)
819 {
820         struct ip_vs_sync_thread_data *tinfo;
821         struct task_struct **realtask, *task;
822         struct socket *sock;
823         char *name, *buf = NULL;
824         int (*threadfn)(void *data);
825         int result = -ENOMEM;
826
827         IP_VS_DBG(7, "%s: pid %d\n", __func__, task_pid_nr(current));
828         IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n",
829                   sizeof(struct ip_vs_sync_conn));
830
831         if (state == IP_VS_STATE_MASTER) {
832                 if (sync_master_thread)
833                         return -EEXIST;
834
835                 strlcpy(ip_vs_master_mcast_ifn, mcast_ifn,
836                         sizeof(ip_vs_master_mcast_ifn));
837                 ip_vs_master_syncid = syncid;
838                 realtask = &sync_master_thread;
839                 name = "ipvs_syncmaster";
840                 threadfn = sync_thread_master;
841                 sock = make_send_sock();
842         } else if (state == IP_VS_STATE_BACKUP) {
843                 if (sync_backup_thread)
844                         return -EEXIST;
845
846                 strlcpy(ip_vs_backup_mcast_ifn, mcast_ifn,
847                         sizeof(ip_vs_backup_mcast_ifn));
848                 ip_vs_backup_syncid = syncid;
849                 realtask = &sync_backup_thread;
850                 name = "ipvs_syncbackup";
851                 threadfn = sync_thread_backup;
852                 sock = make_receive_sock();
853         } else {
854                 return -EINVAL;
855         }
856
857         if (IS_ERR(sock)) {
858                 result = PTR_ERR(sock);
859                 goto out;
860         }
861
862         set_sync_mesg_maxlen(state);
863         if (state == IP_VS_STATE_BACKUP) {
864                 buf = kmalloc(sync_recv_mesg_maxlen, GFP_KERNEL);
865                 if (!buf)
866                         goto outsocket;
867         }
868
869         tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL);
870         if (!tinfo)
871                 goto outbuf;
872
873         tinfo->sock = sock;
874         tinfo->buf = buf;
875
876         task = kthread_run(threadfn, tinfo, name);
877         if (IS_ERR(task)) {
878                 result = PTR_ERR(task);
879                 goto outtinfo;
880         }
881
882         /* mark as active */
883         *realtask = task;
884         ip_vs_sync_state |= state;
885
886         /* increase the module use count */
887         ip_vs_use_count_inc();
888
889         return 0;
890
891 outtinfo:
892         kfree(tinfo);
893 outbuf:
894         kfree(buf);
895 outsocket:
896         sock_release(sock);
897 out:
898         return result;
899 }
900
901
902 int stop_sync_thread(int state)
903 {
904         IP_VS_DBG(7, "%s: pid %d\n", __func__, task_pid_nr(current));
905
906         if (state == IP_VS_STATE_MASTER) {
907                 if (!sync_master_thread)
908                         return -ESRCH;
909
910                 IP_VS_INFO("stopping master sync thread %d ...\n",
911                            task_pid_nr(sync_master_thread));
912
913                 /*
914                  * The lock synchronizes with sb_queue_tail(), so that we don't
915                  * add sync buffers to the queue, when we are already in
916                  * progress of stopping the master sync daemon.
917                  */
918
919                 spin_lock_bh(&ip_vs_sync_lock);
920                 ip_vs_sync_state &= ~IP_VS_STATE_MASTER;
921                 spin_unlock_bh(&ip_vs_sync_lock);
922                 kthread_stop(sync_master_thread);
923                 sync_master_thread = NULL;
924         } else if (state == IP_VS_STATE_BACKUP) {
925                 if (!sync_backup_thread)
926                         return -ESRCH;
927
928                 IP_VS_INFO("stopping backup sync thread %d ...\n",
929                            task_pid_nr(sync_backup_thread));
930
931                 ip_vs_sync_state &= ~IP_VS_STATE_BACKUP;
932                 kthread_stop(sync_backup_thread);
933                 sync_backup_thread = NULL;
934         } else {
935                 return -EINVAL;
936         }
937
938         /* decrease the module use count */
939         ip_vs_use_count_dec();
940
941         return 0;
942 }