Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
[linux-2.6] / net / ipv4 / 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
34 #include <net/ip.h>
35 #include <net/sock.h>
36
37 #include <net/ip_vs.h>
38
39 #define IP_VS_SYNC_GROUP 0xe0000051    /* multicast addr - 224.0.0.81 */
40 #define IP_VS_SYNC_PORT  8848          /* multicast port */
41
42
43 /*
44  *      IPVS sync connection entry
45  */
46 struct ip_vs_sync_conn {
47         __u8                    reserved;
48
49         /* Protocol, addresses and port numbers */
50         __u8                    protocol;       /* Which protocol (TCP/UDP) */
51         __be16                  cport;
52         __be16                  vport;
53         __be16                  dport;
54         __be32                  caddr;          /* client address */
55         __be32                  vaddr;          /* virtual address */
56         __be32                  daddr;          /* destination address */
57
58         /* Flags and state transition */
59         __be16                  flags;          /* status flags */
60         __be16                  state;          /* state info */
61
62         /* The sequence options start here */
63 };
64
65 struct ip_vs_sync_conn_options {
66         struct ip_vs_seq        in_seq;         /* incoming seq. struct */
67         struct ip_vs_seq        out_seq;        /* outgoing seq. struct */
68 };
69
70 struct ip_vs_sync_thread_data {
71         struct socket *sock;
72         char *buf;
73 };
74
75 #define SIMPLE_CONN_SIZE  (sizeof(struct ip_vs_sync_conn))
76 #define FULL_CONN_SIZE  \
77 (sizeof(struct ip_vs_sync_conn) + sizeof(struct ip_vs_sync_conn_options))
78
79
80 /*
81   The master mulitcasts messages to the backup load balancers in the
82   following format.
83
84        0                   1                   2                   3
85        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
86       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
87       |  Count Conns  |    SyncID     |            Size               |
88       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
89       |                                                               |
90       |                    IPVS Sync Connection (1)                   |
91       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
92       |                            .                                  |
93       |                            .                                  |
94       |                            .                                  |
95       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
96       |                                                               |
97       |                    IPVS Sync Connection (n)                   |
98       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
99 */
100
101 #define SYNC_MESG_HEADER_LEN    4
102
103 struct ip_vs_sync_mesg {
104         __u8                    nr_conns;
105         __u8                    syncid;
106         __u16                   size;
107
108         /* ip_vs_sync_conn entries start here */
109 };
110
111 /* the maximum length of sync (sending/receiving) message */
112 static int sync_send_mesg_maxlen;
113 static int sync_recv_mesg_maxlen;
114
115 struct ip_vs_sync_buff {
116         struct list_head        list;
117         unsigned long           firstuse;
118
119         /* pointers for the message data */
120         struct ip_vs_sync_mesg  *mesg;
121         unsigned char           *head;
122         unsigned char           *end;
123 };
124
125
126 /* the sync_buff list head and the lock */
127 static LIST_HEAD(ip_vs_sync_queue);
128 static DEFINE_SPINLOCK(ip_vs_sync_lock);
129
130 /* current sync_buff for accepting new conn entries */
131 static struct ip_vs_sync_buff   *curr_sb = NULL;
132 static DEFINE_SPINLOCK(curr_sb_lock);
133
134 /* ipvs sync daemon state */
135 volatile int ip_vs_sync_state = IP_VS_STATE_NONE;
136 volatile int ip_vs_master_syncid = 0;
137 volatile int ip_vs_backup_syncid = 0;
138
139 /* multicast interface name */
140 char ip_vs_master_mcast_ifn[IP_VS_IFNAME_MAXLEN];
141 char ip_vs_backup_mcast_ifn[IP_VS_IFNAME_MAXLEN];
142
143 /* sync daemon tasks */
144 static struct task_struct *sync_master_thread;
145 static struct task_struct *sync_backup_thread;
146
147 /* multicast addr */
148 static struct sockaddr_in mcast_addr = {
149         .sin_family             = AF_INET,
150         .sin_port               = __constant_htons(IP_VS_SYNC_PORT),
151         .sin_addr.s_addr        = __constant_htonl(IP_VS_SYNC_GROUP),
152 };
153
154
155 static inline struct ip_vs_sync_buff *sb_dequeue(void)
156 {
157         struct ip_vs_sync_buff *sb;
158
159         spin_lock_bh(&ip_vs_sync_lock);
160         if (list_empty(&ip_vs_sync_queue)) {
161                 sb = NULL;
162         } else {
163                 sb = list_entry(ip_vs_sync_queue.next,
164                                 struct ip_vs_sync_buff,
165                                 list);
166                 list_del(&sb->list);
167         }
168         spin_unlock_bh(&ip_vs_sync_lock);
169
170         return sb;
171 }
172
173 static inline struct ip_vs_sync_buff * ip_vs_sync_buff_create(void)
174 {
175         struct ip_vs_sync_buff *sb;
176
177         if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
178                 return NULL;
179
180         if (!(sb->mesg=kmalloc(sync_send_mesg_maxlen, GFP_ATOMIC))) {
181                 kfree(sb);
182                 return NULL;
183         }
184         sb->mesg->nr_conns = 0;
185         sb->mesg->syncid = ip_vs_master_syncid;
186         sb->mesg->size = 4;
187         sb->head = (unsigned char *)sb->mesg + 4;
188         sb->end = (unsigned char *)sb->mesg + sync_send_mesg_maxlen;
189         sb->firstuse = jiffies;
190         return sb;
191 }
192
193 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
194 {
195         kfree(sb->mesg);
196         kfree(sb);
197 }
198
199 static inline void sb_queue_tail(struct ip_vs_sync_buff *sb)
200 {
201         spin_lock(&ip_vs_sync_lock);
202         if (ip_vs_sync_state & IP_VS_STATE_MASTER)
203                 list_add_tail(&sb->list, &ip_vs_sync_queue);
204         else
205                 ip_vs_sync_buff_release(sb);
206         spin_unlock(&ip_vs_sync_lock);
207 }
208
209 /*
210  *      Get the current sync buffer if it has been created for more
211  *      than the specified time or the specified time is zero.
212  */
213 static inline struct ip_vs_sync_buff *
214 get_curr_sync_buff(unsigned long time)
215 {
216         struct ip_vs_sync_buff *sb;
217
218         spin_lock_bh(&curr_sb_lock);
219         if (curr_sb && (time == 0 ||
220                         time_before(jiffies - curr_sb->firstuse, time))) {
221                 sb = curr_sb;
222                 curr_sb = NULL;
223         } else
224                 sb = NULL;
225         spin_unlock_bh(&curr_sb_lock);
226         return sb;
227 }
228
229
230 /*
231  *      Add an ip_vs_conn information into the current sync_buff.
232  *      Called by ip_vs_in.
233  */
234 void ip_vs_sync_conn(struct ip_vs_conn *cp)
235 {
236         struct ip_vs_sync_mesg *m;
237         struct ip_vs_sync_conn *s;
238         int len;
239
240         spin_lock(&curr_sb_lock);
241         if (!curr_sb) {
242                 if (!(curr_sb=ip_vs_sync_buff_create())) {
243                         spin_unlock(&curr_sb_lock);
244                         IP_VS_ERR("ip_vs_sync_buff_create failed.\n");
245                         return;
246                 }
247         }
248
249         len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
250                 SIMPLE_CONN_SIZE;
251         m = curr_sb->mesg;
252         s = (struct ip_vs_sync_conn *)curr_sb->head;
253
254         /* copy members */
255         s->protocol = cp->protocol;
256         s->cport = cp->cport;
257         s->vport = cp->vport;
258         s->dport = cp->dport;
259         s->caddr = cp->caddr;
260         s->vaddr = cp->vaddr;
261         s->daddr = cp->daddr;
262         s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
263         s->state = htons(cp->state);
264         if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
265                 struct ip_vs_sync_conn_options *opt =
266                         (struct ip_vs_sync_conn_options *)&s[1];
267                 memcpy(opt, &cp->in_seq, sizeof(*opt));
268         }
269
270         m->nr_conns++;
271         m->size += len;
272         curr_sb->head += len;
273
274         /* check if there is a space for next one */
275         if (curr_sb->head+FULL_CONN_SIZE > curr_sb->end) {
276                 sb_queue_tail(curr_sb);
277                 curr_sb = NULL;
278         }
279         spin_unlock(&curr_sb_lock);
280
281         /* synchronize its controller if it has */
282         if (cp->control)
283                 ip_vs_sync_conn(cp->control);
284 }
285
286
287 /*
288  *      Process received multicast message and create the corresponding
289  *      ip_vs_conn entries.
290  */
291 static void ip_vs_process_message(const char *buffer, const size_t buflen)
292 {
293         struct ip_vs_sync_mesg *m = (struct ip_vs_sync_mesg *)buffer;
294         struct ip_vs_sync_conn *s;
295         struct ip_vs_sync_conn_options *opt;
296         struct ip_vs_conn *cp;
297         struct ip_vs_protocol *pp;
298         struct ip_vs_dest *dest;
299         char *p;
300         int i;
301
302         if (buflen < sizeof(struct ip_vs_sync_mesg)) {
303                 IP_VS_ERR_RL("sync message header too short\n");
304                 return;
305         }
306
307         /* Convert size back to host byte order */
308         m->size = ntohs(m->size);
309
310         if (buflen != m->size) {
311                 IP_VS_ERR_RL("bogus sync message size\n");
312                 return;
313         }
314
315         /* SyncID sanity check */
316         if (ip_vs_backup_syncid != 0 && m->syncid != ip_vs_backup_syncid) {
317                 IP_VS_DBG(7, "Ignoring incoming msg with syncid = %d\n",
318                           m->syncid);
319                 return;
320         }
321
322         p = (char *)buffer + sizeof(struct ip_vs_sync_mesg);
323         for (i=0; i<m->nr_conns; i++) {
324                 unsigned flags, state;
325
326                 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
327                         IP_VS_ERR_RL("bogus conn in sync message\n");
328                         return;
329                 }
330                 s = (struct ip_vs_sync_conn *) p;
331                 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
332                 flags &= ~IP_VS_CONN_F_HASHED;
333                 if (flags & IP_VS_CONN_F_SEQ_MASK) {
334                         opt = (struct ip_vs_sync_conn_options *)&s[1];
335                         p += FULL_CONN_SIZE;
336                         if (p > buffer+buflen) {
337                                 IP_VS_ERR_RL("bogus conn options in sync message\n");
338                                 return;
339                         }
340                 } else {
341                         opt = NULL;
342                         p += SIMPLE_CONN_SIZE;
343                 }
344
345                 state = ntohs(s->state);
346                 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
347                         pp = ip_vs_proto_get(s->protocol);
348                         if (!pp) {
349                                 IP_VS_ERR_RL("Unsupported protocol %u in sync msg\n",
350                                         s->protocol);
351                                 continue;
352                         }
353                         if (state >= pp->num_states) {
354                                 IP_VS_DBG(2, "Invalid %s state %u in sync msg\n",
355                                         pp->name, state);
356                                 continue;
357                         }
358                 } else {
359                         /* protocol in templates is not used for state/timeout */
360                         pp = NULL;
361                         if (state > 0) {
362                                 IP_VS_DBG(2, "Invalid template state %u in sync msg\n",
363                                         state);
364                                 state = 0;
365                         }
366                 }
367
368                 if (!(flags & IP_VS_CONN_F_TEMPLATE))
369                         cp = ip_vs_conn_in_get(s->protocol,
370                                                s->caddr, s->cport,
371                                                s->vaddr, s->vport);
372                 else
373                         cp = ip_vs_ct_in_get(s->protocol,
374                                                s->caddr, s->cport,
375                                                s->vaddr, s->vport);
376                 if (!cp) {
377                         /*
378                          * Find the appropriate destination for the connection.
379                          * If it is not found the connection will remain unbound
380                          * but still handled.
381                          */
382                         dest = ip_vs_find_dest(s->daddr, s->dport,
383                                                s->vaddr, s->vport,
384                                                s->protocol);
385                         /*  Set the approprite ativity flag */
386                         if (s->protocol == IPPROTO_TCP) {
387                                 if (state != IP_VS_TCP_S_ESTABLISHED)
388                                         flags |= IP_VS_CONN_F_INACTIVE;
389                                 else
390                                         flags &= ~IP_VS_CONN_F_INACTIVE;
391                         }
392                         cp = ip_vs_conn_new(s->protocol,
393                                             s->caddr, s->cport,
394                                             s->vaddr, s->vport,
395                                             s->daddr, s->dport,
396                                             flags, dest);
397                         if (dest)
398                                 atomic_dec(&dest->refcnt);
399                         if (!cp) {
400                                 IP_VS_ERR("ip_vs_conn_new failed\n");
401                                 return;
402                         }
403                 } else if (!cp->dest) {
404                         dest = ip_vs_try_bind_dest(cp);
405                         if (dest)
406                                 atomic_dec(&dest->refcnt);
407                 } else if ((cp->dest) && (cp->protocol == IPPROTO_TCP) &&
408                            (cp->state != state)) {
409                         /* update active/inactive flag for the connection */
410                         dest = cp->dest;
411                         if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
412                                 (state != IP_VS_TCP_S_ESTABLISHED)) {
413                                 atomic_dec(&dest->activeconns);
414                                 atomic_inc(&dest->inactconns);
415                                 cp->flags |= IP_VS_CONN_F_INACTIVE;
416                         } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) &&
417                                 (state == IP_VS_TCP_S_ESTABLISHED)) {
418                                 atomic_inc(&dest->activeconns);
419                                 atomic_dec(&dest->inactconns);
420                                 cp->flags &= ~IP_VS_CONN_F_INACTIVE;
421                         }
422                 }
423
424                 if (opt)
425                         memcpy(&cp->in_seq, opt, sizeof(*opt));
426                 atomic_set(&cp->in_pkts, sysctl_ip_vs_sync_threshold[0]);
427                 cp->state = state;
428                 cp->old_state = cp->state;
429                 /*
430                  * We can not recover the right timeout for templates
431                  * in all cases, we can not find the right fwmark
432                  * virtual service. If needed, we can do it for
433                  * non-fwmark persistent services.
434                  */
435                 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pp->timeout_table)
436                         cp->timeout = pp->timeout_table[state];
437                 else
438                         cp->timeout = (3*60*HZ);
439                 ip_vs_conn_put(cp);
440         }
441 }
442
443
444 /*
445  *      Setup loopback of outgoing multicasts on a sending socket
446  */
447 static void set_mcast_loop(struct sock *sk, u_char loop)
448 {
449         struct inet_sock *inet = inet_sk(sk);
450
451         /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
452         lock_sock(sk);
453         inet->mc_loop = loop ? 1 : 0;
454         release_sock(sk);
455 }
456
457 /*
458  *      Specify TTL for outgoing multicasts on a sending socket
459  */
460 static void set_mcast_ttl(struct sock *sk, u_char ttl)
461 {
462         struct inet_sock *inet = inet_sk(sk);
463
464         /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
465         lock_sock(sk);
466         inet->mc_ttl = ttl;
467         release_sock(sk);
468 }
469
470 /*
471  *      Specifiy default interface for outgoing multicasts
472  */
473 static int set_mcast_if(struct sock *sk, char *ifname)
474 {
475         struct net_device *dev;
476         struct inet_sock *inet = inet_sk(sk);
477
478         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
479                 return -ENODEV;
480
481         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
482                 return -EINVAL;
483
484         lock_sock(sk);
485         inet->mc_index = dev->ifindex;
486         /*  inet->mc_addr  = 0; */
487         release_sock(sk);
488
489         return 0;
490 }
491
492
493 /*
494  *      Set the maximum length of sync message according to the
495  *      specified interface's MTU.
496  */
497 static int set_sync_mesg_maxlen(int sync_state)
498 {
499         struct net_device *dev;
500         int num;
501
502         if (sync_state == IP_VS_STATE_MASTER) {
503                 if ((dev = __dev_get_by_name(&init_net, ip_vs_master_mcast_ifn)) == NULL)
504                         return -ENODEV;
505
506                 num = (dev->mtu - sizeof(struct iphdr) -
507                        sizeof(struct udphdr) -
508                        SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE;
509                 sync_send_mesg_maxlen =
510                         SYNC_MESG_HEADER_LEN + SIMPLE_CONN_SIZE * num;
511                 IP_VS_DBG(7, "setting the maximum length of sync sending "
512                           "message %d.\n", sync_send_mesg_maxlen);
513         } else if (sync_state == IP_VS_STATE_BACKUP) {
514                 if ((dev = __dev_get_by_name(&init_net, ip_vs_backup_mcast_ifn)) == NULL)
515                         return -ENODEV;
516
517                 sync_recv_mesg_maxlen = dev->mtu -
518                         sizeof(struct iphdr) - sizeof(struct udphdr);
519                 IP_VS_DBG(7, "setting the maximum length of sync receiving "
520                           "message %d.\n", sync_recv_mesg_maxlen);
521         }
522
523         return 0;
524 }
525
526
527 /*
528  *      Join a multicast group.
529  *      the group is specified by a class D multicast address 224.0.0.0/8
530  *      in the in_addr structure passed in as a parameter.
531  */
532 static int
533 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
534 {
535         struct ip_mreqn mreq;
536         struct net_device *dev;
537         int ret;
538
539         memset(&mreq, 0, sizeof(mreq));
540         memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
541
542         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
543                 return -ENODEV;
544         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
545                 return -EINVAL;
546
547         mreq.imr_ifindex = dev->ifindex;
548
549         lock_sock(sk);
550         ret = ip_mc_join_group(sk, &mreq);
551         release_sock(sk);
552
553         return ret;
554 }
555
556
557 static int bind_mcastif_addr(struct socket *sock, char *ifname)
558 {
559         struct net_device *dev;
560         __be32 addr;
561         struct sockaddr_in sin;
562
563         if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL)
564                 return -ENODEV;
565
566         addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
567         if (!addr)
568                 IP_VS_ERR("You probably need to specify IP address on "
569                           "multicast interface.\n");
570
571         IP_VS_DBG(7, "binding socket with (%s) %u.%u.%u.%u\n",
572                   ifname, NIPQUAD(addr));
573
574         /* Now bind the socket with the address of multicast interface */
575         sin.sin_family       = AF_INET;
576         sin.sin_addr.s_addr  = addr;
577         sin.sin_port         = 0;
578
579         return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
580 }
581
582 /*
583  *      Set up sending multicast socket over UDP
584  */
585 static struct socket * make_send_sock(void)
586 {
587         struct socket *sock;
588         int result;
589
590         /* First create a socket */
591         result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
592         if (result < 0) {
593                 IP_VS_ERR("Error during creation of socket; terminating\n");
594                 return ERR_PTR(result);
595         }
596
597         result = set_mcast_if(sock->sk, ip_vs_master_mcast_ifn);
598         if (result < 0) {
599                 IP_VS_ERR("Error setting outbound mcast interface\n");
600                 goto error;
601         }
602
603         set_mcast_loop(sock->sk, 0);
604         set_mcast_ttl(sock->sk, 1);
605
606         result = bind_mcastif_addr(sock, ip_vs_master_mcast_ifn);
607         if (result < 0) {
608                 IP_VS_ERR("Error binding address of the mcast interface\n");
609                 goto error;
610         }
611
612         result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr,
613                         sizeof(struct sockaddr), 0);
614         if (result < 0) {
615                 IP_VS_ERR("Error connecting to the multicast addr\n");
616                 goto error;
617         }
618
619         return sock;
620
621   error:
622         sock_release(sock);
623         return ERR_PTR(result);
624 }
625
626
627 /*
628  *      Set up receiving multicast socket over UDP
629  */
630 static struct socket * make_receive_sock(void)
631 {
632         struct socket *sock;
633         int result;
634
635         /* First create a socket */
636         result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
637         if (result < 0) {
638                 IP_VS_ERR("Error during creation of socket; terminating\n");
639                 return ERR_PTR(result);
640         }
641
642         /* it is equivalent to the REUSEADDR option in user-space */
643         sock->sk->sk_reuse = 1;
644
645         result = sock->ops->bind(sock, (struct sockaddr *) &mcast_addr,
646                         sizeof(struct sockaddr));
647         if (result < 0) {
648                 IP_VS_ERR("Error binding to the multicast addr\n");
649                 goto error;
650         }
651
652         /* join the multicast group */
653         result = join_mcast_group(sock->sk,
654                         (struct in_addr *) &mcast_addr.sin_addr,
655                         ip_vs_backup_mcast_ifn);
656         if (result < 0) {
657                 IP_VS_ERR("Error joining to the multicast group\n");
658                 goto error;
659         }
660
661         return sock;
662
663   error:
664         sock_release(sock);
665         return ERR_PTR(result);
666 }
667
668
669 static int
670 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
671 {
672         struct msghdr   msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
673         struct kvec     iov;
674         int             len;
675
676         EnterFunction(7);
677         iov.iov_base     = (void *)buffer;
678         iov.iov_len      = length;
679
680         len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
681
682         LeaveFunction(7);
683         return len;
684 }
685
686 static void
687 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
688 {
689         int msize;
690
691         msize = msg->size;
692
693         /* Put size in network byte order */
694         msg->size = htons(msg->size);
695
696         if (ip_vs_send_async(sock, (char *)msg, msize) != msize)
697                 IP_VS_ERR("ip_vs_send_async error\n");
698 }
699
700 static int
701 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
702 {
703         struct msghdr           msg = {NULL,};
704         struct kvec             iov;
705         int                     len;
706
707         EnterFunction(7);
708
709         /* Receive a packet */
710         iov.iov_base     = buffer;
711         iov.iov_len      = (size_t)buflen;
712
713         len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, 0);
714
715         if (len < 0)
716                 return -1;
717
718         LeaveFunction(7);
719         return len;
720 }
721
722
723 static int sync_thread_master(void *data)
724 {
725         struct ip_vs_sync_thread_data *tinfo = data;
726         struct ip_vs_sync_buff *sb;
727
728         IP_VS_INFO("sync thread started: state = MASTER, mcast_ifn = %s, "
729                    "syncid = %d\n",
730                    ip_vs_master_mcast_ifn, ip_vs_master_syncid);
731
732         while (!kthread_should_stop()) {
733                 while ((sb = sb_dequeue())) {
734                         ip_vs_send_sync_msg(tinfo->sock, sb->mesg);
735                         ip_vs_sync_buff_release(sb);
736                 }
737
738                 /* check if entries stay in curr_sb for 2 seconds */
739                 sb = get_curr_sync_buff(2 * HZ);
740                 if (sb) {
741                         ip_vs_send_sync_msg(tinfo->sock, sb->mesg);
742                         ip_vs_sync_buff_release(sb);
743                 }
744
745                 schedule_timeout_interruptible(HZ);
746         }
747
748         /* clean up the sync_buff queue */
749         while ((sb=sb_dequeue())) {
750                 ip_vs_sync_buff_release(sb);
751         }
752
753         /* clean up the current sync_buff */
754         if ((sb = get_curr_sync_buff(0))) {
755                 ip_vs_sync_buff_release(sb);
756         }
757
758         /* release the sending multicast socket */
759         sock_release(tinfo->sock);
760         kfree(tinfo);
761
762         return 0;
763 }
764
765
766 static int sync_thread_backup(void *data)
767 {
768         struct ip_vs_sync_thread_data *tinfo = data;
769         int len;
770
771         IP_VS_INFO("sync thread started: state = BACKUP, mcast_ifn = %s, "
772                    "syncid = %d\n",
773                    ip_vs_backup_mcast_ifn, ip_vs_backup_syncid);
774
775         while (!kthread_should_stop()) {
776                 wait_event_interruptible(*tinfo->sock->sk->sk_sleep,
777                          !skb_queue_empty(&tinfo->sock->sk->sk_receive_queue)
778                          || kthread_should_stop());
779
780                 /* do we have data now? */
781                 while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) {
782                         len = ip_vs_receive(tinfo->sock, tinfo->buf,
783                                         sync_recv_mesg_maxlen);
784                         if (len <= 0) {
785                                 IP_VS_ERR("receiving message error\n");
786                                 break;
787                         }
788
789                         /* disable bottom half, because it accesses the data
790                            shared by softirq while getting/creating conns */
791                         local_bh_disable();
792                         ip_vs_process_message(tinfo->buf, len);
793                         local_bh_enable();
794                 }
795         }
796
797         /* release the sending multicast socket */
798         sock_release(tinfo->sock);
799         kfree(tinfo->buf);
800         kfree(tinfo);
801
802         return 0;
803 }
804
805
806 int start_sync_thread(int state, char *mcast_ifn, __u8 syncid)
807 {
808         struct ip_vs_sync_thread_data *tinfo;
809         struct task_struct **realtask, *task;
810         struct socket *sock;
811         char *name, *buf = NULL;
812         int (*threadfn)(void *data);
813         int result = -ENOMEM;
814
815         IP_VS_DBG(7, "%s: pid %d\n", __func__, task_pid_nr(current));
816         IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n",
817                   sizeof(struct ip_vs_sync_conn));
818
819         if (state == IP_VS_STATE_MASTER) {
820                 if (sync_master_thread)
821                         return -EEXIST;
822
823                 strlcpy(ip_vs_master_mcast_ifn, mcast_ifn,
824                         sizeof(ip_vs_master_mcast_ifn));
825                 ip_vs_master_syncid = syncid;
826                 realtask = &sync_master_thread;
827                 name = "ipvs_syncmaster";
828                 threadfn = sync_thread_master;
829                 sock = make_send_sock();
830         } else if (state == IP_VS_STATE_BACKUP) {
831                 if (sync_backup_thread)
832                         return -EEXIST;
833
834                 strlcpy(ip_vs_backup_mcast_ifn, mcast_ifn,
835                         sizeof(ip_vs_backup_mcast_ifn));
836                 ip_vs_backup_syncid = syncid;
837                 realtask = &sync_backup_thread;
838                 name = "ipvs_syncbackup";
839                 threadfn = sync_thread_backup;
840                 sock = make_receive_sock();
841         } else {
842                 return -EINVAL;
843         }
844
845         if (IS_ERR(sock)) {
846                 result = PTR_ERR(sock);
847                 goto out;
848         }
849
850         set_sync_mesg_maxlen(state);
851         if (state == IP_VS_STATE_BACKUP) {
852                 buf = kmalloc(sync_recv_mesg_maxlen, GFP_KERNEL);
853                 if (!buf)
854                         goto outsocket;
855         }
856
857         tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL);
858         if (!tinfo)
859                 goto outbuf;
860
861         tinfo->sock = sock;
862         tinfo->buf = buf;
863
864         task = kthread_run(threadfn, tinfo, name);
865         if (IS_ERR(task)) {
866                 result = PTR_ERR(task);
867                 goto outtinfo;
868         }
869
870         /* mark as active */
871         *realtask = task;
872         ip_vs_sync_state |= state;
873
874         /* increase the module use count */
875         ip_vs_use_count_inc();
876
877         return 0;
878
879 outtinfo:
880         kfree(tinfo);
881 outbuf:
882         kfree(buf);
883 outsocket:
884         sock_release(sock);
885 out:
886         return result;
887 }
888
889
890 int stop_sync_thread(int state)
891 {
892         IP_VS_DBG(7, "%s: pid %d\n", __func__, task_pid_nr(current));
893
894         if (state == IP_VS_STATE_MASTER) {
895                 if (!sync_master_thread)
896                         return -ESRCH;
897
898                 IP_VS_INFO("stopping master sync thread %d ...\n",
899                            task_pid_nr(sync_master_thread));
900
901                 /*
902                  * The lock synchronizes with sb_queue_tail(), so that we don't
903                  * add sync buffers to the queue, when we are already in
904                  * progress of stopping the master sync daemon.
905                  */
906
907                 spin_lock_bh(&ip_vs_sync_lock);
908                 ip_vs_sync_state &= ~IP_VS_STATE_MASTER;
909                 spin_unlock_bh(&ip_vs_sync_lock);
910                 kthread_stop(sync_master_thread);
911                 sync_master_thread = NULL;
912         } else if (state == IP_VS_STATE_BACKUP) {
913                 if (!sync_backup_thread)
914                         return -ESRCH;
915
916                 IP_VS_INFO("stopping backup sync thread %d ...\n",
917                            task_pid_nr(sync_backup_thread));
918
919                 ip_vs_sync_state &= ~IP_VS_STATE_BACKUP;
920                 kthread_stop(sync_backup_thread);
921                 sync_backup_thread = NULL;
922         } else {
923                 return -EINVAL;
924         }
925
926         /* decrease the module use count */
927         ip_vs_use_count_dec();
928
929         return 0;
930 }