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
8 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
10 * ip_vs_sync: sync connection info from master load balancer to backups
14 * Alexandre Cassen : Added master & backup support at a time.
15 * Alexandre Cassen : Added SyncID support for incoming sync
17 * Justin Ossevoort : Fix endian problem on sync message size.
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>
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>
37 #include <net/ip_vs.h>
39 #define IP_VS_SYNC_GROUP 0xe0000051 /* multicast addr - 224.0.0.81 */
40 #define IP_VS_SYNC_PORT 8848 /* multicast port */
44 * IPVS sync connection entry
46 struct ip_vs_sync_conn
{
49 /* Protocol, addresses and port numbers */
50 __u8 protocol
; /* Which protocol (TCP/UDP) */
54 __be32 caddr
; /* client address */
55 __be32 vaddr
; /* virtual address */
56 __be32 daddr
; /* destination address */
58 /* Flags and state transition */
59 __be16 flags
; /* status flags */
60 __be16 state
; /* state info */
62 /* The sequence options start here */
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 */
70 struct ip_vs_sync_thread_data
{
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))
81 The master mulitcasts messages to the backup load balancers in the
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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
90 | IPVS Sync Connection (1) |
91 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
95 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
97 | IPVS Sync Connection (n) |
98 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
101 #define SYNC_MESG_HEADER_LEN 4
103 struct ip_vs_sync_mesg
{
108 /* ip_vs_sync_conn entries start here */
111 /* the maximum length of sync (sending/receiving) message */
112 static int sync_send_mesg_maxlen
;
113 static int sync_recv_mesg_maxlen
;
115 struct ip_vs_sync_buff
{
116 struct list_head list
;
117 unsigned long firstuse
;
119 /* pointers for the message data */
120 struct ip_vs_sync_mesg
*mesg
;
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
);
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
);
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;
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
];
143 /* sync daemon tasks */
144 static struct task_struct
*sync_master_thread
;
145 static struct task_struct
*sync_backup_thread
;
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
),
155 static inline struct ip_vs_sync_buff
*sb_dequeue(void)
157 struct ip_vs_sync_buff
*sb
;
159 spin_lock_bh(&ip_vs_sync_lock
);
160 if (list_empty(&ip_vs_sync_queue
)) {
163 sb
= list_entry(ip_vs_sync_queue
.next
,
164 struct ip_vs_sync_buff
,
168 spin_unlock_bh(&ip_vs_sync_lock
);
173 static inline struct ip_vs_sync_buff
* ip_vs_sync_buff_create(void)
175 struct ip_vs_sync_buff
*sb
;
177 if (!(sb
=kmalloc(sizeof(struct ip_vs_sync_buff
), GFP_ATOMIC
)))
180 if (!(sb
->mesg
=kmalloc(sync_send_mesg_maxlen
, GFP_ATOMIC
))) {
184 sb
->mesg
->nr_conns
= 0;
185 sb
->mesg
->syncid
= ip_vs_master_syncid
;
187 sb
->head
= (unsigned char *)sb
->mesg
+ 4;
188 sb
->end
= (unsigned char *)sb
->mesg
+ sync_send_mesg_maxlen
;
189 sb
->firstuse
= jiffies
;
193 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff
*sb
)
199 static inline void sb_queue_tail(struct ip_vs_sync_buff
*sb
)
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
);
205 ip_vs_sync_buff_release(sb
);
206 spin_unlock(&ip_vs_sync_lock
);
210 * Get the current sync buffer if it has been created for more
211 * than the specified time or the specified time is zero.
213 static inline struct ip_vs_sync_buff
*
214 get_curr_sync_buff(unsigned long time
)
216 struct ip_vs_sync_buff
*sb
;
218 spin_lock_bh(&curr_sb_lock
);
219 if (curr_sb
&& (time
== 0 ||
220 time_before(jiffies
- curr_sb
->firstuse
, time
))) {
225 spin_unlock_bh(&curr_sb_lock
);
231 * Add an ip_vs_conn information into the current sync_buff.
232 * Called by ip_vs_in.
234 void ip_vs_sync_conn(struct ip_vs_conn
*cp
)
236 struct ip_vs_sync_mesg
*m
;
237 struct ip_vs_sync_conn
*s
;
240 spin_lock(&curr_sb_lock
);
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");
249 len
= (cp
->flags
& IP_VS_CONN_F_SEQ_MASK
) ? FULL_CONN_SIZE
:
252 s
= (struct ip_vs_sync_conn
*)curr_sb
->head
;
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
.ip
;
260 s
->vaddr
= cp
->vaddr
.ip
;
261 s
->daddr
= cp
->daddr
.ip
;
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
));
272 curr_sb
->head
+= len
;
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
);
279 spin_unlock(&curr_sb_lock
);
281 /* synchronize its controller if it has */
283 ip_vs_sync_conn(cp
->control
);
288 * Process received multicast message and create the corresponding
289 * ip_vs_conn entries.
291 static void ip_vs_process_message(const char *buffer
, const size_t buflen
)
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
;
302 if (buflen
< sizeof(struct ip_vs_sync_mesg
)) {
303 IP_VS_ERR_RL("sync message header too short\n");
307 /* Convert size back to host byte order */
308 m
->size
= ntohs(m
->size
);
310 if (buflen
!= m
->size
) {
311 IP_VS_ERR_RL("bogus sync message size\n");
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",
322 p
= (char *)buffer
+ sizeof(struct ip_vs_sync_mesg
);
323 for (i
=0; i
<m
->nr_conns
; i
++) {
324 unsigned flags
, state
;
326 if (p
+ SIMPLE_CONN_SIZE
> buffer
+buflen
) {
327 IP_VS_ERR_RL("bogus conn in sync message\n");
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];
336 if (p
> buffer
+buflen
) {
337 IP_VS_ERR_RL("bogus conn options in sync message\n");
342 p
+= SIMPLE_CONN_SIZE
;
345 state
= ntohs(s
->state
);
346 if (!(flags
& IP_VS_CONN_F_TEMPLATE
)) {
347 pp
= ip_vs_proto_get(s
->protocol
);
349 IP_VS_ERR_RL("Unsupported protocol %u in sync msg\n",
353 if (state
>= pp
->num_states
) {
354 IP_VS_DBG(2, "Invalid %s state %u in sync msg\n",
359 /* protocol in templates is not used for state/timeout */
362 IP_VS_DBG(2, "Invalid template state %u in sync msg\n",
368 if (!(flags
& IP_VS_CONN_F_TEMPLATE
))
369 cp
= ip_vs_conn_in_get(AF_INET
, s
->protocol
,
370 (union nf_inet_addr
*)&s
->caddr
,
372 (union nf_inet_addr
*)&s
->vaddr
,
375 cp
= ip_vs_ct_in_get(AF_INET
, s
->protocol
,
376 (union nf_inet_addr
*)&s
->caddr
,
378 (union nf_inet_addr
*)&s
->vaddr
,
382 * Find the appropriate destination for the connection.
383 * If it is not found the connection will remain unbound
386 dest
= ip_vs_find_dest(AF_INET
,
387 (union nf_inet_addr
*)&s
->daddr
,
389 (union nf_inet_addr
*)&s
->vaddr
,
392 /* Set the approprite ativity flag */
393 if (s
->protocol
== IPPROTO_TCP
) {
394 if (state
!= IP_VS_TCP_S_ESTABLISHED
)
395 flags
|= IP_VS_CONN_F_INACTIVE
;
397 flags
&= ~IP_VS_CONN_F_INACTIVE
;
399 cp
= ip_vs_conn_new(AF_INET
, s
->protocol
,
400 (union nf_inet_addr
*)s
->caddr
,
402 (union nf_inet_addr
*)s
->vaddr
,
404 (union nf_inet_addr
*)s
->daddr
,
408 atomic_dec(&dest
->refcnt
);
410 IP_VS_ERR("ip_vs_conn_new failed\n");
413 } else if (!cp
->dest
) {
414 dest
= ip_vs_try_bind_dest(cp
);
416 atomic_dec(&dest
->refcnt
);
417 } else if ((cp
->dest
) && (cp
->protocol
== IPPROTO_TCP
) &&
418 (cp
->state
!= state
)) {
419 /* update active/inactive flag for the connection */
421 if (!(cp
->flags
& IP_VS_CONN_F_INACTIVE
) &&
422 (state
!= IP_VS_TCP_S_ESTABLISHED
)) {
423 atomic_dec(&dest
->activeconns
);
424 atomic_inc(&dest
->inactconns
);
425 cp
->flags
|= IP_VS_CONN_F_INACTIVE
;
426 } else if ((cp
->flags
& IP_VS_CONN_F_INACTIVE
) &&
427 (state
== IP_VS_TCP_S_ESTABLISHED
)) {
428 atomic_inc(&dest
->activeconns
);
429 atomic_dec(&dest
->inactconns
);
430 cp
->flags
&= ~IP_VS_CONN_F_INACTIVE
;
435 memcpy(&cp
->in_seq
, opt
, sizeof(*opt
));
436 atomic_set(&cp
->in_pkts
, sysctl_ip_vs_sync_threshold
[0]);
438 cp
->old_state
= cp
->state
;
440 * We can not recover the right timeout for templates
441 * in all cases, we can not find the right fwmark
442 * virtual service. If needed, we can do it for
443 * non-fwmark persistent services.
445 if (!(flags
& IP_VS_CONN_F_TEMPLATE
) && pp
->timeout_table
)
446 cp
->timeout
= pp
->timeout_table
[state
];
448 cp
->timeout
= (3*60*HZ
);
455 * Setup loopback of outgoing multicasts on a sending socket
457 static void set_mcast_loop(struct sock
*sk
, u_char loop
)
459 struct inet_sock
*inet
= inet_sk(sk
);
461 /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
463 inet
->mc_loop
= loop
? 1 : 0;
468 * Specify TTL for outgoing multicasts on a sending socket
470 static void set_mcast_ttl(struct sock
*sk
, u_char ttl
)
472 struct inet_sock
*inet
= inet_sk(sk
);
474 /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
481 * Specifiy default interface for outgoing multicasts
483 static int set_mcast_if(struct sock
*sk
, char *ifname
)
485 struct net_device
*dev
;
486 struct inet_sock
*inet
= inet_sk(sk
);
488 if ((dev
= __dev_get_by_name(&init_net
, ifname
)) == NULL
)
491 if (sk
->sk_bound_dev_if
&& dev
->ifindex
!= sk
->sk_bound_dev_if
)
495 inet
->mc_index
= dev
->ifindex
;
496 /* inet->mc_addr = 0; */
504 * Set the maximum length of sync message according to the
505 * specified interface's MTU.
507 static int set_sync_mesg_maxlen(int sync_state
)
509 struct net_device
*dev
;
512 if (sync_state
== IP_VS_STATE_MASTER
) {
513 if ((dev
= __dev_get_by_name(&init_net
, ip_vs_master_mcast_ifn
)) == NULL
)
516 num
= (dev
->mtu
- sizeof(struct iphdr
) -
517 sizeof(struct udphdr
) -
518 SYNC_MESG_HEADER_LEN
- 20) / SIMPLE_CONN_SIZE
;
519 sync_send_mesg_maxlen
=
520 SYNC_MESG_HEADER_LEN
+ SIMPLE_CONN_SIZE
* num
;
521 IP_VS_DBG(7, "setting the maximum length of sync sending "
522 "message %d.\n", sync_send_mesg_maxlen
);
523 } else if (sync_state
== IP_VS_STATE_BACKUP
) {
524 if ((dev
= __dev_get_by_name(&init_net
, ip_vs_backup_mcast_ifn
)) == NULL
)
527 sync_recv_mesg_maxlen
= dev
->mtu
-
528 sizeof(struct iphdr
) - sizeof(struct udphdr
);
529 IP_VS_DBG(7, "setting the maximum length of sync receiving "
530 "message %d.\n", sync_recv_mesg_maxlen
);
538 * Join a multicast group.
539 * the group is specified by a class D multicast address 224.0.0.0/8
540 * in the in_addr structure passed in as a parameter.
543 join_mcast_group(struct sock
*sk
, struct in_addr
*addr
, char *ifname
)
545 struct ip_mreqn mreq
;
546 struct net_device
*dev
;
549 memset(&mreq
, 0, sizeof(mreq
));
550 memcpy(&mreq
.imr_multiaddr
, addr
, sizeof(struct in_addr
));
552 if ((dev
= __dev_get_by_name(&init_net
, ifname
)) == NULL
)
554 if (sk
->sk_bound_dev_if
&& dev
->ifindex
!= sk
->sk_bound_dev_if
)
557 mreq
.imr_ifindex
= dev
->ifindex
;
560 ret
= ip_mc_join_group(sk
, &mreq
);
567 static int bind_mcastif_addr(struct socket
*sock
, char *ifname
)
569 struct net_device
*dev
;
571 struct sockaddr_in sin
;
573 if ((dev
= __dev_get_by_name(&init_net
, ifname
)) == NULL
)
576 addr
= inet_select_addr(dev
, 0, RT_SCOPE_UNIVERSE
);
578 IP_VS_ERR("You probably need to specify IP address on "
579 "multicast interface.\n");
581 IP_VS_DBG(7, "binding socket with (%s) %u.%u.%u.%u\n",
582 ifname
, NIPQUAD(addr
));
584 /* Now bind the socket with the address of multicast interface */
585 sin
.sin_family
= AF_INET
;
586 sin
.sin_addr
.s_addr
= addr
;
589 return sock
->ops
->bind(sock
, (struct sockaddr
*)&sin
, sizeof(sin
));
593 * Set up sending multicast socket over UDP
595 static struct socket
* make_send_sock(void)
600 /* First create a socket */
601 result
= sock_create_kern(PF_INET
, SOCK_DGRAM
, IPPROTO_UDP
, &sock
);
603 IP_VS_ERR("Error during creation of socket; terminating\n");
604 return ERR_PTR(result
);
607 result
= set_mcast_if(sock
->sk
, ip_vs_master_mcast_ifn
);
609 IP_VS_ERR("Error setting outbound mcast interface\n");
613 set_mcast_loop(sock
->sk
, 0);
614 set_mcast_ttl(sock
->sk
, 1);
616 result
= bind_mcastif_addr(sock
, ip_vs_master_mcast_ifn
);
618 IP_VS_ERR("Error binding address of the mcast interface\n");
622 result
= sock
->ops
->connect(sock
, (struct sockaddr
*) &mcast_addr
,
623 sizeof(struct sockaddr
), 0);
625 IP_VS_ERR("Error connecting to the multicast addr\n");
633 return ERR_PTR(result
);
638 * Set up receiving multicast socket over UDP
640 static struct socket
* make_receive_sock(void)
645 /* First create a socket */
646 result
= sock_create_kern(PF_INET
, SOCK_DGRAM
, IPPROTO_UDP
, &sock
);
648 IP_VS_ERR("Error during creation of socket; terminating\n");
649 return ERR_PTR(result
);
652 /* it is equivalent to the REUSEADDR option in user-space */
653 sock
->sk
->sk_reuse
= 1;
655 result
= sock
->ops
->bind(sock
, (struct sockaddr
*) &mcast_addr
,
656 sizeof(struct sockaddr
));
658 IP_VS_ERR("Error binding to the multicast addr\n");
662 /* join the multicast group */
663 result
= join_mcast_group(sock
->sk
,
664 (struct in_addr
*) &mcast_addr
.sin_addr
,
665 ip_vs_backup_mcast_ifn
);
667 IP_VS_ERR("Error joining to the multicast group\n");
675 return ERR_PTR(result
);
680 ip_vs_send_async(struct socket
*sock
, const char *buffer
, const size_t length
)
682 struct msghdr msg
= {.msg_flags
= MSG_DONTWAIT
|MSG_NOSIGNAL
};
687 iov
.iov_base
= (void *)buffer
;
688 iov
.iov_len
= length
;
690 len
= kernel_sendmsg(sock
, &msg
, &iov
, 1, (size_t)(length
));
697 ip_vs_send_sync_msg(struct socket
*sock
, struct ip_vs_sync_mesg
*msg
)
703 /* Put size in network byte order */
704 msg
->size
= htons(msg
->size
);
706 if (ip_vs_send_async(sock
, (char *)msg
, msize
) != msize
)
707 IP_VS_ERR("ip_vs_send_async error\n");
711 ip_vs_receive(struct socket
*sock
, char *buffer
, const size_t buflen
)
713 struct msghdr msg
= {NULL
,};
719 /* Receive a packet */
720 iov
.iov_base
= buffer
;
721 iov
.iov_len
= (size_t)buflen
;
723 len
= kernel_recvmsg(sock
, &msg
, &iov
, 1, buflen
, 0);
733 static int sync_thread_master(void *data
)
735 struct ip_vs_sync_thread_data
*tinfo
= data
;
736 struct ip_vs_sync_buff
*sb
;
738 IP_VS_INFO("sync thread started: state = MASTER, mcast_ifn = %s, "
740 ip_vs_master_mcast_ifn
, ip_vs_master_syncid
);
742 while (!kthread_should_stop()) {
743 while ((sb
= sb_dequeue())) {
744 ip_vs_send_sync_msg(tinfo
->sock
, sb
->mesg
);
745 ip_vs_sync_buff_release(sb
);
748 /* check if entries stay in curr_sb for 2 seconds */
749 sb
= get_curr_sync_buff(2 * HZ
);
751 ip_vs_send_sync_msg(tinfo
->sock
, sb
->mesg
);
752 ip_vs_sync_buff_release(sb
);
755 schedule_timeout_interruptible(HZ
);
758 /* clean up the sync_buff queue */
759 while ((sb
=sb_dequeue())) {
760 ip_vs_sync_buff_release(sb
);
763 /* clean up the current sync_buff */
764 if ((sb
= get_curr_sync_buff(0))) {
765 ip_vs_sync_buff_release(sb
);
768 /* release the sending multicast socket */
769 sock_release(tinfo
->sock
);
776 static int sync_thread_backup(void *data
)
778 struct ip_vs_sync_thread_data
*tinfo
= data
;
781 IP_VS_INFO("sync thread started: state = BACKUP, mcast_ifn = %s, "
783 ip_vs_backup_mcast_ifn
, ip_vs_backup_syncid
);
785 while (!kthread_should_stop()) {
786 wait_event_interruptible(*tinfo
->sock
->sk
->sk_sleep
,
787 !skb_queue_empty(&tinfo
->sock
->sk
->sk_receive_queue
)
788 || kthread_should_stop());
790 /* do we have data now? */
791 while (!skb_queue_empty(&(tinfo
->sock
->sk
->sk_receive_queue
))) {
792 len
= ip_vs_receive(tinfo
->sock
, tinfo
->buf
,
793 sync_recv_mesg_maxlen
);
795 IP_VS_ERR("receiving message error\n");
799 /* disable bottom half, because it accesses the data
800 shared by softirq while getting/creating conns */
802 ip_vs_process_message(tinfo
->buf
, len
);
807 /* release the sending multicast socket */
808 sock_release(tinfo
->sock
);
816 int start_sync_thread(int state
, char *mcast_ifn
, __u8 syncid
)
818 struct ip_vs_sync_thread_data
*tinfo
;
819 struct task_struct
**realtask
, *task
;
821 char *name
, *buf
= NULL
;
822 int (*threadfn
)(void *data
);
823 int result
= -ENOMEM
;
825 IP_VS_DBG(7, "%s: pid %d\n", __func__
, task_pid_nr(current
));
826 IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n",
827 sizeof(struct ip_vs_sync_conn
));
829 if (state
== IP_VS_STATE_MASTER
) {
830 if (sync_master_thread
)
833 strlcpy(ip_vs_master_mcast_ifn
, mcast_ifn
,
834 sizeof(ip_vs_master_mcast_ifn
));
835 ip_vs_master_syncid
= syncid
;
836 realtask
= &sync_master_thread
;
837 name
= "ipvs_syncmaster";
838 threadfn
= sync_thread_master
;
839 sock
= make_send_sock();
840 } else if (state
== IP_VS_STATE_BACKUP
) {
841 if (sync_backup_thread
)
844 strlcpy(ip_vs_backup_mcast_ifn
, mcast_ifn
,
845 sizeof(ip_vs_backup_mcast_ifn
));
846 ip_vs_backup_syncid
= syncid
;
847 realtask
= &sync_backup_thread
;
848 name
= "ipvs_syncbackup";
849 threadfn
= sync_thread_backup
;
850 sock
= make_receive_sock();
856 result
= PTR_ERR(sock
);
860 set_sync_mesg_maxlen(state
);
861 if (state
== IP_VS_STATE_BACKUP
) {
862 buf
= kmalloc(sync_recv_mesg_maxlen
, GFP_KERNEL
);
867 tinfo
= kmalloc(sizeof(*tinfo
), GFP_KERNEL
);
874 task
= kthread_run(threadfn
, tinfo
, name
);
876 result
= PTR_ERR(task
);
882 ip_vs_sync_state
|= state
;
884 /* increase the module use count */
885 ip_vs_use_count_inc();
900 int stop_sync_thread(int state
)
902 IP_VS_DBG(7, "%s: pid %d\n", __func__
, task_pid_nr(current
));
904 if (state
== IP_VS_STATE_MASTER
) {
905 if (!sync_master_thread
)
908 IP_VS_INFO("stopping master sync thread %d ...\n",
909 task_pid_nr(sync_master_thread
));
912 * The lock synchronizes with sb_queue_tail(), so that we don't
913 * add sync buffers to the queue, when we are already in
914 * progress of stopping the master sync daemon.
917 spin_lock_bh(&ip_vs_sync_lock
);
918 ip_vs_sync_state
&= ~IP_VS_STATE_MASTER
;
919 spin_unlock_bh(&ip_vs_sync_lock
);
920 kthread_stop(sync_master_thread
);
921 sync_master_thread
= NULL
;
922 } else if (state
== IP_VS_STATE_BACKUP
) {
923 if (!sync_backup_thread
)
926 IP_VS_INFO("stopping backup sync thread %d ...\n",
927 task_pid_nr(sync_backup_thread
));
929 ip_vs_sync_state
&= ~IP_VS_STATE_BACKUP
;
930 kthread_stop(sync_backup_thread
);
931 sync_backup_thread
= NULL
;
936 /* decrease the module use count */
937 ip_vs_use_count_dec();