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 * 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
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.
20 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
22 * ip_vs_sync: sync connection info from master load balancer to backups
26 * Alexandre Cassen : Added master & backup support at a time.
27 * Alexandre Cassen : Added SyncID support for incoming sync
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.
34 #define KMSG_COMPONENT "IPVS"
35 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
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>
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>
52 #include <asm/unaligned.h> /* Used for ntoh_seq and hton_seq */
57 #include <net/ip_vs.h>
59 #define IP_VS_SYNC_GROUP 0xe0000051 /* multicast addr - 224.0.0.81 */
60 #define IP_VS_SYNC_PORT 8848 /* multicast port */
62 #define SYNC_PROTO_VER 1 /* Protocol version in header */
64 static struct lock_class_key __ipvs_sync_key
;
66 * IPVS sync connection entry
67 * Version 0, i.e. original version.
69 struct ip_vs_sync_conn_v0
{
72 /* Protocol, addresses and port numbers */
73 __u8 protocol
; /* Which protocol (TCP/UDP) */
77 __be32 caddr
; /* client address */
78 __be32 vaddr
; /* virtual address */
79 __be32 daddr
; /* destination address */
81 /* Flags and state transition */
82 __be16 flags
; /* status flags */
83 __be16 state
; /* state info */
85 /* The sequence options start here */
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 */
94 Sync Connection format (sync_conn)
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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
102 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
104 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
106 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
108 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
109 | timeout (in sec.) |
110 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
112 | IP-Addresses (v4 or v6) |
114 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
116 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
117 | Param. Type | Param. Length | Param. data |
118 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
120 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
121 | | Param Type | Param. Length |
122 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
124 | Last Param data should be padded for 32 bit alignment |
125 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
129 * Type 0, IPv4 sync connection format
131 struct ip_vs_sync_v4
{
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 */
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 */
151 * Type 2 messages IPv6
153 struct ip_vs_sync_v6
{
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 */
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 */
173 union ip_vs_sync_conn
{
174 struct ip_vs_sync_v4 v4
;
175 struct ip_vs_sync_v6 v6
;
178 /* Bits in Type field in above */
179 #define STYPE_INET6 0
180 #define STYPE_F_INET6 (1 << STYPE_INET6)
182 #define SVER_SHIFT 12 /* Shift to get version */
183 #define SVER_MASK 0x0fff /* Mask to strip version */
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
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))
195 struct ip_vs_sync_thread_data
{
201 /* Version 0 definition of packet sizes */
202 #define SIMPLE_CONN_SIZE (sizeof(struct ip_vs_sync_conn_v0))
203 #define FULL_CONN_SIZE \
204 (sizeof(struct ip_vs_sync_conn_v0) + sizeof(struct ip_vs_sync_conn_options))
208 The master mulitcasts messages (Datagrams) to the backup load balancers
209 in the following format.
212 Note, first byte should be Zero, so ver 0 receivers will drop the packet.
215 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
216 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
217 | 0 | SyncID | Size |
218 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
219 | Count Conns | Version | Reserved, set to Zero |
220 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
222 | IPVS Sync Connection (1) |
223 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
227 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
229 | IPVS Sync Connection (n) |
230 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
234 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
235 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
236 | Count Conns | SyncID | Size |
237 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
238 | IPVS Sync Connection (1) |
241 #define SYNC_MESG_HEADER_LEN 4
242 #define MAX_CONNS_PER_SYNCBUFF 255 /* nr_conns in ip_vs_sync_mesg is 8 bit */
244 /* Version 0 header */
245 struct ip_vs_sync_mesg_v0
{
250 /* ip_vs_sync_conn entries start here */
253 /* Version 1 header */
254 struct ip_vs_sync_mesg
{
255 __u8 reserved
; /* must be zero */
259 __s8 version
; /* SYNC_PROTO_VER */
261 /* ip_vs_sync_conn entries start here */
264 struct ip_vs_sync_buff
{
265 struct list_head list
;
266 unsigned long firstuse
;
268 /* pointers for the message data */
269 struct ip_vs_sync_mesg
*mesg
;
275 static struct sockaddr_in mcast_addr
= {
276 .sin_family
= AF_INET
,
277 .sin_port
= cpu_to_be16(IP_VS_SYNC_PORT
),
278 .sin_addr
.s_addr
= cpu_to_be32(IP_VS_SYNC_GROUP
),
282 * Copy of struct ip_vs_seq
283 * From unaligned network order to aligned host order
285 static void ntoh_seq(struct ip_vs_seq
*no
, struct ip_vs_seq
*ho
)
287 ho
->init_seq
= get_unaligned_be32(&no
->init_seq
);
288 ho
->delta
= get_unaligned_be32(&no
->delta
);
289 ho
->previous_delta
= get_unaligned_be32(&no
->previous_delta
);
293 * Copy of struct ip_vs_seq
294 * From Aligned host order to unaligned network order
296 static void hton_seq(struct ip_vs_seq
*ho
, struct ip_vs_seq
*no
)
298 put_unaligned_be32(ho
->init_seq
, &no
->init_seq
);
299 put_unaligned_be32(ho
->delta
, &no
->delta
);
300 put_unaligned_be32(ho
->previous_delta
, &no
->previous_delta
);
303 static inline struct ip_vs_sync_buff
*sb_dequeue(struct netns_ipvs
*ipvs
)
305 struct ip_vs_sync_buff
*sb
;
307 spin_lock_bh(&ipvs
->sync_lock
);
308 if (list_empty(&ipvs
->sync_queue
)) {
311 sb
= list_entry(ipvs
->sync_queue
.next
,
312 struct ip_vs_sync_buff
,
316 spin_unlock_bh(&ipvs
->sync_lock
);
322 * Create a new sync buffer for Version 1 proto.
324 static inline struct ip_vs_sync_buff
*
325 ip_vs_sync_buff_create(struct netns_ipvs
*ipvs
)
327 struct ip_vs_sync_buff
*sb
;
329 if (!(sb
=kmalloc(sizeof(struct ip_vs_sync_buff
), GFP_ATOMIC
)))
332 sb
->mesg
= kmalloc(ipvs
->send_mesg_maxlen
, GFP_ATOMIC
);
337 sb
->mesg
->reserved
= 0; /* old nr_conns i.e. must be zeo now */
338 sb
->mesg
->version
= SYNC_PROTO_VER
;
339 sb
->mesg
->syncid
= ipvs
->master_syncid
;
340 sb
->mesg
->size
= sizeof(struct ip_vs_sync_mesg
);
341 sb
->mesg
->nr_conns
= 0;
343 sb
->head
= (unsigned char *)sb
->mesg
+ sizeof(struct ip_vs_sync_mesg
);
344 sb
->end
= (unsigned char *)sb
->mesg
+ ipvs
->send_mesg_maxlen
;
346 sb
->firstuse
= jiffies
;
350 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff
*sb
)
356 static inline void sb_queue_tail(struct netns_ipvs
*ipvs
)
358 struct ip_vs_sync_buff
*sb
= ipvs
->sync_buff
;
360 spin_lock(&ipvs
->sync_lock
);
361 if (ipvs
->sync_state
& IP_VS_STATE_MASTER
)
362 list_add_tail(&sb
->list
, &ipvs
->sync_queue
);
364 ip_vs_sync_buff_release(sb
);
365 spin_unlock(&ipvs
->sync_lock
);
369 * Get the current sync buffer if it has been created for more
370 * than the specified time or the specified time is zero.
372 static inline struct ip_vs_sync_buff
*
373 get_curr_sync_buff(struct netns_ipvs
*ipvs
, unsigned long time
)
375 struct ip_vs_sync_buff
*sb
;
377 spin_lock_bh(&ipvs
->sync_buff_lock
);
378 if (ipvs
->sync_buff
&&
379 time_after_eq(jiffies
- ipvs
->sync_buff
->firstuse
, time
)) {
380 sb
= ipvs
->sync_buff
;
381 ipvs
->sync_buff
= NULL
;
384 spin_unlock_bh(&ipvs
->sync_buff_lock
);
389 * Switch mode from sending version 0 or 1
390 * - must handle sync_buf
392 void ip_vs_sync_switch_mode(struct net
*net
, int mode
)
394 struct netns_ipvs
*ipvs
= net_ipvs(net
);
396 if (!(ipvs
->sync_state
& IP_VS_STATE_MASTER
))
398 if (mode
== sysctl_sync_ver(ipvs
) || !ipvs
->sync_buff
)
401 spin_lock_bh(&ipvs
->sync_buff_lock
);
402 /* Buffer empty ? then let buf_create do the job */
403 if (ipvs
->sync_buff
->mesg
->size
<= sizeof(struct ip_vs_sync_mesg
)) {
404 kfree(ipvs
->sync_buff
);
405 ipvs
->sync_buff
= NULL
;
407 spin_lock_bh(&ipvs
->sync_lock
);
408 if (ipvs
->sync_state
& IP_VS_STATE_MASTER
)
409 list_add_tail(&ipvs
->sync_buff
->list
,
412 ip_vs_sync_buff_release(ipvs
->sync_buff
);
413 spin_unlock_bh(&ipvs
->sync_lock
);
415 spin_unlock_bh(&ipvs
->sync_buff_lock
);
419 * Create a new sync buffer for Version 0 proto.
421 static inline struct ip_vs_sync_buff
*
422 ip_vs_sync_buff_create_v0(struct netns_ipvs
*ipvs
)
424 struct ip_vs_sync_buff
*sb
;
425 struct ip_vs_sync_mesg_v0
*mesg
;
427 if (!(sb
=kmalloc(sizeof(struct ip_vs_sync_buff
), GFP_ATOMIC
)))
430 sb
->mesg
= kmalloc(ipvs
->send_mesg_maxlen
, GFP_ATOMIC
);
435 mesg
= (struct ip_vs_sync_mesg_v0
*)sb
->mesg
;
437 mesg
->syncid
= ipvs
->master_syncid
;
438 mesg
->size
= sizeof(struct ip_vs_sync_mesg_v0
);
439 sb
->head
= (unsigned char *)mesg
+ sizeof(struct ip_vs_sync_mesg_v0
);
440 sb
->end
= (unsigned char *)mesg
+ ipvs
->send_mesg_maxlen
;
441 sb
->firstuse
= jiffies
;
446 * Version 0 , could be switched in by sys_ctl.
447 * Add an ip_vs_conn information into the current sync_buff.
449 void ip_vs_sync_conn_v0(struct net
*net
, struct ip_vs_conn
*cp
)
451 struct netns_ipvs
*ipvs
= net_ipvs(net
);
452 struct ip_vs_sync_mesg_v0
*m
;
453 struct ip_vs_sync_conn_v0
*s
;
456 if (unlikely(cp
->af
!= AF_INET
))
458 /* Do not sync ONE PACKET */
459 if (cp
->flags
& IP_VS_CONN_F_ONE_PACKET
)
462 spin_lock(&ipvs
->sync_buff_lock
);
463 if (!ipvs
->sync_buff
) {
465 ip_vs_sync_buff_create_v0(ipvs
);
466 if (!ipvs
->sync_buff
) {
467 spin_unlock(&ipvs
->sync_buff_lock
);
468 pr_err("ip_vs_sync_buff_create failed.\n");
473 len
= (cp
->flags
& IP_VS_CONN_F_SEQ_MASK
) ? FULL_CONN_SIZE
:
475 m
= (struct ip_vs_sync_mesg_v0
*)ipvs
->sync_buff
->mesg
;
476 s
= (struct ip_vs_sync_conn_v0
*)ipvs
->sync_buff
->head
;
480 s
->protocol
= cp
->protocol
;
481 s
->cport
= cp
->cport
;
482 s
->vport
= cp
->vport
;
483 s
->dport
= cp
->dport
;
484 s
->caddr
= cp
->caddr
.ip
;
485 s
->vaddr
= cp
->vaddr
.ip
;
486 s
->daddr
= cp
->daddr
.ip
;
487 s
->flags
= htons(cp
->flags
& ~IP_VS_CONN_F_HASHED
);
488 s
->state
= htons(cp
->state
);
489 if (cp
->flags
& IP_VS_CONN_F_SEQ_MASK
) {
490 struct ip_vs_sync_conn_options
*opt
=
491 (struct ip_vs_sync_conn_options
*)&s
[1];
492 memcpy(opt
, &cp
->in_seq
, sizeof(*opt
));
497 ipvs
->sync_buff
->head
+= len
;
499 /* check if there is a space for next one */
500 if (ipvs
->sync_buff
->head
+ FULL_CONN_SIZE
> ipvs
->sync_buff
->end
) {
502 ipvs
->sync_buff
= NULL
;
504 spin_unlock(&ipvs
->sync_buff_lock
);
506 /* synchronize its controller if it has */
508 ip_vs_sync_conn(net
, cp
->control
);
512 * Add an ip_vs_conn information into the current sync_buff.
513 * Called by ip_vs_in.
514 * Sending Version 1 messages
516 void ip_vs_sync_conn(struct net
*net
, struct ip_vs_conn
*cp
)
518 struct netns_ipvs
*ipvs
= net_ipvs(net
);
519 struct ip_vs_sync_mesg
*m
;
520 union ip_vs_sync_conn
*s
;
522 unsigned int len
, pe_name_len
, pad
;
524 /* Handle old version of the protocol */
525 if (sysctl_sync_ver(ipvs
) == 0) {
526 ip_vs_sync_conn_v0(net
, cp
);
529 /* Do not sync ONE PACKET */
530 if (cp
->flags
& IP_VS_CONN_F_ONE_PACKET
)
535 if (cp
->pe_data_len
) {
536 if (!cp
->pe_data
|| !cp
->dest
) {
537 IP_VS_ERR_RL("SYNC, connection pe_data invalid\n");
540 pe_name_len
= strnlen(cp
->pe
->name
, IP_VS_PENAME_MAXLEN
);
543 spin_lock(&ipvs
->sync_buff_lock
);
545 #ifdef CONFIG_IP_VS_IPV6
546 if (cp
->af
== AF_INET6
)
547 len
= sizeof(struct ip_vs_sync_v6
);
550 len
= sizeof(struct ip_vs_sync_v4
);
552 if (cp
->flags
& IP_VS_CONN_F_SEQ_MASK
)
553 len
+= sizeof(struct ip_vs_sync_conn_options
) + 2;
556 len
+= cp
->pe_data_len
+ 2; /* + Param hdr field */
558 len
+= pe_name_len
+ 2;
560 /* check if there is a space for this one */
562 if (ipvs
->sync_buff
) {
563 pad
= (4 - (size_t)ipvs
->sync_buff
->head
) & 3;
564 if (ipvs
->sync_buff
->head
+ len
+ pad
> ipvs
->sync_buff
->end
) {
566 ipvs
->sync_buff
= NULL
;
571 if (!ipvs
->sync_buff
) {
572 ipvs
->sync_buff
= ip_vs_sync_buff_create(ipvs
);
573 if (!ipvs
->sync_buff
) {
574 spin_unlock(&ipvs
->sync_buff_lock
);
575 pr_err("ip_vs_sync_buff_create failed.\n");
580 m
= ipvs
->sync_buff
->mesg
;
581 p
= ipvs
->sync_buff
->head
;
582 ipvs
->sync_buff
->head
+= pad
+ len
;
583 m
->size
+= pad
+ len
;
584 /* Add ev. padding from prev. sync_conn */
588 s
= (union ip_vs_sync_conn
*)p
;
590 /* Set message type & copy members */
591 s
->v4
.type
= (cp
->af
== AF_INET6
? STYPE_F_INET6
: 0);
592 s
->v4
.ver_size
= htons(len
& SVER_MASK
); /* Version 0 */
593 s
->v4
.flags
= htonl(cp
->flags
& ~IP_VS_CONN_F_HASHED
);
594 s
->v4
.state
= htons(cp
->state
);
595 s
->v4
.protocol
= cp
->protocol
;
596 s
->v4
.cport
= cp
->cport
;
597 s
->v4
.vport
= cp
->vport
;
598 s
->v4
.dport
= cp
->dport
;
599 s
->v4
.fwmark
= htonl(cp
->fwmark
);
600 s
->v4
.timeout
= htonl(cp
->timeout
/ HZ
);
603 #ifdef CONFIG_IP_VS_IPV6
604 if (cp
->af
== AF_INET6
) {
605 p
+= sizeof(struct ip_vs_sync_v6
);
606 ipv6_addr_copy(&s
->v6
.caddr
, &cp
->caddr
.in6
);
607 ipv6_addr_copy(&s
->v6
.vaddr
, &cp
->vaddr
.in6
);
608 ipv6_addr_copy(&s
->v6
.daddr
, &cp
->daddr
.in6
);
612 p
+= sizeof(struct ip_vs_sync_v4
); /* options ptr */
613 s
->v4
.caddr
= cp
->caddr
.ip
;
614 s
->v4
.vaddr
= cp
->vaddr
.ip
;
615 s
->v4
.daddr
= cp
->daddr
.ip
;
617 if (cp
->flags
& IP_VS_CONN_F_SEQ_MASK
) {
618 *(p
++) = IPVS_OPT_SEQ_DATA
;
619 *(p
++) = sizeof(struct ip_vs_sync_conn_options
);
620 hton_seq((struct ip_vs_seq
*)p
, &cp
->in_seq
);
621 p
+= sizeof(struct ip_vs_seq
);
622 hton_seq((struct ip_vs_seq
*)p
, &cp
->out_seq
);
623 p
+= sizeof(struct ip_vs_seq
);
626 if (cp
->pe_data_len
&& cp
->pe_data
) {
627 *(p
++) = IPVS_OPT_PE_DATA
;
628 *(p
++) = cp
->pe_data_len
;
629 memcpy(p
, cp
->pe_data
, cp
->pe_data_len
);
630 p
+= cp
->pe_data_len
;
633 *(p
++) = IPVS_OPT_PE_NAME
;
634 *(p
++) = pe_name_len
;
635 memcpy(p
, cp
->pe
->name
, pe_name_len
);
640 spin_unlock(&ipvs
->sync_buff_lock
);
643 /* synchronize its controller if it has */
648 * Reduce sync rate for templates
649 * i.e only increment in_pkts for Templates.
651 if (cp
->flags
& IP_VS_CONN_F_TEMPLATE
) {
652 int pkts
= atomic_add_return(1, &cp
->in_pkts
);
654 if (pkts
% sysctl_sync_period(ipvs
) != 1)
661 * fill_param used by version 1
664 ip_vs_conn_fill_param_sync(struct net
*net
, int af
, union ip_vs_sync_conn
*sc
,
665 struct ip_vs_conn_param
*p
,
666 __u8
*pe_data
, unsigned int pe_data_len
,
667 __u8
*pe_name
, unsigned int pe_name_len
)
669 #ifdef CONFIG_IP_VS_IPV6
671 ip_vs_conn_fill_param(net
, af
, sc
->v6
.protocol
,
672 (const union nf_inet_addr
*)&sc
->v6
.caddr
,
674 (const union nf_inet_addr
*)&sc
->v6
.vaddr
,
678 ip_vs_conn_fill_param(net
, af
, sc
->v4
.protocol
,
679 (const union nf_inet_addr
*)&sc
->v4
.caddr
,
681 (const union nf_inet_addr
*)&sc
->v4
.vaddr
,
686 char buff
[IP_VS_PENAME_MAXLEN
+1];
688 memcpy(buff
, pe_name
, pe_name_len
);
690 p
->pe
= __ip_vs_pe_getbyname(buff
);
692 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n",
697 IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n");
701 p
->pe_data
= kmemdup(pe_data
, pe_data_len
, GFP_ATOMIC
);
704 module_put(p
->pe
->module
);
707 p
->pe_data_len
= pe_data_len
;
713 * Connection Add / Update.
714 * Common for version 0 and 1 reception of backup sync_conns.
718 static void ip_vs_proc_conn(struct net
*net
, struct ip_vs_conn_param
*param
,
719 unsigned int flags
, unsigned int state
,
720 unsigned int protocol
, unsigned int type
,
721 const union nf_inet_addr
*daddr
, __be16 dport
,
722 unsigned long timeout
, __u32 fwmark
,
723 struct ip_vs_sync_conn_options
*opt
)
725 struct ip_vs_dest
*dest
;
726 struct ip_vs_conn
*cp
;
727 struct netns_ipvs
*ipvs
= net_ipvs(net
);
729 if (!(flags
& IP_VS_CONN_F_TEMPLATE
))
730 cp
= ip_vs_conn_in_get(param
);
732 cp
= ip_vs_ct_in_get(param
);
734 if (cp
&& param
->pe_data
) /* Free pe_data */
735 kfree(param
->pe_data
);
738 * Find the appropriate destination for the connection.
739 * If it is not found the connection will remain unbound
742 dest
= ip_vs_find_dest(net
, type
, daddr
, dport
, param
->vaddr
,
743 param
->vport
, protocol
, fwmark
);
745 /* Set the approprite ativity flag */
746 if (protocol
== IPPROTO_TCP
) {
747 if (state
!= IP_VS_TCP_S_ESTABLISHED
)
748 flags
|= IP_VS_CONN_F_INACTIVE
;
750 flags
&= ~IP_VS_CONN_F_INACTIVE
;
751 } else if (protocol
== IPPROTO_SCTP
) {
752 if (state
!= IP_VS_SCTP_S_ESTABLISHED
)
753 flags
|= IP_VS_CONN_F_INACTIVE
;
755 flags
&= ~IP_VS_CONN_F_INACTIVE
;
757 cp
= ip_vs_conn_new(param
, daddr
, dport
, flags
, dest
, fwmark
);
759 atomic_dec(&dest
->refcnt
);
762 kfree(param
->pe_data
);
763 IP_VS_DBG(2, "BACKUP, add new conn. failed\n");
766 } else if (!cp
->dest
) {
767 dest
= ip_vs_try_bind_dest(cp
);
769 atomic_dec(&dest
->refcnt
);
770 } else if ((cp
->dest
) && (cp
->protocol
== IPPROTO_TCP
) &&
771 (cp
->state
!= state
)) {
772 /* update active/inactive flag for the connection */
774 if (!(cp
->flags
& IP_VS_CONN_F_INACTIVE
) &&
775 (state
!= IP_VS_TCP_S_ESTABLISHED
)) {
776 atomic_dec(&dest
->activeconns
);
777 atomic_inc(&dest
->inactconns
);
778 cp
->flags
|= IP_VS_CONN_F_INACTIVE
;
779 } else if ((cp
->flags
& IP_VS_CONN_F_INACTIVE
) &&
780 (state
== IP_VS_TCP_S_ESTABLISHED
)) {
781 atomic_inc(&dest
->activeconns
);
782 atomic_dec(&dest
->inactconns
);
783 cp
->flags
&= ~IP_VS_CONN_F_INACTIVE
;
785 } else if ((cp
->dest
) && (cp
->protocol
== IPPROTO_SCTP
) &&
786 (cp
->state
!= state
)) {
788 if (!(cp
->flags
& IP_VS_CONN_F_INACTIVE
) &&
789 (state
!= IP_VS_SCTP_S_ESTABLISHED
)) {
790 atomic_dec(&dest
->activeconns
);
791 atomic_inc(&dest
->inactconns
);
792 cp
->flags
&= ~IP_VS_CONN_F_INACTIVE
;
797 memcpy(&cp
->in_seq
, opt
, sizeof(*opt
));
798 atomic_set(&cp
->in_pkts
, sysctl_sync_threshold(ipvs
));
800 cp
->old_state
= cp
->state
;
802 * For Ver 0 messages style
803 * - Not possible to recover the right timeout for templates
804 * - can not find the right fwmark
805 * virtual service. If needed, we can do it for
806 * non-fwmark persistent services.
807 * Ver 1 messages style.
811 if (timeout
> MAX_SCHEDULE_TIMEOUT
/ HZ
)
812 timeout
= MAX_SCHEDULE_TIMEOUT
/ HZ
;
813 cp
->timeout
= timeout
*HZ
;
815 struct ip_vs_proto_data
*pd
;
817 pd
= ip_vs_proto_data_get(net
, protocol
);
818 if (!(flags
& IP_VS_CONN_F_TEMPLATE
) && pd
&& pd
->timeout_table
)
819 cp
->timeout
= pd
->timeout_table
[state
];
821 cp
->timeout
= (3*60*HZ
);
827 * Process received multicast message for Version 0
829 static void ip_vs_process_message_v0(struct net
*net
, const char *buffer
,
832 struct ip_vs_sync_mesg_v0
*m
= (struct ip_vs_sync_mesg_v0
*)buffer
;
833 struct ip_vs_sync_conn_v0
*s
;
834 struct ip_vs_sync_conn_options
*opt
;
835 struct ip_vs_protocol
*pp
;
836 struct ip_vs_conn_param param
;
840 p
= (char *)buffer
+ sizeof(struct ip_vs_sync_mesg_v0
);
841 for (i
=0; i
<m
->nr_conns
; i
++) {
842 unsigned flags
, state
;
844 if (p
+ SIMPLE_CONN_SIZE
> buffer
+buflen
) {
845 IP_VS_ERR_RL("BACKUP v0, bogus conn\n");
848 s
= (struct ip_vs_sync_conn_v0
*) p
;
849 flags
= ntohs(s
->flags
) | IP_VS_CONN_F_SYNC
;
850 flags
&= ~IP_VS_CONN_F_HASHED
;
851 if (flags
& IP_VS_CONN_F_SEQ_MASK
) {
852 opt
= (struct ip_vs_sync_conn_options
*)&s
[1];
854 if (p
> buffer
+buflen
) {
855 IP_VS_ERR_RL("BACKUP v0, Dropping buffer bogus conn options\n");
860 p
+= SIMPLE_CONN_SIZE
;
863 state
= ntohs(s
->state
);
864 if (!(flags
& IP_VS_CONN_F_TEMPLATE
)) {
865 pp
= ip_vs_proto_get(s
->protocol
);
867 IP_VS_DBG(2, "BACKUP v0, Unsupported protocol %u\n",
871 if (state
>= pp
->num_states
) {
872 IP_VS_DBG(2, "BACKUP v0, Invalid %s state %u\n",
877 /* protocol in templates is not used for state/timeout */
879 IP_VS_DBG(2, "BACKUP v0, Invalid template state %u\n",
885 ip_vs_conn_fill_param(net
, AF_INET
, s
->protocol
,
886 (const union nf_inet_addr
*)&s
->caddr
,
888 (const union nf_inet_addr
*)&s
->vaddr
,
891 /* Send timeout as Zero */
892 ip_vs_proc_conn(net
, ¶m
, flags
, state
, s
->protocol
, AF_INET
,
893 (union nf_inet_addr
*)&s
->daddr
, s
->dport
,
901 static inline int ip_vs_proc_seqopt(__u8
*p
, unsigned int plen
,
903 struct ip_vs_sync_conn_options
*opt
)
905 struct ip_vs_sync_conn_options
*topt
;
907 topt
= (struct ip_vs_sync_conn_options
*)p
;
909 if (plen
!= sizeof(struct ip_vs_sync_conn_options
)) {
910 IP_VS_DBG(2, "BACKUP, bogus conn options length\n");
913 if (*opt_flags
& IPVS_OPT_F_SEQ_DATA
) {
914 IP_VS_DBG(2, "BACKUP, conn options found twice\n");
917 ntoh_seq(&topt
->in_seq
, &opt
->in_seq
);
918 ntoh_seq(&topt
->out_seq
, &opt
->out_seq
);
919 *opt_flags
|= IPVS_OPT_F_SEQ_DATA
;
923 static int ip_vs_proc_str(__u8
*p
, unsigned int plen
, unsigned int *data_len
,
924 __u8
**data
, unsigned int maxlen
,
925 __u32
*opt_flags
, __u32 flag
)
928 IP_VS_DBG(2, "BACKUP, bogus par.data len > %d\n", maxlen
);
931 if (*opt_flags
& flag
) {
932 IP_VS_DBG(2, "BACKUP, Par.data found twice 0x%x\n", flag
);
941 * Process a Version 1 sync. connection
943 static inline int ip_vs_proc_sync_conn(struct net
*net
, __u8
*p
, __u8
*msg_end
)
945 struct ip_vs_sync_conn_options opt
;
946 union ip_vs_sync_conn
*s
;
947 struct ip_vs_protocol
*pp
;
948 struct ip_vs_conn_param param
;
950 unsigned int af
, state
, pe_data_len
=0, pe_name_len
=0;
951 __u8
*pe_data
=NULL
, *pe_name
=NULL
;
955 s
= (union ip_vs_sync_conn
*) p
;
957 if (s
->v6
.type
& STYPE_F_INET6
) {
958 #ifdef CONFIG_IP_VS_IPV6
960 p
+= sizeof(struct ip_vs_sync_v6
);
962 IP_VS_DBG(3,"BACKUP, IPv6 msg received, and IPVS is not compiled for IPv6\n");
966 } else if (!s
->v4
.type
) {
968 p
+= sizeof(struct ip_vs_sync_v4
);
975 /* Process optional params check Type & Len. */
976 while (p
< msg_end
) {
985 if (!plen
|| ((p
+ plen
) > msg_end
))
987 /* Handle seq option p = param data */
988 switch (ptype
& ~IPVS_OPT_F_PARAM
) {
989 case IPVS_OPT_SEQ_DATA
:
990 if (ip_vs_proc_seqopt(p
, plen
, &opt_flags
, &opt
))
994 case IPVS_OPT_PE_DATA
:
995 if (ip_vs_proc_str(p
, plen
, &pe_data_len
, &pe_data
,
996 IP_VS_PEDATA_MAXLEN
, &opt_flags
,
1001 case IPVS_OPT_PE_NAME
:
1002 if (ip_vs_proc_str(p
, plen
,&pe_name_len
, &pe_name
,
1003 IP_VS_PENAME_MAXLEN
, &opt_flags
,
1004 IPVS_OPT_F_PE_NAME
))
1009 /* Param data mandatory ? */
1010 if (!(ptype
& IPVS_OPT_F_PARAM
)) {
1011 IP_VS_DBG(3, "BACKUP, Unknown mandatory param %d found\n",
1012 ptype
& ~IPVS_OPT_F_PARAM
);
1017 p
+= plen
; /* Next option */
1020 /* Get flags and Mask off unsupported */
1021 flags
= ntohl(s
->v4
.flags
) & IP_VS_CONN_F_BACKUP_MASK
;
1022 flags
|= IP_VS_CONN_F_SYNC
;
1023 state
= ntohs(s
->v4
.state
);
1025 if (!(flags
& IP_VS_CONN_F_TEMPLATE
)) {
1026 pp
= ip_vs_proto_get(s
->v4
.protocol
);
1028 IP_VS_DBG(3,"BACKUP, Unsupported protocol %u\n",
1033 if (state
>= pp
->num_states
) {
1034 IP_VS_DBG(3, "BACKUP, Invalid %s state %u\n",
1040 /* protocol in templates is not used for state/timeout */
1042 IP_VS_DBG(3, "BACKUP, Invalid template state %u\n",
1047 if (ip_vs_conn_fill_param_sync(net
, af
, s
, ¶m
, pe_data
,
1048 pe_data_len
, pe_name
, pe_name_len
)) {
1052 /* If only IPv4, just silent skip IPv6 */
1054 ip_vs_proc_conn(net
, ¶m
, flags
, state
, s
->v4
.protocol
, af
,
1055 (union nf_inet_addr
*)&s
->v4
.daddr
, s
->v4
.dport
,
1056 ntohl(s
->v4
.timeout
), ntohl(s
->v4
.fwmark
),
1057 (opt_flags
& IPVS_OPT_F_SEQ_DATA
? &opt
: NULL
)
1059 #ifdef CONFIG_IP_VS_IPV6
1061 ip_vs_proc_conn(net
, ¶m
, flags
, state
, s
->v6
.protocol
, af
,
1062 (union nf_inet_addr
*)&s
->v6
.daddr
, s
->v6
.dport
,
1063 ntohl(s
->v6
.timeout
), ntohl(s
->v6
.fwmark
),
1064 (opt_flags
& IPVS_OPT_F_SEQ_DATA
? &opt
: NULL
)
1070 IP_VS_DBG(2, "BACKUP, Single msg dropped err:%d\n", retc
);
1075 * Process received multicast message and create the corresponding
1076 * ip_vs_conn entries.
1077 * Handles Version 0 & 1
1079 static void ip_vs_process_message(struct net
*net
, __u8
*buffer
,
1080 const size_t buflen
)
1082 struct netns_ipvs
*ipvs
= net_ipvs(net
);
1083 struct ip_vs_sync_mesg
*m2
= (struct ip_vs_sync_mesg
*)buffer
;
1087 if (buflen
< sizeof(struct ip_vs_sync_mesg_v0
)) {
1088 IP_VS_DBG(2, "BACKUP, message header too short\n");
1091 /* Convert size back to host byte order */
1092 m2
->size
= ntohs(m2
->size
);
1094 if (buflen
!= m2
->size
) {
1095 IP_VS_DBG(2, "BACKUP, bogus message size\n");
1098 /* SyncID sanity check */
1099 if (ipvs
->backup_syncid
!= 0 && m2
->syncid
!= ipvs
->backup_syncid
) {
1100 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2
->syncid
);
1103 /* Handle version 1 message */
1104 if ((m2
->version
== SYNC_PROTO_VER
) && (m2
->reserved
== 0)
1105 && (m2
->spare
== 0)) {
1107 msg_end
= buffer
+ sizeof(struct ip_vs_sync_mesg
);
1108 nr_conns
= m2
->nr_conns
;
1110 for (i
=0; i
<nr_conns
; i
++) {
1111 union ip_vs_sync_conn
*s
;
1116 if (p
+ sizeof(s
->v4
) > buffer
+buflen
) {
1117 IP_VS_ERR_RL("BACKUP, Dropping buffer, to small\n");
1120 s
= (union ip_vs_sync_conn
*)p
;
1121 size
= ntohs(s
->v4
.ver_size
) & SVER_MASK
;
1123 /* Basic sanity checks */
1124 if (msg_end
> buffer
+buflen
) {
1125 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n");
1128 if (ntohs(s
->v4
.ver_size
) >> SVER_SHIFT
) {
1129 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n",
1130 ntohs(s
->v4
.ver_size
) >> SVER_SHIFT
);
1133 /* Process a single sync_conn */
1134 retc
= ip_vs_proc_sync_conn(net
, p
, msg_end
);
1136 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n",
1140 /* Make sure we have 32 bit alignment */
1141 msg_end
= p
+ ((size
+ 3) & ~3);
1144 /* Old type of message */
1145 ip_vs_process_message_v0(net
, buffer
, buflen
);
1152 * Setup loopback of outgoing multicasts on a sending socket
1154 static void set_mcast_loop(struct sock
*sk
, u_char loop
)
1156 struct inet_sock
*inet
= inet_sk(sk
);
1158 /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
1160 inet
->mc_loop
= loop
? 1 : 0;
1165 * Specify TTL for outgoing multicasts on a sending socket
1167 static void set_mcast_ttl(struct sock
*sk
, u_char ttl
)
1169 struct inet_sock
*inet
= inet_sk(sk
);
1171 /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
1178 * Specifiy default interface for outgoing multicasts
1180 static int set_mcast_if(struct sock
*sk
, char *ifname
)
1182 struct net_device
*dev
;
1183 struct inet_sock
*inet
= inet_sk(sk
);
1184 struct net
*net
= sock_net(sk
);
1186 dev
= __dev_get_by_name(net
, ifname
);
1190 if (sk
->sk_bound_dev_if
&& dev
->ifindex
!= sk
->sk_bound_dev_if
)
1194 inet
->mc_index
= dev
->ifindex
;
1195 /* inet->mc_addr = 0; */
1203 * Set the maximum length of sync message according to the
1204 * specified interface's MTU.
1206 static int set_sync_mesg_maxlen(struct net
*net
, int sync_state
)
1208 struct netns_ipvs
*ipvs
= net_ipvs(net
);
1209 struct net_device
*dev
;
1212 if (sync_state
== IP_VS_STATE_MASTER
) {
1213 dev
= __dev_get_by_name(net
, ipvs
->master_mcast_ifn
);
1217 num
= (dev
->mtu
- sizeof(struct iphdr
) -
1218 sizeof(struct udphdr
) -
1219 SYNC_MESG_HEADER_LEN
- 20) / SIMPLE_CONN_SIZE
;
1220 ipvs
->send_mesg_maxlen
= SYNC_MESG_HEADER_LEN
+
1221 SIMPLE_CONN_SIZE
* min(num
, MAX_CONNS_PER_SYNCBUFF
);
1222 IP_VS_DBG(7, "setting the maximum length of sync sending "
1223 "message %d.\n", ipvs
->send_mesg_maxlen
);
1224 } else if (sync_state
== IP_VS_STATE_BACKUP
) {
1225 dev
= __dev_get_by_name(net
, ipvs
->backup_mcast_ifn
);
1229 ipvs
->recv_mesg_maxlen
= dev
->mtu
-
1230 sizeof(struct iphdr
) - sizeof(struct udphdr
);
1231 IP_VS_DBG(7, "setting the maximum length of sync receiving "
1232 "message %d.\n", ipvs
->recv_mesg_maxlen
);
1240 * Join a multicast group.
1241 * the group is specified by a class D multicast address 224.0.0.0/8
1242 * in the in_addr structure passed in as a parameter.
1245 join_mcast_group(struct sock
*sk
, struct in_addr
*addr
, char *ifname
)
1247 struct net
*net
= sock_net(sk
);
1248 struct ip_mreqn mreq
;
1249 struct net_device
*dev
;
1252 memset(&mreq
, 0, sizeof(mreq
));
1253 memcpy(&mreq
.imr_multiaddr
, addr
, sizeof(struct in_addr
));
1255 dev
= __dev_get_by_name(net
, ifname
);
1258 if (sk
->sk_bound_dev_if
&& dev
->ifindex
!= sk
->sk_bound_dev_if
)
1261 mreq
.imr_ifindex
= dev
->ifindex
;
1264 ret
= ip_mc_join_group(sk
, &mreq
);
1271 static int bind_mcastif_addr(struct socket
*sock
, char *ifname
)
1273 struct net
*net
= sock_net(sock
->sk
);
1274 struct net_device
*dev
;
1276 struct sockaddr_in sin
;
1278 dev
= __dev_get_by_name(net
, ifname
);
1282 addr
= inet_select_addr(dev
, 0, RT_SCOPE_UNIVERSE
);
1284 pr_err("You probably need to specify IP address on "
1285 "multicast interface.\n");
1287 IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
1290 /* Now bind the socket with the address of multicast interface */
1291 sin
.sin_family
= AF_INET
;
1292 sin
.sin_addr
.s_addr
= addr
;
1295 return sock
->ops
->bind(sock
, (struct sockaddr
*)&sin
, sizeof(sin
));
1299 * Set up sending multicast socket over UDP
1301 static struct socket
*make_send_sock(struct net
*net
)
1303 struct netns_ipvs
*ipvs
= net_ipvs(net
);
1304 struct socket
*sock
;
1307 /* First create a socket move it to right name space later */
1308 result
= sock_create_kern(PF_INET
, SOCK_DGRAM
, IPPROTO_UDP
, &sock
);
1310 pr_err("Error during creation of socket; terminating\n");
1311 return ERR_PTR(result
);
1314 * Kernel sockets that are a part of a namespace, should not
1315 * hold a reference to a namespace in order to allow to stop it.
1316 * After sk_change_net should be released using sk_release_kernel.
1318 sk_change_net(sock
->sk
, net
);
1319 result
= set_mcast_if(sock
->sk
, ipvs
->master_mcast_ifn
);
1321 pr_err("Error setting outbound mcast interface\n");
1325 set_mcast_loop(sock
->sk
, 0);
1326 set_mcast_ttl(sock
->sk
, 1);
1328 result
= bind_mcastif_addr(sock
, ipvs
->master_mcast_ifn
);
1330 pr_err("Error binding address of the mcast interface\n");
1334 result
= sock
->ops
->connect(sock
, (struct sockaddr
*) &mcast_addr
,
1335 sizeof(struct sockaddr
), 0);
1337 pr_err("Error connecting to the multicast addr\n");
1344 sk_release_kernel(sock
->sk
);
1345 return ERR_PTR(result
);
1350 * Set up receiving multicast socket over UDP
1352 static struct socket
*make_receive_sock(struct net
*net
)
1354 struct netns_ipvs
*ipvs
= net_ipvs(net
);
1355 struct socket
*sock
;
1358 /* First create a socket */
1359 result
= sock_create_kern(PF_INET
, SOCK_DGRAM
, IPPROTO_UDP
, &sock
);
1361 pr_err("Error during creation of socket; terminating\n");
1362 return ERR_PTR(result
);
1365 * Kernel sockets that are a part of a namespace, should not
1366 * hold a reference to a namespace in order to allow to stop it.
1367 * After sk_change_net should be released using sk_release_kernel.
1369 sk_change_net(sock
->sk
, net
);
1370 /* it is equivalent to the REUSEADDR option in user-space */
1371 sock
->sk
->sk_reuse
= 1;
1373 result
= sock
->ops
->bind(sock
, (struct sockaddr
*) &mcast_addr
,
1374 sizeof(struct sockaddr
));
1376 pr_err("Error binding to the multicast addr\n");
1380 /* join the multicast group */
1381 result
= join_mcast_group(sock
->sk
,
1382 (struct in_addr
*) &mcast_addr
.sin_addr
,
1383 ipvs
->backup_mcast_ifn
);
1385 pr_err("Error joining to the multicast group\n");
1392 sk_release_kernel(sock
->sk
);
1393 return ERR_PTR(result
);
1398 ip_vs_send_async(struct socket
*sock
, const char *buffer
, const size_t length
)
1400 struct msghdr msg
= {.msg_flags
= MSG_DONTWAIT
|MSG_NOSIGNAL
};
1405 iov
.iov_base
= (void *)buffer
;
1406 iov
.iov_len
= length
;
1408 len
= kernel_sendmsg(sock
, &msg
, &iov
, 1, (size_t)(length
));
1415 ip_vs_send_sync_msg(struct socket
*sock
, struct ip_vs_sync_mesg
*msg
)
1421 /* Put size in network byte order */
1422 msg
->size
= htons(msg
->size
);
1424 if (ip_vs_send_async(sock
, (char *)msg
, msize
) != msize
)
1425 pr_err("ip_vs_send_async error\n");
1429 ip_vs_receive(struct socket
*sock
, char *buffer
, const size_t buflen
)
1431 struct msghdr msg
= {NULL
,};
1437 /* Receive a packet */
1438 iov
.iov_base
= buffer
;
1439 iov
.iov_len
= (size_t)buflen
;
1441 len
= kernel_recvmsg(sock
, &msg
, &iov
, 1, buflen
, 0);
1451 static int sync_thread_master(void *data
)
1453 struct ip_vs_sync_thread_data
*tinfo
= data
;
1454 struct netns_ipvs
*ipvs
= net_ipvs(tinfo
->net
);
1455 struct ip_vs_sync_buff
*sb
;
1457 pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
1459 ipvs
->master_mcast_ifn
, ipvs
->master_syncid
);
1461 while (!kthread_should_stop()) {
1462 while ((sb
= sb_dequeue(ipvs
))) {
1463 ip_vs_send_sync_msg(tinfo
->sock
, sb
->mesg
);
1464 ip_vs_sync_buff_release(sb
);
1467 /* check if entries stay in ipvs->sync_buff for 2 seconds */
1468 sb
= get_curr_sync_buff(ipvs
, 2 * HZ
);
1470 ip_vs_send_sync_msg(tinfo
->sock
, sb
->mesg
);
1471 ip_vs_sync_buff_release(sb
);
1474 schedule_timeout_interruptible(HZ
);
1477 /* clean up the sync_buff queue */
1478 while ((sb
= sb_dequeue(ipvs
)))
1479 ip_vs_sync_buff_release(sb
);
1481 /* clean up the current sync_buff */
1482 sb
= get_curr_sync_buff(ipvs
, 0);
1484 ip_vs_sync_buff_release(sb
);
1486 /* release the sending multicast socket */
1487 sk_release_kernel(tinfo
->sock
->sk
);
1494 static int sync_thread_backup(void *data
)
1496 struct ip_vs_sync_thread_data
*tinfo
= data
;
1497 struct netns_ipvs
*ipvs
= net_ipvs(tinfo
->net
);
1500 pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
1502 ipvs
->backup_mcast_ifn
, ipvs
->backup_syncid
);
1504 while (!kthread_should_stop()) {
1505 wait_event_interruptible(*sk_sleep(tinfo
->sock
->sk
),
1506 !skb_queue_empty(&tinfo
->sock
->sk
->sk_receive_queue
)
1507 || kthread_should_stop());
1509 /* do we have data now? */
1510 while (!skb_queue_empty(&(tinfo
->sock
->sk
->sk_receive_queue
))) {
1511 len
= ip_vs_receive(tinfo
->sock
, tinfo
->buf
,
1512 ipvs
->recv_mesg_maxlen
);
1514 pr_err("receiving message error\n");
1518 /* disable bottom half, because it accesses the data
1519 shared by softirq while getting/creating conns */
1521 ip_vs_process_message(tinfo
->net
, tinfo
->buf
, len
);
1526 /* release the sending multicast socket */
1527 sk_release_kernel(tinfo
->sock
->sk
);
1535 int start_sync_thread(struct net
*net
, int state
, char *mcast_ifn
, __u8 syncid
)
1537 struct ip_vs_sync_thread_data
*tinfo
;
1538 struct task_struct
**realtask
, *task
;
1539 struct socket
*sock
;
1540 struct netns_ipvs
*ipvs
= net_ipvs(net
);
1541 char *name
, *buf
= NULL
;
1542 int (*threadfn
)(void *data
);
1543 int result
= -ENOMEM
;
1545 IP_VS_DBG(7, "%s(): pid %d\n", __func__
, task_pid_nr(current
));
1546 IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n",
1547 sizeof(struct ip_vs_sync_conn_v0
));
1550 if (state
== IP_VS_STATE_MASTER
) {
1551 if (ipvs
->master_thread
)
1554 strlcpy(ipvs
->master_mcast_ifn
, mcast_ifn
,
1555 sizeof(ipvs
->master_mcast_ifn
));
1556 ipvs
->master_syncid
= syncid
;
1557 realtask
= &ipvs
->master_thread
;
1558 name
= "ipvs_master:%d";
1559 threadfn
= sync_thread_master
;
1560 sock
= make_send_sock(net
);
1561 } else if (state
== IP_VS_STATE_BACKUP
) {
1562 if (ipvs
->backup_thread
)
1565 strlcpy(ipvs
->backup_mcast_ifn
, mcast_ifn
,
1566 sizeof(ipvs
->backup_mcast_ifn
));
1567 ipvs
->backup_syncid
= syncid
;
1568 realtask
= &ipvs
->backup_thread
;
1569 name
= "ipvs_backup:%d";
1570 threadfn
= sync_thread_backup
;
1571 sock
= make_receive_sock(net
);
1577 result
= PTR_ERR(sock
);
1581 set_sync_mesg_maxlen(net
, state
);
1582 if (state
== IP_VS_STATE_BACKUP
) {
1583 buf
= kmalloc(ipvs
->recv_mesg_maxlen
, GFP_KERNEL
);
1588 tinfo
= kmalloc(sizeof(*tinfo
), GFP_KERNEL
);
1596 task
= kthread_run(threadfn
, tinfo
, name
, ipvs
->gen
);
1598 result
= PTR_ERR(task
);
1602 /* mark as active */
1604 ipvs
->sync_state
|= state
;
1606 /* increase the module use count */
1607 ip_vs_use_count_inc();
1616 sk_release_kernel(sock
->sk
);
1622 int stop_sync_thread(struct net
*net
, int state
)
1624 struct netns_ipvs
*ipvs
= net_ipvs(net
);
1627 IP_VS_DBG(7, "%s(): pid %d\n", __func__
, task_pid_nr(current
));
1629 if (state
== IP_VS_STATE_MASTER
) {
1630 if (!ipvs
->master_thread
)
1633 pr_info("stopping master sync thread %d ...\n",
1634 task_pid_nr(ipvs
->master_thread
));
1637 * The lock synchronizes with sb_queue_tail(), so that we don't
1638 * add sync buffers to the queue, when we are already in
1639 * progress of stopping the master sync daemon.
1642 spin_lock_bh(&ipvs
->sync_lock
);
1643 ipvs
->sync_state
&= ~IP_VS_STATE_MASTER
;
1644 spin_unlock_bh(&ipvs
->sync_lock
);
1645 retc
= kthread_stop(ipvs
->master_thread
);
1646 ipvs
->master_thread
= NULL
;
1647 } else if (state
== IP_VS_STATE_BACKUP
) {
1648 if (!ipvs
->backup_thread
)
1651 pr_info("stopping backup sync thread %d ...\n",
1652 task_pid_nr(ipvs
->backup_thread
));
1654 ipvs
->sync_state
&= ~IP_VS_STATE_BACKUP
;
1655 retc
= kthread_stop(ipvs
->backup_thread
);
1656 ipvs
->backup_thread
= NULL
;
1659 /* decrease the module use count */
1660 ip_vs_use_count_dec();
1666 * Initialize data struct for each netns
1668 int __net_init
ip_vs_sync_net_init(struct net
*net
)
1670 struct netns_ipvs
*ipvs
= net_ipvs(net
);
1672 __mutex_init(&ipvs
->sync_mutex
, "ipvs->sync_mutex", &__ipvs_sync_key
);
1673 INIT_LIST_HEAD(&ipvs
->sync_queue
);
1674 spin_lock_init(&ipvs
->sync_lock
);
1675 spin_lock_init(&ipvs
->sync_buff_lock
);
1677 ipvs
->sync_mcast_addr
.sin_family
= AF_INET
;
1678 ipvs
->sync_mcast_addr
.sin_port
= cpu_to_be16(IP_VS_SYNC_PORT
);
1679 ipvs
->sync_mcast_addr
.sin_addr
.s_addr
= cpu_to_be32(IP_VS_SYNC_GROUP
);
1683 void ip_vs_sync_net_cleanup(struct net
*net
)
1686 struct netns_ipvs
*ipvs
= net_ipvs(net
);
1688 mutex_lock(&ipvs
->sync_mutex
);
1689 retc
= stop_sync_thread(net
, IP_VS_STATE_MASTER
);
1690 if (retc
&& retc
!= -ESRCH
)
1691 pr_err("Failed to stop Master Daemon\n");
1693 retc
= stop_sync_thread(net
, IP_VS_STATE_BACKUP
);
1694 if (retc
&& retc
!= -ESRCH
)
1695 pr_err("Failed to stop Backup Daemon\n");
1696 mutex_unlock(&ipvs
->sync_mutex
);