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
{
202 /* Version 0 definition of packet sizes */
203 #define SIMPLE_CONN_SIZE (sizeof(struct ip_vs_sync_conn_v0))
204 #define FULL_CONN_SIZE \
205 (sizeof(struct ip_vs_sync_conn_v0) + sizeof(struct ip_vs_sync_conn_options))
209 The master mulitcasts messages (Datagrams) to the backup load balancers
210 in the following format.
213 Note, first byte should be Zero, so ver 0 receivers will drop the packet.
216 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
217 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
218 | 0 | SyncID | Size |
219 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
220 | Count Conns | Version | Reserved, set to Zero |
221 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
223 | IPVS Sync Connection (1) |
224 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
228 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
230 | IPVS Sync Connection (n) |
231 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
235 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
236 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
237 | Count Conns | SyncID | Size |
238 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
239 | IPVS Sync Connection (1) |
242 #define SYNC_MESG_HEADER_LEN 4
243 #define MAX_CONNS_PER_SYNCBUFF 255 /* nr_conns in ip_vs_sync_mesg is 8 bit */
245 /* Version 0 header */
246 struct ip_vs_sync_mesg_v0
{
251 /* ip_vs_sync_conn entries start here */
254 /* Version 1 header */
255 struct ip_vs_sync_mesg
{
256 __u8 reserved
; /* must be zero */
260 __s8 version
; /* SYNC_PROTO_VER */
262 /* ip_vs_sync_conn entries start here */
265 union ipvs_sockaddr
{
266 struct sockaddr_in in
;
267 struct sockaddr_in6 in6
;
270 struct ip_vs_sync_buff
{
271 struct list_head list
;
272 unsigned long firstuse
;
274 /* pointers for the message data */
275 struct ip_vs_sync_mesg
*mesg
;
281 * Copy of struct ip_vs_seq
282 * From unaligned network order to aligned host order
284 static void ntoh_seq(struct ip_vs_seq
*no
, struct ip_vs_seq
*ho
)
286 ho
->init_seq
= get_unaligned_be32(&no
->init_seq
);
287 ho
->delta
= get_unaligned_be32(&no
->delta
);
288 ho
->previous_delta
= get_unaligned_be32(&no
->previous_delta
);
292 * Copy of struct ip_vs_seq
293 * From Aligned host order to unaligned network order
295 static void hton_seq(struct ip_vs_seq
*ho
, struct ip_vs_seq
*no
)
297 put_unaligned_be32(ho
->init_seq
, &no
->init_seq
);
298 put_unaligned_be32(ho
->delta
, &no
->delta
);
299 put_unaligned_be32(ho
->previous_delta
, &no
->previous_delta
);
302 static inline struct ip_vs_sync_buff
*
303 sb_dequeue(struct netns_ipvs
*ipvs
, struct ipvs_master_sync_state
*ms
)
305 struct ip_vs_sync_buff
*sb
;
307 spin_lock_bh(&ipvs
->sync_lock
);
308 if (list_empty(&ms
->sync_queue
)) {
310 __set_current_state(TASK_INTERRUPTIBLE
);
312 sb
= list_entry(ms
->sync_queue
.next
, struct ip_vs_sync_buff
,
315 ms
->sync_queue_len
--;
316 if (!ms
->sync_queue_len
)
317 ms
->sync_queue_delay
= 0;
319 spin_unlock_bh(&ipvs
->sync_lock
);
325 * Create a new sync buffer for Version 1 proto.
327 static inline struct ip_vs_sync_buff
*
328 ip_vs_sync_buff_create(struct netns_ipvs
*ipvs
, unsigned int len
)
330 struct ip_vs_sync_buff
*sb
;
332 if (!(sb
=kmalloc(sizeof(struct ip_vs_sync_buff
), GFP_ATOMIC
)))
335 len
= max_t(unsigned int, len
+ sizeof(struct ip_vs_sync_mesg
),
336 ipvs
->mcfg
.sync_maxlen
);
337 sb
->mesg
= kmalloc(len
, GFP_ATOMIC
);
342 sb
->mesg
->reserved
= 0; /* old nr_conns i.e. must be zero now */
343 sb
->mesg
->version
= SYNC_PROTO_VER
;
344 sb
->mesg
->syncid
= ipvs
->mcfg
.syncid
;
345 sb
->mesg
->size
= htons(sizeof(struct ip_vs_sync_mesg
));
346 sb
->mesg
->nr_conns
= 0;
348 sb
->head
= (unsigned char *)sb
->mesg
+ sizeof(struct ip_vs_sync_mesg
);
349 sb
->end
= (unsigned char *)sb
->mesg
+ len
;
351 sb
->firstuse
= jiffies
;
355 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff
*sb
)
361 static inline void sb_queue_tail(struct netns_ipvs
*ipvs
,
362 struct ipvs_master_sync_state
*ms
)
364 struct ip_vs_sync_buff
*sb
= ms
->sync_buff
;
366 spin_lock(&ipvs
->sync_lock
);
367 if (ipvs
->sync_state
& IP_VS_STATE_MASTER
&&
368 ms
->sync_queue_len
< sysctl_sync_qlen_max(ipvs
)) {
369 if (!ms
->sync_queue_len
)
370 schedule_delayed_work(&ms
->master_wakeup_work
,
371 max(IPVS_SYNC_SEND_DELAY
, 1));
372 ms
->sync_queue_len
++;
373 list_add_tail(&sb
->list
, &ms
->sync_queue
);
374 if ((++ms
->sync_queue_delay
) == IPVS_SYNC_WAKEUP_RATE
)
375 wake_up_process(ms
->master_thread
);
377 ip_vs_sync_buff_release(sb
);
378 spin_unlock(&ipvs
->sync_lock
);
382 * Get the current sync buffer if it has been created for more
383 * than the specified time or the specified time is zero.
385 static inline struct ip_vs_sync_buff
*
386 get_curr_sync_buff(struct netns_ipvs
*ipvs
, struct ipvs_master_sync_state
*ms
,
389 struct ip_vs_sync_buff
*sb
;
391 spin_lock_bh(&ipvs
->sync_buff_lock
);
393 if (sb
&& time_after_eq(jiffies
- sb
->firstuse
, time
)) {
394 ms
->sync_buff
= NULL
;
395 __set_current_state(TASK_RUNNING
);
398 spin_unlock_bh(&ipvs
->sync_buff_lock
);
403 select_master_thread_id(struct netns_ipvs
*ipvs
, struct ip_vs_conn
*cp
)
405 return ((long) cp
>> (1 + ilog2(sizeof(*cp
)))) & ipvs
->threads_mask
;
409 * Create a new sync buffer for Version 0 proto.
411 static inline struct ip_vs_sync_buff
*
412 ip_vs_sync_buff_create_v0(struct netns_ipvs
*ipvs
, unsigned int len
)
414 struct ip_vs_sync_buff
*sb
;
415 struct ip_vs_sync_mesg_v0
*mesg
;
417 if (!(sb
=kmalloc(sizeof(struct ip_vs_sync_buff
), GFP_ATOMIC
)))
420 len
= max_t(unsigned int, len
+ sizeof(struct ip_vs_sync_mesg_v0
),
421 ipvs
->mcfg
.sync_maxlen
);
422 sb
->mesg
= kmalloc(len
, GFP_ATOMIC
);
427 mesg
= (struct ip_vs_sync_mesg_v0
*)sb
->mesg
;
429 mesg
->syncid
= ipvs
->mcfg
.syncid
;
430 mesg
->size
= htons(sizeof(struct ip_vs_sync_mesg_v0
));
431 sb
->head
= (unsigned char *)mesg
+ sizeof(struct ip_vs_sync_mesg_v0
);
432 sb
->end
= (unsigned char *)mesg
+ len
;
433 sb
->firstuse
= jiffies
;
437 /* Check if connection is controlled by persistence */
438 static inline bool in_persistence(struct ip_vs_conn
*cp
)
440 for (cp
= cp
->control
; cp
; cp
= cp
->control
) {
441 if (cp
->flags
& IP_VS_CONN_F_TEMPLATE
)
447 /* Check if conn should be synced.
448 * pkts: conn packets, use sysctl_sync_threshold to avoid packet check
449 * - (1) sync_refresh_period: reduce sync rate. Additionally, retry
450 * sync_retries times with period of sync_refresh_period/8
451 * - (2) if both sync_refresh_period and sync_period are 0 send sync only
452 * for state changes or only once when pkts matches sync_threshold
453 * - (3) templates: rate can be reduced only with sync_refresh_period or
456 static int ip_vs_sync_conn_needed(struct netns_ipvs
*ipvs
,
457 struct ip_vs_conn
*cp
, int pkts
)
459 unsigned long orig
= ACCESS_ONCE(cp
->sync_endtime
);
460 unsigned long now
= jiffies
;
461 unsigned long n
= (now
+ cp
->timeout
) & ~3UL;
462 unsigned int sync_refresh_period
;
466 /* Check if we sync in current state */
467 if (unlikely(cp
->flags
& IP_VS_CONN_F_TEMPLATE
))
469 else if (unlikely(sysctl_sync_persist_mode(ipvs
) && in_persistence(cp
)))
471 else if (likely(cp
->protocol
== IPPROTO_TCP
)) {
472 if (!((1 << cp
->state
) &
473 ((1 << IP_VS_TCP_S_ESTABLISHED
) |
474 (1 << IP_VS_TCP_S_FIN_WAIT
) |
475 (1 << IP_VS_TCP_S_CLOSE
) |
476 (1 << IP_VS_TCP_S_CLOSE_WAIT
) |
477 (1 << IP_VS_TCP_S_TIME_WAIT
))))
479 force
= cp
->state
!= cp
->old_state
;
480 if (force
&& cp
->state
!= IP_VS_TCP_S_ESTABLISHED
)
482 } else if (unlikely(cp
->protocol
== IPPROTO_SCTP
)) {
483 if (!((1 << cp
->state
) &
484 ((1 << IP_VS_SCTP_S_ESTABLISHED
) |
485 (1 << IP_VS_SCTP_S_SHUTDOWN_SENT
) |
486 (1 << IP_VS_SCTP_S_SHUTDOWN_RECEIVED
) |
487 (1 << IP_VS_SCTP_S_SHUTDOWN_ACK_SENT
) |
488 (1 << IP_VS_SCTP_S_CLOSED
))))
490 force
= cp
->state
!= cp
->old_state
;
491 if (force
&& cp
->state
!= IP_VS_SCTP_S_ESTABLISHED
)
494 /* UDP or another protocol with single state */
498 sync_refresh_period
= sysctl_sync_refresh_period(ipvs
);
499 if (sync_refresh_period
> 0) {
500 long diff
= n
- orig
;
501 long min_diff
= max(cp
->timeout
>> 1, 10UL * HZ
);
503 /* Avoid sync if difference is below sync_refresh_period
504 * and below the half timeout.
506 if (abs(diff
) < min_t(long, sync_refresh_period
, min_diff
)) {
507 int retries
= orig
& 3;
509 if (retries
>= sysctl_sync_retries(ipvs
))
511 if (time_before(now
, orig
- cp
->timeout
+
512 (sync_refresh_period
>> 3)))
517 sync_period
= sysctl_sync_period(ipvs
);
518 if (sync_period
> 0) {
519 if (!(cp
->flags
& IP_VS_CONN_F_TEMPLATE
) &&
520 pkts
% sync_period
!= sysctl_sync_threshold(ipvs
))
522 } else if (sync_refresh_period
<= 0 &&
523 pkts
!= sysctl_sync_threshold(ipvs
))
527 cp
->old_state
= cp
->state
;
528 n
= cmpxchg(&cp
->sync_endtime
, orig
, n
);
529 return n
== orig
|| force
;
533 * Version 0 , could be switched in by sys_ctl.
534 * Add an ip_vs_conn information into the current sync_buff.
536 static void ip_vs_sync_conn_v0(struct net
*net
, struct ip_vs_conn
*cp
,
539 struct netns_ipvs
*ipvs
= net_ipvs(net
);
540 struct ip_vs_sync_mesg_v0
*m
;
541 struct ip_vs_sync_conn_v0
*s
;
542 struct ip_vs_sync_buff
*buff
;
543 struct ipvs_master_sync_state
*ms
;
547 if (unlikely(cp
->af
!= AF_INET
))
549 /* Do not sync ONE PACKET */
550 if (cp
->flags
& IP_VS_CONN_F_ONE_PACKET
)
553 if (!ip_vs_sync_conn_needed(ipvs
, cp
, pkts
))
556 spin_lock_bh(&ipvs
->sync_buff_lock
);
557 if (!(ipvs
->sync_state
& IP_VS_STATE_MASTER
)) {
558 spin_unlock_bh(&ipvs
->sync_buff_lock
);
562 id
= select_master_thread_id(ipvs
, cp
);
564 buff
= ms
->sync_buff
;
565 len
= (cp
->flags
& IP_VS_CONN_F_SEQ_MASK
) ? FULL_CONN_SIZE
:
568 m
= (struct ip_vs_sync_mesg_v0
*) buff
->mesg
;
569 /* Send buffer if it is for v1 */
570 if (buff
->head
+ len
> buff
->end
|| !m
->nr_conns
) {
571 sb_queue_tail(ipvs
, ms
);
572 ms
->sync_buff
= NULL
;
577 buff
= ip_vs_sync_buff_create_v0(ipvs
, len
);
579 spin_unlock_bh(&ipvs
->sync_buff_lock
);
580 pr_err("ip_vs_sync_buff_create failed.\n");
583 ms
->sync_buff
= buff
;
586 m
= (struct ip_vs_sync_mesg_v0
*) buff
->mesg
;
587 s
= (struct ip_vs_sync_conn_v0
*) buff
->head
;
591 s
->protocol
= cp
->protocol
;
592 s
->cport
= cp
->cport
;
593 s
->vport
= cp
->vport
;
594 s
->dport
= cp
->dport
;
595 s
->caddr
= cp
->caddr
.ip
;
596 s
->vaddr
= cp
->vaddr
.ip
;
597 s
->daddr
= cp
->daddr
.ip
;
598 s
->flags
= htons(cp
->flags
& ~IP_VS_CONN_F_HASHED
);
599 s
->state
= htons(cp
->state
);
600 if (cp
->flags
& IP_VS_CONN_F_SEQ_MASK
) {
601 struct ip_vs_sync_conn_options
*opt
=
602 (struct ip_vs_sync_conn_options
*)&s
[1];
603 memcpy(opt
, &cp
->in_seq
, sizeof(*opt
));
607 m
->size
= htons(ntohs(m
->size
) + len
);
609 spin_unlock_bh(&ipvs
->sync_buff_lock
);
611 /* synchronize its controller if it has */
614 if (cp
->flags
& IP_VS_CONN_F_TEMPLATE
)
615 pkts
= atomic_add_return(1, &cp
->in_pkts
);
617 pkts
= sysctl_sync_threshold(ipvs
);
618 ip_vs_sync_conn(net
, cp
, pkts
);
623 * Add an ip_vs_conn information into the current sync_buff.
624 * Called by ip_vs_in.
625 * Sending Version 1 messages
627 void ip_vs_sync_conn(struct net
*net
, struct ip_vs_conn
*cp
, int pkts
)
629 struct netns_ipvs
*ipvs
= net_ipvs(net
);
630 struct ip_vs_sync_mesg
*m
;
631 union ip_vs_sync_conn
*s
;
632 struct ip_vs_sync_buff
*buff
;
633 struct ipvs_master_sync_state
*ms
;
636 unsigned int len
, pe_name_len
, pad
;
638 /* Handle old version of the protocol */
639 if (sysctl_sync_ver(ipvs
) == 0) {
640 ip_vs_sync_conn_v0(net
, cp
, pkts
);
643 /* Do not sync ONE PACKET */
644 if (cp
->flags
& IP_VS_CONN_F_ONE_PACKET
)
647 if (!ip_vs_sync_conn_needed(ipvs
, cp
, pkts
))
652 if (cp
->pe_data_len
) {
653 if (!cp
->pe_data
|| !cp
->dest
) {
654 IP_VS_ERR_RL("SYNC, connection pe_data invalid\n");
657 pe_name_len
= strnlen(cp
->pe
->name
, IP_VS_PENAME_MAXLEN
);
660 spin_lock_bh(&ipvs
->sync_buff_lock
);
661 if (!(ipvs
->sync_state
& IP_VS_STATE_MASTER
)) {
662 spin_unlock_bh(&ipvs
->sync_buff_lock
);
666 id
= select_master_thread_id(ipvs
, cp
);
669 #ifdef CONFIG_IP_VS_IPV6
670 if (cp
->af
== AF_INET6
)
671 len
= sizeof(struct ip_vs_sync_v6
);
674 len
= sizeof(struct ip_vs_sync_v4
);
676 if (cp
->flags
& IP_VS_CONN_F_SEQ_MASK
)
677 len
+= sizeof(struct ip_vs_sync_conn_options
) + 2;
680 len
+= cp
->pe_data_len
+ 2; /* + Param hdr field */
682 len
+= pe_name_len
+ 2;
684 /* check if there is a space for this one */
686 buff
= ms
->sync_buff
;
689 pad
= (4 - (size_t) buff
->head
) & 3;
690 /* Send buffer if it is for v0 */
691 if (buff
->head
+ len
+ pad
> buff
->end
|| m
->reserved
) {
692 sb_queue_tail(ipvs
, ms
);
693 ms
->sync_buff
= NULL
;
700 buff
= ip_vs_sync_buff_create(ipvs
, len
);
702 spin_unlock_bh(&ipvs
->sync_buff_lock
);
703 pr_err("ip_vs_sync_buff_create failed.\n");
706 ms
->sync_buff
= buff
;
711 buff
->head
+= pad
+ len
;
712 m
->size
= htons(ntohs(m
->size
) + pad
+ len
);
713 /* Add ev. padding from prev. sync_conn */
717 s
= (union ip_vs_sync_conn
*)p
;
719 /* Set message type & copy members */
720 s
->v4
.type
= (cp
->af
== AF_INET6
? STYPE_F_INET6
: 0);
721 s
->v4
.ver_size
= htons(len
& SVER_MASK
); /* Version 0 */
722 s
->v4
.flags
= htonl(cp
->flags
& ~IP_VS_CONN_F_HASHED
);
723 s
->v4
.state
= htons(cp
->state
);
724 s
->v4
.protocol
= cp
->protocol
;
725 s
->v4
.cport
= cp
->cport
;
726 s
->v4
.vport
= cp
->vport
;
727 s
->v4
.dport
= cp
->dport
;
728 s
->v4
.fwmark
= htonl(cp
->fwmark
);
729 s
->v4
.timeout
= htonl(cp
->timeout
/ HZ
);
732 #ifdef CONFIG_IP_VS_IPV6
733 if (cp
->af
== AF_INET6
) {
734 p
+= sizeof(struct ip_vs_sync_v6
);
735 s
->v6
.caddr
= cp
->caddr
.in6
;
736 s
->v6
.vaddr
= cp
->vaddr
.in6
;
737 s
->v6
.daddr
= cp
->daddr
.in6
;
741 p
+= sizeof(struct ip_vs_sync_v4
); /* options ptr */
742 s
->v4
.caddr
= cp
->caddr
.ip
;
743 s
->v4
.vaddr
= cp
->vaddr
.ip
;
744 s
->v4
.daddr
= cp
->daddr
.ip
;
746 if (cp
->flags
& IP_VS_CONN_F_SEQ_MASK
) {
747 *(p
++) = IPVS_OPT_SEQ_DATA
;
748 *(p
++) = sizeof(struct ip_vs_sync_conn_options
);
749 hton_seq((struct ip_vs_seq
*)p
, &cp
->in_seq
);
750 p
+= sizeof(struct ip_vs_seq
);
751 hton_seq((struct ip_vs_seq
*)p
, &cp
->out_seq
);
752 p
+= sizeof(struct ip_vs_seq
);
755 if (cp
->pe_data_len
&& cp
->pe_data
) {
756 *(p
++) = IPVS_OPT_PE_DATA
;
757 *(p
++) = cp
->pe_data_len
;
758 memcpy(p
, cp
->pe_data
, cp
->pe_data_len
);
759 p
+= cp
->pe_data_len
;
762 *(p
++) = IPVS_OPT_PE_NAME
;
763 *(p
++) = pe_name_len
;
764 memcpy(p
, cp
->pe
->name
, pe_name_len
);
769 spin_unlock_bh(&ipvs
->sync_buff_lock
);
772 /* synchronize its controller if it has */
776 if (cp
->flags
& IP_VS_CONN_F_TEMPLATE
)
777 pkts
= atomic_add_return(1, &cp
->in_pkts
);
779 pkts
= sysctl_sync_threshold(ipvs
);
784 * fill_param used by version 1
787 ip_vs_conn_fill_param_sync(struct netns_ipvs
*ipvs
, int af
, union ip_vs_sync_conn
*sc
,
788 struct ip_vs_conn_param
*p
,
789 __u8
*pe_data
, unsigned int pe_data_len
,
790 __u8
*pe_name
, unsigned int pe_name_len
)
792 #ifdef CONFIG_IP_VS_IPV6
794 ip_vs_conn_fill_param(ipvs
, af
, sc
->v6
.protocol
,
795 (const union nf_inet_addr
*)&sc
->v6
.caddr
,
797 (const union nf_inet_addr
*)&sc
->v6
.vaddr
,
801 ip_vs_conn_fill_param(ipvs
, af
, sc
->v4
.protocol
,
802 (const union nf_inet_addr
*)&sc
->v4
.caddr
,
804 (const union nf_inet_addr
*)&sc
->v4
.vaddr
,
809 char buff
[IP_VS_PENAME_MAXLEN
+1];
811 memcpy(buff
, pe_name
, pe_name_len
);
813 p
->pe
= __ip_vs_pe_getbyname(buff
);
815 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n",
820 IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n");
824 p
->pe_data
= kmemdup(pe_data
, pe_data_len
, GFP_ATOMIC
);
826 module_put(p
->pe
->module
);
829 p
->pe_data_len
= pe_data_len
;
835 * Connection Add / Update.
836 * Common for version 0 and 1 reception of backup sync_conns.
840 static void ip_vs_proc_conn(struct net
*net
, struct ip_vs_conn_param
*param
,
841 unsigned int flags
, unsigned int state
,
842 unsigned int protocol
, unsigned int type
,
843 const union nf_inet_addr
*daddr
, __be16 dport
,
844 unsigned long timeout
, __u32 fwmark
,
845 struct ip_vs_sync_conn_options
*opt
)
847 struct ip_vs_dest
*dest
;
848 struct ip_vs_conn
*cp
;
849 struct netns_ipvs
*ipvs
= net_ipvs(net
);
851 if (!(flags
& IP_VS_CONN_F_TEMPLATE
)) {
852 cp
= ip_vs_conn_in_get(param
);
853 if (cp
&& ((cp
->dport
!= dport
) ||
854 !ip_vs_addr_equal(cp
->daf
, &cp
->daddr
, daddr
))) {
855 if (!(flags
& IP_VS_CONN_F_INACTIVE
)) {
856 ip_vs_conn_expire_now(cp
);
857 __ip_vs_conn_put(cp
);
860 /* This is the expiration message for the
861 * connection that was already replaced, so we
864 __ip_vs_conn_put(cp
);
865 kfree(param
->pe_data
);
870 cp
= ip_vs_ct_in_get(param
);
875 kfree(param
->pe_data
);
878 spin_lock_bh(&cp
->lock
);
879 if ((cp
->flags
^ flags
) & IP_VS_CONN_F_INACTIVE
&&
880 !(flags
& IP_VS_CONN_F_TEMPLATE
) && dest
) {
881 if (flags
& IP_VS_CONN_F_INACTIVE
) {
882 atomic_dec(&dest
->activeconns
);
883 atomic_inc(&dest
->inactconns
);
885 atomic_inc(&dest
->activeconns
);
886 atomic_dec(&dest
->inactconns
);
889 flags
&= IP_VS_CONN_F_BACKUP_UPD_MASK
;
890 flags
|= cp
->flags
& ~IP_VS_CONN_F_BACKUP_UPD_MASK
;
892 spin_unlock_bh(&cp
->lock
);
894 ip_vs_try_bind_dest(cp
);
897 * Find the appropriate destination for the connection.
898 * If it is not found the connection will remain unbound
902 /* This function is only invoked by the synchronization
903 * code. We do not currently support heterogeneous pools
904 * with synchronization, so we can make the assumption that
905 * the svc_af is the same as the dest_af
907 dest
= ip_vs_find_dest(ipvs
, type
, type
, daddr
, dport
,
908 param
->vaddr
, param
->vport
, protocol
,
911 cp
= ip_vs_conn_new(param
, type
, daddr
, dport
, flags
, dest
,
915 kfree(param
->pe_data
);
916 IP_VS_DBG(2, "BACKUP, add new conn. failed\n");
919 if (!(flags
& IP_VS_CONN_F_TEMPLATE
))
920 kfree(param
->pe_data
);
924 memcpy(&cp
->in_seq
, opt
, sizeof(*opt
));
925 atomic_set(&cp
->in_pkts
, sysctl_sync_threshold(ipvs
));
927 cp
->old_state
= cp
->state
;
929 * For Ver 0 messages style
930 * - Not possible to recover the right timeout for templates
931 * - can not find the right fwmark
932 * virtual service. If needed, we can do it for
933 * non-fwmark persistent services.
934 * Ver 1 messages style.
938 if (timeout
> MAX_SCHEDULE_TIMEOUT
/ HZ
)
939 timeout
= MAX_SCHEDULE_TIMEOUT
/ HZ
;
940 cp
->timeout
= timeout
*HZ
;
942 struct ip_vs_proto_data
*pd
;
944 pd
= ip_vs_proto_data_get(ipvs
, protocol
);
945 if (!(flags
& IP_VS_CONN_F_TEMPLATE
) && pd
&& pd
->timeout_table
)
946 cp
->timeout
= pd
->timeout_table
[state
];
948 cp
->timeout
= (3*60*HZ
);
954 * Process received multicast message for Version 0
956 static void ip_vs_process_message_v0(struct netns_ipvs
*ipvs
, const char *buffer
,
959 struct ip_vs_sync_mesg_v0
*m
= (struct ip_vs_sync_mesg_v0
*)buffer
;
960 struct ip_vs_sync_conn_v0
*s
;
961 struct ip_vs_sync_conn_options
*opt
;
962 struct ip_vs_protocol
*pp
;
963 struct ip_vs_conn_param param
;
967 p
= (char *)buffer
+ sizeof(struct ip_vs_sync_mesg_v0
);
968 for (i
=0; i
<m
->nr_conns
; i
++) {
969 unsigned int flags
, state
;
971 if (p
+ SIMPLE_CONN_SIZE
> buffer
+buflen
) {
972 IP_VS_ERR_RL("BACKUP v0, bogus conn\n");
975 s
= (struct ip_vs_sync_conn_v0
*) p
;
976 flags
= ntohs(s
->flags
) | IP_VS_CONN_F_SYNC
;
977 flags
&= ~IP_VS_CONN_F_HASHED
;
978 if (flags
& IP_VS_CONN_F_SEQ_MASK
) {
979 opt
= (struct ip_vs_sync_conn_options
*)&s
[1];
981 if (p
> buffer
+buflen
) {
982 IP_VS_ERR_RL("BACKUP v0, Dropping buffer bogus conn options\n");
987 p
+= SIMPLE_CONN_SIZE
;
990 state
= ntohs(s
->state
);
991 if (!(flags
& IP_VS_CONN_F_TEMPLATE
)) {
992 pp
= ip_vs_proto_get(s
->protocol
);
994 IP_VS_DBG(2, "BACKUP v0, Unsupported protocol %u\n",
998 if (state
>= pp
->num_states
) {
999 IP_VS_DBG(2, "BACKUP v0, Invalid %s state %u\n",
1004 /* protocol in templates is not used for state/timeout */
1006 IP_VS_DBG(2, "BACKUP v0, Invalid template state %u\n",
1012 ip_vs_conn_fill_param(ipvs
, AF_INET
, s
->protocol
,
1013 (const union nf_inet_addr
*)&s
->caddr
,
1015 (const union nf_inet_addr
*)&s
->vaddr
,
1018 /* Send timeout as Zero */
1019 ip_vs_proc_conn(ipvs
->net
, ¶m
, flags
, state
, s
->protocol
, AF_INET
,
1020 (union nf_inet_addr
*)&s
->daddr
, s
->dport
,
1028 static inline int ip_vs_proc_seqopt(__u8
*p
, unsigned int plen
,
1030 struct ip_vs_sync_conn_options
*opt
)
1032 struct ip_vs_sync_conn_options
*topt
;
1034 topt
= (struct ip_vs_sync_conn_options
*)p
;
1036 if (plen
!= sizeof(struct ip_vs_sync_conn_options
)) {
1037 IP_VS_DBG(2, "BACKUP, bogus conn options length\n");
1040 if (*opt_flags
& IPVS_OPT_F_SEQ_DATA
) {
1041 IP_VS_DBG(2, "BACKUP, conn options found twice\n");
1044 ntoh_seq(&topt
->in_seq
, &opt
->in_seq
);
1045 ntoh_seq(&topt
->out_seq
, &opt
->out_seq
);
1046 *opt_flags
|= IPVS_OPT_F_SEQ_DATA
;
1050 static int ip_vs_proc_str(__u8
*p
, unsigned int plen
, unsigned int *data_len
,
1051 __u8
**data
, unsigned int maxlen
,
1052 __u32
*opt_flags
, __u32 flag
)
1054 if (plen
> maxlen
) {
1055 IP_VS_DBG(2, "BACKUP, bogus par.data len > %d\n", maxlen
);
1058 if (*opt_flags
& flag
) {
1059 IP_VS_DBG(2, "BACKUP, Par.data found twice 0x%x\n", flag
);
1068 * Process a Version 1 sync. connection
1070 static inline int ip_vs_proc_sync_conn(struct net
*net
, __u8
*p
, __u8
*msg_end
)
1072 struct ip_vs_sync_conn_options opt
;
1073 union ip_vs_sync_conn
*s
;
1074 struct ip_vs_protocol
*pp
;
1075 struct ip_vs_conn_param param
;
1077 unsigned int af
, state
, pe_data_len
=0, pe_name_len
=0;
1078 __u8
*pe_data
=NULL
, *pe_name
=NULL
;
1082 s
= (union ip_vs_sync_conn
*) p
;
1084 if (s
->v6
.type
& STYPE_F_INET6
) {
1085 #ifdef CONFIG_IP_VS_IPV6
1087 p
+= sizeof(struct ip_vs_sync_v6
);
1089 IP_VS_DBG(3,"BACKUP, IPv6 msg received, and IPVS is not compiled for IPv6\n");
1093 } else if (!s
->v4
.type
) {
1095 p
+= sizeof(struct ip_vs_sync_v4
);
1102 /* Process optional params check Type & Len. */
1103 while (p
< msg_end
) {
1112 if (!plen
|| ((p
+ plen
) > msg_end
))
1114 /* Handle seq option p = param data */
1115 switch (ptype
& ~IPVS_OPT_F_PARAM
) {
1116 case IPVS_OPT_SEQ_DATA
:
1117 if (ip_vs_proc_seqopt(p
, plen
, &opt_flags
, &opt
))
1121 case IPVS_OPT_PE_DATA
:
1122 if (ip_vs_proc_str(p
, plen
, &pe_data_len
, &pe_data
,
1123 IP_VS_PEDATA_MAXLEN
, &opt_flags
,
1124 IPVS_OPT_F_PE_DATA
))
1128 case IPVS_OPT_PE_NAME
:
1129 if (ip_vs_proc_str(p
, plen
,&pe_name_len
, &pe_name
,
1130 IP_VS_PENAME_MAXLEN
, &opt_flags
,
1131 IPVS_OPT_F_PE_NAME
))
1136 /* Param data mandatory ? */
1137 if (!(ptype
& IPVS_OPT_F_PARAM
)) {
1138 IP_VS_DBG(3, "BACKUP, Unknown mandatory param %d found\n",
1139 ptype
& ~IPVS_OPT_F_PARAM
);
1144 p
+= plen
; /* Next option */
1147 /* Get flags and Mask off unsupported */
1148 flags
= ntohl(s
->v4
.flags
) & IP_VS_CONN_F_BACKUP_MASK
;
1149 flags
|= IP_VS_CONN_F_SYNC
;
1150 state
= ntohs(s
->v4
.state
);
1152 if (!(flags
& IP_VS_CONN_F_TEMPLATE
)) {
1153 pp
= ip_vs_proto_get(s
->v4
.protocol
);
1155 IP_VS_DBG(3,"BACKUP, Unsupported protocol %u\n",
1160 if (state
>= pp
->num_states
) {
1161 IP_VS_DBG(3, "BACKUP, Invalid %s state %u\n",
1167 /* protocol in templates is not used for state/timeout */
1169 IP_VS_DBG(3, "BACKUP, Invalid template state %u\n",
1174 if (ip_vs_conn_fill_param_sync(net_ipvs(net
), af
, s
, ¶m
, pe_data
,
1175 pe_data_len
, pe_name
, pe_name_len
)) {
1179 /* If only IPv4, just silent skip IPv6 */
1181 ip_vs_proc_conn(net
, ¶m
, flags
, state
, s
->v4
.protocol
, af
,
1182 (union nf_inet_addr
*)&s
->v4
.daddr
, s
->v4
.dport
,
1183 ntohl(s
->v4
.timeout
), ntohl(s
->v4
.fwmark
),
1184 (opt_flags
& IPVS_OPT_F_SEQ_DATA
? &opt
: NULL
)
1186 #ifdef CONFIG_IP_VS_IPV6
1188 ip_vs_proc_conn(net
, ¶m
, flags
, state
, s
->v6
.protocol
, af
,
1189 (union nf_inet_addr
*)&s
->v6
.daddr
, s
->v6
.dport
,
1190 ntohl(s
->v6
.timeout
), ntohl(s
->v6
.fwmark
),
1191 (opt_flags
& IPVS_OPT_F_SEQ_DATA
? &opt
: NULL
)
1194 ip_vs_pe_put(param
.pe
);
1198 IP_VS_DBG(2, "BACKUP, Single msg dropped err:%d\n", retc
);
1203 * Process received multicast message and create the corresponding
1204 * ip_vs_conn entries.
1205 * Handles Version 0 & 1
1207 static void ip_vs_process_message(struct net
*net
, __u8
*buffer
,
1208 const size_t buflen
)
1210 struct netns_ipvs
*ipvs
= net_ipvs(net
);
1211 struct ip_vs_sync_mesg
*m2
= (struct ip_vs_sync_mesg
*)buffer
;
1215 if (buflen
< sizeof(struct ip_vs_sync_mesg_v0
)) {
1216 IP_VS_DBG(2, "BACKUP, message header too short\n");
1220 if (buflen
!= ntohs(m2
->size
)) {
1221 IP_VS_DBG(2, "BACKUP, bogus message size\n");
1224 /* SyncID sanity check */
1225 if (ipvs
->bcfg
.syncid
!= 0 && m2
->syncid
!= ipvs
->bcfg
.syncid
) {
1226 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2
->syncid
);
1229 /* Handle version 1 message */
1230 if ((m2
->version
== SYNC_PROTO_VER
) && (m2
->reserved
== 0)
1231 && (m2
->spare
== 0)) {
1233 msg_end
= buffer
+ sizeof(struct ip_vs_sync_mesg
);
1234 nr_conns
= m2
->nr_conns
;
1236 for (i
=0; i
<nr_conns
; i
++) {
1237 union ip_vs_sync_conn
*s
;
1242 if (p
+ sizeof(s
->v4
) > buffer
+buflen
) {
1243 IP_VS_ERR_RL("BACKUP, Dropping buffer, to small\n");
1246 s
= (union ip_vs_sync_conn
*)p
;
1247 size
= ntohs(s
->v4
.ver_size
) & SVER_MASK
;
1249 /* Basic sanity checks */
1250 if (msg_end
> buffer
+buflen
) {
1251 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n");
1254 if (ntohs(s
->v4
.ver_size
) >> SVER_SHIFT
) {
1255 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n",
1256 ntohs(s
->v4
.ver_size
) >> SVER_SHIFT
);
1259 /* Process a single sync_conn */
1260 retc
= ip_vs_proc_sync_conn(net
, p
, msg_end
);
1262 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n",
1266 /* Make sure we have 32 bit alignment */
1267 msg_end
= p
+ ((size
+ 3) & ~3);
1270 /* Old type of message */
1271 ip_vs_process_message_v0(ipvs
, buffer
, buflen
);
1278 * Setup sndbuf (mode=1) or rcvbuf (mode=0)
1280 static void set_sock_size(struct sock
*sk
, int mode
, int val
)
1282 /* setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &val, sizeof(val)); */
1283 /* setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)); */
1286 val
= clamp_t(int, val
, (SOCK_MIN_SNDBUF
+ 1) / 2,
1288 sk
->sk_sndbuf
= val
* 2;
1289 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
1291 val
= clamp_t(int, val
, (SOCK_MIN_RCVBUF
+ 1) / 2,
1293 sk
->sk_rcvbuf
= val
* 2;
1294 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
1300 * Setup loopback of outgoing multicasts on a sending socket
1302 static void set_mcast_loop(struct sock
*sk
, u_char loop
)
1304 struct inet_sock
*inet
= inet_sk(sk
);
1306 /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
1308 inet
->mc_loop
= loop
? 1 : 0;
1309 #ifdef CONFIG_IP_VS_IPV6
1310 if (sk
->sk_family
== AF_INET6
) {
1311 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1313 /* IPV6_MULTICAST_LOOP */
1314 np
->mc_loop
= loop
? 1 : 0;
1321 * Specify TTL for outgoing multicasts on a sending socket
1323 static void set_mcast_ttl(struct sock
*sk
, u_char ttl
)
1325 struct inet_sock
*inet
= inet_sk(sk
);
1327 /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
1330 #ifdef CONFIG_IP_VS_IPV6
1331 if (sk
->sk_family
== AF_INET6
) {
1332 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1334 /* IPV6_MULTICAST_HOPS */
1335 np
->mcast_hops
= ttl
;
1341 /* Control fragmentation of messages */
1342 static void set_mcast_pmtudisc(struct sock
*sk
, int val
)
1344 struct inet_sock
*inet
= inet_sk(sk
);
1346 /* setsockopt(sock, SOL_IP, IP_MTU_DISCOVER, &val, sizeof(val)); */
1348 inet
->pmtudisc
= val
;
1349 #ifdef CONFIG_IP_VS_IPV6
1350 if (sk
->sk_family
== AF_INET6
) {
1351 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1353 /* IPV6_MTU_DISCOVER */
1361 * Specifiy default interface for outgoing multicasts
1363 static int set_mcast_if(struct sock
*sk
, char *ifname
)
1365 struct net_device
*dev
;
1366 struct inet_sock
*inet
= inet_sk(sk
);
1367 struct net
*net
= sock_net(sk
);
1369 dev
= __dev_get_by_name(net
, ifname
);
1373 if (sk
->sk_bound_dev_if
&& dev
->ifindex
!= sk
->sk_bound_dev_if
)
1377 inet
->mc_index
= dev
->ifindex
;
1378 /* inet->mc_addr = 0; */
1379 #ifdef CONFIG_IP_VS_IPV6
1380 if (sk
->sk_family
== AF_INET6
) {
1381 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1383 /* IPV6_MULTICAST_IF */
1384 np
->mcast_oif
= dev
->ifindex
;
1394 * Join a multicast group.
1395 * the group is specified by a class D multicast address 224.0.0.0/8
1396 * in the in_addr structure passed in as a parameter.
1399 join_mcast_group(struct sock
*sk
, struct in_addr
*addr
, char *ifname
)
1401 struct net
*net
= sock_net(sk
);
1402 struct ip_mreqn mreq
;
1403 struct net_device
*dev
;
1406 memset(&mreq
, 0, sizeof(mreq
));
1407 memcpy(&mreq
.imr_multiaddr
, addr
, sizeof(struct in_addr
));
1409 dev
= __dev_get_by_name(net
, ifname
);
1412 if (sk
->sk_bound_dev_if
&& dev
->ifindex
!= sk
->sk_bound_dev_if
)
1415 mreq
.imr_ifindex
= dev
->ifindex
;
1418 ret
= ip_mc_join_group(sk
, &mreq
);
1424 #ifdef CONFIG_IP_VS_IPV6
1425 static int join_mcast_group6(struct sock
*sk
, struct in6_addr
*addr
,
1428 struct net
*net
= sock_net(sk
);
1429 struct net_device
*dev
;
1432 dev
= __dev_get_by_name(net
, ifname
);
1435 if (sk
->sk_bound_dev_if
&& dev
->ifindex
!= sk
->sk_bound_dev_if
)
1439 ret
= ipv6_sock_mc_join(sk
, dev
->ifindex
, addr
);
1446 static int bind_mcastif_addr(struct socket
*sock
, char *ifname
)
1448 struct net
*net
= sock_net(sock
->sk
);
1449 struct net_device
*dev
;
1451 struct sockaddr_in sin
;
1453 dev
= __dev_get_by_name(net
, ifname
);
1457 addr
= inet_select_addr(dev
, 0, RT_SCOPE_UNIVERSE
);
1459 pr_err("You probably need to specify IP address on "
1460 "multicast interface.\n");
1462 IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
1465 /* Now bind the socket with the address of multicast interface */
1466 sin
.sin_family
= AF_INET
;
1467 sin
.sin_addr
.s_addr
= addr
;
1470 return sock
->ops
->bind(sock
, (struct sockaddr
*)&sin
, sizeof(sin
));
1473 static void get_mcast_sockaddr(union ipvs_sockaddr
*sa
, int *salen
,
1474 struct ipvs_sync_daemon_cfg
*c
, int id
)
1476 if (AF_INET6
== c
->mcast_af
) {
1477 sa
->in6
= (struct sockaddr_in6
) {
1478 .sin6_family
= AF_INET6
,
1479 .sin6_port
= htons(c
->mcast_port
+ id
),
1481 sa
->in6
.sin6_addr
= c
->mcast_group
.in6
;
1482 *salen
= sizeof(sa
->in6
);
1484 sa
->in
= (struct sockaddr_in
) {
1485 .sin_family
= AF_INET
,
1486 .sin_port
= htons(c
->mcast_port
+ id
),
1488 sa
->in
.sin_addr
= c
->mcast_group
.in
;
1489 *salen
= sizeof(sa
->in
);
1494 * Set up sending multicast socket over UDP
1496 static struct socket
*make_send_sock(struct net
*net
, int id
)
1498 struct netns_ipvs
*ipvs
= net_ipvs(net
);
1499 /* multicast addr */
1500 union ipvs_sockaddr mcast_addr
;
1501 struct socket
*sock
;
1504 /* First create a socket */
1505 result
= sock_create_kern(net
, ipvs
->mcfg
.mcast_af
, SOCK_DGRAM
,
1506 IPPROTO_UDP
, &sock
);
1508 pr_err("Error during creation of socket; terminating\n");
1509 return ERR_PTR(result
);
1511 result
= set_mcast_if(sock
->sk
, ipvs
->mcfg
.mcast_ifn
);
1513 pr_err("Error setting outbound mcast interface\n");
1517 set_mcast_loop(sock
->sk
, 0);
1518 set_mcast_ttl(sock
->sk
, ipvs
->mcfg
.mcast_ttl
);
1519 /* Allow fragmentation if MTU changes */
1520 set_mcast_pmtudisc(sock
->sk
, IP_PMTUDISC_DONT
);
1521 result
= sysctl_sync_sock_size(ipvs
);
1523 set_sock_size(sock
->sk
, 1, result
);
1525 if (AF_INET
== ipvs
->mcfg
.mcast_af
)
1526 result
= bind_mcastif_addr(sock
, ipvs
->mcfg
.mcast_ifn
);
1530 pr_err("Error binding address of the mcast interface\n");
1534 get_mcast_sockaddr(&mcast_addr
, &salen
, &ipvs
->mcfg
, id
);
1535 result
= sock
->ops
->connect(sock
, (struct sockaddr
*) &mcast_addr
,
1538 pr_err("Error connecting to the multicast addr\n");
1546 return ERR_PTR(result
);
1551 * Set up receiving multicast socket over UDP
1553 static struct socket
*make_receive_sock(struct net
*net
, int id
)
1555 struct netns_ipvs
*ipvs
= net_ipvs(net
);
1556 /* multicast addr */
1557 union ipvs_sockaddr mcast_addr
;
1558 struct socket
*sock
;
1561 /* First create a socket */
1562 result
= sock_create_kern(net
, ipvs
->bcfg
.mcast_af
, SOCK_DGRAM
,
1563 IPPROTO_UDP
, &sock
);
1565 pr_err("Error during creation of socket; terminating\n");
1566 return ERR_PTR(result
);
1568 /* it is equivalent to the REUSEADDR option in user-space */
1569 sock
->sk
->sk_reuse
= SK_CAN_REUSE
;
1570 result
= sysctl_sync_sock_size(ipvs
);
1572 set_sock_size(sock
->sk
, 0, result
);
1574 get_mcast_sockaddr(&mcast_addr
, &salen
, &ipvs
->bcfg
, id
);
1575 result
= sock
->ops
->bind(sock
, (struct sockaddr
*)&mcast_addr
, salen
);
1577 pr_err("Error binding to the multicast addr\n");
1581 /* join the multicast group */
1582 #ifdef CONFIG_IP_VS_IPV6
1583 if (ipvs
->bcfg
.mcast_af
== AF_INET6
)
1584 result
= join_mcast_group6(sock
->sk
, &mcast_addr
.in6
.sin6_addr
,
1585 ipvs
->bcfg
.mcast_ifn
);
1588 result
= join_mcast_group(sock
->sk
, &mcast_addr
.in
.sin_addr
,
1589 ipvs
->bcfg
.mcast_ifn
);
1591 pr_err("Error joining to the multicast group\n");
1599 return ERR_PTR(result
);
1604 ip_vs_send_async(struct socket
*sock
, const char *buffer
, const size_t length
)
1606 struct msghdr msg
= {.msg_flags
= MSG_DONTWAIT
|MSG_NOSIGNAL
};
1611 iov
.iov_base
= (void *)buffer
;
1612 iov
.iov_len
= length
;
1614 len
= kernel_sendmsg(sock
, &msg
, &iov
, 1, (size_t)(length
));
1621 ip_vs_send_sync_msg(struct socket
*sock
, struct ip_vs_sync_mesg
*msg
)
1626 msize
= ntohs(msg
->size
);
1628 ret
= ip_vs_send_async(sock
, (char *)msg
, msize
);
1629 if (ret
>= 0 || ret
== -EAGAIN
)
1631 pr_err("ip_vs_send_async error %d\n", ret
);
1636 ip_vs_receive(struct socket
*sock
, char *buffer
, const size_t buflen
)
1638 struct msghdr msg
= {NULL
,};
1644 /* Receive a packet */
1645 iov
.iov_base
= buffer
;
1646 iov
.iov_len
= (size_t)buflen
;
1648 len
= kernel_recvmsg(sock
, &msg
, &iov
, 1, buflen
, MSG_DONTWAIT
);
1657 /* Wakeup the master thread for sending */
1658 static void master_wakeup_work_handler(struct work_struct
*work
)
1660 struct ipvs_master_sync_state
*ms
=
1661 container_of(work
, struct ipvs_master_sync_state
,
1662 master_wakeup_work
.work
);
1663 struct netns_ipvs
*ipvs
= ms
->ipvs
;
1665 spin_lock_bh(&ipvs
->sync_lock
);
1666 if (ms
->sync_queue_len
&&
1667 ms
->sync_queue_delay
< IPVS_SYNC_WAKEUP_RATE
) {
1668 ms
->sync_queue_delay
= IPVS_SYNC_WAKEUP_RATE
;
1669 wake_up_process(ms
->master_thread
);
1671 spin_unlock_bh(&ipvs
->sync_lock
);
1674 /* Get next buffer to send */
1675 static inline struct ip_vs_sync_buff
*
1676 next_sync_buff(struct netns_ipvs
*ipvs
, struct ipvs_master_sync_state
*ms
)
1678 struct ip_vs_sync_buff
*sb
;
1680 sb
= sb_dequeue(ipvs
, ms
);
1683 /* Do not delay entries in buffer for more than 2 seconds */
1684 return get_curr_sync_buff(ipvs
, ms
, IPVS_SYNC_FLUSH_TIME
);
1687 static int sync_thread_master(void *data
)
1689 struct ip_vs_sync_thread_data
*tinfo
= data
;
1690 struct netns_ipvs
*ipvs
= net_ipvs(tinfo
->net
);
1691 struct ipvs_master_sync_state
*ms
= &ipvs
->ms
[tinfo
->id
];
1692 struct sock
*sk
= tinfo
->sock
->sk
;
1693 struct ip_vs_sync_buff
*sb
;
1695 pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
1696 "syncid = %d, id = %d\n",
1697 ipvs
->mcfg
.mcast_ifn
, ipvs
->mcfg
.syncid
, tinfo
->id
);
1700 sb
= next_sync_buff(ipvs
, ms
);
1701 if (unlikely(kthread_should_stop()))
1704 schedule_timeout(IPVS_SYNC_CHECK_PERIOD
);
1707 while (ip_vs_send_sync_msg(tinfo
->sock
, sb
->mesg
) < 0) {
1708 /* (Ab)use interruptible sleep to avoid increasing
1711 __wait_event_interruptible(*sk_sleep(sk
),
1712 sock_writeable(sk
) ||
1713 kthread_should_stop());
1714 if (unlikely(kthread_should_stop()))
1717 ip_vs_sync_buff_release(sb
);
1721 __set_current_state(TASK_RUNNING
);
1723 ip_vs_sync_buff_release(sb
);
1725 /* clean up the sync_buff queue */
1726 while ((sb
= sb_dequeue(ipvs
, ms
)))
1727 ip_vs_sync_buff_release(sb
);
1728 __set_current_state(TASK_RUNNING
);
1730 /* clean up the current sync_buff */
1731 sb
= get_curr_sync_buff(ipvs
, ms
, 0);
1733 ip_vs_sync_buff_release(sb
);
1735 /* release the sending multicast socket */
1736 sock_release(tinfo
->sock
);
1743 static int sync_thread_backup(void *data
)
1745 struct ip_vs_sync_thread_data
*tinfo
= data
;
1746 struct netns_ipvs
*ipvs
= net_ipvs(tinfo
->net
);
1749 pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
1750 "syncid = %d, id = %d\n",
1751 ipvs
->bcfg
.mcast_ifn
, ipvs
->bcfg
.syncid
, tinfo
->id
);
1753 while (!kthread_should_stop()) {
1754 wait_event_interruptible(*sk_sleep(tinfo
->sock
->sk
),
1755 !skb_queue_empty(&tinfo
->sock
->sk
->sk_receive_queue
)
1756 || kthread_should_stop());
1758 /* do we have data now? */
1759 while (!skb_queue_empty(&(tinfo
->sock
->sk
->sk_receive_queue
))) {
1760 len
= ip_vs_receive(tinfo
->sock
, tinfo
->buf
,
1761 ipvs
->bcfg
.sync_maxlen
);
1764 pr_err("receiving message error\n");
1768 ip_vs_process_message(tinfo
->net
, tinfo
->buf
, len
);
1772 /* release the sending multicast socket */
1773 sock_release(tinfo
->sock
);
1781 int start_sync_thread(struct netns_ipvs
*ipvs
, struct ipvs_sync_daemon_cfg
*c
,
1784 struct ip_vs_sync_thread_data
*tinfo
;
1785 struct task_struct
**array
= NULL
, *task
;
1786 struct socket
*sock
;
1787 struct net_device
*dev
;
1789 int (*threadfn
)(void *data
);
1790 int id
, count
, hlen
;
1791 int result
= -ENOMEM
;
1794 IP_VS_DBG(7, "%s(): pid %d\n", __func__
, task_pid_nr(current
));
1795 IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n",
1796 sizeof(struct ip_vs_sync_conn_v0
));
1798 if (!ipvs
->sync_state
) {
1799 count
= clamp(sysctl_sync_ports(ipvs
), 1, IPVS_SYNC_PORTS_MAX
);
1800 ipvs
->threads_mask
= count
- 1;
1802 count
= ipvs
->threads_mask
+ 1;
1804 if (c
->mcast_af
== AF_UNSPEC
) {
1805 c
->mcast_af
= AF_INET
;
1806 c
->mcast_group
.ip
= cpu_to_be32(IP_VS_SYNC_GROUP
);
1809 c
->mcast_port
= IP_VS_SYNC_PORT
;
1813 dev
= __dev_get_by_name(ipvs
->net
, c
->mcast_ifn
);
1815 pr_err("Unknown mcast interface: %s\n", c
->mcast_ifn
);
1818 hlen
= (AF_INET6
== c
->mcast_af
) ?
1819 sizeof(struct ipv6hdr
) + sizeof(struct udphdr
) :
1820 sizeof(struct iphdr
) + sizeof(struct udphdr
);
1821 mtu
= (state
== IP_VS_STATE_BACKUP
) ?
1822 clamp(dev
->mtu
, 1500U, 65535U) : 1500U;
1823 min_mtu
= (state
== IP_VS_STATE_BACKUP
) ? 1024 : 1;
1826 c
->sync_maxlen
= clamp_t(unsigned int,
1827 c
->sync_maxlen
, min_mtu
,
1830 c
->sync_maxlen
= mtu
- hlen
;
1832 if (state
== IP_VS_STATE_MASTER
) {
1837 name
= "ipvs-m:%d:%d";
1838 threadfn
= sync_thread_master
;
1839 } else if (state
== IP_VS_STATE_BACKUP
) {
1840 if (ipvs
->backup_threads
)
1844 name
= "ipvs-b:%d:%d";
1845 threadfn
= sync_thread_backup
;
1850 if (state
== IP_VS_STATE_MASTER
) {
1851 struct ipvs_master_sync_state
*ms
;
1853 ipvs
->ms
= kzalloc(count
* sizeof(ipvs
->ms
[0]), GFP_KERNEL
);
1857 for (id
= 0; id
< count
; id
++, ms
++) {
1858 INIT_LIST_HEAD(&ms
->sync_queue
);
1859 ms
->sync_queue_len
= 0;
1860 ms
->sync_queue_delay
= 0;
1861 INIT_DELAYED_WORK(&ms
->master_wakeup_work
,
1862 master_wakeup_work_handler
);
1866 array
= kzalloc(count
* sizeof(struct task_struct
*),
1873 for (id
= 0; id
< count
; id
++) {
1874 if (state
== IP_VS_STATE_MASTER
)
1875 sock
= make_send_sock(ipvs
->net
, id
);
1877 sock
= make_receive_sock(ipvs
->net
, id
);
1879 result
= PTR_ERR(sock
);
1882 tinfo
= kmalloc(sizeof(*tinfo
), GFP_KERNEL
);
1885 tinfo
->net
= ipvs
->net
;
1887 if (state
== IP_VS_STATE_BACKUP
) {
1888 tinfo
->buf
= kmalloc(ipvs
->bcfg
.sync_maxlen
,
1897 task
= kthread_run(threadfn
, tinfo
, name
, ipvs
->gen
, id
);
1899 result
= PTR_ERR(task
);
1903 if (state
== IP_VS_STATE_MASTER
)
1904 ipvs
->ms
[id
].master_thread
= task
;
1909 /* mark as active */
1911 if (state
== IP_VS_STATE_BACKUP
)
1912 ipvs
->backup_threads
= array
;
1913 spin_lock_bh(&ipvs
->sync_buff_lock
);
1914 ipvs
->sync_state
|= state
;
1915 spin_unlock_bh(&ipvs
->sync_buff_lock
);
1917 /* increase the module use count */
1918 ip_vs_use_count_inc();
1927 sock_release(tinfo
->sock
);
1932 while (count
-- > 0) {
1933 if (state
== IP_VS_STATE_MASTER
)
1934 kthread_stop(ipvs
->ms
[count
].master_thread
);
1936 kthread_stop(array
[count
]);
1941 if (!(ipvs
->sync_state
& IP_VS_STATE_MASTER
)) {
1949 int stop_sync_thread(struct net
*net
, int state
)
1951 struct netns_ipvs
*ipvs
= net_ipvs(net
);
1952 struct task_struct
**array
;
1956 IP_VS_DBG(7, "%s(): pid %d\n", __func__
, task_pid_nr(current
));
1958 if (state
== IP_VS_STATE_MASTER
) {
1963 * The lock synchronizes with sb_queue_tail(), so that we don't
1964 * add sync buffers to the queue, when we are already in
1965 * progress of stopping the master sync daemon.
1968 spin_lock_bh(&ipvs
->sync_buff_lock
);
1969 spin_lock(&ipvs
->sync_lock
);
1970 ipvs
->sync_state
&= ~IP_VS_STATE_MASTER
;
1971 spin_unlock(&ipvs
->sync_lock
);
1972 spin_unlock_bh(&ipvs
->sync_buff_lock
);
1975 for (id
= ipvs
->threads_mask
; id
>= 0; id
--) {
1976 struct ipvs_master_sync_state
*ms
= &ipvs
->ms
[id
];
1979 pr_info("stopping master sync thread %d ...\n",
1980 task_pid_nr(ms
->master_thread
));
1981 cancel_delayed_work_sync(&ms
->master_wakeup_work
);
1982 ret
= kthread_stop(ms
->master_thread
);
1988 } else if (state
== IP_VS_STATE_BACKUP
) {
1989 if (!ipvs
->backup_threads
)
1992 ipvs
->sync_state
&= ~IP_VS_STATE_BACKUP
;
1993 array
= ipvs
->backup_threads
;
1995 for (id
= ipvs
->threads_mask
; id
>= 0; id
--) {
1998 pr_info("stopping backup sync thread %d ...\n",
1999 task_pid_nr(array
[id
]));
2000 ret
= kthread_stop(array
[id
]);
2005 ipvs
->backup_threads
= NULL
;
2008 /* decrease the module use count */
2009 ip_vs_use_count_dec();
2015 * Initialize data struct for each netns
2017 int __net_init
ip_vs_sync_net_init(struct net
*net
)
2019 struct netns_ipvs
*ipvs
= net_ipvs(net
);
2021 __mutex_init(&ipvs
->sync_mutex
, "ipvs->sync_mutex", &__ipvs_sync_key
);
2022 spin_lock_init(&ipvs
->sync_lock
);
2023 spin_lock_init(&ipvs
->sync_buff_lock
);
2027 void ip_vs_sync_net_cleanup(struct net
*net
)
2030 struct netns_ipvs
*ipvs
= net_ipvs(net
);
2032 mutex_lock(&ipvs
->sync_mutex
);
2033 retc
= stop_sync_thread(net
, IP_VS_STATE_MASTER
);
2034 if (retc
&& retc
!= -ESRCH
)
2035 pr_err("Failed to stop Master Daemon\n");
2037 retc
= stop_sync_thread(net
, IP_VS_STATE_BACKUP
);
2038 if (retc
&& retc
!= -ESRCH
)
2039 pr_err("Failed to stop Backup Daemon\n");
2040 mutex_unlock(&ipvs
->sync_mutex
);