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
{
196 struct netns_ipvs
*ipvs
;
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 memset(ho
, 0, sizeof(*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
*
304 sb_dequeue(struct netns_ipvs
*ipvs
, struct ipvs_master_sync_state
*ms
)
306 struct ip_vs_sync_buff
*sb
;
308 spin_lock_bh(&ipvs
->sync_lock
);
309 if (list_empty(&ms
->sync_queue
)) {
311 __set_current_state(TASK_INTERRUPTIBLE
);
313 sb
= list_entry(ms
->sync_queue
.next
, struct ip_vs_sync_buff
,
316 ms
->sync_queue_len
--;
317 if (!ms
->sync_queue_len
)
318 ms
->sync_queue_delay
= 0;
320 spin_unlock_bh(&ipvs
->sync_lock
);
326 * Create a new sync buffer for Version 1 proto.
328 static inline struct ip_vs_sync_buff
*
329 ip_vs_sync_buff_create(struct netns_ipvs
*ipvs
, unsigned int len
)
331 struct ip_vs_sync_buff
*sb
;
333 if (!(sb
=kmalloc(sizeof(struct ip_vs_sync_buff
), GFP_ATOMIC
)))
336 len
= max_t(unsigned int, len
+ sizeof(struct ip_vs_sync_mesg
),
337 ipvs
->mcfg
.sync_maxlen
);
338 sb
->mesg
= kmalloc(len
, GFP_ATOMIC
);
343 sb
->mesg
->reserved
= 0; /* old nr_conns i.e. must be zero now */
344 sb
->mesg
->version
= SYNC_PROTO_VER
;
345 sb
->mesg
->syncid
= ipvs
->mcfg
.syncid
;
346 sb
->mesg
->size
= htons(sizeof(struct ip_vs_sync_mesg
));
347 sb
->mesg
->nr_conns
= 0;
349 sb
->head
= (unsigned char *)sb
->mesg
+ sizeof(struct ip_vs_sync_mesg
);
350 sb
->end
= (unsigned char *)sb
->mesg
+ len
;
352 sb
->firstuse
= jiffies
;
356 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff
*sb
)
362 static inline void sb_queue_tail(struct netns_ipvs
*ipvs
,
363 struct ipvs_master_sync_state
*ms
)
365 struct ip_vs_sync_buff
*sb
= ms
->sync_buff
;
367 spin_lock(&ipvs
->sync_lock
);
368 if (ipvs
->sync_state
& IP_VS_STATE_MASTER
&&
369 ms
->sync_queue_len
< sysctl_sync_qlen_max(ipvs
)) {
370 if (!ms
->sync_queue_len
)
371 schedule_delayed_work(&ms
->master_wakeup_work
,
372 max(IPVS_SYNC_SEND_DELAY
, 1));
373 ms
->sync_queue_len
++;
374 list_add_tail(&sb
->list
, &ms
->sync_queue
);
375 if ((++ms
->sync_queue_delay
) == IPVS_SYNC_WAKEUP_RATE
)
376 wake_up_process(ms
->master_thread
);
378 ip_vs_sync_buff_release(sb
);
379 spin_unlock(&ipvs
->sync_lock
);
383 * Get the current sync buffer if it has been created for more
384 * than the specified time or the specified time is zero.
386 static inline struct ip_vs_sync_buff
*
387 get_curr_sync_buff(struct netns_ipvs
*ipvs
, struct ipvs_master_sync_state
*ms
,
390 struct ip_vs_sync_buff
*sb
;
392 spin_lock_bh(&ipvs
->sync_buff_lock
);
394 if (sb
&& time_after_eq(jiffies
- sb
->firstuse
, time
)) {
395 ms
->sync_buff
= NULL
;
396 __set_current_state(TASK_RUNNING
);
399 spin_unlock_bh(&ipvs
->sync_buff_lock
);
404 select_master_thread_id(struct netns_ipvs
*ipvs
, struct ip_vs_conn
*cp
)
406 return ((long) cp
>> (1 + ilog2(sizeof(*cp
)))) & ipvs
->threads_mask
;
410 * Create a new sync buffer for Version 0 proto.
412 static inline struct ip_vs_sync_buff
*
413 ip_vs_sync_buff_create_v0(struct netns_ipvs
*ipvs
, unsigned int len
)
415 struct ip_vs_sync_buff
*sb
;
416 struct ip_vs_sync_mesg_v0
*mesg
;
418 if (!(sb
=kmalloc(sizeof(struct ip_vs_sync_buff
), GFP_ATOMIC
)))
421 len
= max_t(unsigned int, len
+ sizeof(struct ip_vs_sync_mesg_v0
),
422 ipvs
->mcfg
.sync_maxlen
);
423 sb
->mesg
= kmalloc(len
, GFP_ATOMIC
);
428 mesg
= (struct ip_vs_sync_mesg_v0
*)sb
->mesg
;
430 mesg
->syncid
= ipvs
->mcfg
.syncid
;
431 mesg
->size
= htons(sizeof(struct ip_vs_sync_mesg_v0
));
432 sb
->head
= (unsigned char *)mesg
+ sizeof(struct ip_vs_sync_mesg_v0
);
433 sb
->end
= (unsigned char *)mesg
+ len
;
434 sb
->firstuse
= jiffies
;
438 /* Check if connection is controlled by persistence */
439 static inline bool in_persistence(struct ip_vs_conn
*cp
)
441 for (cp
= cp
->control
; cp
; cp
= cp
->control
) {
442 if (cp
->flags
& IP_VS_CONN_F_TEMPLATE
)
448 /* Check if conn should be synced.
449 * pkts: conn packets, use sysctl_sync_threshold to avoid packet check
450 * - (1) sync_refresh_period: reduce sync rate. Additionally, retry
451 * sync_retries times with period of sync_refresh_period/8
452 * - (2) if both sync_refresh_period and sync_period are 0 send sync only
453 * for state changes or only once when pkts matches sync_threshold
454 * - (3) templates: rate can be reduced only with sync_refresh_period or
457 static int ip_vs_sync_conn_needed(struct netns_ipvs
*ipvs
,
458 struct ip_vs_conn
*cp
, int pkts
)
460 unsigned long orig
= ACCESS_ONCE(cp
->sync_endtime
);
461 unsigned long now
= jiffies
;
462 unsigned long n
= (now
+ cp
->timeout
) & ~3UL;
463 unsigned int sync_refresh_period
;
467 /* Check if we sync in current state */
468 if (unlikely(cp
->flags
& IP_VS_CONN_F_TEMPLATE
))
470 else if (unlikely(sysctl_sync_persist_mode(ipvs
) && in_persistence(cp
)))
472 else if (likely(cp
->protocol
== IPPROTO_TCP
)) {
473 if (!((1 << cp
->state
) &
474 ((1 << IP_VS_TCP_S_ESTABLISHED
) |
475 (1 << IP_VS_TCP_S_FIN_WAIT
) |
476 (1 << IP_VS_TCP_S_CLOSE
) |
477 (1 << IP_VS_TCP_S_CLOSE_WAIT
) |
478 (1 << IP_VS_TCP_S_TIME_WAIT
))))
480 force
= cp
->state
!= cp
->old_state
;
481 if (force
&& cp
->state
!= IP_VS_TCP_S_ESTABLISHED
)
483 } else if (unlikely(cp
->protocol
== IPPROTO_SCTP
)) {
484 if (!((1 << cp
->state
) &
485 ((1 << IP_VS_SCTP_S_ESTABLISHED
) |
486 (1 << IP_VS_SCTP_S_SHUTDOWN_SENT
) |
487 (1 << IP_VS_SCTP_S_SHUTDOWN_RECEIVED
) |
488 (1 << IP_VS_SCTP_S_SHUTDOWN_ACK_SENT
) |
489 (1 << IP_VS_SCTP_S_CLOSED
))))
491 force
= cp
->state
!= cp
->old_state
;
492 if (force
&& cp
->state
!= IP_VS_SCTP_S_ESTABLISHED
)
495 /* UDP or another protocol with single state */
499 sync_refresh_period
= sysctl_sync_refresh_period(ipvs
);
500 if (sync_refresh_period
> 0) {
501 long diff
= n
- orig
;
502 long min_diff
= max(cp
->timeout
>> 1, 10UL * HZ
);
504 /* Avoid sync if difference is below sync_refresh_period
505 * and below the half timeout.
507 if (abs(diff
) < min_t(long, sync_refresh_period
, min_diff
)) {
508 int retries
= orig
& 3;
510 if (retries
>= sysctl_sync_retries(ipvs
))
512 if (time_before(now
, orig
- cp
->timeout
+
513 (sync_refresh_period
>> 3)))
518 sync_period
= sysctl_sync_period(ipvs
);
519 if (sync_period
> 0) {
520 if (!(cp
->flags
& IP_VS_CONN_F_TEMPLATE
) &&
521 pkts
% sync_period
!= sysctl_sync_threshold(ipvs
))
523 } else if (!sync_refresh_period
&&
524 pkts
!= sysctl_sync_threshold(ipvs
))
528 cp
->old_state
= cp
->state
;
529 n
= cmpxchg(&cp
->sync_endtime
, orig
, n
);
530 return n
== orig
|| force
;
534 * Version 0 , could be switched in by sys_ctl.
535 * Add an ip_vs_conn information into the current sync_buff.
537 static void ip_vs_sync_conn_v0(struct netns_ipvs
*ipvs
, struct ip_vs_conn
*cp
,
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(ipvs
, 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 netns_ipvs
*ipvs
, struct ip_vs_conn
*cp
, int pkts
)
629 struct ip_vs_sync_mesg
*m
;
630 union ip_vs_sync_conn
*s
;
631 struct ip_vs_sync_buff
*buff
;
632 struct ipvs_master_sync_state
*ms
;
635 unsigned int len
, pe_name_len
, pad
;
637 /* Handle old version of the protocol */
638 if (sysctl_sync_ver(ipvs
) == 0) {
639 ip_vs_sync_conn_v0(ipvs
, cp
, pkts
);
642 /* Do not sync ONE PACKET */
643 if (cp
->flags
& IP_VS_CONN_F_ONE_PACKET
)
646 if (!ip_vs_sync_conn_needed(ipvs
, cp
, pkts
))
651 if (cp
->pe_data_len
) {
652 if (!cp
->pe_data
|| !cp
->dest
) {
653 IP_VS_ERR_RL("SYNC, connection pe_data invalid\n");
656 pe_name_len
= strnlen(cp
->pe
->name
, IP_VS_PENAME_MAXLEN
);
659 spin_lock_bh(&ipvs
->sync_buff_lock
);
660 if (!(ipvs
->sync_state
& IP_VS_STATE_MASTER
)) {
661 spin_unlock_bh(&ipvs
->sync_buff_lock
);
665 id
= select_master_thread_id(ipvs
, cp
);
668 #ifdef CONFIG_IP_VS_IPV6
669 if (cp
->af
== AF_INET6
)
670 len
= sizeof(struct ip_vs_sync_v6
);
673 len
= sizeof(struct ip_vs_sync_v4
);
675 if (cp
->flags
& IP_VS_CONN_F_SEQ_MASK
)
676 len
+= sizeof(struct ip_vs_sync_conn_options
) + 2;
679 len
+= cp
->pe_data_len
+ 2; /* + Param hdr field */
681 len
+= pe_name_len
+ 2;
683 /* check if there is a space for this one */
685 buff
= ms
->sync_buff
;
688 pad
= (4 - (size_t) buff
->head
) & 3;
689 /* Send buffer if it is for v0 */
690 if (buff
->head
+ len
+ pad
> buff
->end
|| m
->reserved
) {
691 sb_queue_tail(ipvs
, ms
);
692 ms
->sync_buff
= NULL
;
699 buff
= ip_vs_sync_buff_create(ipvs
, len
);
701 spin_unlock_bh(&ipvs
->sync_buff_lock
);
702 pr_err("ip_vs_sync_buff_create failed.\n");
705 ms
->sync_buff
= buff
;
710 buff
->head
+= pad
+ len
;
711 m
->size
= htons(ntohs(m
->size
) + pad
+ len
);
712 /* Add ev. padding from prev. sync_conn */
716 s
= (union ip_vs_sync_conn
*)p
;
718 /* Set message type & copy members */
719 s
->v4
.type
= (cp
->af
== AF_INET6
? STYPE_F_INET6
: 0);
720 s
->v4
.ver_size
= htons(len
& SVER_MASK
); /* Version 0 */
721 s
->v4
.flags
= htonl(cp
->flags
& ~IP_VS_CONN_F_HASHED
);
722 s
->v4
.state
= htons(cp
->state
);
723 s
->v4
.protocol
= cp
->protocol
;
724 s
->v4
.cport
= cp
->cport
;
725 s
->v4
.vport
= cp
->vport
;
726 s
->v4
.dport
= cp
->dport
;
727 s
->v4
.fwmark
= htonl(cp
->fwmark
);
728 s
->v4
.timeout
= htonl(cp
->timeout
/ HZ
);
731 #ifdef CONFIG_IP_VS_IPV6
732 if (cp
->af
== AF_INET6
) {
733 p
+= sizeof(struct ip_vs_sync_v6
);
734 s
->v6
.caddr
= cp
->caddr
.in6
;
735 s
->v6
.vaddr
= cp
->vaddr
.in6
;
736 s
->v6
.daddr
= cp
->daddr
.in6
;
740 p
+= sizeof(struct ip_vs_sync_v4
); /* options ptr */
741 s
->v4
.caddr
= cp
->caddr
.ip
;
742 s
->v4
.vaddr
= cp
->vaddr
.ip
;
743 s
->v4
.daddr
= cp
->daddr
.ip
;
745 if (cp
->flags
& IP_VS_CONN_F_SEQ_MASK
) {
746 *(p
++) = IPVS_OPT_SEQ_DATA
;
747 *(p
++) = sizeof(struct ip_vs_sync_conn_options
);
748 hton_seq((struct ip_vs_seq
*)p
, &cp
->in_seq
);
749 p
+= sizeof(struct ip_vs_seq
);
750 hton_seq((struct ip_vs_seq
*)p
, &cp
->out_seq
);
751 p
+= sizeof(struct ip_vs_seq
);
754 if (cp
->pe_data_len
&& cp
->pe_data
) {
755 *(p
++) = IPVS_OPT_PE_DATA
;
756 *(p
++) = cp
->pe_data_len
;
757 memcpy(p
, cp
->pe_data
, cp
->pe_data_len
);
758 p
+= cp
->pe_data_len
;
761 *(p
++) = IPVS_OPT_PE_NAME
;
762 *(p
++) = pe_name_len
;
763 memcpy(p
, cp
->pe
->name
, pe_name_len
);
768 spin_unlock_bh(&ipvs
->sync_buff_lock
);
771 /* synchronize its controller if it has */
775 if (cp
->flags
& IP_VS_CONN_F_TEMPLATE
)
776 pkts
= atomic_add_return(1, &cp
->in_pkts
);
778 pkts
= sysctl_sync_threshold(ipvs
);
783 * fill_param used by version 1
786 ip_vs_conn_fill_param_sync(struct netns_ipvs
*ipvs
, int af
, union ip_vs_sync_conn
*sc
,
787 struct ip_vs_conn_param
*p
,
788 __u8
*pe_data
, unsigned int pe_data_len
,
789 __u8
*pe_name
, unsigned int pe_name_len
)
791 #ifdef CONFIG_IP_VS_IPV6
793 ip_vs_conn_fill_param(ipvs
, af
, sc
->v6
.protocol
,
794 (const union nf_inet_addr
*)&sc
->v6
.caddr
,
796 (const union nf_inet_addr
*)&sc
->v6
.vaddr
,
800 ip_vs_conn_fill_param(ipvs
, af
, sc
->v4
.protocol
,
801 (const union nf_inet_addr
*)&sc
->v4
.caddr
,
803 (const union nf_inet_addr
*)&sc
->v4
.vaddr
,
808 char buff
[IP_VS_PENAME_MAXLEN
+1];
810 memcpy(buff
, pe_name
, pe_name_len
);
812 p
->pe
= __ip_vs_pe_getbyname(buff
);
814 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n",
819 IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n");
823 p
->pe_data
= kmemdup(pe_data
, pe_data_len
, GFP_ATOMIC
);
825 module_put(p
->pe
->module
);
828 p
->pe_data_len
= pe_data_len
;
834 * Connection Add / Update.
835 * Common for version 0 and 1 reception of backup sync_conns.
839 static void ip_vs_proc_conn(struct netns_ipvs
*ipvs
, struct ip_vs_conn_param
*param
,
840 unsigned int flags
, unsigned int state
,
841 unsigned int protocol
, unsigned int type
,
842 const union nf_inet_addr
*daddr
, __be16 dport
,
843 unsigned long timeout
, __u32 fwmark
,
844 struct ip_vs_sync_conn_options
*opt
)
846 struct ip_vs_dest
*dest
;
847 struct ip_vs_conn
*cp
;
849 if (!(flags
& IP_VS_CONN_F_TEMPLATE
)) {
850 cp
= ip_vs_conn_in_get(param
);
851 if (cp
&& ((cp
->dport
!= dport
) ||
852 !ip_vs_addr_equal(cp
->daf
, &cp
->daddr
, daddr
))) {
853 if (!(flags
& IP_VS_CONN_F_INACTIVE
)) {
854 ip_vs_conn_expire_now(cp
);
855 __ip_vs_conn_put(cp
);
858 /* This is the expiration message for the
859 * connection that was already replaced, so we
862 __ip_vs_conn_put(cp
);
863 kfree(param
->pe_data
);
868 cp
= ip_vs_ct_in_get(param
);
873 kfree(param
->pe_data
);
876 spin_lock_bh(&cp
->lock
);
877 if ((cp
->flags
^ flags
) & IP_VS_CONN_F_INACTIVE
&&
878 !(flags
& IP_VS_CONN_F_TEMPLATE
) && dest
) {
879 if (flags
& IP_VS_CONN_F_INACTIVE
) {
880 atomic_dec(&dest
->activeconns
);
881 atomic_inc(&dest
->inactconns
);
883 atomic_inc(&dest
->activeconns
);
884 atomic_dec(&dest
->inactconns
);
887 flags
&= IP_VS_CONN_F_BACKUP_UPD_MASK
;
888 flags
|= cp
->flags
& ~IP_VS_CONN_F_BACKUP_UPD_MASK
;
890 spin_unlock_bh(&cp
->lock
);
892 ip_vs_try_bind_dest(cp
);
895 * Find the appropriate destination for the connection.
896 * If it is not found the connection will remain unbound
900 /* This function is only invoked by the synchronization
901 * code. We do not currently support heterogeneous pools
902 * with synchronization, so we can make the assumption that
903 * the svc_af is the same as the dest_af
905 dest
= ip_vs_find_dest(ipvs
, type
, type
, daddr
, dport
,
906 param
->vaddr
, param
->vport
, protocol
,
909 cp
= ip_vs_conn_new(param
, type
, daddr
, dport
, flags
, dest
,
913 kfree(param
->pe_data
);
914 IP_VS_DBG(2, "BACKUP, add new conn. failed\n");
917 if (!(flags
& IP_VS_CONN_F_TEMPLATE
))
918 kfree(param
->pe_data
);
922 cp
->in_seq
= opt
->in_seq
;
923 cp
->out_seq
= opt
->out_seq
;
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
, ¶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 netns_ipvs
*ipvs
, __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(ipvs
, 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(ipvs
, ¶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(ipvs
, ¶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 netns_ipvs
*ipvs
, __u8
*buffer
,
1208 const size_t buflen
)
1210 struct ip_vs_sync_mesg
*m2
= (struct ip_vs_sync_mesg
*)buffer
;
1214 if (buflen
< sizeof(struct ip_vs_sync_mesg_v0
)) {
1215 IP_VS_DBG(2, "BACKUP, message header too short\n");
1219 if (buflen
!= ntohs(m2
->size
)) {
1220 IP_VS_DBG(2, "BACKUP, bogus message size\n");
1223 /* SyncID sanity check */
1224 if (ipvs
->bcfg
.syncid
!= 0 && m2
->syncid
!= ipvs
->bcfg
.syncid
) {
1225 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2
->syncid
);
1228 /* Handle version 1 message */
1229 if ((m2
->version
== SYNC_PROTO_VER
) && (m2
->reserved
== 0)
1230 && (m2
->spare
== 0)) {
1232 msg_end
= buffer
+ sizeof(struct ip_vs_sync_mesg
);
1233 nr_conns
= m2
->nr_conns
;
1235 for (i
=0; i
<nr_conns
; i
++) {
1236 union ip_vs_sync_conn
*s
;
1241 if (p
+ sizeof(s
->v4
) > buffer
+buflen
) {
1242 IP_VS_ERR_RL("BACKUP, Dropping buffer, to small\n");
1245 s
= (union ip_vs_sync_conn
*)p
;
1246 size
= ntohs(s
->v4
.ver_size
) & SVER_MASK
;
1248 /* Basic sanity checks */
1249 if (msg_end
> buffer
+buflen
) {
1250 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n");
1253 if (ntohs(s
->v4
.ver_size
) >> SVER_SHIFT
) {
1254 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n",
1255 ntohs(s
->v4
.ver_size
) >> SVER_SHIFT
);
1258 /* Process a single sync_conn */
1259 retc
= ip_vs_proc_sync_conn(ipvs
, p
, msg_end
);
1261 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n",
1265 /* Make sure we have 32 bit alignment */
1266 msg_end
= p
+ ((size
+ 3) & ~3);
1269 /* Old type of message */
1270 ip_vs_process_message_v0(ipvs
, buffer
, buflen
);
1277 * Setup sndbuf (mode=1) or rcvbuf (mode=0)
1279 static void set_sock_size(struct sock
*sk
, int mode
, int val
)
1281 /* setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &val, sizeof(val)); */
1282 /* setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)); */
1285 val
= clamp_t(int, val
, (SOCK_MIN_SNDBUF
+ 1) / 2,
1287 sk
->sk_sndbuf
= val
* 2;
1288 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
1290 val
= clamp_t(int, val
, (SOCK_MIN_RCVBUF
+ 1) / 2,
1292 sk
->sk_rcvbuf
= val
* 2;
1293 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
1299 * Setup loopback of outgoing multicasts on a sending socket
1301 static void set_mcast_loop(struct sock
*sk
, u_char loop
)
1303 struct inet_sock
*inet
= inet_sk(sk
);
1305 /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
1307 inet
->mc_loop
= loop
? 1 : 0;
1308 #ifdef CONFIG_IP_VS_IPV6
1309 if (sk
->sk_family
== AF_INET6
) {
1310 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1312 /* IPV6_MULTICAST_LOOP */
1313 np
->mc_loop
= loop
? 1 : 0;
1320 * Specify TTL for outgoing multicasts on a sending socket
1322 static void set_mcast_ttl(struct sock
*sk
, u_char ttl
)
1324 struct inet_sock
*inet
= inet_sk(sk
);
1326 /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
1329 #ifdef CONFIG_IP_VS_IPV6
1330 if (sk
->sk_family
== AF_INET6
) {
1331 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1333 /* IPV6_MULTICAST_HOPS */
1334 np
->mcast_hops
= ttl
;
1340 /* Control fragmentation of messages */
1341 static void set_mcast_pmtudisc(struct sock
*sk
, int val
)
1343 struct inet_sock
*inet
= inet_sk(sk
);
1345 /* setsockopt(sock, SOL_IP, IP_MTU_DISCOVER, &val, sizeof(val)); */
1347 inet
->pmtudisc
= val
;
1348 #ifdef CONFIG_IP_VS_IPV6
1349 if (sk
->sk_family
== AF_INET6
) {
1350 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1352 /* IPV6_MTU_DISCOVER */
1360 * Specifiy default interface for outgoing multicasts
1362 static int set_mcast_if(struct sock
*sk
, char *ifname
)
1364 struct net_device
*dev
;
1365 struct inet_sock
*inet
= inet_sk(sk
);
1366 struct net
*net
= sock_net(sk
);
1368 dev
= __dev_get_by_name(net
, ifname
);
1372 if (sk
->sk_bound_dev_if
&& dev
->ifindex
!= sk
->sk_bound_dev_if
)
1376 inet
->mc_index
= dev
->ifindex
;
1377 /* inet->mc_addr = 0; */
1378 #ifdef CONFIG_IP_VS_IPV6
1379 if (sk
->sk_family
== AF_INET6
) {
1380 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1382 /* IPV6_MULTICAST_IF */
1383 np
->mcast_oif
= dev
->ifindex
;
1393 * Join a multicast group.
1394 * the group is specified by a class D multicast address 224.0.0.0/8
1395 * in the in_addr structure passed in as a parameter.
1398 join_mcast_group(struct sock
*sk
, struct in_addr
*addr
, char *ifname
)
1400 struct net
*net
= sock_net(sk
);
1401 struct ip_mreqn mreq
;
1402 struct net_device
*dev
;
1405 memset(&mreq
, 0, sizeof(mreq
));
1406 memcpy(&mreq
.imr_multiaddr
, addr
, sizeof(struct in_addr
));
1408 dev
= __dev_get_by_name(net
, ifname
);
1411 if (sk
->sk_bound_dev_if
&& dev
->ifindex
!= sk
->sk_bound_dev_if
)
1414 mreq
.imr_ifindex
= dev
->ifindex
;
1417 ret
= ip_mc_join_group(sk
, &mreq
);
1423 #ifdef CONFIG_IP_VS_IPV6
1424 static int join_mcast_group6(struct sock
*sk
, struct in6_addr
*addr
,
1427 struct net
*net
= sock_net(sk
);
1428 struct net_device
*dev
;
1431 dev
= __dev_get_by_name(net
, ifname
);
1434 if (sk
->sk_bound_dev_if
&& dev
->ifindex
!= sk
->sk_bound_dev_if
)
1438 ret
= ipv6_sock_mc_join(sk
, dev
->ifindex
, addr
);
1445 static int bind_mcastif_addr(struct socket
*sock
, char *ifname
)
1447 struct net
*net
= sock_net(sock
->sk
);
1448 struct net_device
*dev
;
1450 struct sockaddr_in sin
;
1452 dev
= __dev_get_by_name(net
, ifname
);
1456 addr
= inet_select_addr(dev
, 0, RT_SCOPE_UNIVERSE
);
1458 pr_err("You probably need to specify IP address on "
1459 "multicast interface.\n");
1461 IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
1464 /* Now bind the socket with the address of multicast interface */
1465 sin
.sin_family
= AF_INET
;
1466 sin
.sin_addr
.s_addr
= addr
;
1469 return sock
->ops
->bind(sock
, (struct sockaddr
*)&sin
, sizeof(sin
));
1472 static void get_mcast_sockaddr(union ipvs_sockaddr
*sa
, int *salen
,
1473 struct ipvs_sync_daemon_cfg
*c
, int id
)
1475 if (AF_INET6
== c
->mcast_af
) {
1476 sa
->in6
= (struct sockaddr_in6
) {
1477 .sin6_family
= AF_INET6
,
1478 .sin6_port
= htons(c
->mcast_port
+ id
),
1480 sa
->in6
.sin6_addr
= c
->mcast_group
.in6
;
1481 *salen
= sizeof(sa
->in6
);
1483 sa
->in
= (struct sockaddr_in
) {
1484 .sin_family
= AF_INET
,
1485 .sin_port
= htons(c
->mcast_port
+ id
),
1487 sa
->in
.sin_addr
= c
->mcast_group
.in
;
1488 *salen
= sizeof(sa
->in
);
1493 * Set up sending multicast socket over UDP
1495 static struct socket
*make_send_sock(struct netns_ipvs
*ipvs
, int id
)
1497 /* multicast addr */
1498 union ipvs_sockaddr mcast_addr
;
1499 struct socket
*sock
;
1502 /* First create a socket */
1503 result
= sock_create_kern(ipvs
->net
, ipvs
->mcfg
.mcast_af
, SOCK_DGRAM
,
1504 IPPROTO_UDP
, &sock
);
1506 pr_err("Error during creation of socket; terminating\n");
1507 return ERR_PTR(result
);
1509 result
= set_mcast_if(sock
->sk
, ipvs
->mcfg
.mcast_ifn
);
1511 pr_err("Error setting outbound mcast interface\n");
1515 set_mcast_loop(sock
->sk
, 0);
1516 set_mcast_ttl(sock
->sk
, ipvs
->mcfg
.mcast_ttl
);
1517 /* Allow fragmentation if MTU changes */
1518 set_mcast_pmtudisc(sock
->sk
, IP_PMTUDISC_DONT
);
1519 result
= sysctl_sync_sock_size(ipvs
);
1521 set_sock_size(sock
->sk
, 1, result
);
1523 if (AF_INET
== ipvs
->mcfg
.mcast_af
)
1524 result
= bind_mcastif_addr(sock
, ipvs
->mcfg
.mcast_ifn
);
1528 pr_err("Error binding address of the mcast interface\n");
1532 get_mcast_sockaddr(&mcast_addr
, &salen
, &ipvs
->mcfg
, id
);
1533 result
= sock
->ops
->connect(sock
, (struct sockaddr
*) &mcast_addr
,
1536 pr_err("Error connecting to the multicast addr\n");
1544 return ERR_PTR(result
);
1549 * Set up receiving multicast socket over UDP
1551 static struct socket
*make_receive_sock(struct netns_ipvs
*ipvs
, int id
,
1554 /* multicast addr */
1555 union ipvs_sockaddr mcast_addr
;
1556 struct socket
*sock
;
1559 /* First create a socket */
1560 result
= sock_create_kern(ipvs
->net
, ipvs
->bcfg
.mcast_af
, SOCK_DGRAM
,
1561 IPPROTO_UDP
, &sock
);
1563 pr_err("Error during creation of socket; terminating\n");
1564 return ERR_PTR(result
);
1566 /* it is equivalent to the REUSEADDR option in user-space */
1567 sock
->sk
->sk_reuse
= SK_CAN_REUSE
;
1568 result
= sysctl_sync_sock_size(ipvs
);
1570 set_sock_size(sock
->sk
, 0, result
);
1572 get_mcast_sockaddr(&mcast_addr
, &salen
, &ipvs
->bcfg
, id
);
1573 sock
->sk
->sk_bound_dev_if
= ifindex
;
1574 result
= sock
->ops
->bind(sock
, (struct sockaddr
*)&mcast_addr
, salen
);
1576 pr_err("Error binding to the multicast addr\n");
1580 /* join the multicast group */
1581 #ifdef CONFIG_IP_VS_IPV6
1582 if (ipvs
->bcfg
.mcast_af
== AF_INET6
)
1583 result
= join_mcast_group6(sock
->sk
, &mcast_addr
.in6
.sin6_addr
,
1584 ipvs
->bcfg
.mcast_ifn
);
1587 result
= join_mcast_group(sock
->sk
, &mcast_addr
.in
.sin_addr
,
1588 ipvs
->bcfg
.mcast_ifn
);
1590 pr_err("Error joining to the multicast group\n");
1598 return ERR_PTR(result
);
1603 ip_vs_send_async(struct socket
*sock
, const char *buffer
, const size_t length
)
1605 struct msghdr msg
= {.msg_flags
= MSG_DONTWAIT
|MSG_NOSIGNAL
};
1610 iov
.iov_base
= (void *)buffer
;
1611 iov
.iov_len
= length
;
1613 len
= kernel_sendmsg(sock
, &msg
, &iov
, 1, (size_t)(length
));
1620 ip_vs_send_sync_msg(struct socket
*sock
, struct ip_vs_sync_mesg
*msg
)
1625 msize
= ntohs(msg
->size
);
1627 ret
= ip_vs_send_async(sock
, (char *)msg
, msize
);
1628 if (ret
>= 0 || ret
== -EAGAIN
)
1630 pr_err("ip_vs_send_async error %d\n", ret
);
1635 ip_vs_receive(struct socket
*sock
, char *buffer
, const size_t buflen
)
1637 struct msghdr msg
= {NULL
,};
1643 /* Receive a packet */
1644 iov
.iov_base
= buffer
;
1645 iov
.iov_len
= (size_t)buflen
;
1647 len
= kernel_recvmsg(sock
, &msg
, &iov
, 1, buflen
, MSG_DONTWAIT
);
1656 /* Wakeup the master thread for sending */
1657 static void master_wakeup_work_handler(struct work_struct
*work
)
1659 struct ipvs_master_sync_state
*ms
=
1660 container_of(work
, struct ipvs_master_sync_state
,
1661 master_wakeup_work
.work
);
1662 struct netns_ipvs
*ipvs
= ms
->ipvs
;
1664 spin_lock_bh(&ipvs
->sync_lock
);
1665 if (ms
->sync_queue_len
&&
1666 ms
->sync_queue_delay
< IPVS_SYNC_WAKEUP_RATE
) {
1667 ms
->sync_queue_delay
= IPVS_SYNC_WAKEUP_RATE
;
1668 wake_up_process(ms
->master_thread
);
1670 spin_unlock_bh(&ipvs
->sync_lock
);
1673 /* Get next buffer to send */
1674 static inline struct ip_vs_sync_buff
*
1675 next_sync_buff(struct netns_ipvs
*ipvs
, struct ipvs_master_sync_state
*ms
)
1677 struct ip_vs_sync_buff
*sb
;
1679 sb
= sb_dequeue(ipvs
, ms
);
1682 /* Do not delay entries in buffer for more than 2 seconds */
1683 return get_curr_sync_buff(ipvs
, ms
, IPVS_SYNC_FLUSH_TIME
);
1686 static int sync_thread_master(void *data
)
1688 struct ip_vs_sync_thread_data
*tinfo
= data
;
1689 struct netns_ipvs
*ipvs
= tinfo
->ipvs
;
1690 struct ipvs_master_sync_state
*ms
= &ipvs
->ms
[tinfo
->id
];
1691 struct sock
*sk
= tinfo
->sock
->sk
;
1692 struct ip_vs_sync_buff
*sb
;
1694 pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
1695 "syncid = %d, id = %d\n",
1696 ipvs
->mcfg
.mcast_ifn
, ipvs
->mcfg
.syncid
, tinfo
->id
);
1699 sb
= next_sync_buff(ipvs
, ms
);
1700 if (unlikely(kthread_should_stop()))
1703 schedule_timeout(IPVS_SYNC_CHECK_PERIOD
);
1706 while (ip_vs_send_sync_msg(tinfo
->sock
, sb
->mesg
) < 0) {
1707 /* (Ab)use interruptible sleep to avoid increasing
1710 __wait_event_interruptible(*sk_sleep(sk
),
1711 sock_writeable(sk
) ||
1712 kthread_should_stop());
1713 if (unlikely(kthread_should_stop()))
1716 ip_vs_sync_buff_release(sb
);
1720 __set_current_state(TASK_RUNNING
);
1722 ip_vs_sync_buff_release(sb
);
1724 /* clean up the sync_buff queue */
1725 while ((sb
= sb_dequeue(ipvs
, ms
)))
1726 ip_vs_sync_buff_release(sb
);
1727 __set_current_state(TASK_RUNNING
);
1729 /* clean up the current sync_buff */
1730 sb
= get_curr_sync_buff(ipvs
, ms
, 0);
1732 ip_vs_sync_buff_release(sb
);
1734 /* release the sending multicast socket */
1735 sock_release(tinfo
->sock
);
1742 static int sync_thread_backup(void *data
)
1744 struct ip_vs_sync_thread_data
*tinfo
= data
;
1745 struct netns_ipvs
*ipvs
= tinfo
->ipvs
;
1748 pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
1749 "syncid = %d, id = %d\n",
1750 ipvs
->bcfg
.mcast_ifn
, ipvs
->bcfg
.syncid
, tinfo
->id
);
1752 while (!kthread_should_stop()) {
1753 wait_event_interruptible(*sk_sleep(tinfo
->sock
->sk
),
1754 !skb_queue_empty(&tinfo
->sock
->sk
->sk_receive_queue
)
1755 || kthread_should_stop());
1757 /* do we have data now? */
1758 while (!skb_queue_empty(&(tinfo
->sock
->sk
->sk_receive_queue
))) {
1759 len
= ip_vs_receive(tinfo
->sock
, tinfo
->buf
,
1760 ipvs
->bcfg
.sync_maxlen
);
1763 pr_err("receiving message error\n");
1767 ip_vs_process_message(ipvs
, tinfo
->buf
, len
);
1771 /* release the sending multicast socket */
1772 sock_release(tinfo
->sock
);
1780 int start_sync_thread(struct netns_ipvs
*ipvs
, struct ipvs_sync_daemon_cfg
*c
,
1783 struct ip_vs_sync_thread_data
*tinfo
;
1784 struct task_struct
**array
= NULL
, *task
;
1785 struct socket
*sock
;
1786 struct net_device
*dev
;
1788 int (*threadfn
)(void *data
);
1789 int id
, count
, hlen
;
1790 int result
= -ENOMEM
;
1793 IP_VS_DBG(7, "%s(): pid %d\n", __func__
, task_pid_nr(current
));
1794 IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %zd bytes\n",
1795 sizeof(struct ip_vs_sync_conn_v0
));
1797 if (!ipvs
->sync_state
) {
1798 count
= clamp(sysctl_sync_ports(ipvs
), 1, IPVS_SYNC_PORTS_MAX
);
1799 ipvs
->threads_mask
= count
- 1;
1801 count
= ipvs
->threads_mask
+ 1;
1803 if (c
->mcast_af
== AF_UNSPEC
) {
1804 c
->mcast_af
= AF_INET
;
1805 c
->mcast_group
.ip
= cpu_to_be32(IP_VS_SYNC_GROUP
);
1808 c
->mcast_port
= IP_VS_SYNC_PORT
;
1812 dev
= __dev_get_by_name(ipvs
->net
, c
->mcast_ifn
);
1814 pr_err("Unknown mcast interface: %s\n", c
->mcast_ifn
);
1817 hlen
= (AF_INET6
== c
->mcast_af
) ?
1818 sizeof(struct ipv6hdr
) + sizeof(struct udphdr
) :
1819 sizeof(struct iphdr
) + sizeof(struct udphdr
);
1820 mtu
= (state
== IP_VS_STATE_BACKUP
) ?
1821 clamp(dev
->mtu
, 1500U, 65535U) : 1500U;
1822 min_mtu
= (state
== IP_VS_STATE_BACKUP
) ? 1024 : 1;
1825 c
->sync_maxlen
= clamp_t(unsigned int,
1826 c
->sync_maxlen
, min_mtu
,
1829 c
->sync_maxlen
= mtu
- hlen
;
1831 if (state
== IP_VS_STATE_MASTER
) {
1836 name
= "ipvs-m:%d:%d";
1837 threadfn
= sync_thread_master
;
1838 } else if (state
== IP_VS_STATE_BACKUP
) {
1839 if (ipvs
->backup_threads
)
1843 name
= "ipvs-b:%d:%d";
1844 threadfn
= sync_thread_backup
;
1849 if (state
== IP_VS_STATE_MASTER
) {
1850 struct ipvs_master_sync_state
*ms
;
1852 ipvs
->ms
= kcalloc(count
, sizeof(ipvs
->ms
[0]), GFP_KERNEL
);
1856 for (id
= 0; id
< count
; id
++, ms
++) {
1857 INIT_LIST_HEAD(&ms
->sync_queue
);
1858 ms
->sync_queue_len
= 0;
1859 ms
->sync_queue_delay
= 0;
1860 INIT_DELAYED_WORK(&ms
->master_wakeup_work
,
1861 master_wakeup_work_handler
);
1865 array
= kcalloc(count
, sizeof(struct task_struct
*),
1872 for (id
= 0; id
< count
; id
++) {
1873 if (state
== IP_VS_STATE_MASTER
)
1874 sock
= make_send_sock(ipvs
, id
);
1876 sock
= make_receive_sock(ipvs
, id
, dev
->ifindex
);
1878 result
= PTR_ERR(sock
);
1881 tinfo
= kmalloc(sizeof(*tinfo
), GFP_KERNEL
);
1886 if (state
== IP_VS_STATE_BACKUP
) {
1887 tinfo
->buf
= kmalloc(ipvs
->bcfg
.sync_maxlen
,
1896 task
= kthread_run(threadfn
, tinfo
, name
, ipvs
->gen
, id
);
1898 result
= PTR_ERR(task
);
1902 if (state
== IP_VS_STATE_MASTER
)
1903 ipvs
->ms
[id
].master_thread
= task
;
1908 /* mark as active */
1910 if (state
== IP_VS_STATE_BACKUP
)
1911 ipvs
->backup_threads
= array
;
1912 spin_lock_bh(&ipvs
->sync_buff_lock
);
1913 ipvs
->sync_state
|= state
;
1914 spin_unlock_bh(&ipvs
->sync_buff_lock
);
1916 /* increase the module use count */
1917 ip_vs_use_count_inc();
1926 sock_release(tinfo
->sock
);
1931 while (count
-- > 0) {
1932 if (state
== IP_VS_STATE_MASTER
)
1933 kthread_stop(ipvs
->ms
[count
].master_thread
);
1935 kthread_stop(array
[count
]);
1940 if (!(ipvs
->sync_state
& IP_VS_STATE_MASTER
)) {
1948 int stop_sync_thread(struct netns_ipvs
*ipvs
, int state
)
1950 struct task_struct
**array
;
1954 IP_VS_DBG(7, "%s(): pid %d\n", __func__
, task_pid_nr(current
));
1956 if (state
== IP_VS_STATE_MASTER
) {
1961 * The lock synchronizes with sb_queue_tail(), so that we don't
1962 * add sync buffers to the queue, when we are already in
1963 * progress of stopping the master sync daemon.
1966 spin_lock_bh(&ipvs
->sync_buff_lock
);
1967 spin_lock(&ipvs
->sync_lock
);
1968 ipvs
->sync_state
&= ~IP_VS_STATE_MASTER
;
1969 spin_unlock(&ipvs
->sync_lock
);
1970 spin_unlock_bh(&ipvs
->sync_buff_lock
);
1973 for (id
= ipvs
->threads_mask
; id
>= 0; id
--) {
1974 struct ipvs_master_sync_state
*ms
= &ipvs
->ms
[id
];
1977 pr_info("stopping master sync thread %d ...\n",
1978 task_pid_nr(ms
->master_thread
));
1979 cancel_delayed_work_sync(&ms
->master_wakeup_work
);
1980 ret
= kthread_stop(ms
->master_thread
);
1986 } else if (state
== IP_VS_STATE_BACKUP
) {
1987 if (!ipvs
->backup_threads
)
1990 ipvs
->sync_state
&= ~IP_VS_STATE_BACKUP
;
1991 array
= ipvs
->backup_threads
;
1993 for (id
= ipvs
->threads_mask
; id
>= 0; id
--) {
1996 pr_info("stopping backup sync thread %d ...\n",
1997 task_pid_nr(array
[id
]));
1998 ret
= kthread_stop(array
[id
]);
2003 ipvs
->backup_threads
= NULL
;
2006 /* decrease the module use count */
2007 ip_vs_use_count_dec();
2013 * Initialize data struct for each netns
2015 int __net_init
ip_vs_sync_net_init(struct netns_ipvs
*ipvs
)
2017 __mutex_init(&ipvs
->sync_mutex
, "ipvs->sync_mutex", &__ipvs_sync_key
);
2018 spin_lock_init(&ipvs
->sync_lock
);
2019 spin_lock_init(&ipvs
->sync_buff_lock
);
2023 void ip_vs_sync_net_cleanup(struct netns_ipvs
*ipvs
)
2027 mutex_lock(&ipvs
->sync_mutex
);
2028 retc
= stop_sync_thread(ipvs
, IP_VS_STATE_MASTER
);
2029 if (retc
&& retc
!= -ESRCH
)
2030 pr_err("Failed to stop Master Daemon\n");
2032 retc
= stop_sync_thread(ipvs
, IP_VS_STATE_BACKUP
);
2033 if (retc
&& retc
!= -ESRCH
)
2034 pr_err("Failed to stop Backup Daemon\n");
2035 mutex_unlock(&ipvs
->sync_mutex
);