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 struct ip_vs_sync_buff {
266 struct list_head list;
267 unsigned long firstuse;
269 /* pointers for the message data */
270 struct ip_vs_sync_mesg *mesg;
276 * Copy of struct ip_vs_seq
277 * From unaligned network order to aligned host order
279 static void ntoh_seq(struct ip_vs_seq *no, struct ip_vs_seq *ho)
281 ho->init_seq = get_unaligned_be32(&no->init_seq);
282 ho->delta = get_unaligned_be32(&no->delta);
283 ho->previous_delta = get_unaligned_be32(&no->previous_delta);
287 * Copy of struct ip_vs_seq
288 * From Aligned host order to unaligned network order
290 static void hton_seq(struct ip_vs_seq *ho, struct ip_vs_seq *no)
292 put_unaligned_be32(ho->init_seq, &no->init_seq);
293 put_unaligned_be32(ho->delta, &no->delta);
294 put_unaligned_be32(ho->previous_delta, &no->previous_delta);
297 static inline struct ip_vs_sync_buff *
298 sb_dequeue(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
300 struct ip_vs_sync_buff *sb;
302 spin_lock_bh(&ipvs->sync_lock);
303 if (list_empty(&ms->sync_queue)) {
305 __set_current_state(TASK_INTERRUPTIBLE);
307 sb = list_entry(ms->sync_queue.next, struct ip_vs_sync_buff,
310 ms->sync_queue_len--;
311 if (!ms->sync_queue_len)
312 ms->sync_queue_delay = 0;
314 spin_unlock_bh(&ipvs->sync_lock);
320 * Create a new sync buffer for Version 1 proto.
322 static inline struct ip_vs_sync_buff *
323 ip_vs_sync_buff_create(struct netns_ipvs *ipvs)
325 struct ip_vs_sync_buff *sb;
327 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
330 sb->mesg = kmalloc(ipvs->send_mesg_maxlen, GFP_ATOMIC);
335 sb->mesg->reserved = 0; /* old nr_conns i.e. must be zero now */
336 sb->mesg->version = SYNC_PROTO_VER;
337 sb->mesg->syncid = ipvs->master_syncid;
338 sb->mesg->size = htons(sizeof(struct ip_vs_sync_mesg));
339 sb->mesg->nr_conns = 0;
341 sb->head = (unsigned char *)sb->mesg + sizeof(struct ip_vs_sync_mesg);
342 sb->end = (unsigned char *)sb->mesg + ipvs->send_mesg_maxlen;
344 sb->firstuse = jiffies;
348 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
354 static inline void sb_queue_tail(struct netns_ipvs *ipvs,
355 struct ipvs_master_sync_state *ms)
357 struct ip_vs_sync_buff *sb = ms->sync_buff;
359 spin_lock(&ipvs->sync_lock);
360 if (ipvs->sync_state & IP_VS_STATE_MASTER &&
361 ms->sync_queue_len < sysctl_sync_qlen_max(ipvs)) {
362 if (!ms->sync_queue_len)
363 schedule_delayed_work(&ms->master_wakeup_work,
364 max(IPVS_SYNC_SEND_DELAY, 1));
365 ms->sync_queue_len++;
366 list_add_tail(&sb->list, &ms->sync_queue);
367 if ((++ms->sync_queue_delay) == IPVS_SYNC_WAKEUP_RATE)
368 wake_up_process(ms->master_thread);
370 ip_vs_sync_buff_release(sb);
371 spin_unlock(&ipvs->sync_lock);
375 * Get the current sync buffer if it has been created for more
376 * than the specified time or the specified time is zero.
378 static inline struct ip_vs_sync_buff *
379 get_curr_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms,
382 struct ip_vs_sync_buff *sb;
384 spin_lock_bh(&ipvs->sync_buff_lock);
386 if (sb && time_after_eq(jiffies - sb->firstuse, time)) {
387 ms->sync_buff = NULL;
388 __set_current_state(TASK_RUNNING);
391 spin_unlock_bh(&ipvs->sync_buff_lock);
396 select_master_thread_id(struct netns_ipvs *ipvs, struct ip_vs_conn *cp)
398 return ((long) cp >> (1 + ilog2(sizeof(*cp)))) & ipvs->threads_mask;
402 * Create a new sync buffer for Version 0 proto.
404 static inline struct ip_vs_sync_buff *
405 ip_vs_sync_buff_create_v0(struct netns_ipvs *ipvs)
407 struct ip_vs_sync_buff *sb;
408 struct ip_vs_sync_mesg_v0 *mesg;
410 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
413 sb->mesg = kmalloc(ipvs->send_mesg_maxlen, GFP_ATOMIC);
418 mesg = (struct ip_vs_sync_mesg_v0 *)sb->mesg;
420 mesg->syncid = ipvs->master_syncid;
421 mesg->size = htons(sizeof(struct ip_vs_sync_mesg_v0));
422 sb->head = (unsigned char *)mesg + sizeof(struct ip_vs_sync_mesg_v0);
423 sb->end = (unsigned char *)mesg + ipvs->send_mesg_maxlen;
424 sb->firstuse = jiffies;
428 /* Check if conn should be synced.
429 * pkts: conn packets, use sysctl_sync_threshold to avoid packet check
430 * - (1) sync_refresh_period: reduce sync rate. Additionally, retry
431 * sync_retries times with period of sync_refresh_period/8
432 * - (2) if both sync_refresh_period and sync_period are 0 send sync only
433 * for state changes or only once when pkts matches sync_threshold
434 * - (3) templates: rate can be reduced only with sync_refresh_period or
437 static int ip_vs_sync_conn_needed(struct netns_ipvs *ipvs,
438 struct ip_vs_conn *cp, int pkts)
440 unsigned long orig = ACCESS_ONCE(cp->sync_endtime);
441 unsigned long now = jiffies;
442 unsigned long n = (now + cp->timeout) & ~3UL;
443 unsigned int sync_refresh_period;
447 /* Check if we sync in current state */
448 if (unlikely(cp->flags & IP_VS_CONN_F_TEMPLATE))
450 else if (likely(cp->protocol == IPPROTO_TCP)) {
451 if (!((1 << cp->state) &
452 ((1 << IP_VS_TCP_S_ESTABLISHED) |
453 (1 << IP_VS_TCP_S_FIN_WAIT) |
454 (1 << IP_VS_TCP_S_CLOSE) |
455 (1 << IP_VS_TCP_S_CLOSE_WAIT) |
456 (1 << IP_VS_TCP_S_TIME_WAIT))))
458 force = cp->state != cp->old_state;
459 if (force && cp->state != IP_VS_TCP_S_ESTABLISHED)
461 } else if (unlikely(cp->protocol == IPPROTO_SCTP)) {
462 if (!((1 << cp->state) &
463 ((1 << IP_VS_SCTP_S_ESTABLISHED) |
464 (1 << IP_VS_SCTP_S_SHUTDOWN_SENT) |
465 (1 << IP_VS_SCTP_S_SHUTDOWN_RECEIVED) |
466 (1 << IP_VS_SCTP_S_SHUTDOWN_ACK_SENT) |
467 (1 << IP_VS_SCTP_S_CLOSED))))
469 force = cp->state != cp->old_state;
470 if (force && cp->state != IP_VS_SCTP_S_ESTABLISHED)
473 /* UDP or another protocol with single state */
477 sync_refresh_period = sysctl_sync_refresh_period(ipvs);
478 if (sync_refresh_period > 0) {
479 long diff = n - orig;
480 long min_diff = max(cp->timeout >> 1, 10UL * HZ);
482 /* Avoid sync if difference is below sync_refresh_period
483 * and below the half timeout.
485 if (abs(diff) < min_t(long, sync_refresh_period, min_diff)) {
486 int retries = orig & 3;
488 if (retries >= sysctl_sync_retries(ipvs))
490 if (time_before(now, orig - cp->timeout +
491 (sync_refresh_period >> 3)))
496 sync_period = sysctl_sync_period(ipvs);
497 if (sync_period > 0) {
498 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE) &&
499 pkts % sync_period != sysctl_sync_threshold(ipvs))
501 } else if (sync_refresh_period <= 0 &&
502 pkts != sysctl_sync_threshold(ipvs))
506 cp->old_state = cp->state;
507 n = cmpxchg(&cp->sync_endtime, orig, n);
508 return n == orig || force;
512 * Version 0 , could be switched in by sys_ctl.
513 * Add an ip_vs_conn information into the current sync_buff.
515 static void ip_vs_sync_conn_v0(struct net *net, struct ip_vs_conn *cp,
518 struct netns_ipvs *ipvs = net_ipvs(net);
519 struct ip_vs_sync_mesg_v0 *m;
520 struct ip_vs_sync_conn_v0 *s;
521 struct ip_vs_sync_buff *buff;
522 struct ipvs_master_sync_state *ms;
526 if (unlikely(cp->af != AF_INET))
528 /* Do not sync ONE PACKET */
529 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
532 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
535 spin_lock_bh(&ipvs->sync_buff_lock);
536 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
537 spin_unlock_bh(&ipvs->sync_buff_lock);
541 id = select_master_thread_id(ipvs, cp);
543 buff = ms->sync_buff;
545 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
546 /* Send buffer if it is for v1 */
548 sb_queue_tail(ipvs, ms);
549 ms->sync_buff = NULL;
554 buff = ip_vs_sync_buff_create_v0(ipvs);
556 spin_unlock_bh(&ipvs->sync_buff_lock);
557 pr_err("ip_vs_sync_buff_create failed.\n");
560 ms->sync_buff = buff;
563 len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
565 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
566 s = (struct ip_vs_sync_conn_v0 *) buff->head;
570 s->protocol = cp->protocol;
571 s->cport = cp->cport;
572 s->vport = cp->vport;
573 s->dport = cp->dport;
574 s->caddr = cp->caddr.ip;
575 s->vaddr = cp->vaddr.ip;
576 s->daddr = cp->daddr.ip;
577 s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
578 s->state = htons(cp->state);
579 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
580 struct ip_vs_sync_conn_options *opt =
581 (struct ip_vs_sync_conn_options *)&s[1];
582 memcpy(opt, &cp->in_seq, sizeof(*opt));
586 m->size = htons(ntohs(m->size) + len);
589 /* check if there is a space for next one */
590 if (buff->head + FULL_CONN_SIZE > buff->end) {
591 sb_queue_tail(ipvs, ms);
592 ms->sync_buff = NULL;
594 spin_unlock_bh(&ipvs->sync_buff_lock);
596 /* synchronize its controller if it has */
599 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
600 pkts = atomic_add_return(1, &cp->in_pkts);
602 pkts = sysctl_sync_threshold(ipvs);
603 ip_vs_sync_conn(net, cp->control, pkts);
608 * Add an ip_vs_conn information into the current sync_buff.
609 * Called by ip_vs_in.
610 * Sending Version 1 messages
612 void ip_vs_sync_conn(struct net *net, struct ip_vs_conn *cp, int pkts)
614 struct netns_ipvs *ipvs = net_ipvs(net);
615 struct ip_vs_sync_mesg *m;
616 union ip_vs_sync_conn *s;
617 struct ip_vs_sync_buff *buff;
618 struct ipvs_master_sync_state *ms;
621 unsigned int len, pe_name_len, pad;
623 /* Handle old version of the protocol */
624 if (sysctl_sync_ver(ipvs) == 0) {
625 ip_vs_sync_conn_v0(net, cp, pkts);
628 /* Do not sync ONE PACKET */
629 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
632 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
637 if (cp->pe_data_len) {
638 if (!cp->pe_data || !cp->dest) {
639 IP_VS_ERR_RL("SYNC, connection pe_data invalid\n");
642 pe_name_len = strnlen(cp->pe->name, IP_VS_PENAME_MAXLEN);
645 spin_lock_bh(&ipvs->sync_buff_lock);
646 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
647 spin_unlock_bh(&ipvs->sync_buff_lock);
651 id = select_master_thread_id(ipvs, cp);
654 #ifdef CONFIG_IP_VS_IPV6
655 if (cp->af == AF_INET6)
656 len = sizeof(struct ip_vs_sync_v6);
659 len = sizeof(struct ip_vs_sync_v4);
661 if (cp->flags & IP_VS_CONN_F_SEQ_MASK)
662 len += sizeof(struct ip_vs_sync_conn_options) + 2;
665 len += cp->pe_data_len + 2; /* + Param hdr field */
667 len += pe_name_len + 2;
669 /* check if there is a space for this one */
671 buff = ms->sync_buff;
674 pad = (4 - (size_t) buff->head) & 3;
675 /* Send buffer if it is for v0 */
676 if (buff->head + len + pad > buff->end || m->reserved) {
677 sb_queue_tail(ipvs, ms);
678 ms->sync_buff = NULL;
685 buff = ip_vs_sync_buff_create(ipvs);
687 spin_unlock_bh(&ipvs->sync_buff_lock);
688 pr_err("ip_vs_sync_buff_create failed.\n");
691 ms->sync_buff = buff;
696 buff->head += pad + len;
697 m->size = htons(ntohs(m->size) + pad + len);
698 /* Add ev. padding from prev. sync_conn */
702 s = (union ip_vs_sync_conn *)p;
704 /* Set message type & copy members */
705 s->v4.type = (cp->af == AF_INET6 ? STYPE_F_INET6 : 0);
706 s->v4.ver_size = htons(len & SVER_MASK); /* Version 0 */
707 s->v4.flags = htonl(cp->flags & ~IP_VS_CONN_F_HASHED);
708 s->v4.state = htons(cp->state);
709 s->v4.protocol = cp->protocol;
710 s->v4.cport = cp->cport;
711 s->v4.vport = cp->vport;
712 s->v4.dport = cp->dport;
713 s->v4.fwmark = htonl(cp->fwmark);
714 s->v4.timeout = htonl(cp->timeout / HZ);
717 #ifdef CONFIG_IP_VS_IPV6
718 if (cp->af == AF_INET6) {
719 p += sizeof(struct ip_vs_sync_v6);
720 s->v6.caddr = cp->caddr.in6;
721 s->v6.vaddr = cp->vaddr.in6;
722 s->v6.daddr = cp->daddr.in6;
726 p += sizeof(struct ip_vs_sync_v4); /* options ptr */
727 s->v4.caddr = cp->caddr.ip;
728 s->v4.vaddr = cp->vaddr.ip;
729 s->v4.daddr = cp->daddr.ip;
731 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
732 *(p++) = IPVS_OPT_SEQ_DATA;
733 *(p++) = sizeof(struct ip_vs_sync_conn_options);
734 hton_seq((struct ip_vs_seq *)p, &cp->in_seq);
735 p += sizeof(struct ip_vs_seq);
736 hton_seq((struct ip_vs_seq *)p, &cp->out_seq);
737 p += sizeof(struct ip_vs_seq);
740 if (cp->pe_data_len && cp->pe_data) {
741 *(p++) = IPVS_OPT_PE_DATA;
742 *(p++) = cp->pe_data_len;
743 memcpy(p, cp->pe_data, cp->pe_data_len);
744 p += cp->pe_data_len;
747 *(p++) = IPVS_OPT_PE_NAME;
748 *(p++) = pe_name_len;
749 memcpy(p, cp->pe->name, pe_name_len);
754 spin_unlock_bh(&ipvs->sync_buff_lock);
757 /* synchronize its controller if it has */
761 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
762 pkts = atomic_add_return(1, &cp->in_pkts);
764 pkts = sysctl_sync_threshold(ipvs);
769 * fill_param used by version 1
772 ip_vs_conn_fill_param_sync(struct net *net, int af, union ip_vs_sync_conn *sc,
773 struct ip_vs_conn_param *p,
774 __u8 *pe_data, unsigned int pe_data_len,
775 __u8 *pe_name, unsigned int pe_name_len)
777 #ifdef CONFIG_IP_VS_IPV6
779 ip_vs_conn_fill_param(net, af, sc->v6.protocol,
780 (const union nf_inet_addr *)&sc->v6.caddr,
782 (const union nf_inet_addr *)&sc->v6.vaddr,
786 ip_vs_conn_fill_param(net, af, sc->v4.protocol,
787 (const union nf_inet_addr *)&sc->v4.caddr,
789 (const union nf_inet_addr *)&sc->v4.vaddr,
794 char buff[IP_VS_PENAME_MAXLEN+1];
796 memcpy(buff, pe_name, pe_name_len);
798 p->pe = __ip_vs_pe_getbyname(buff);
800 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n",
805 IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n");
809 p->pe_data = kmemdup(pe_data, pe_data_len, GFP_ATOMIC);
812 module_put(p->pe->module);
815 p->pe_data_len = pe_data_len;
821 * Connection Add / Update.
822 * Common for version 0 and 1 reception of backup sync_conns.
826 static void ip_vs_proc_conn(struct net *net, struct ip_vs_conn_param *param,
827 unsigned int flags, unsigned int state,
828 unsigned int protocol, unsigned int type,
829 const union nf_inet_addr *daddr, __be16 dport,
830 unsigned long timeout, __u32 fwmark,
831 struct ip_vs_sync_conn_options *opt)
833 struct ip_vs_dest *dest;
834 struct ip_vs_conn *cp;
835 struct netns_ipvs *ipvs = net_ipvs(net);
837 if (!(flags & IP_VS_CONN_F_TEMPLATE))
838 cp = ip_vs_conn_in_get(param);
840 cp = ip_vs_ct_in_get(param);
844 kfree(param->pe_data);
847 spin_lock_bh(&cp->lock);
848 if ((cp->flags ^ flags) & IP_VS_CONN_F_INACTIVE &&
849 !(flags & IP_VS_CONN_F_TEMPLATE) && dest) {
850 if (flags & IP_VS_CONN_F_INACTIVE) {
851 atomic_dec(&dest->activeconns);
852 atomic_inc(&dest->inactconns);
854 atomic_inc(&dest->activeconns);
855 atomic_dec(&dest->inactconns);
858 flags &= IP_VS_CONN_F_BACKUP_UPD_MASK;
859 flags |= cp->flags & ~IP_VS_CONN_F_BACKUP_UPD_MASK;
861 spin_unlock_bh(&cp->lock);
863 ip_vs_try_bind_dest(cp);
866 * Find the appropriate destination for the connection.
867 * If it is not found the connection will remain unbound
871 dest = ip_vs_find_dest(net, type, daddr, dport, param->vaddr,
872 param->vport, protocol, fwmark, flags);
874 cp = ip_vs_conn_new(param, daddr, dport, flags, dest, fwmark);
878 kfree(param->pe_data);
879 IP_VS_DBG(2, "BACKUP, add new conn. failed\n");
885 memcpy(&cp->in_seq, opt, sizeof(*opt));
886 atomic_set(&cp->in_pkts, sysctl_sync_threshold(ipvs));
888 cp->old_state = cp->state;
890 * For Ver 0 messages style
891 * - Not possible to recover the right timeout for templates
892 * - can not find the right fwmark
893 * virtual service. If needed, we can do it for
894 * non-fwmark persistent services.
895 * Ver 1 messages style.
899 if (timeout > MAX_SCHEDULE_TIMEOUT / HZ)
900 timeout = MAX_SCHEDULE_TIMEOUT / HZ;
901 cp->timeout = timeout*HZ;
903 struct ip_vs_proto_data *pd;
905 pd = ip_vs_proto_data_get(net, protocol);
906 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pd && pd->timeout_table)
907 cp->timeout = pd->timeout_table[state];
909 cp->timeout = (3*60*HZ);
915 * Process received multicast message for Version 0
917 static void ip_vs_process_message_v0(struct net *net, const char *buffer,
920 struct ip_vs_sync_mesg_v0 *m = (struct ip_vs_sync_mesg_v0 *)buffer;
921 struct ip_vs_sync_conn_v0 *s;
922 struct ip_vs_sync_conn_options *opt;
923 struct ip_vs_protocol *pp;
924 struct ip_vs_conn_param param;
928 p = (char *)buffer + sizeof(struct ip_vs_sync_mesg_v0);
929 for (i=0; i<m->nr_conns; i++) {
930 unsigned int flags, state;
932 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
933 IP_VS_ERR_RL("BACKUP v0, bogus conn\n");
936 s = (struct ip_vs_sync_conn_v0 *) p;
937 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
938 flags &= ~IP_VS_CONN_F_HASHED;
939 if (flags & IP_VS_CONN_F_SEQ_MASK) {
940 opt = (struct ip_vs_sync_conn_options *)&s[1];
942 if (p > buffer+buflen) {
943 IP_VS_ERR_RL("BACKUP v0, Dropping buffer bogus conn options\n");
948 p += SIMPLE_CONN_SIZE;
951 state = ntohs(s->state);
952 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
953 pp = ip_vs_proto_get(s->protocol);
955 IP_VS_DBG(2, "BACKUP v0, Unsupported protocol %u\n",
959 if (state >= pp->num_states) {
960 IP_VS_DBG(2, "BACKUP v0, Invalid %s state %u\n",
965 /* protocol in templates is not used for state/timeout */
967 IP_VS_DBG(2, "BACKUP v0, Invalid template state %u\n",
973 ip_vs_conn_fill_param(net, AF_INET, s->protocol,
974 (const union nf_inet_addr *)&s->caddr,
976 (const union nf_inet_addr *)&s->vaddr,
979 /* Send timeout as Zero */
980 ip_vs_proc_conn(net, ¶m, flags, state, s->protocol, AF_INET,
981 (union nf_inet_addr *)&s->daddr, s->dport,
989 static inline int ip_vs_proc_seqopt(__u8 *p, unsigned int plen,
991 struct ip_vs_sync_conn_options *opt)
993 struct ip_vs_sync_conn_options *topt;
995 topt = (struct ip_vs_sync_conn_options *)p;
997 if (plen != sizeof(struct ip_vs_sync_conn_options)) {
998 IP_VS_DBG(2, "BACKUP, bogus conn options length\n");
1001 if (*opt_flags & IPVS_OPT_F_SEQ_DATA) {
1002 IP_VS_DBG(2, "BACKUP, conn options found twice\n");
1005 ntoh_seq(&topt->in_seq, &opt->in_seq);
1006 ntoh_seq(&topt->out_seq, &opt->out_seq);
1007 *opt_flags |= IPVS_OPT_F_SEQ_DATA;
1011 static int ip_vs_proc_str(__u8 *p, unsigned int plen, unsigned int *data_len,
1012 __u8 **data, unsigned int maxlen,
1013 __u32 *opt_flags, __u32 flag)
1015 if (plen > maxlen) {
1016 IP_VS_DBG(2, "BACKUP, bogus par.data len > %d\n", maxlen);
1019 if (*opt_flags & flag) {
1020 IP_VS_DBG(2, "BACKUP, Par.data found twice 0x%x\n", flag);
1029 * Process a Version 1 sync. connection
1031 static inline int ip_vs_proc_sync_conn(struct net *net, __u8 *p, __u8 *msg_end)
1033 struct ip_vs_sync_conn_options opt;
1034 union ip_vs_sync_conn *s;
1035 struct ip_vs_protocol *pp;
1036 struct ip_vs_conn_param param;
1038 unsigned int af, state, pe_data_len=0, pe_name_len=0;
1039 __u8 *pe_data=NULL, *pe_name=NULL;
1043 s = (union ip_vs_sync_conn *) p;
1045 if (s->v6.type & STYPE_F_INET6) {
1046 #ifdef CONFIG_IP_VS_IPV6
1048 p += sizeof(struct ip_vs_sync_v6);
1050 IP_VS_DBG(3,"BACKUP, IPv6 msg received, and IPVS is not compiled for IPv6\n");
1054 } else if (!s->v4.type) {
1056 p += sizeof(struct ip_vs_sync_v4);
1063 /* Process optional params check Type & Len. */
1064 while (p < msg_end) {
1073 if (!plen || ((p + plen) > msg_end))
1075 /* Handle seq option p = param data */
1076 switch (ptype & ~IPVS_OPT_F_PARAM) {
1077 case IPVS_OPT_SEQ_DATA:
1078 if (ip_vs_proc_seqopt(p, plen, &opt_flags, &opt))
1082 case IPVS_OPT_PE_DATA:
1083 if (ip_vs_proc_str(p, plen, &pe_data_len, &pe_data,
1084 IP_VS_PEDATA_MAXLEN, &opt_flags,
1085 IPVS_OPT_F_PE_DATA))
1089 case IPVS_OPT_PE_NAME:
1090 if (ip_vs_proc_str(p, plen,&pe_name_len, &pe_name,
1091 IP_VS_PENAME_MAXLEN, &opt_flags,
1092 IPVS_OPT_F_PE_NAME))
1097 /* Param data mandatory ? */
1098 if (!(ptype & IPVS_OPT_F_PARAM)) {
1099 IP_VS_DBG(3, "BACKUP, Unknown mandatory param %d found\n",
1100 ptype & ~IPVS_OPT_F_PARAM);
1105 p += plen; /* Next option */
1108 /* Get flags and Mask off unsupported */
1109 flags = ntohl(s->v4.flags) & IP_VS_CONN_F_BACKUP_MASK;
1110 flags |= IP_VS_CONN_F_SYNC;
1111 state = ntohs(s->v4.state);
1113 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
1114 pp = ip_vs_proto_get(s->v4.protocol);
1116 IP_VS_DBG(3,"BACKUP, Unsupported protocol %u\n",
1121 if (state >= pp->num_states) {
1122 IP_VS_DBG(3, "BACKUP, Invalid %s state %u\n",
1128 /* protocol in templates is not used for state/timeout */
1130 IP_VS_DBG(3, "BACKUP, Invalid template state %u\n",
1135 if (ip_vs_conn_fill_param_sync(net, af, s, ¶m, pe_data,
1136 pe_data_len, pe_name, pe_name_len)) {
1140 /* If only IPv4, just silent skip IPv6 */
1142 ip_vs_proc_conn(net, ¶m, flags, state, s->v4.protocol, af,
1143 (union nf_inet_addr *)&s->v4.daddr, s->v4.dport,
1144 ntohl(s->v4.timeout), ntohl(s->v4.fwmark),
1145 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1147 #ifdef CONFIG_IP_VS_IPV6
1149 ip_vs_proc_conn(net, ¶m, flags, state, s->v6.protocol, af,
1150 (union nf_inet_addr *)&s->v6.daddr, s->v6.dport,
1151 ntohl(s->v6.timeout), ntohl(s->v6.fwmark),
1152 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1158 IP_VS_DBG(2, "BACKUP, Single msg dropped err:%d\n", retc);
1163 * Process received multicast message and create the corresponding
1164 * ip_vs_conn entries.
1165 * Handles Version 0 & 1
1167 static void ip_vs_process_message(struct net *net, __u8 *buffer,
1168 const size_t buflen)
1170 struct netns_ipvs *ipvs = net_ipvs(net);
1171 struct ip_vs_sync_mesg *m2 = (struct ip_vs_sync_mesg *)buffer;
1175 if (buflen < sizeof(struct ip_vs_sync_mesg_v0)) {
1176 IP_VS_DBG(2, "BACKUP, message header too short\n");
1180 if (buflen != ntohs(m2->size)) {
1181 IP_VS_DBG(2, "BACKUP, bogus message size\n");
1184 /* SyncID sanity check */
1185 if (ipvs->backup_syncid != 0 && m2->syncid != ipvs->backup_syncid) {
1186 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2->syncid);
1189 /* Handle version 1 message */
1190 if ((m2->version == SYNC_PROTO_VER) && (m2->reserved == 0)
1191 && (m2->spare == 0)) {
1193 msg_end = buffer + sizeof(struct ip_vs_sync_mesg);
1194 nr_conns = m2->nr_conns;
1196 for (i=0; i<nr_conns; i++) {
1197 union ip_vs_sync_conn *s;
1202 if (p + sizeof(s->v4) > buffer+buflen) {
1203 IP_VS_ERR_RL("BACKUP, Dropping buffer, to small\n");
1206 s = (union ip_vs_sync_conn *)p;
1207 size = ntohs(s->v4.ver_size) & SVER_MASK;
1209 /* Basic sanity checks */
1210 if (msg_end > buffer+buflen) {
1211 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n");
1214 if (ntohs(s->v4.ver_size) >> SVER_SHIFT) {
1215 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n",
1216 ntohs(s->v4.ver_size) >> SVER_SHIFT);
1219 /* Process a single sync_conn */
1220 retc = ip_vs_proc_sync_conn(net, p, msg_end);
1222 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n",
1226 /* Make sure we have 32 bit alignment */
1227 msg_end = p + ((size + 3) & ~3);
1230 /* Old type of message */
1231 ip_vs_process_message_v0(net, buffer, buflen);
1238 * Setup sndbuf (mode=1) or rcvbuf (mode=0)
1240 static void set_sock_size(struct sock *sk, int mode, int val)
1242 /* setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &val, sizeof(val)); */
1243 /* setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)); */
1246 val = clamp_t(int, val, (SOCK_MIN_SNDBUF + 1) / 2,
1248 sk->sk_sndbuf = val * 2;
1249 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1251 val = clamp_t(int, val, (SOCK_MIN_RCVBUF + 1) / 2,
1253 sk->sk_rcvbuf = val * 2;
1254 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1260 * Setup loopback of outgoing multicasts on a sending socket
1262 static void set_mcast_loop(struct sock *sk, u_char loop)
1264 struct inet_sock *inet = inet_sk(sk);
1266 /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
1268 inet->mc_loop = loop ? 1 : 0;
1273 * Specify TTL for outgoing multicasts on a sending socket
1275 static void set_mcast_ttl(struct sock *sk, u_char ttl)
1277 struct inet_sock *inet = inet_sk(sk);
1279 /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
1286 * Specifiy default interface for outgoing multicasts
1288 static int set_mcast_if(struct sock *sk, char *ifname)
1290 struct net_device *dev;
1291 struct inet_sock *inet = inet_sk(sk);
1292 struct net *net = sock_net(sk);
1294 dev = __dev_get_by_name(net, ifname);
1298 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1302 inet->mc_index = dev->ifindex;
1303 /* inet->mc_addr = 0; */
1311 * Set the maximum length of sync message according to the
1312 * specified interface's MTU.
1314 static int set_sync_mesg_maxlen(struct net *net, int sync_state)
1316 struct netns_ipvs *ipvs = net_ipvs(net);
1317 struct net_device *dev;
1320 if (sync_state == IP_VS_STATE_MASTER) {
1321 dev = __dev_get_by_name(net, ipvs->master_mcast_ifn);
1325 num = (dev->mtu - sizeof(struct iphdr) -
1326 sizeof(struct udphdr) -
1327 SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE;
1328 ipvs->send_mesg_maxlen = SYNC_MESG_HEADER_LEN +
1329 SIMPLE_CONN_SIZE * min(num, MAX_CONNS_PER_SYNCBUFF);
1330 IP_VS_DBG(7, "setting the maximum length of sync sending "
1331 "message %d.\n", ipvs->send_mesg_maxlen);
1332 } else if (sync_state == IP_VS_STATE_BACKUP) {
1333 dev = __dev_get_by_name(net, ipvs->backup_mcast_ifn);
1337 ipvs->recv_mesg_maxlen = dev->mtu -
1338 sizeof(struct iphdr) - sizeof(struct udphdr);
1339 IP_VS_DBG(7, "setting the maximum length of sync receiving "
1340 "message %d.\n", ipvs->recv_mesg_maxlen);
1348 * Join a multicast group.
1349 * the group is specified by a class D multicast address 224.0.0.0/8
1350 * in the in_addr structure passed in as a parameter.
1353 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
1355 struct net *net = sock_net(sk);
1356 struct ip_mreqn mreq;
1357 struct net_device *dev;
1360 memset(&mreq, 0, sizeof(mreq));
1361 memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
1363 dev = __dev_get_by_name(net, ifname);
1366 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1369 mreq.imr_ifindex = dev->ifindex;
1372 ret = ip_mc_join_group(sk, &mreq);
1379 static int bind_mcastif_addr(struct socket *sock, char *ifname)
1381 struct net *net = sock_net(sock->sk);
1382 struct net_device *dev;
1384 struct sockaddr_in sin;
1386 dev = __dev_get_by_name(net, ifname);
1390 addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
1392 pr_err("You probably need to specify IP address on "
1393 "multicast interface.\n");
1395 IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
1398 /* Now bind the socket with the address of multicast interface */
1399 sin.sin_family = AF_INET;
1400 sin.sin_addr.s_addr = addr;
1403 return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
1407 * Set up sending multicast socket over UDP
1409 static struct socket *make_send_sock(struct net *net, int id)
1411 struct netns_ipvs *ipvs = net_ipvs(net);
1412 /* multicast addr */
1413 struct sockaddr_in mcast_addr = {
1414 .sin_family = AF_INET,
1415 .sin_port = cpu_to_be16(IP_VS_SYNC_PORT + id),
1416 .sin_addr.s_addr = cpu_to_be32(IP_VS_SYNC_GROUP),
1418 struct socket *sock;
1421 /* First create a socket move it to right name space later */
1422 result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1424 pr_err("Error during creation of socket; terminating\n");
1425 return ERR_PTR(result);
1428 * Kernel sockets that are a part of a namespace, should not
1429 * hold a reference to a namespace in order to allow to stop it.
1430 * After sk_change_net should be released using sk_release_kernel.
1432 sk_change_net(sock->sk, net);
1433 result = set_mcast_if(sock->sk, ipvs->master_mcast_ifn);
1435 pr_err("Error setting outbound mcast interface\n");
1439 set_mcast_loop(sock->sk, 0);
1440 set_mcast_ttl(sock->sk, 1);
1441 result = sysctl_sync_sock_size(ipvs);
1443 set_sock_size(sock->sk, 1, result);
1445 result = bind_mcastif_addr(sock, ipvs->master_mcast_ifn);
1447 pr_err("Error binding address of the mcast interface\n");
1451 result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr,
1452 sizeof(struct sockaddr), 0);
1454 pr_err("Error connecting to the multicast addr\n");
1461 sk_release_kernel(sock->sk);
1462 return ERR_PTR(result);
1467 * Set up receiving multicast socket over UDP
1469 static struct socket *make_receive_sock(struct net *net, int id)
1471 struct netns_ipvs *ipvs = net_ipvs(net);
1472 /* multicast addr */
1473 struct sockaddr_in mcast_addr = {
1474 .sin_family = AF_INET,
1475 .sin_port = cpu_to_be16(IP_VS_SYNC_PORT + id),
1476 .sin_addr.s_addr = cpu_to_be32(IP_VS_SYNC_GROUP),
1478 struct socket *sock;
1481 /* First create a socket */
1482 result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1484 pr_err("Error during creation of socket; terminating\n");
1485 return ERR_PTR(result);
1488 * Kernel sockets that are a part of a namespace, should not
1489 * hold a reference to a namespace in order to allow to stop it.
1490 * After sk_change_net should be released using sk_release_kernel.
1492 sk_change_net(sock->sk, net);
1493 /* it is equivalent to the REUSEADDR option in user-space */
1494 sock->sk->sk_reuse = SK_CAN_REUSE;
1495 result = sysctl_sync_sock_size(ipvs);
1497 set_sock_size(sock->sk, 0, result);
1499 result = sock->ops->bind(sock, (struct sockaddr *) &mcast_addr,
1500 sizeof(struct sockaddr));
1502 pr_err("Error binding to the multicast addr\n");
1506 /* join the multicast group */
1507 result = join_mcast_group(sock->sk,
1508 (struct in_addr *) &mcast_addr.sin_addr,
1509 ipvs->backup_mcast_ifn);
1511 pr_err("Error joining to the multicast group\n");
1518 sk_release_kernel(sock->sk);
1519 return ERR_PTR(result);
1524 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
1526 struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
1531 iov.iov_base = (void *)buffer;
1532 iov.iov_len = length;
1534 len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
1541 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
1546 msize = ntohs(msg->size);
1548 ret = ip_vs_send_async(sock, (char *)msg, msize);
1549 if (ret >= 0 || ret == -EAGAIN)
1551 pr_err("ip_vs_send_async error %d\n", ret);
1556 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
1558 struct msghdr msg = {NULL,};
1564 /* Receive a packet */
1565 iov.iov_base = buffer;
1566 iov.iov_len = (size_t)buflen;
1568 len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, MSG_DONTWAIT);
1577 /* Wakeup the master thread for sending */
1578 static void master_wakeup_work_handler(struct work_struct *work)
1580 struct ipvs_master_sync_state *ms =
1581 container_of(work, struct ipvs_master_sync_state,
1582 master_wakeup_work.work);
1583 struct netns_ipvs *ipvs = ms->ipvs;
1585 spin_lock_bh(&ipvs->sync_lock);
1586 if (ms->sync_queue_len &&
1587 ms->sync_queue_delay < IPVS_SYNC_WAKEUP_RATE) {
1588 ms->sync_queue_delay = IPVS_SYNC_WAKEUP_RATE;
1589 wake_up_process(ms->master_thread);
1591 spin_unlock_bh(&ipvs->sync_lock);
1594 /* Get next buffer to send */
1595 static inline struct ip_vs_sync_buff *
1596 next_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
1598 struct ip_vs_sync_buff *sb;
1600 sb = sb_dequeue(ipvs, ms);
1603 /* Do not delay entries in buffer for more than 2 seconds */
1604 return get_curr_sync_buff(ipvs, ms, IPVS_SYNC_FLUSH_TIME);
1607 static int sync_thread_master(void *data)
1609 struct ip_vs_sync_thread_data *tinfo = data;
1610 struct netns_ipvs *ipvs = net_ipvs(tinfo->net);
1611 struct ipvs_master_sync_state *ms = &ipvs->ms[tinfo->id];
1612 struct sock *sk = tinfo->sock->sk;
1613 struct ip_vs_sync_buff *sb;
1615 pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
1616 "syncid = %d, id = %d\n",
1617 ipvs->master_mcast_ifn, ipvs->master_syncid, tinfo->id);
1620 sb = next_sync_buff(ipvs, ms);
1621 if (unlikely(kthread_should_stop()))
1624 schedule_timeout(IPVS_SYNC_CHECK_PERIOD);
1627 while (ip_vs_send_sync_msg(tinfo->sock, sb->mesg) < 0) {
1630 __wait_event_interruptible(*sk_sleep(sk),
1631 sock_writeable(sk) ||
1632 kthread_should_stop(),
1634 if (unlikely(kthread_should_stop()))
1637 ip_vs_sync_buff_release(sb);
1641 __set_current_state(TASK_RUNNING);
1643 ip_vs_sync_buff_release(sb);
1645 /* clean up the sync_buff queue */
1646 while ((sb = sb_dequeue(ipvs, ms)))
1647 ip_vs_sync_buff_release(sb);
1648 __set_current_state(TASK_RUNNING);
1650 /* clean up the current sync_buff */
1651 sb = get_curr_sync_buff(ipvs, ms, 0);
1653 ip_vs_sync_buff_release(sb);
1655 /* release the sending multicast socket */
1656 sk_release_kernel(tinfo->sock->sk);
1663 static int sync_thread_backup(void *data)
1665 struct ip_vs_sync_thread_data *tinfo = data;
1666 struct netns_ipvs *ipvs = net_ipvs(tinfo->net);
1669 pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
1670 "syncid = %d, id = %d\n",
1671 ipvs->backup_mcast_ifn, ipvs->backup_syncid, tinfo->id);
1673 while (!kthread_should_stop()) {
1674 wait_event_interruptible(*sk_sleep(tinfo->sock->sk),
1675 !skb_queue_empty(&tinfo->sock->sk->sk_receive_queue)
1676 || kthread_should_stop());
1678 /* do we have data now? */
1679 while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) {
1680 len = ip_vs_receive(tinfo->sock, tinfo->buf,
1681 ipvs->recv_mesg_maxlen);
1684 pr_err("receiving message error\n");
1688 ip_vs_process_message(tinfo->net, tinfo->buf, len);
1692 /* release the sending multicast socket */
1693 sk_release_kernel(tinfo->sock->sk);
1701 int start_sync_thread(struct net *net, int state, char *mcast_ifn, __u8 syncid)
1703 struct ip_vs_sync_thread_data *tinfo;
1704 struct task_struct **array = NULL, *task;
1705 struct socket *sock;
1706 struct netns_ipvs *ipvs = net_ipvs(net);
1708 int (*threadfn)(void *data);
1710 int result = -ENOMEM;
1712 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1713 IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n",
1714 sizeof(struct ip_vs_sync_conn_v0));
1716 if (!ipvs->sync_state) {
1717 count = clamp(sysctl_sync_ports(ipvs), 1, IPVS_SYNC_PORTS_MAX);
1718 ipvs->threads_mask = count - 1;
1720 count = ipvs->threads_mask + 1;
1722 if (state == IP_VS_STATE_MASTER) {
1726 strlcpy(ipvs->master_mcast_ifn, mcast_ifn,
1727 sizeof(ipvs->master_mcast_ifn));
1728 ipvs->master_syncid = syncid;
1729 name = "ipvs-m:%d:%d";
1730 threadfn = sync_thread_master;
1731 } else if (state == IP_VS_STATE_BACKUP) {
1732 if (ipvs->backup_threads)
1735 strlcpy(ipvs->backup_mcast_ifn, mcast_ifn,
1736 sizeof(ipvs->backup_mcast_ifn));
1737 ipvs->backup_syncid = syncid;
1738 name = "ipvs-b:%d:%d";
1739 threadfn = sync_thread_backup;
1744 if (state == IP_VS_STATE_MASTER) {
1745 struct ipvs_master_sync_state *ms;
1747 ipvs->ms = kzalloc(count * sizeof(ipvs->ms[0]), GFP_KERNEL);
1751 for (id = 0; id < count; id++, ms++) {
1752 INIT_LIST_HEAD(&ms->sync_queue);
1753 ms->sync_queue_len = 0;
1754 ms->sync_queue_delay = 0;
1755 INIT_DELAYED_WORK(&ms->master_wakeup_work,
1756 master_wakeup_work_handler);
1760 array = kzalloc(count * sizeof(struct task_struct *),
1765 set_sync_mesg_maxlen(net, state);
1768 for (id = 0; id < count; id++) {
1769 if (state == IP_VS_STATE_MASTER)
1770 sock = make_send_sock(net, id);
1772 sock = make_receive_sock(net, id);
1774 result = PTR_ERR(sock);
1777 tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL);
1782 if (state == IP_VS_STATE_BACKUP) {
1783 tinfo->buf = kmalloc(ipvs->recv_mesg_maxlen,
1792 task = kthread_run(threadfn, tinfo, name, ipvs->gen, id);
1794 result = PTR_ERR(task);
1798 if (state == IP_VS_STATE_MASTER)
1799 ipvs->ms[id].master_thread = task;
1804 /* mark as active */
1806 if (state == IP_VS_STATE_BACKUP)
1807 ipvs->backup_threads = array;
1808 spin_lock_bh(&ipvs->sync_buff_lock);
1809 ipvs->sync_state |= state;
1810 spin_unlock_bh(&ipvs->sync_buff_lock);
1812 /* increase the module use count */
1813 ip_vs_use_count_inc();
1818 sk_release_kernel(sock->sk);
1822 sk_release_kernel(tinfo->sock->sk);
1827 while (count-- > 0) {
1828 if (state == IP_VS_STATE_MASTER)
1829 kthread_stop(ipvs->ms[count].master_thread);
1831 kthread_stop(array[count]);
1836 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
1844 int stop_sync_thread(struct net *net, int state)
1846 struct netns_ipvs *ipvs = net_ipvs(net);
1847 struct task_struct **array;
1851 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1853 if (state == IP_VS_STATE_MASTER) {
1858 * The lock synchronizes with sb_queue_tail(), so that we don't
1859 * add sync buffers to the queue, when we are already in
1860 * progress of stopping the master sync daemon.
1863 spin_lock_bh(&ipvs->sync_buff_lock);
1864 spin_lock(&ipvs->sync_lock);
1865 ipvs->sync_state &= ~IP_VS_STATE_MASTER;
1866 spin_unlock(&ipvs->sync_lock);
1867 spin_unlock_bh(&ipvs->sync_buff_lock);
1870 for (id = ipvs->threads_mask; id >= 0; id--) {
1871 struct ipvs_master_sync_state *ms = &ipvs->ms[id];
1874 pr_info("stopping master sync thread %d ...\n",
1875 task_pid_nr(ms->master_thread));
1876 cancel_delayed_work_sync(&ms->master_wakeup_work);
1877 ret = kthread_stop(ms->master_thread);
1883 } else if (state == IP_VS_STATE_BACKUP) {
1884 if (!ipvs->backup_threads)
1887 ipvs->sync_state &= ~IP_VS_STATE_BACKUP;
1888 array = ipvs->backup_threads;
1890 for (id = ipvs->threads_mask; id >= 0; id--) {
1893 pr_info("stopping backup sync thread %d ...\n",
1894 task_pid_nr(array[id]));
1895 ret = kthread_stop(array[id]);
1900 ipvs->backup_threads = NULL;
1903 /* decrease the module use count */
1904 ip_vs_use_count_dec();
1910 * Initialize data struct for each netns
1912 int __net_init ip_vs_sync_net_init(struct net *net)
1914 struct netns_ipvs *ipvs = net_ipvs(net);
1916 __mutex_init(&ipvs->sync_mutex, "ipvs->sync_mutex", &__ipvs_sync_key);
1917 spin_lock_init(&ipvs->sync_lock);
1918 spin_lock_init(&ipvs->sync_buff_lock);
1922 void ip_vs_sync_net_cleanup(struct net *net)
1925 struct netns_ipvs *ipvs = net_ipvs(net);
1927 mutex_lock(&ipvs->sync_mutex);
1928 retc = stop_sync_thread(net, IP_VS_STATE_MASTER);
1929 if (retc && retc != -ESRCH)
1930 pr_err("Failed to stop Master Daemon\n");
1932 retc = stop_sync_thread(net, IP_VS_STATE_BACKUP);
1933 if (retc && retc != -ESRCH)
1934 pr_err("Failed to stop Backup Daemon\n");
1935 mutex_unlock(&ipvs->sync_mutex);