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 ho->init_seq = get_unaligned_be32(&no->init_seq);
287 ho->delta = get_unaligned_be32(&no->delta);
288 ho->previous_delta = get_unaligned_be32(&no->previous_delta);
292 * Copy of struct ip_vs_seq
293 * From Aligned host order to unaligned network order
295 static void hton_seq(struct ip_vs_seq *ho, struct ip_vs_seq *no)
297 put_unaligned_be32(ho->init_seq, &no->init_seq);
298 put_unaligned_be32(ho->delta, &no->delta);
299 put_unaligned_be32(ho->previous_delta, &no->previous_delta);
302 static inline struct ip_vs_sync_buff *
303 sb_dequeue(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
305 struct ip_vs_sync_buff *sb;
307 spin_lock_bh(&ipvs->sync_lock);
308 if (list_empty(&ms->sync_queue)) {
310 __set_current_state(TASK_INTERRUPTIBLE);
312 sb = list_entry(ms->sync_queue.next, struct ip_vs_sync_buff,
315 ms->sync_queue_len--;
316 if (!ms->sync_queue_len)
317 ms->sync_queue_delay = 0;
319 spin_unlock_bh(&ipvs->sync_lock);
325 * Create a new sync buffer for Version 1 proto.
327 static inline struct ip_vs_sync_buff *
328 ip_vs_sync_buff_create(struct netns_ipvs *ipvs, unsigned int len)
330 struct ip_vs_sync_buff *sb;
332 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
335 len = max_t(unsigned int, len + sizeof(struct ip_vs_sync_mesg),
336 ipvs->mcfg.sync_maxlen);
337 sb->mesg = kmalloc(len, GFP_ATOMIC);
342 sb->mesg->reserved = 0; /* old nr_conns i.e. must be zero now */
343 sb->mesg->version = SYNC_PROTO_VER;
344 sb->mesg->syncid = ipvs->mcfg.syncid;
345 sb->mesg->size = htons(sizeof(struct ip_vs_sync_mesg));
346 sb->mesg->nr_conns = 0;
348 sb->head = (unsigned char *)sb->mesg + sizeof(struct ip_vs_sync_mesg);
349 sb->end = (unsigned char *)sb->mesg + len;
351 sb->firstuse = jiffies;
355 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
361 static inline void sb_queue_tail(struct netns_ipvs *ipvs,
362 struct ipvs_master_sync_state *ms)
364 struct ip_vs_sync_buff *sb = ms->sync_buff;
366 spin_lock(&ipvs->sync_lock);
367 if (ipvs->sync_state & IP_VS_STATE_MASTER &&
368 ms->sync_queue_len < sysctl_sync_qlen_max(ipvs)) {
369 if (!ms->sync_queue_len)
370 schedule_delayed_work(&ms->master_wakeup_work,
371 max(IPVS_SYNC_SEND_DELAY, 1));
372 ms->sync_queue_len++;
373 list_add_tail(&sb->list, &ms->sync_queue);
374 if ((++ms->sync_queue_delay) == IPVS_SYNC_WAKEUP_RATE)
375 wake_up_process(ms->master_thread);
377 ip_vs_sync_buff_release(sb);
378 spin_unlock(&ipvs->sync_lock);
382 * Get the current sync buffer if it has been created for more
383 * than the specified time or the specified time is zero.
385 static inline struct ip_vs_sync_buff *
386 get_curr_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms,
389 struct ip_vs_sync_buff *sb;
391 spin_lock_bh(&ipvs->sync_buff_lock);
393 if (sb && time_after_eq(jiffies - sb->firstuse, time)) {
394 ms->sync_buff = NULL;
395 __set_current_state(TASK_RUNNING);
398 spin_unlock_bh(&ipvs->sync_buff_lock);
403 select_master_thread_id(struct netns_ipvs *ipvs, struct ip_vs_conn *cp)
405 return ((long) cp >> (1 + ilog2(sizeof(*cp)))) & ipvs->threads_mask;
409 * Create a new sync buffer for Version 0 proto.
411 static inline struct ip_vs_sync_buff *
412 ip_vs_sync_buff_create_v0(struct netns_ipvs *ipvs, unsigned int len)
414 struct ip_vs_sync_buff *sb;
415 struct ip_vs_sync_mesg_v0 *mesg;
417 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
420 len = max_t(unsigned int, len + sizeof(struct ip_vs_sync_mesg_v0),
421 ipvs->mcfg.sync_maxlen);
422 sb->mesg = kmalloc(len, GFP_ATOMIC);
427 mesg = (struct ip_vs_sync_mesg_v0 *)sb->mesg;
429 mesg->syncid = ipvs->mcfg.syncid;
430 mesg->size = htons(sizeof(struct ip_vs_sync_mesg_v0));
431 sb->head = (unsigned char *)mesg + sizeof(struct ip_vs_sync_mesg_v0);
432 sb->end = (unsigned char *)mesg + len;
433 sb->firstuse = jiffies;
437 /* Check if connection is controlled by persistence */
438 static inline bool in_persistence(struct ip_vs_conn *cp)
440 for (cp = cp->control; cp; cp = cp->control) {
441 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
447 /* Check if conn should be synced.
448 * pkts: conn packets, use sysctl_sync_threshold to avoid packet check
449 * - (1) sync_refresh_period: reduce sync rate. Additionally, retry
450 * sync_retries times with period of sync_refresh_period/8
451 * - (2) if both sync_refresh_period and sync_period are 0 send sync only
452 * for state changes or only once when pkts matches sync_threshold
453 * - (3) templates: rate can be reduced only with sync_refresh_period or
456 static int ip_vs_sync_conn_needed(struct netns_ipvs *ipvs,
457 struct ip_vs_conn *cp, int pkts)
459 unsigned long orig = ACCESS_ONCE(cp->sync_endtime);
460 unsigned long now = jiffies;
461 unsigned long n = (now + cp->timeout) & ~3UL;
462 unsigned int sync_refresh_period;
466 /* Check if we sync in current state */
467 if (unlikely(cp->flags & IP_VS_CONN_F_TEMPLATE))
469 else if (unlikely(sysctl_sync_persist_mode(ipvs) && in_persistence(cp)))
471 else if (likely(cp->protocol == IPPROTO_TCP)) {
472 if (!((1 << cp->state) &
473 ((1 << IP_VS_TCP_S_ESTABLISHED) |
474 (1 << IP_VS_TCP_S_FIN_WAIT) |
475 (1 << IP_VS_TCP_S_CLOSE) |
476 (1 << IP_VS_TCP_S_CLOSE_WAIT) |
477 (1 << IP_VS_TCP_S_TIME_WAIT))))
479 force = cp->state != cp->old_state;
480 if (force && cp->state != IP_VS_TCP_S_ESTABLISHED)
482 } else if (unlikely(cp->protocol == IPPROTO_SCTP)) {
483 if (!((1 << cp->state) &
484 ((1 << IP_VS_SCTP_S_ESTABLISHED) |
485 (1 << IP_VS_SCTP_S_SHUTDOWN_SENT) |
486 (1 << IP_VS_SCTP_S_SHUTDOWN_RECEIVED) |
487 (1 << IP_VS_SCTP_S_SHUTDOWN_ACK_SENT) |
488 (1 << IP_VS_SCTP_S_CLOSED))))
490 force = cp->state != cp->old_state;
491 if (force && cp->state != IP_VS_SCTP_S_ESTABLISHED)
494 /* UDP or another protocol with single state */
498 sync_refresh_period = sysctl_sync_refresh_period(ipvs);
499 if (sync_refresh_period > 0) {
500 long diff = n - orig;
501 long min_diff = max(cp->timeout >> 1, 10UL * HZ);
503 /* Avoid sync if difference is below sync_refresh_period
504 * and below the half timeout.
506 if (abs(diff) < min_t(long, sync_refresh_period, min_diff)) {
507 int retries = orig & 3;
509 if (retries >= sysctl_sync_retries(ipvs))
511 if (time_before(now, orig - cp->timeout +
512 (sync_refresh_period >> 3)))
517 sync_period = sysctl_sync_period(ipvs);
518 if (sync_period > 0) {
519 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE) &&
520 pkts % sync_period != sysctl_sync_threshold(ipvs))
522 } else if (sync_refresh_period <= 0 &&
523 pkts != sysctl_sync_threshold(ipvs))
527 cp->old_state = cp->state;
528 n = cmpxchg(&cp->sync_endtime, orig, n);
529 return n == orig || force;
533 * Version 0 , could be switched in by sys_ctl.
534 * Add an ip_vs_conn information into the current sync_buff.
536 static void ip_vs_sync_conn_v0(struct netns_ipvs *ipvs, struct ip_vs_conn *cp,
539 struct ip_vs_sync_mesg_v0 *m;
540 struct ip_vs_sync_conn_v0 *s;
541 struct ip_vs_sync_buff *buff;
542 struct ipvs_master_sync_state *ms;
546 if (unlikely(cp->af != AF_INET))
548 /* Do not sync ONE PACKET */
549 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
552 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
555 spin_lock_bh(&ipvs->sync_buff_lock);
556 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
557 spin_unlock_bh(&ipvs->sync_buff_lock);
561 id = select_master_thread_id(ipvs, cp);
563 buff = ms->sync_buff;
564 len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
567 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
568 /* Send buffer if it is for v1 */
569 if (buff->head + len > buff->end || !m->nr_conns) {
570 sb_queue_tail(ipvs, ms);
571 ms->sync_buff = NULL;
576 buff = ip_vs_sync_buff_create_v0(ipvs, len);
578 spin_unlock_bh(&ipvs->sync_buff_lock);
579 pr_err("ip_vs_sync_buff_create failed.\n");
582 ms->sync_buff = buff;
585 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
586 s = (struct ip_vs_sync_conn_v0 *) buff->head;
590 s->protocol = cp->protocol;
591 s->cport = cp->cport;
592 s->vport = cp->vport;
593 s->dport = cp->dport;
594 s->caddr = cp->caddr.ip;
595 s->vaddr = cp->vaddr.ip;
596 s->daddr = cp->daddr.ip;
597 s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
598 s->state = htons(cp->state);
599 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
600 struct ip_vs_sync_conn_options *opt =
601 (struct ip_vs_sync_conn_options *)&s[1];
602 memcpy(opt, &cp->in_seq, sizeof(*opt));
606 m->size = htons(ntohs(m->size) + len);
608 spin_unlock_bh(&ipvs->sync_buff_lock);
610 /* synchronize its controller if it has */
613 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
614 pkts = atomic_add_return(1, &cp->in_pkts);
616 pkts = sysctl_sync_threshold(ipvs);
617 ip_vs_sync_conn(ipvs, cp, pkts);
622 * Add an ip_vs_conn information into the current sync_buff.
623 * Called by ip_vs_in.
624 * Sending Version 1 messages
626 void ip_vs_sync_conn(struct netns_ipvs *ipvs, struct ip_vs_conn *cp, int pkts)
628 struct ip_vs_sync_mesg *m;
629 union ip_vs_sync_conn *s;
630 struct ip_vs_sync_buff *buff;
631 struct ipvs_master_sync_state *ms;
634 unsigned int len, pe_name_len, pad;
636 /* Handle old version of the protocol */
637 if (sysctl_sync_ver(ipvs) == 0) {
638 ip_vs_sync_conn_v0(ipvs, cp, pkts);
641 /* Do not sync ONE PACKET */
642 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
645 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
650 if (cp->pe_data_len) {
651 if (!cp->pe_data || !cp->dest) {
652 IP_VS_ERR_RL("SYNC, connection pe_data invalid\n");
655 pe_name_len = strnlen(cp->pe->name, IP_VS_PENAME_MAXLEN);
658 spin_lock_bh(&ipvs->sync_buff_lock);
659 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
660 spin_unlock_bh(&ipvs->sync_buff_lock);
664 id = select_master_thread_id(ipvs, cp);
667 #ifdef CONFIG_IP_VS_IPV6
668 if (cp->af == AF_INET6)
669 len = sizeof(struct ip_vs_sync_v6);
672 len = sizeof(struct ip_vs_sync_v4);
674 if (cp->flags & IP_VS_CONN_F_SEQ_MASK)
675 len += sizeof(struct ip_vs_sync_conn_options) + 2;
678 len += cp->pe_data_len + 2; /* + Param hdr field */
680 len += pe_name_len + 2;
682 /* check if there is a space for this one */
684 buff = ms->sync_buff;
687 pad = (4 - (size_t) buff->head) & 3;
688 /* Send buffer if it is for v0 */
689 if (buff->head + len + pad > buff->end || m->reserved) {
690 sb_queue_tail(ipvs, ms);
691 ms->sync_buff = NULL;
698 buff = ip_vs_sync_buff_create(ipvs, len);
700 spin_unlock_bh(&ipvs->sync_buff_lock);
701 pr_err("ip_vs_sync_buff_create failed.\n");
704 ms->sync_buff = buff;
709 buff->head += pad + len;
710 m->size = htons(ntohs(m->size) + pad + len);
711 /* Add ev. padding from prev. sync_conn */
715 s = (union ip_vs_sync_conn *)p;
717 /* Set message type & copy members */
718 s->v4.type = (cp->af == AF_INET6 ? STYPE_F_INET6 : 0);
719 s->v4.ver_size = htons(len & SVER_MASK); /* Version 0 */
720 s->v4.flags = htonl(cp->flags & ~IP_VS_CONN_F_HASHED);
721 s->v4.state = htons(cp->state);
722 s->v4.protocol = cp->protocol;
723 s->v4.cport = cp->cport;
724 s->v4.vport = cp->vport;
725 s->v4.dport = cp->dport;
726 s->v4.fwmark = htonl(cp->fwmark);
727 s->v4.timeout = htonl(cp->timeout / HZ);
730 #ifdef CONFIG_IP_VS_IPV6
731 if (cp->af == AF_INET6) {
732 p += sizeof(struct ip_vs_sync_v6);
733 s->v6.caddr = cp->caddr.in6;
734 s->v6.vaddr = cp->vaddr.in6;
735 s->v6.daddr = cp->daddr.in6;
739 p += sizeof(struct ip_vs_sync_v4); /* options ptr */
740 s->v4.caddr = cp->caddr.ip;
741 s->v4.vaddr = cp->vaddr.ip;
742 s->v4.daddr = cp->daddr.ip;
744 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
745 *(p++) = IPVS_OPT_SEQ_DATA;
746 *(p++) = sizeof(struct ip_vs_sync_conn_options);
747 hton_seq((struct ip_vs_seq *)p, &cp->in_seq);
748 p += sizeof(struct ip_vs_seq);
749 hton_seq((struct ip_vs_seq *)p, &cp->out_seq);
750 p += sizeof(struct ip_vs_seq);
753 if (cp->pe_data_len && cp->pe_data) {
754 *(p++) = IPVS_OPT_PE_DATA;
755 *(p++) = cp->pe_data_len;
756 memcpy(p, cp->pe_data, cp->pe_data_len);
757 p += cp->pe_data_len;
760 *(p++) = IPVS_OPT_PE_NAME;
761 *(p++) = pe_name_len;
762 memcpy(p, cp->pe->name, pe_name_len);
767 spin_unlock_bh(&ipvs->sync_buff_lock);
770 /* synchronize its controller if it has */
774 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
775 pkts = atomic_add_return(1, &cp->in_pkts);
777 pkts = sysctl_sync_threshold(ipvs);
782 * fill_param used by version 1
785 ip_vs_conn_fill_param_sync(struct netns_ipvs *ipvs, int af, union ip_vs_sync_conn *sc,
786 struct ip_vs_conn_param *p,
787 __u8 *pe_data, unsigned int pe_data_len,
788 __u8 *pe_name, unsigned int pe_name_len)
790 #ifdef CONFIG_IP_VS_IPV6
792 ip_vs_conn_fill_param(ipvs, af, sc->v6.protocol,
793 (const union nf_inet_addr *)&sc->v6.caddr,
795 (const union nf_inet_addr *)&sc->v6.vaddr,
799 ip_vs_conn_fill_param(ipvs, af, sc->v4.protocol,
800 (const union nf_inet_addr *)&sc->v4.caddr,
802 (const union nf_inet_addr *)&sc->v4.vaddr,
807 char buff[IP_VS_PENAME_MAXLEN+1];
809 memcpy(buff, pe_name, pe_name_len);
811 p->pe = __ip_vs_pe_getbyname(buff);
813 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n",
818 IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n");
822 p->pe_data = kmemdup(pe_data, pe_data_len, GFP_ATOMIC);
824 module_put(p->pe->module);
827 p->pe_data_len = pe_data_len;
833 * Connection Add / Update.
834 * Common for version 0 and 1 reception of backup sync_conns.
838 static void ip_vs_proc_conn(struct netns_ipvs *ipvs, struct ip_vs_conn_param *param,
839 unsigned int flags, unsigned int state,
840 unsigned int protocol, unsigned int type,
841 const union nf_inet_addr *daddr, __be16 dport,
842 unsigned long timeout, __u32 fwmark,
843 struct ip_vs_sync_conn_options *opt)
845 struct ip_vs_dest *dest;
846 struct ip_vs_conn *cp;
848 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
849 cp = ip_vs_conn_in_get(param);
850 if (cp && ((cp->dport != dport) ||
851 !ip_vs_addr_equal(cp->daf, &cp->daddr, daddr))) {
852 if (!(flags & IP_VS_CONN_F_INACTIVE)) {
853 ip_vs_conn_expire_now(cp);
854 __ip_vs_conn_put(cp);
857 /* This is the expiration message for the
858 * connection that was already replaced, so we
861 __ip_vs_conn_put(cp);
862 kfree(param->pe_data);
867 cp = ip_vs_ct_in_get(param);
872 kfree(param->pe_data);
875 spin_lock_bh(&cp->lock);
876 if ((cp->flags ^ flags) & IP_VS_CONN_F_INACTIVE &&
877 !(flags & IP_VS_CONN_F_TEMPLATE) && dest) {
878 if (flags & IP_VS_CONN_F_INACTIVE) {
879 atomic_dec(&dest->activeconns);
880 atomic_inc(&dest->inactconns);
882 atomic_inc(&dest->activeconns);
883 atomic_dec(&dest->inactconns);
886 flags &= IP_VS_CONN_F_BACKUP_UPD_MASK;
887 flags |= cp->flags & ~IP_VS_CONN_F_BACKUP_UPD_MASK;
889 spin_unlock_bh(&cp->lock);
891 ip_vs_try_bind_dest(cp);
894 * Find the appropriate destination for the connection.
895 * If it is not found the connection will remain unbound
899 /* This function is only invoked by the synchronization
900 * code. We do not currently support heterogeneous pools
901 * with synchronization, so we can make the assumption that
902 * the svc_af is the same as the dest_af
904 dest = ip_vs_find_dest(ipvs, type, type, daddr, dport,
905 param->vaddr, param->vport, protocol,
908 cp = ip_vs_conn_new(param, type, daddr, dport, flags, dest,
912 kfree(param->pe_data);
913 IP_VS_DBG(2, "BACKUP, add new conn. failed\n");
916 if (!(flags & IP_VS_CONN_F_TEMPLATE))
917 kfree(param->pe_data);
921 memcpy(&cp->in_seq, opt, sizeof(*opt));
922 atomic_set(&cp->in_pkts, sysctl_sync_threshold(ipvs));
924 cp->old_state = cp->state;
926 * For Ver 0 messages style
927 * - Not possible to recover the right timeout for templates
928 * - can not find the right fwmark
929 * virtual service. If needed, we can do it for
930 * non-fwmark persistent services.
931 * Ver 1 messages style.
935 if (timeout > MAX_SCHEDULE_TIMEOUT / HZ)
936 timeout = MAX_SCHEDULE_TIMEOUT / HZ;
937 cp->timeout = timeout*HZ;
939 struct ip_vs_proto_data *pd;
941 pd = ip_vs_proto_data_get(ipvs, protocol);
942 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pd && pd->timeout_table)
943 cp->timeout = pd->timeout_table[state];
945 cp->timeout = (3*60*HZ);
951 * Process received multicast message for Version 0
953 static void ip_vs_process_message_v0(struct netns_ipvs *ipvs, const char *buffer,
956 struct ip_vs_sync_mesg_v0 *m = (struct ip_vs_sync_mesg_v0 *)buffer;
957 struct ip_vs_sync_conn_v0 *s;
958 struct ip_vs_sync_conn_options *opt;
959 struct ip_vs_protocol *pp;
960 struct ip_vs_conn_param param;
964 p = (char *)buffer + sizeof(struct ip_vs_sync_mesg_v0);
965 for (i=0; i<m->nr_conns; i++) {
966 unsigned int flags, state;
968 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
969 IP_VS_ERR_RL("BACKUP v0, bogus conn\n");
972 s = (struct ip_vs_sync_conn_v0 *) p;
973 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
974 flags &= ~IP_VS_CONN_F_HASHED;
975 if (flags & IP_VS_CONN_F_SEQ_MASK) {
976 opt = (struct ip_vs_sync_conn_options *)&s[1];
978 if (p > buffer+buflen) {
979 IP_VS_ERR_RL("BACKUP v0, Dropping buffer bogus conn options\n");
984 p += SIMPLE_CONN_SIZE;
987 state = ntohs(s->state);
988 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
989 pp = ip_vs_proto_get(s->protocol);
991 IP_VS_DBG(2, "BACKUP v0, Unsupported protocol %u\n",
995 if (state >= pp->num_states) {
996 IP_VS_DBG(2, "BACKUP v0, Invalid %s state %u\n",
1001 /* protocol in templates is not used for state/timeout */
1003 IP_VS_DBG(2, "BACKUP v0, Invalid template state %u\n",
1009 ip_vs_conn_fill_param(ipvs, AF_INET, s->protocol,
1010 (const union nf_inet_addr *)&s->caddr,
1012 (const union nf_inet_addr *)&s->vaddr,
1015 /* Send timeout as Zero */
1016 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->protocol, AF_INET,
1017 (union nf_inet_addr *)&s->daddr, s->dport,
1025 static inline int ip_vs_proc_seqopt(__u8 *p, unsigned int plen,
1027 struct ip_vs_sync_conn_options *opt)
1029 struct ip_vs_sync_conn_options *topt;
1031 topt = (struct ip_vs_sync_conn_options *)p;
1033 if (plen != sizeof(struct ip_vs_sync_conn_options)) {
1034 IP_VS_DBG(2, "BACKUP, bogus conn options length\n");
1037 if (*opt_flags & IPVS_OPT_F_SEQ_DATA) {
1038 IP_VS_DBG(2, "BACKUP, conn options found twice\n");
1041 ntoh_seq(&topt->in_seq, &opt->in_seq);
1042 ntoh_seq(&topt->out_seq, &opt->out_seq);
1043 *opt_flags |= IPVS_OPT_F_SEQ_DATA;
1047 static int ip_vs_proc_str(__u8 *p, unsigned int plen, unsigned int *data_len,
1048 __u8 **data, unsigned int maxlen,
1049 __u32 *opt_flags, __u32 flag)
1051 if (plen > maxlen) {
1052 IP_VS_DBG(2, "BACKUP, bogus par.data len > %d\n", maxlen);
1055 if (*opt_flags & flag) {
1056 IP_VS_DBG(2, "BACKUP, Par.data found twice 0x%x\n", flag);
1065 * Process a Version 1 sync. connection
1067 static inline int ip_vs_proc_sync_conn(struct netns_ipvs *ipvs, __u8 *p, __u8 *msg_end)
1069 struct ip_vs_sync_conn_options opt;
1070 union ip_vs_sync_conn *s;
1071 struct ip_vs_protocol *pp;
1072 struct ip_vs_conn_param param;
1074 unsigned int af, state, pe_data_len=0, pe_name_len=0;
1075 __u8 *pe_data=NULL, *pe_name=NULL;
1079 s = (union ip_vs_sync_conn *) p;
1081 if (s->v6.type & STYPE_F_INET6) {
1082 #ifdef CONFIG_IP_VS_IPV6
1084 p += sizeof(struct ip_vs_sync_v6);
1086 IP_VS_DBG(3,"BACKUP, IPv6 msg received, and IPVS is not compiled for IPv6\n");
1090 } else if (!s->v4.type) {
1092 p += sizeof(struct ip_vs_sync_v4);
1099 /* Process optional params check Type & Len. */
1100 while (p < msg_end) {
1109 if (!plen || ((p + plen) > msg_end))
1111 /* Handle seq option p = param data */
1112 switch (ptype & ~IPVS_OPT_F_PARAM) {
1113 case IPVS_OPT_SEQ_DATA:
1114 if (ip_vs_proc_seqopt(p, plen, &opt_flags, &opt))
1118 case IPVS_OPT_PE_DATA:
1119 if (ip_vs_proc_str(p, plen, &pe_data_len, &pe_data,
1120 IP_VS_PEDATA_MAXLEN, &opt_flags,
1121 IPVS_OPT_F_PE_DATA))
1125 case IPVS_OPT_PE_NAME:
1126 if (ip_vs_proc_str(p, plen,&pe_name_len, &pe_name,
1127 IP_VS_PENAME_MAXLEN, &opt_flags,
1128 IPVS_OPT_F_PE_NAME))
1133 /* Param data mandatory ? */
1134 if (!(ptype & IPVS_OPT_F_PARAM)) {
1135 IP_VS_DBG(3, "BACKUP, Unknown mandatory param %d found\n",
1136 ptype & ~IPVS_OPT_F_PARAM);
1141 p += plen; /* Next option */
1144 /* Get flags and Mask off unsupported */
1145 flags = ntohl(s->v4.flags) & IP_VS_CONN_F_BACKUP_MASK;
1146 flags |= IP_VS_CONN_F_SYNC;
1147 state = ntohs(s->v4.state);
1149 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
1150 pp = ip_vs_proto_get(s->v4.protocol);
1152 IP_VS_DBG(3,"BACKUP, Unsupported protocol %u\n",
1157 if (state >= pp->num_states) {
1158 IP_VS_DBG(3, "BACKUP, Invalid %s state %u\n",
1164 /* protocol in templates is not used for state/timeout */
1166 IP_VS_DBG(3, "BACKUP, Invalid template state %u\n",
1171 if (ip_vs_conn_fill_param_sync(ipvs, af, s, ¶m, pe_data,
1172 pe_data_len, pe_name, pe_name_len)) {
1176 /* If only IPv4, just silent skip IPv6 */
1178 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->v4.protocol, af,
1179 (union nf_inet_addr *)&s->v4.daddr, s->v4.dport,
1180 ntohl(s->v4.timeout), ntohl(s->v4.fwmark),
1181 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1183 #ifdef CONFIG_IP_VS_IPV6
1185 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->v6.protocol, af,
1186 (union nf_inet_addr *)&s->v6.daddr, s->v6.dport,
1187 ntohl(s->v6.timeout), ntohl(s->v6.fwmark),
1188 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1191 ip_vs_pe_put(param.pe);
1195 IP_VS_DBG(2, "BACKUP, Single msg dropped err:%d\n", retc);
1200 * Process received multicast message and create the corresponding
1201 * ip_vs_conn entries.
1202 * Handles Version 0 & 1
1204 static void ip_vs_process_message(struct netns_ipvs *ipvs, __u8 *buffer,
1205 const size_t buflen)
1207 struct ip_vs_sync_mesg *m2 = (struct ip_vs_sync_mesg *)buffer;
1211 if (buflen < sizeof(struct ip_vs_sync_mesg_v0)) {
1212 IP_VS_DBG(2, "BACKUP, message header too short\n");
1216 if (buflen != ntohs(m2->size)) {
1217 IP_VS_DBG(2, "BACKUP, bogus message size\n");
1220 /* SyncID sanity check */
1221 if (ipvs->bcfg.syncid != 0 && m2->syncid != ipvs->bcfg.syncid) {
1222 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2->syncid);
1225 /* Handle version 1 message */
1226 if ((m2->version == SYNC_PROTO_VER) && (m2->reserved == 0)
1227 && (m2->spare == 0)) {
1229 msg_end = buffer + sizeof(struct ip_vs_sync_mesg);
1230 nr_conns = m2->nr_conns;
1232 for (i=0; i<nr_conns; i++) {
1233 union ip_vs_sync_conn *s;
1238 if (p + sizeof(s->v4) > buffer+buflen) {
1239 IP_VS_ERR_RL("BACKUP, Dropping buffer, to small\n");
1242 s = (union ip_vs_sync_conn *)p;
1243 size = ntohs(s->v4.ver_size) & SVER_MASK;
1245 /* Basic sanity checks */
1246 if (msg_end > buffer+buflen) {
1247 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n");
1250 if (ntohs(s->v4.ver_size) >> SVER_SHIFT) {
1251 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n",
1252 ntohs(s->v4.ver_size) >> SVER_SHIFT);
1255 /* Process a single sync_conn */
1256 retc = ip_vs_proc_sync_conn(ipvs, p, msg_end);
1258 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n",
1262 /* Make sure we have 32 bit alignment */
1263 msg_end = p + ((size + 3) & ~3);
1266 /* Old type of message */
1267 ip_vs_process_message_v0(ipvs, buffer, buflen);
1274 * Setup sndbuf (mode=1) or rcvbuf (mode=0)
1276 static void set_sock_size(struct sock *sk, int mode, int val)
1278 /* setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &val, sizeof(val)); */
1279 /* setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)); */
1282 val = clamp_t(int, val, (SOCK_MIN_SNDBUF + 1) / 2,
1284 sk->sk_sndbuf = val * 2;
1285 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1287 val = clamp_t(int, val, (SOCK_MIN_RCVBUF + 1) / 2,
1289 sk->sk_rcvbuf = val * 2;
1290 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1296 * Setup loopback of outgoing multicasts on a sending socket
1298 static void set_mcast_loop(struct sock *sk, u_char loop)
1300 struct inet_sock *inet = inet_sk(sk);
1302 /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
1304 inet->mc_loop = loop ? 1 : 0;
1305 #ifdef CONFIG_IP_VS_IPV6
1306 if (sk->sk_family == AF_INET6) {
1307 struct ipv6_pinfo *np = inet6_sk(sk);
1309 /* IPV6_MULTICAST_LOOP */
1310 np->mc_loop = loop ? 1 : 0;
1317 * Specify TTL for outgoing multicasts on a sending socket
1319 static void set_mcast_ttl(struct sock *sk, u_char ttl)
1321 struct inet_sock *inet = inet_sk(sk);
1323 /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
1326 #ifdef CONFIG_IP_VS_IPV6
1327 if (sk->sk_family == AF_INET6) {
1328 struct ipv6_pinfo *np = inet6_sk(sk);
1330 /* IPV6_MULTICAST_HOPS */
1331 np->mcast_hops = ttl;
1337 /* Control fragmentation of messages */
1338 static void set_mcast_pmtudisc(struct sock *sk, int val)
1340 struct inet_sock *inet = inet_sk(sk);
1342 /* setsockopt(sock, SOL_IP, IP_MTU_DISCOVER, &val, sizeof(val)); */
1344 inet->pmtudisc = val;
1345 #ifdef CONFIG_IP_VS_IPV6
1346 if (sk->sk_family == AF_INET6) {
1347 struct ipv6_pinfo *np = inet6_sk(sk);
1349 /* IPV6_MTU_DISCOVER */
1357 * Specifiy default interface for outgoing multicasts
1359 static int set_mcast_if(struct sock *sk, char *ifname)
1361 struct net_device *dev;
1362 struct inet_sock *inet = inet_sk(sk);
1363 struct net *net = sock_net(sk);
1365 dev = __dev_get_by_name(net, ifname);
1369 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1373 inet->mc_index = dev->ifindex;
1374 /* inet->mc_addr = 0; */
1375 #ifdef CONFIG_IP_VS_IPV6
1376 if (sk->sk_family == AF_INET6) {
1377 struct ipv6_pinfo *np = inet6_sk(sk);
1379 /* IPV6_MULTICAST_IF */
1380 np->mcast_oif = dev->ifindex;
1390 * Join a multicast group.
1391 * the group is specified by a class D multicast address 224.0.0.0/8
1392 * in the in_addr structure passed in as a parameter.
1395 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
1397 struct net *net = sock_net(sk);
1398 struct ip_mreqn mreq;
1399 struct net_device *dev;
1402 memset(&mreq, 0, sizeof(mreq));
1403 memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
1405 dev = __dev_get_by_name(net, ifname);
1408 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1411 mreq.imr_ifindex = dev->ifindex;
1414 ret = ip_mc_join_group(sk, &mreq);
1420 #ifdef CONFIG_IP_VS_IPV6
1421 static int join_mcast_group6(struct sock *sk, struct in6_addr *addr,
1424 struct net *net = sock_net(sk);
1425 struct net_device *dev;
1428 dev = __dev_get_by_name(net, ifname);
1431 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1435 ret = ipv6_sock_mc_join(sk, dev->ifindex, addr);
1442 static int bind_mcastif_addr(struct socket *sock, char *ifname)
1444 struct net *net = sock_net(sock->sk);
1445 struct net_device *dev;
1447 struct sockaddr_in sin;
1449 dev = __dev_get_by_name(net, ifname);
1453 addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
1455 pr_err("You probably need to specify IP address on "
1456 "multicast interface.\n");
1458 IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
1461 /* Now bind the socket with the address of multicast interface */
1462 sin.sin_family = AF_INET;
1463 sin.sin_addr.s_addr = addr;
1466 return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
1469 static void get_mcast_sockaddr(union ipvs_sockaddr *sa, int *salen,
1470 struct ipvs_sync_daemon_cfg *c, int id)
1472 if (AF_INET6 == c->mcast_af) {
1473 sa->in6 = (struct sockaddr_in6) {
1474 .sin6_family = AF_INET6,
1475 .sin6_port = htons(c->mcast_port + id),
1477 sa->in6.sin6_addr = c->mcast_group.in6;
1478 *salen = sizeof(sa->in6);
1480 sa->in = (struct sockaddr_in) {
1481 .sin_family = AF_INET,
1482 .sin_port = htons(c->mcast_port + id),
1484 sa->in.sin_addr = c->mcast_group.in;
1485 *salen = sizeof(sa->in);
1490 * Set up sending multicast socket over UDP
1492 static struct socket *make_send_sock(struct netns_ipvs *ipvs, int id)
1494 /* multicast addr */
1495 union ipvs_sockaddr mcast_addr;
1496 struct socket *sock;
1499 /* First create a socket */
1500 result = sock_create_kern(ipvs->net, ipvs->mcfg.mcast_af, SOCK_DGRAM,
1501 IPPROTO_UDP, &sock);
1503 pr_err("Error during creation of socket; terminating\n");
1504 return ERR_PTR(result);
1506 result = set_mcast_if(sock->sk, ipvs->mcfg.mcast_ifn);
1508 pr_err("Error setting outbound mcast interface\n");
1512 set_mcast_loop(sock->sk, 0);
1513 set_mcast_ttl(sock->sk, ipvs->mcfg.mcast_ttl);
1514 /* Allow fragmentation if MTU changes */
1515 set_mcast_pmtudisc(sock->sk, IP_PMTUDISC_DONT);
1516 result = sysctl_sync_sock_size(ipvs);
1518 set_sock_size(sock->sk, 1, result);
1520 if (AF_INET == ipvs->mcfg.mcast_af)
1521 result = bind_mcastif_addr(sock, ipvs->mcfg.mcast_ifn);
1525 pr_err("Error binding address of the mcast interface\n");
1529 get_mcast_sockaddr(&mcast_addr, &salen, &ipvs->mcfg, id);
1530 result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr,
1533 pr_err("Error connecting to the multicast addr\n");
1541 return ERR_PTR(result);
1546 * Set up receiving multicast socket over UDP
1548 static struct socket *make_receive_sock(struct netns_ipvs *ipvs, int id,
1551 /* multicast addr */
1552 union ipvs_sockaddr mcast_addr;
1553 struct socket *sock;
1556 /* First create a socket */
1557 result = sock_create_kern(ipvs->net, ipvs->bcfg.mcast_af, SOCK_DGRAM,
1558 IPPROTO_UDP, &sock);
1560 pr_err("Error during creation of socket; terminating\n");
1561 return ERR_PTR(result);
1563 /* it is equivalent to the REUSEADDR option in user-space */
1564 sock->sk->sk_reuse = SK_CAN_REUSE;
1565 result = sysctl_sync_sock_size(ipvs);
1567 set_sock_size(sock->sk, 0, result);
1569 get_mcast_sockaddr(&mcast_addr, &salen, &ipvs->bcfg, id);
1570 sock->sk->sk_bound_dev_if = ifindex;
1571 result = sock->ops->bind(sock, (struct sockaddr *)&mcast_addr, salen);
1573 pr_err("Error binding to the multicast addr\n");
1577 /* join the multicast group */
1578 #ifdef CONFIG_IP_VS_IPV6
1579 if (ipvs->bcfg.mcast_af == AF_INET6)
1580 result = join_mcast_group6(sock->sk, &mcast_addr.in6.sin6_addr,
1581 ipvs->bcfg.mcast_ifn);
1584 result = join_mcast_group(sock->sk, &mcast_addr.in.sin_addr,
1585 ipvs->bcfg.mcast_ifn);
1587 pr_err("Error joining to the multicast group\n");
1595 return ERR_PTR(result);
1600 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
1602 struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
1607 iov.iov_base = (void *)buffer;
1608 iov.iov_len = length;
1610 len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
1617 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
1622 msize = ntohs(msg->size);
1624 ret = ip_vs_send_async(sock, (char *)msg, msize);
1625 if (ret >= 0 || ret == -EAGAIN)
1627 pr_err("ip_vs_send_async error %d\n", ret);
1632 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
1634 struct msghdr msg = {NULL,};
1640 /* Receive a packet */
1641 iov.iov_base = buffer;
1642 iov.iov_len = (size_t)buflen;
1644 len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, MSG_DONTWAIT);
1653 /* Wakeup the master thread for sending */
1654 static void master_wakeup_work_handler(struct work_struct *work)
1656 struct ipvs_master_sync_state *ms =
1657 container_of(work, struct ipvs_master_sync_state,
1658 master_wakeup_work.work);
1659 struct netns_ipvs *ipvs = ms->ipvs;
1661 spin_lock_bh(&ipvs->sync_lock);
1662 if (ms->sync_queue_len &&
1663 ms->sync_queue_delay < IPVS_SYNC_WAKEUP_RATE) {
1664 ms->sync_queue_delay = IPVS_SYNC_WAKEUP_RATE;
1665 wake_up_process(ms->master_thread);
1667 spin_unlock_bh(&ipvs->sync_lock);
1670 /* Get next buffer to send */
1671 static inline struct ip_vs_sync_buff *
1672 next_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
1674 struct ip_vs_sync_buff *sb;
1676 sb = sb_dequeue(ipvs, ms);
1679 /* Do not delay entries in buffer for more than 2 seconds */
1680 return get_curr_sync_buff(ipvs, ms, IPVS_SYNC_FLUSH_TIME);
1683 static int sync_thread_master(void *data)
1685 struct ip_vs_sync_thread_data *tinfo = data;
1686 struct netns_ipvs *ipvs = tinfo->ipvs;
1687 struct ipvs_master_sync_state *ms = &ipvs->ms[tinfo->id];
1688 struct sock *sk = tinfo->sock->sk;
1689 struct ip_vs_sync_buff *sb;
1691 pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
1692 "syncid = %d, id = %d\n",
1693 ipvs->mcfg.mcast_ifn, ipvs->mcfg.syncid, tinfo->id);
1696 sb = next_sync_buff(ipvs, ms);
1697 if (unlikely(kthread_should_stop()))
1700 schedule_timeout(IPVS_SYNC_CHECK_PERIOD);
1703 while (ip_vs_send_sync_msg(tinfo->sock, sb->mesg) < 0) {
1704 /* (Ab)use interruptible sleep to avoid increasing
1707 __wait_event_interruptible(*sk_sleep(sk),
1708 sock_writeable(sk) ||
1709 kthread_should_stop());
1710 if (unlikely(kthread_should_stop()))
1713 ip_vs_sync_buff_release(sb);
1717 __set_current_state(TASK_RUNNING);
1719 ip_vs_sync_buff_release(sb);
1721 /* clean up the sync_buff queue */
1722 while ((sb = sb_dequeue(ipvs, ms)))
1723 ip_vs_sync_buff_release(sb);
1724 __set_current_state(TASK_RUNNING);
1726 /* clean up the current sync_buff */
1727 sb = get_curr_sync_buff(ipvs, ms, 0);
1729 ip_vs_sync_buff_release(sb);
1731 /* release the sending multicast socket */
1732 sock_release(tinfo->sock);
1739 static int sync_thread_backup(void *data)
1741 struct ip_vs_sync_thread_data *tinfo = data;
1742 struct netns_ipvs *ipvs = tinfo->ipvs;
1745 pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
1746 "syncid = %d, id = %d\n",
1747 ipvs->bcfg.mcast_ifn, ipvs->bcfg.syncid, tinfo->id);
1749 while (!kthread_should_stop()) {
1750 wait_event_interruptible(*sk_sleep(tinfo->sock->sk),
1751 !skb_queue_empty(&tinfo->sock->sk->sk_receive_queue)
1752 || kthread_should_stop());
1754 /* do we have data now? */
1755 while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) {
1756 len = ip_vs_receive(tinfo->sock, tinfo->buf,
1757 ipvs->bcfg.sync_maxlen);
1760 pr_err("receiving message error\n");
1764 ip_vs_process_message(ipvs, tinfo->buf, len);
1768 /* release the sending multicast socket */
1769 sock_release(tinfo->sock);
1777 int start_sync_thread(struct netns_ipvs *ipvs, struct ipvs_sync_daemon_cfg *c,
1780 struct ip_vs_sync_thread_data *tinfo;
1781 struct task_struct **array = NULL, *task;
1782 struct socket *sock;
1783 struct net_device *dev;
1785 int (*threadfn)(void *data);
1786 int id, count, hlen;
1787 int result = -ENOMEM;
1790 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1791 IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n",
1792 sizeof(struct ip_vs_sync_conn_v0));
1794 if (!ipvs->sync_state) {
1795 count = clamp(sysctl_sync_ports(ipvs), 1, IPVS_SYNC_PORTS_MAX);
1796 ipvs->threads_mask = count - 1;
1798 count = ipvs->threads_mask + 1;
1800 if (c->mcast_af == AF_UNSPEC) {
1801 c->mcast_af = AF_INET;
1802 c->mcast_group.ip = cpu_to_be32(IP_VS_SYNC_GROUP);
1805 c->mcast_port = IP_VS_SYNC_PORT;
1809 dev = __dev_get_by_name(ipvs->net, c->mcast_ifn);
1811 pr_err("Unknown mcast interface: %s\n", c->mcast_ifn);
1814 hlen = (AF_INET6 == c->mcast_af) ?
1815 sizeof(struct ipv6hdr) + sizeof(struct udphdr) :
1816 sizeof(struct iphdr) + sizeof(struct udphdr);
1817 mtu = (state == IP_VS_STATE_BACKUP) ?
1818 clamp(dev->mtu, 1500U, 65535U) : 1500U;
1819 min_mtu = (state == IP_VS_STATE_BACKUP) ? 1024 : 1;
1822 c->sync_maxlen = clamp_t(unsigned int,
1823 c->sync_maxlen, min_mtu,
1826 c->sync_maxlen = mtu - hlen;
1828 if (state == IP_VS_STATE_MASTER) {
1833 name = "ipvs-m:%d:%d";
1834 threadfn = sync_thread_master;
1835 } else if (state == IP_VS_STATE_BACKUP) {
1836 if (ipvs->backup_threads)
1840 name = "ipvs-b:%d:%d";
1841 threadfn = sync_thread_backup;
1846 if (state == IP_VS_STATE_MASTER) {
1847 struct ipvs_master_sync_state *ms;
1849 ipvs->ms = kzalloc(count * sizeof(ipvs->ms[0]), GFP_KERNEL);
1853 for (id = 0; id < count; id++, ms++) {
1854 INIT_LIST_HEAD(&ms->sync_queue);
1855 ms->sync_queue_len = 0;
1856 ms->sync_queue_delay = 0;
1857 INIT_DELAYED_WORK(&ms->master_wakeup_work,
1858 master_wakeup_work_handler);
1862 array = kzalloc(count * sizeof(struct task_struct *),
1869 for (id = 0; id < count; id++) {
1870 if (state == IP_VS_STATE_MASTER)
1871 sock = make_send_sock(ipvs, id);
1873 sock = make_receive_sock(ipvs, id, dev->ifindex);
1875 result = PTR_ERR(sock);
1878 tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL);
1883 if (state == IP_VS_STATE_BACKUP) {
1884 tinfo->buf = kmalloc(ipvs->bcfg.sync_maxlen,
1893 task = kthread_run(threadfn, tinfo, name, ipvs->gen, id);
1895 result = PTR_ERR(task);
1899 if (state == IP_VS_STATE_MASTER)
1900 ipvs->ms[id].master_thread = task;
1905 /* mark as active */
1907 if (state == IP_VS_STATE_BACKUP)
1908 ipvs->backup_threads = array;
1909 spin_lock_bh(&ipvs->sync_buff_lock);
1910 ipvs->sync_state |= state;
1911 spin_unlock_bh(&ipvs->sync_buff_lock);
1913 /* increase the module use count */
1914 ip_vs_use_count_inc();
1923 sock_release(tinfo->sock);
1928 while (count-- > 0) {
1929 if (state == IP_VS_STATE_MASTER)
1930 kthread_stop(ipvs->ms[count].master_thread);
1932 kthread_stop(array[count]);
1937 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
1945 int stop_sync_thread(struct netns_ipvs *ipvs, int state)
1947 struct task_struct **array;
1951 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1953 if (state == IP_VS_STATE_MASTER) {
1958 * The lock synchronizes with sb_queue_tail(), so that we don't
1959 * add sync buffers to the queue, when we are already in
1960 * progress of stopping the master sync daemon.
1963 spin_lock_bh(&ipvs->sync_buff_lock);
1964 spin_lock(&ipvs->sync_lock);
1965 ipvs->sync_state &= ~IP_VS_STATE_MASTER;
1966 spin_unlock(&ipvs->sync_lock);
1967 spin_unlock_bh(&ipvs->sync_buff_lock);
1970 for (id = ipvs->threads_mask; id >= 0; id--) {
1971 struct ipvs_master_sync_state *ms = &ipvs->ms[id];
1974 pr_info("stopping master sync thread %d ...\n",
1975 task_pid_nr(ms->master_thread));
1976 cancel_delayed_work_sync(&ms->master_wakeup_work);
1977 ret = kthread_stop(ms->master_thread);
1983 } else if (state == IP_VS_STATE_BACKUP) {
1984 if (!ipvs->backup_threads)
1987 ipvs->sync_state &= ~IP_VS_STATE_BACKUP;
1988 array = ipvs->backup_threads;
1990 for (id = ipvs->threads_mask; id >= 0; id--) {
1993 pr_info("stopping backup sync thread %d ...\n",
1994 task_pid_nr(array[id]));
1995 ret = kthread_stop(array[id]);
2000 ipvs->backup_threads = NULL;
2003 /* decrease the module use count */
2004 ip_vs_use_count_dec();
2010 * Initialize data struct for each netns
2012 int __net_init ip_vs_sync_net_init(struct netns_ipvs *ipvs)
2014 __mutex_init(&ipvs->sync_mutex, "ipvs->sync_mutex", &__ipvs_sync_key);
2015 spin_lock_init(&ipvs->sync_lock);
2016 spin_lock_init(&ipvs->sync_buff_lock);
2020 void ip_vs_sync_net_cleanup(struct netns_ipvs *ipvs)
2024 mutex_lock(&ipvs->sync_mutex);
2025 retc = stop_sync_thread(ipvs, IP_VS_STATE_MASTER);
2026 if (retc && retc != -ESRCH)
2027 pr_err("Failed to stop Master Daemon\n");
2029 retc = stop_sync_thread(ipvs, IP_VS_STATE_BACKUP);
2030 if (retc && retc != -ESRCH)
2031 pr_err("Failed to stop Backup Daemon\n");
2032 mutex_unlock(&ipvs->sync_mutex);