1 // SPDX-License-Identifier: GPL-2.0
3 * IPVS An implementation of the IP virtual server support for the
4 * LINUX operating system. IPVS is now implemented as a module
5 * over the NetFilter framework. IPVS can be used to build a
6 * high-performance and highly available server based on a
9 * Version 1, is capable of handling both version 0 and 1 messages.
10 * Version 0 is the plain old format.
11 * Note Version 0 receivers will just drop Ver 1 messages.
12 * Version 1 is capable of handle IPv6, Persistence data,
13 * time-outs, and firewall marks.
14 * In ver.1 "ip_vs_sync_conn_options" will be sent in netw. order.
15 * Ver. 0 can be turned on by sysctl -w net.ipv4.vs.sync_version=0
17 * Definitions Message: is a complete datagram
18 * Sync_conn: is a part of a Message
19 * Param Data is an option to a Sync_conn.
21 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
23 * ip_vs_sync: sync connection info from master load balancer to backups
27 * Alexandre Cassen : Added master & backup support at a time.
28 * Alexandre Cassen : Added SyncID support for incoming sync
30 * Justin Ossevoort : Fix endian problem on sync message size.
31 * Hans Schillstrom : Added Version 1: i.e. IPv6,
32 * Persistence support, fwmark and time-out.
35 #define KMSG_COMPONENT "IPVS"
36 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
38 #include <linux/module.h>
39 #include <linux/slab.h>
40 #include <linux/inetdevice.h>
41 #include <linux/net.h>
42 #include <linux/completion.h>
43 #include <linux/delay.h>
44 #include <linux/skbuff.h>
46 #include <linux/igmp.h> /* for ip_mc_join_group */
47 #include <linux/udp.h>
48 #include <linux/err.h>
49 #include <linux/kthread.h>
50 #include <linux/wait.h>
51 #include <linux/kernel.h>
52 #include <linux/sched/signal.h>
54 #include <asm/unaligned.h> /* Used for ntoh_seq and hton_seq */
59 #include <net/ip_vs.h>
61 #define IP_VS_SYNC_GROUP 0xe0000051 /* multicast addr - 224.0.0.81 */
62 #define IP_VS_SYNC_PORT 8848 /* multicast port */
64 #define SYNC_PROTO_VER 1 /* Protocol version in header */
66 static struct lock_class_key __ipvs_sync_key;
68 * IPVS sync connection entry
69 * Version 0, i.e. original version.
71 struct ip_vs_sync_conn_v0 {
74 /* Protocol, addresses and port numbers */
75 __u8 protocol; /* Which protocol (TCP/UDP) */
79 __be32 caddr; /* client address */
80 __be32 vaddr; /* virtual address */
81 __be32 daddr; /* destination address */
83 /* Flags and state transition */
84 __be16 flags; /* status flags */
85 __be16 state; /* state info */
87 /* The sequence options start here */
90 struct ip_vs_sync_conn_options {
91 struct ip_vs_seq in_seq; /* incoming seq. struct */
92 struct ip_vs_seq out_seq; /* outgoing seq. struct */
96 Sync Connection format (sync_conn)
99 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
100 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
101 | Type | Protocol | Ver. | Size |
102 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
104 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
106 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
108 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
110 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
111 | timeout (in sec.) |
112 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
114 | IP-Addresses (v4 or v6) |
116 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
118 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
119 | Param. Type | Param. Length | Param. data |
120 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
122 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
123 | | Param Type | Param. Length |
124 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
126 | Last Param data should be padded for 32 bit alignment |
127 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
131 * Type 0, IPv4 sync connection format
133 struct ip_vs_sync_v4 {
135 __u8 protocol; /* Which protocol (TCP/UDP) */
136 __be16 ver_size; /* Version msb 4 bits */
137 /* Flags and state transition */
138 __be32 flags; /* status flags */
139 __be16 state; /* state info */
140 /* Protocol, addresses and port numbers */
144 __be32 fwmark; /* Firewall mark from skb */
145 __be32 timeout; /* cp timeout */
146 __be32 caddr; /* client address */
147 __be32 vaddr; /* virtual address */
148 __be32 daddr; /* destination address */
149 /* The sequence options start here */
150 /* PE data padded to 32bit alignment after seq. options */
153 * Type 2 messages IPv6
155 struct ip_vs_sync_v6 {
157 __u8 protocol; /* Which protocol (TCP/UDP) */
158 __be16 ver_size; /* Version msb 4 bits */
159 /* Flags and state transition */
160 __be32 flags; /* status flags */
161 __be16 state; /* state info */
162 /* Protocol, addresses and port numbers */
166 __be32 fwmark; /* Firewall mark from skb */
167 __be32 timeout; /* cp timeout */
168 struct in6_addr caddr; /* client address */
169 struct in6_addr vaddr; /* virtual address */
170 struct in6_addr daddr; /* destination address */
171 /* The sequence options start here */
172 /* PE data padded to 32bit alignment after seq. options */
175 union ip_vs_sync_conn {
176 struct ip_vs_sync_v4 v4;
177 struct ip_vs_sync_v6 v6;
180 /* Bits in Type field in above */
181 #define STYPE_INET6 0
182 #define STYPE_F_INET6 (1 << STYPE_INET6)
184 #define SVER_SHIFT 12 /* Shift to get version */
185 #define SVER_MASK 0x0fff /* Mask to strip version */
187 #define IPVS_OPT_SEQ_DATA 1
188 #define IPVS_OPT_PE_DATA 2
189 #define IPVS_OPT_PE_NAME 3
190 #define IPVS_OPT_PARAM 7
192 #define IPVS_OPT_F_SEQ_DATA (1 << (IPVS_OPT_SEQ_DATA-1))
193 #define IPVS_OPT_F_PE_DATA (1 << (IPVS_OPT_PE_DATA-1))
194 #define IPVS_OPT_F_PE_NAME (1 << (IPVS_OPT_PE_NAME-1))
195 #define IPVS_OPT_F_PARAM (1 << (IPVS_OPT_PARAM-1))
197 struct ip_vs_sync_thread_data {
198 struct netns_ipvs *ipvs;
204 /* Version 0 definition of packet sizes */
205 #define SIMPLE_CONN_SIZE (sizeof(struct ip_vs_sync_conn_v0))
206 #define FULL_CONN_SIZE \
207 (sizeof(struct ip_vs_sync_conn_v0) + sizeof(struct ip_vs_sync_conn_options))
211 The master mulitcasts messages (Datagrams) to the backup load balancers
212 in the following format.
215 Note, first byte should be Zero, so ver 0 receivers will drop the packet.
218 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
219 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
220 | 0 | SyncID | Size |
221 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
222 | Count Conns | Version | Reserved, set to Zero |
223 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
225 | IPVS Sync Connection (1) |
226 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
230 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
232 | IPVS Sync Connection (n) |
233 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
237 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
238 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
239 | Count Conns | SyncID | Size |
240 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
241 | IPVS Sync Connection (1) |
244 #define SYNC_MESG_HEADER_LEN 4
245 #define MAX_CONNS_PER_SYNCBUFF 255 /* nr_conns in ip_vs_sync_mesg is 8 bit */
247 /* Version 0 header */
248 struct ip_vs_sync_mesg_v0 {
253 /* ip_vs_sync_conn entries start here */
256 /* Version 1 header */
257 struct ip_vs_sync_mesg {
258 __u8 reserved; /* must be zero */
262 __s8 version; /* SYNC_PROTO_VER */
264 /* ip_vs_sync_conn entries start here */
267 union ipvs_sockaddr {
268 struct sockaddr_in in;
269 struct sockaddr_in6 in6;
272 struct ip_vs_sync_buff {
273 struct list_head list;
274 unsigned long firstuse;
276 /* pointers for the message data */
277 struct ip_vs_sync_mesg *mesg;
283 * Copy of struct ip_vs_seq
284 * From unaligned network order to aligned host order
286 static void ntoh_seq(struct ip_vs_seq *no, struct ip_vs_seq *ho)
288 memset(ho, 0, sizeof(*ho));
289 ho->init_seq = get_unaligned_be32(&no->init_seq);
290 ho->delta = get_unaligned_be32(&no->delta);
291 ho->previous_delta = get_unaligned_be32(&no->previous_delta);
295 * Copy of struct ip_vs_seq
296 * From Aligned host order to unaligned network order
298 static void hton_seq(struct ip_vs_seq *ho, struct ip_vs_seq *no)
300 put_unaligned_be32(ho->init_seq, &no->init_seq);
301 put_unaligned_be32(ho->delta, &no->delta);
302 put_unaligned_be32(ho->previous_delta, &no->previous_delta);
305 static inline struct ip_vs_sync_buff *
306 sb_dequeue(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
308 struct ip_vs_sync_buff *sb;
310 spin_lock_bh(&ipvs->sync_lock);
311 if (list_empty(&ms->sync_queue)) {
313 __set_current_state(TASK_INTERRUPTIBLE);
315 sb = list_entry(ms->sync_queue.next, struct ip_vs_sync_buff,
318 ms->sync_queue_len--;
319 if (!ms->sync_queue_len)
320 ms->sync_queue_delay = 0;
322 spin_unlock_bh(&ipvs->sync_lock);
328 * Create a new sync buffer for Version 1 proto.
330 static inline struct ip_vs_sync_buff *
331 ip_vs_sync_buff_create(struct netns_ipvs *ipvs, unsigned int len)
333 struct ip_vs_sync_buff *sb;
335 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
338 len = max_t(unsigned int, len + sizeof(struct ip_vs_sync_mesg),
339 ipvs->mcfg.sync_maxlen);
340 sb->mesg = kmalloc(len, GFP_ATOMIC);
345 sb->mesg->reserved = 0; /* old nr_conns i.e. must be zero now */
346 sb->mesg->version = SYNC_PROTO_VER;
347 sb->mesg->syncid = ipvs->mcfg.syncid;
348 sb->mesg->size = htons(sizeof(struct ip_vs_sync_mesg));
349 sb->mesg->nr_conns = 0;
351 sb->head = (unsigned char *)sb->mesg + sizeof(struct ip_vs_sync_mesg);
352 sb->end = (unsigned char *)sb->mesg + len;
354 sb->firstuse = jiffies;
358 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
364 static inline void sb_queue_tail(struct netns_ipvs *ipvs,
365 struct ipvs_master_sync_state *ms)
367 struct ip_vs_sync_buff *sb = ms->sync_buff;
369 spin_lock(&ipvs->sync_lock);
370 if (ipvs->sync_state & IP_VS_STATE_MASTER &&
371 ms->sync_queue_len < sysctl_sync_qlen_max(ipvs)) {
372 if (!ms->sync_queue_len)
373 schedule_delayed_work(&ms->master_wakeup_work,
374 max(IPVS_SYNC_SEND_DELAY, 1));
375 ms->sync_queue_len++;
376 list_add_tail(&sb->list, &ms->sync_queue);
377 if ((++ms->sync_queue_delay) == IPVS_SYNC_WAKEUP_RATE)
378 wake_up_process(ms->master_thread);
380 ip_vs_sync_buff_release(sb);
381 spin_unlock(&ipvs->sync_lock);
385 * Get the current sync buffer if it has been created for more
386 * than the specified time or the specified time is zero.
388 static inline struct ip_vs_sync_buff *
389 get_curr_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms,
392 struct ip_vs_sync_buff *sb;
394 spin_lock_bh(&ipvs->sync_buff_lock);
396 if (sb && time_after_eq(jiffies - sb->firstuse, time)) {
397 ms->sync_buff = NULL;
398 __set_current_state(TASK_RUNNING);
401 spin_unlock_bh(&ipvs->sync_buff_lock);
406 select_master_thread_id(struct netns_ipvs *ipvs, struct ip_vs_conn *cp)
408 return ((long) cp >> (1 + ilog2(sizeof(*cp)))) & ipvs->threads_mask;
412 * Create a new sync buffer for Version 0 proto.
414 static inline struct ip_vs_sync_buff *
415 ip_vs_sync_buff_create_v0(struct netns_ipvs *ipvs, unsigned int len)
417 struct ip_vs_sync_buff *sb;
418 struct ip_vs_sync_mesg_v0 *mesg;
420 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
423 len = max_t(unsigned int, len + sizeof(struct ip_vs_sync_mesg_v0),
424 ipvs->mcfg.sync_maxlen);
425 sb->mesg = kmalloc(len, GFP_ATOMIC);
430 mesg = (struct ip_vs_sync_mesg_v0 *)sb->mesg;
432 mesg->syncid = ipvs->mcfg.syncid;
433 mesg->size = htons(sizeof(struct ip_vs_sync_mesg_v0));
434 sb->head = (unsigned char *)mesg + sizeof(struct ip_vs_sync_mesg_v0);
435 sb->end = (unsigned char *)mesg + len;
436 sb->firstuse = jiffies;
440 /* Check if connection is controlled by persistence */
441 static inline bool in_persistence(struct ip_vs_conn *cp)
443 for (cp = cp->control; cp; cp = cp->control) {
444 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
450 /* Check if conn should be synced.
451 * pkts: conn packets, use sysctl_sync_threshold to avoid packet check
452 * - (1) sync_refresh_period: reduce sync rate. Additionally, retry
453 * sync_retries times with period of sync_refresh_period/8
454 * - (2) if both sync_refresh_period and sync_period are 0 send sync only
455 * for state changes or only once when pkts matches sync_threshold
456 * - (3) templates: rate can be reduced only with sync_refresh_period or
459 static int ip_vs_sync_conn_needed(struct netns_ipvs *ipvs,
460 struct ip_vs_conn *cp, int pkts)
462 unsigned long orig = READ_ONCE(cp->sync_endtime);
463 unsigned long now = jiffies;
464 unsigned long n = (now + cp->timeout) & ~3UL;
465 unsigned int sync_refresh_period;
469 /* Check if we sync in current state */
470 if (unlikely(cp->flags & IP_VS_CONN_F_TEMPLATE))
472 else if (unlikely(sysctl_sync_persist_mode(ipvs) && in_persistence(cp)))
474 else if (likely(cp->protocol == IPPROTO_TCP)) {
475 if (!((1 << cp->state) &
476 ((1 << IP_VS_TCP_S_ESTABLISHED) |
477 (1 << IP_VS_TCP_S_FIN_WAIT) |
478 (1 << IP_VS_TCP_S_CLOSE) |
479 (1 << IP_VS_TCP_S_CLOSE_WAIT) |
480 (1 << IP_VS_TCP_S_TIME_WAIT))))
482 force = cp->state != cp->old_state;
483 if (force && cp->state != IP_VS_TCP_S_ESTABLISHED)
485 } else if (unlikely(cp->protocol == IPPROTO_SCTP)) {
486 if (!((1 << cp->state) &
487 ((1 << IP_VS_SCTP_S_ESTABLISHED) |
488 (1 << IP_VS_SCTP_S_SHUTDOWN_SENT) |
489 (1 << IP_VS_SCTP_S_SHUTDOWN_RECEIVED) |
490 (1 << IP_VS_SCTP_S_SHUTDOWN_ACK_SENT) |
491 (1 << IP_VS_SCTP_S_CLOSED))))
493 force = cp->state != cp->old_state;
494 if (force && cp->state != IP_VS_SCTP_S_ESTABLISHED)
497 /* UDP or another protocol with single state */
501 sync_refresh_period = sysctl_sync_refresh_period(ipvs);
502 if (sync_refresh_period > 0) {
503 long diff = n - orig;
504 long min_diff = max(cp->timeout >> 1, 10UL * HZ);
506 /* Avoid sync if difference is below sync_refresh_period
507 * and below the half timeout.
509 if (abs(diff) < min_t(long, sync_refresh_period, min_diff)) {
510 int retries = orig & 3;
512 if (retries >= sysctl_sync_retries(ipvs))
514 if (time_before(now, orig - cp->timeout +
515 (sync_refresh_period >> 3)))
520 sync_period = sysctl_sync_period(ipvs);
521 if (sync_period > 0) {
522 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE) &&
523 pkts % sync_period != sysctl_sync_threshold(ipvs))
525 } else if (!sync_refresh_period &&
526 pkts != sysctl_sync_threshold(ipvs))
530 cp->old_state = cp->state;
531 n = cmpxchg(&cp->sync_endtime, orig, n);
532 return n == orig || force;
536 * Version 0 , could be switched in by sys_ctl.
537 * Add an ip_vs_conn information into the current sync_buff.
539 static void ip_vs_sync_conn_v0(struct netns_ipvs *ipvs, struct ip_vs_conn *cp,
542 struct ip_vs_sync_mesg_v0 *m;
543 struct ip_vs_sync_conn_v0 *s;
544 struct ip_vs_sync_buff *buff;
545 struct ipvs_master_sync_state *ms;
549 if (unlikely(cp->af != AF_INET))
551 /* Do not sync ONE PACKET */
552 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
555 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
558 spin_lock_bh(&ipvs->sync_buff_lock);
559 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
560 spin_unlock_bh(&ipvs->sync_buff_lock);
564 id = select_master_thread_id(ipvs, cp);
566 buff = ms->sync_buff;
567 len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
570 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
571 /* Send buffer if it is for v1 */
572 if (buff->head + len > buff->end || !m->nr_conns) {
573 sb_queue_tail(ipvs, ms);
574 ms->sync_buff = NULL;
579 buff = ip_vs_sync_buff_create_v0(ipvs, len);
581 spin_unlock_bh(&ipvs->sync_buff_lock);
582 pr_err("ip_vs_sync_buff_create failed.\n");
585 ms->sync_buff = buff;
588 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
589 s = (struct ip_vs_sync_conn_v0 *) buff->head;
593 s->protocol = cp->protocol;
594 s->cport = cp->cport;
595 s->vport = cp->vport;
596 s->dport = cp->dport;
597 s->caddr = cp->caddr.ip;
598 s->vaddr = cp->vaddr.ip;
599 s->daddr = cp->daddr.ip;
600 s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
601 s->state = htons(cp->state);
602 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
603 struct ip_vs_sync_conn_options *opt =
604 (struct ip_vs_sync_conn_options *)&s[1];
605 memcpy(opt, &cp->in_seq, sizeof(*opt));
609 m->size = htons(ntohs(m->size) + len);
611 spin_unlock_bh(&ipvs->sync_buff_lock);
613 /* synchronize its controller if it has */
616 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
617 pkts = atomic_add_return(1, &cp->in_pkts);
619 pkts = sysctl_sync_threshold(ipvs);
620 ip_vs_sync_conn(ipvs, cp, pkts);
625 * Add an ip_vs_conn information into the current sync_buff.
626 * Called by ip_vs_in.
627 * Sending Version 1 messages
629 void ip_vs_sync_conn(struct netns_ipvs *ipvs, struct ip_vs_conn *cp, int pkts)
631 struct ip_vs_sync_mesg *m;
632 union ip_vs_sync_conn *s;
633 struct ip_vs_sync_buff *buff;
634 struct ipvs_master_sync_state *ms;
637 unsigned int len, pe_name_len, pad;
639 /* Handle old version of the protocol */
640 if (sysctl_sync_ver(ipvs) == 0) {
641 ip_vs_sync_conn_v0(ipvs, cp, pkts);
644 /* Do not sync ONE PACKET */
645 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
648 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
653 if (cp->pe_data_len) {
654 if (!cp->pe_data || !cp->dest) {
655 IP_VS_ERR_RL("SYNC, connection pe_data invalid\n");
658 pe_name_len = strnlen(cp->pe->name, IP_VS_PENAME_MAXLEN);
661 spin_lock_bh(&ipvs->sync_buff_lock);
662 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
663 spin_unlock_bh(&ipvs->sync_buff_lock);
667 id = select_master_thread_id(ipvs, cp);
670 #ifdef CONFIG_IP_VS_IPV6
671 if (cp->af == AF_INET6)
672 len = sizeof(struct ip_vs_sync_v6);
675 len = sizeof(struct ip_vs_sync_v4);
677 if (cp->flags & IP_VS_CONN_F_SEQ_MASK)
678 len += sizeof(struct ip_vs_sync_conn_options) + 2;
681 len += cp->pe_data_len + 2; /* + Param hdr field */
683 len += pe_name_len + 2;
685 /* check if there is a space for this one */
687 buff = ms->sync_buff;
690 pad = (4 - (size_t) buff->head) & 3;
691 /* Send buffer if it is for v0 */
692 if (buff->head + len + pad > buff->end || m->reserved) {
693 sb_queue_tail(ipvs, ms);
694 ms->sync_buff = NULL;
701 buff = ip_vs_sync_buff_create(ipvs, len);
703 spin_unlock_bh(&ipvs->sync_buff_lock);
704 pr_err("ip_vs_sync_buff_create failed.\n");
707 ms->sync_buff = buff;
712 buff->head += pad + len;
713 m->size = htons(ntohs(m->size) + pad + len);
714 /* Add ev. padding from prev. sync_conn */
718 s = (union ip_vs_sync_conn *)p;
720 /* Set message type & copy members */
721 s->v4.type = (cp->af == AF_INET6 ? STYPE_F_INET6 : 0);
722 s->v4.ver_size = htons(len & SVER_MASK); /* Version 0 */
723 s->v4.flags = htonl(cp->flags & ~IP_VS_CONN_F_HASHED);
724 s->v4.state = htons(cp->state);
725 s->v4.protocol = cp->protocol;
726 s->v4.cport = cp->cport;
727 s->v4.vport = cp->vport;
728 s->v4.dport = cp->dport;
729 s->v4.fwmark = htonl(cp->fwmark);
730 s->v4.timeout = htonl(cp->timeout / HZ);
733 #ifdef CONFIG_IP_VS_IPV6
734 if (cp->af == AF_INET6) {
735 p += sizeof(struct ip_vs_sync_v6);
736 s->v6.caddr = cp->caddr.in6;
737 s->v6.vaddr = cp->vaddr.in6;
738 s->v6.daddr = cp->daddr.in6;
742 p += sizeof(struct ip_vs_sync_v4); /* options ptr */
743 s->v4.caddr = cp->caddr.ip;
744 s->v4.vaddr = cp->vaddr.ip;
745 s->v4.daddr = cp->daddr.ip;
747 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
748 *(p++) = IPVS_OPT_SEQ_DATA;
749 *(p++) = sizeof(struct ip_vs_sync_conn_options);
750 hton_seq((struct ip_vs_seq *)p, &cp->in_seq);
751 p += sizeof(struct ip_vs_seq);
752 hton_seq((struct ip_vs_seq *)p, &cp->out_seq);
753 p += sizeof(struct ip_vs_seq);
756 if (cp->pe_data_len && cp->pe_data) {
757 *(p++) = IPVS_OPT_PE_DATA;
758 *(p++) = cp->pe_data_len;
759 memcpy(p, cp->pe_data, cp->pe_data_len);
760 p += cp->pe_data_len;
763 *(p++) = IPVS_OPT_PE_NAME;
764 *(p++) = pe_name_len;
765 memcpy(p, cp->pe->name, pe_name_len);
770 spin_unlock_bh(&ipvs->sync_buff_lock);
773 /* synchronize its controller if it has */
777 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
778 pkts = atomic_add_return(1, &cp->in_pkts);
780 pkts = sysctl_sync_threshold(ipvs);
785 * fill_param used by version 1
788 ip_vs_conn_fill_param_sync(struct netns_ipvs *ipvs, int af, union ip_vs_sync_conn *sc,
789 struct ip_vs_conn_param *p,
790 __u8 *pe_data, unsigned int pe_data_len,
791 __u8 *pe_name, unsigned int pe_name_len)
793 #ifdef CONFIG_IP_VS_IPV6
795 ip_vs_conn_fill_param(ipvs, af, sc->v6.protocol,
796 (const union nf_inet_addr *)&sc->v6.caddr,
798 (const union nf_inet_addr *)&sc->v6.vaddr,
802 ip_vs_conn_fill_param(ipvs, af, sc->v4.protocol,
803 (const union nf_inet_addr *)&sc->v4.caddr,
805 (const union nf_inet_addr *)&sc->v4.vaddr,
810 char buff[IP_VS_PENAME_MAXLEN+1];
812 memcpy(buff, pe_name, pe_name_len);
814 p->pe = __ip_vs_pe_getbyname(buff);
816 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n",
821 IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n");
825 p->pe_data = kmemdup(pe_data, pe_data_len, GFP_ATOMIC);
827 module_put(p->pe->module);
830 p->pe_data_len = pe_data_len;
836 * Connection Add / Update.
837 * Common for version 0 and 1 reception of backup sync_conns.
841 static void ip_vs_proc_conn(struct netns_ipvs *ipvs, struct ip_vs_conn_param *param,
842 unsigned int flags, unsigned int state,
843 unsigned int protocol, unsigned int type,
844 const union nf_inet_addr *daddr, __be16 dport,
845 unsigned long timeout, __u32 fwmark,
846 struct ip_vs_sync_conn_options *opt)
848 struct ip_vs_dest *dest;
849 struct ip_vs_conn *cp;
851 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
852 cp = ip_vs_conn_in_get(param);
853 if (cp && ((cp->dport != dport) ||
854 !ip_vs_addr_equal(cp->daf, &cp->daddr, daddr))) {
855 if (!(flags & IP_VS_CONN_F_INACTIVE)) {
856 ip_vs_conn_expire_now(cp);
857 __ip_vs_conn_put(cp);
860 /* This is the expiration message for the
861 * connection that was already replaced, so we
864 __ip_vs_conn_put(cp);
865 kfree(param->pe_data);
870 cp = ip_vs_ct_in_get(param);
875 kfree(param->pe_data);
878 spin_lock_bh(&cp->lock);
879 if ((cp->flags ^ flags) & IP_VS_CONN_F_INACTIVE &&
880 !(flags & IP_VS_CONN_F_TEMPLATE) && dest) {
881 if (flags & IP_VS_CONN_F_INACTIVE) {
882 atomic_dec(&dest->activeconns);
883 atomic_inc(&dest->inactconns);
885 atomic_inc(&dest->activeconns);
886 atomic_dec(&dest->inactconns);
889 flags &= IP_VS_CONN_F_BACKUP_UPD_MASK;
890 flags |= cp->flags & ~IP_VS_CONN_F_BACKUP_UPD_MASK;
892 spin_unlock_bh(&cp->lock);
894 ip_vs_try_bind_dest(cp);
897 * Find the appropriate destination for the connection.
898 * If it is not found the connection will remain unbound
902 /* This function is only invoked by the synchronization
903 * code. We do not currently support heterogeneous pools
904 * with synchronization, so we can make the assumption that
905 * the svc_af is the same as the dest_af
907 dest = ip_vs_find_dest(ipvs, type, type, daddr, dport,
908 param->vaddr, param->vport, protocol,
911 cp = ip_vs_conn_new(param, type, daddr, dport, flags, dest,
915 kfree(param->pe_data);
916 IP_VS_DBG(2, "BACKUP, add new conn. failed\n");
919 if (!(flags & IP_VS_CONN_F_TEMPLATE))
920 kfree(param->pe_data);
924 cp->in_seq = opt->in_seq;
925 cp->out_seq = opt->out_seq;
927 atomic_set(&cp->in_pkts, sysctl_sync_threshold(ipvs));
929 cp->old_state = cp->state;
931 * For Ver 0 messages style
932 * - Not possible to recover the right timeout for templates
933 * - can not find the right fwmark
934 * virtual service. If needed, we can do it for
935 * non-fwmark persistent services.
936 * Ver 1 messages style.
940 if (timeout > MAX_SCHEDULE_TIMEOUT / HZ)
941 timeout = MAX_SCHEDULE_TIMEOUT / HZ;
942 cp->timeout = timeout*HZ;
944 struct ip_vs_proto_data *pd;
946 pd = ip_vs_proto_data_get(ipvs, protocol);
947 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pd && pd->timeout_table)
948 cp->timeout = pd->timeout_table[state];
950 cp->timeout = (3*60*HZ);
956 * Process received multicast message for Version 0
958 static void ip_vs_process_message_v0(struct netns_ipvs *ipvs, const char *buffer,
961 struct ip_vs_sync_mesg_v0 *m = (struct ip_vs_sync_mesg_v0 *)buffer;
962 struct ip_vs_sync_conn_v0 *s;
963 struct ip_vs_sync_conn_options *opt;
964 struct ip_vs_protocol *pp;
965 struct ip_vs_conn_param param;
969 p = (char *)buffer + sizeof(struct ip_vs_sync_mesg_v0);
970 for (i=0; i<m->nr_conns; i++) {
971 unsigned int flags, state;
973 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
974 IP_VS_ERR_RL("BACKUP v0, bogus conn\n");
977 s = (struct ip_vs_sync_conn_v0 *) p;
978 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
979 flags &= ~IP_VS_CONN_F_HASHED;
980 if (flags & IP_VS_CONN_F_SEQ_MASK) {
981 opt = (struct ip_vs_sync_conn_options *)&s[1];
983 if (p > buffer+buflen) {
984 IP_VS_ERR_RL("BACKUP v0, Dropping buffer bogus conn options\n");
989 p += SIMPLE_CONN_SIZE;
992 state = ntohs(s->state);
993 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
994 pp = ip_vs_proto_get(s->protocol);
996 IP_VS_DBG(2, "BACKUP v0, Unsupported protocol %u\n",
1000 if (state >= pp->num_states) {
1001 IP_VS_DBG(2, "BACKUP v0, Invalid %s state %u\n",
1006 /* protocol in templates is not used for state/timeout */
1008 IP_VS_DBG(2, "BACKUP v0, Invalid template state %u\n",
1014 ip_vs_conn_fill_param(ipvs, AF_INET, s->protocol,
1015 (const union nf_inet_addr *)&s->caddr,
1017 (const union nf_inet_addr *)&s->vaddr,
1020 /* Send timeout as Zero */
1021 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->protocol, AF_INET,
1022 (union nf_inet_addr *)&s->daddr, s->dport,
1030 static inline int ip_vs_proc_seqopt(__u8 *p, unsigned int plen,
1032 struct ip_vs_sync_conn_options *opt)
1034 struct ip_vs_sync_conn_options *topt;
1036 topt = (struct ip_vs_sync_conn_options *)p;
1038 if (plen != sizeof(struct ip_vs_sync_conn_options)) {
1039 IP_VS_DBG(2, "BACKUP, bogus conn options length\n");
1042 if (*opt_flags & IPVS_OPT_F_SEQ_DATA) {
1043 IP_VS_DBG(2, "BACKUP, conn options found twice\n");
1046 ntoh_seq(&topt->in_seq, &opt->in_seq);
1047 ntoh_seq(&topt->out_seq, &opt->out_seq);
1048 *opt_flags |= IPVS_OPT_F_SEQ_DATA;
1052 static int ip_vs_proc_str(__u8 *p, unsigned int plen, unsigned int *data_len,
1053 __u8 **data, unsigned int maxlen,
1054 __u32 *opt_flags, __u32 flag)
1056 if (plen > maxlen) {
1057 IP_VS_DBG(2, "BACKUP, bogus par.data len > %d\n", maxlen);
1060 if (*opt_flags & flag) {
1061 IP_VS_DBG(2, "BACKUP, Par.data found twice 0x%x\n", flag);
1070 * Process a Version 1 sync. connection
1072 static inline int ip_vs_proc_sync_conn(struct netns_ipvs *ipvs, __u8 *p, __u8 *msg_end)
1074 struct ip_vs_sync_conn_options opt;
1075 union ip_vs_sync_conn *s;
1076 struct ip_vs_protocol *pp;
1077 struct ip_vs_conn_param param;
1079 unsigned int af, state, pe_data_len=0, pe_name_len=0;
1080 __u8 *pe_data=NULL, *pe_name=NULL;
1084 s = (union ip_vs_sync_conn *) p;
1086 if (s->v6.type & STYPE_F_INET6) {
1087 #ifdef CONFIG_IP_VS_IPV6
1089 p += sizeof(struct ip_vs_sync_v6);
1091 IP_VS_DBG(3,"BACKUP, IPv6 msg received, and IPVS is not compiled for IPv6\n");
1095 } else if (!s->v4.type) {
1097 p += sizeof(struct ip_vs_sync_v4);
1104 /* Process optional params check Type & Len. */
1105 while (p < msg_end) {
1114 if (!plen || ((p + plen) > msg_end))
1116 /* Handle seq option p = param data */
1117 switch (ptype & ~IPVS_OPT_F_PARAM) {
1118 case IPVS_OPT_SEQ_DATA:
1119 if (ip_vs_proc_seqopt(p, plen, &opt_flags, &opt))
1123 case IPVS_OPT_PE_DATA:
1124 if (ip_vs_proc_str(p, plen, &pe_data_len, &pe_data,
1125 IP_VS_PEDATA_MAXLEN, &opt_flags,
1126 IPVS_OPT_F_PE_DATA))
1130 case IPVS_OPT_PE_NAME:
1131 if (ip_vs_proc_str(p, plen,&pe_name_len, &pe_name,
1132 IP_VS_PENAME_MAXLEN, &opt_flags,
1133 IPVS_OPT_F_PE_NAME))
1138 /* Param data mandatory ? */
1139 if (!(ptype & IPVS_OPT_F_PARAM)) {
1140 IP_VS_DBG(3, "BACKUP, Unknown mandatory param %d found\n",
1141 ptype & ~IPVS_OPT_F_PARAM);
1146 p += plen; /* Next option */
1149 /* Get flags and Mask off unsupported */
1150 flags = ntohl(s->v4.flags) & IP_VS_CONN_F_BACKUP_MASK;
1151 flags |= IP_VS_CONN_F_SYNC;
1152 state = ntohs(s->v4.state);
1154 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
1155 pp = ip_vs_proto_get(s->v4.protocol);
1157 IP_VS_DBG(3,"BACKUP, Unsupported protocol %u\n",
1162 if (state >= pp->num_states) {
1163 IP_VS_DBG(3, "BACKUP, Invalid %s state %u\n",
1169 /* protocol in templates is not used for state/timeout */
1171 IP_VS_DBG(3, "BACKUP, Invalid template state %u\n",
1176 if (ip_vs_conn_fill_param_sync(ipvs, af, s, ¶m, pe_data,
1177 pe_data_len, pe_name, pe_name_len)) {
1181 /* If only IPv4, just silent skip IPv6 */
1183 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->v4.protocol, af,
1184 (union nf_inet_addr *)&s->v4.daddr, s->v4.dport,
1185 ntohl(s->v4.timeout), ntohl(s->v4.fwmark),
1186 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1188 #ifdef CONFIG_IP_VS_IPV6
1190 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->v6.protocol, af,
1191 (union nf_inet_addr *)&s->v6.daddr, s->v6.dport,
1192 ntohl(s->v6.timeout), ntohl(s->v6.fwmark),
1193 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1196 ip_vs_pe_put(param.pe);
1200 IP_VS_DBG(2, "BACKUP, Single msg dropped err:%d\n", retc);
1205 * Process received multicast message and create the corresponding
1206 * ip_vs_conn entries.
1207 * Handles Version 0 & 1
1209 static void ip_vs_process_message(struct netns_ipvs *ipvs, __u8 *buffer,
1210 const size_t buflen)
1212 struct ip_vs_sync_mesg *m2 = (struct ip_vs_sync_mesg *)buffer;
1216 if (buflen < sizeof(struct ip_vs_sync_mesg_v0)) {
1217 IP_VS_DBG(2, "BACKUP, message header too short\n");
1221 if (buflen != ntohs(m2->size)) {
1222 IP_VS_DBG(2, "BACKUP, bogus message size\n");
1225 /* SyncID sanity check */
1226 if (ipvs->bcfg.syncid != 0 && m2->syncid != ipvs->bcfg.syncid) {
1227 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2->syncid);
1230 /* Handle version 1 message */
1231 if ((m2->version == SYNC_PROTO_VER) && (m2->reserved == 0)
1232 && (m2->spare == 0)) {
1234 msg_end = buffer + sizeof(struct ip_vs_sync_mesg);
1235 nr_conns = m2->nr_conns;
1237 for (i=0; i<nr_conns; i++) {
1238 union ip_vs_sync_conn *s;
1243 if (p + sizeof(s->v4) > buffer+buflen) {
1244 IP_VS_ERR_RL("BACKUP, Dropping buffer, to small\n");
1247 s = (union ip_vs_sync_conn *)p;
1248 size = ntohs(s->v4.ver_size) & SVER_MASK;
1250 /* Basic sanity checks */
1251 if (msg_end > buffer+buflen) {
1252 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n");
1255 if (ntohs(s->v4.ver_size) >> SVER_SHIFT) {
1256 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n",
1257 ntohs(s->v4.ver_size) >> SVER_SHIFT);
1260 /* Process a single sync_conn */
1261 retc = ip_vs_proc_sync_conn(ipvs, p, msg_end);
1263 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n",
1267 /* Make sure we have 32 bit alignment */
1268 msg_end = p + ((size + 3) & ~3);
1271 /* Old type of message */
1272 ip_vs_process_message_v0(ipvs, buffer, buflen);
1279 * Setup sndbuf (mode=1) or rcvbuf (mode=0)
1281 static void set_sock_size(struct sock *sk, int mode, int val)
1283 /* setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &val, sizeof(val)); */
1284 /* setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)); */
1287 val = clamp_t(int, val, (SOCK_MIN_SNDBUF + 1) / 2,
1289 sk->sk_sndbuf = val * 2;
1290 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1292 val = clamp_t(int, val, (SOCK_MIN_RCVBUF + 1) / 2,
1294 sk->sk_rcvbuf = val * 2;
1295 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1301 * Setup loopback of outgoing multicasts on a sending socket
1303 static void set_mcast_loop(struct sock *sk, u_char loop)
1305 struct inet_sock *inet = inet_sk(sk);
1307 /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
1309 inet->mc_loop = loop ? 1 : 0;
1310 #ifdef CONFIG_IP_VS_IPV6
1311 if (sk->sk_family == AF_INET6) {
1312 struct ipv6_pinfo *np = inet6_sk(sk);
1314 /* IPV6_MULTICAST_LOOP */
1315 np->mc_loop = loop ? 1 : 0;
1322 * Specify TTL for outgoing multicasts on a sending socket
1324 static void set_mcast_ttl(struct sock *sk, u_char ttl)
1326 struct inet_sock *inet = inet_sk(sk);
1328 /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
1331 #ifdef CONFIG_IP_VS_IPV6
1332 if (sk->sk_family == AF_INET6) {
1333 struct ipv6_pinfo *np = inet6_sk(sk);
1335 /* IPV6_MULTICAST_HOPS */
1336 np->mcast_hops = ttl;
1342 /* Control fragmentation of messages */
1343 static void set_mcast_pmtudisc(struct sock *sk, int val)
1345 struct inet_sock *inet = inet_sk(sk);
1347 /* setsockopt(sock, SOL_IP, IP_MTU_DISCOVER, &val, sizeof(val)); */
1349 inet->pmtudisc = val;
1350 #ifdef CONFIG_IP_VS_IPV6
1351 if (sk->sk_family == AF_INET6) {
1352 struct ipv6_pinfo *np = inet6_sk(sk);
1354 /* IPV6_MTU_DISCOVER */
1362 * Specifiy default interface for outgoing multicasts
1364 static int set_mcast_if(struct sock *sk, struct net_device *dev)
1366 struct inet_sock *inet = inet_sk(sk);
1368 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1372 inet->mc_index = dev->ifindex;
1373 /* inet->mc_addr = 0; */
1374 #ifdef CONFIG_IP_VS_IPV6
1375 if (sk->sk_family == AF_INET6) {
1376 struct ipv6_pinfo *np = inet6_sk(sk);
1378 /* IPV6_MULTICAST_IF */
1379 np->mcast_oif = dev->ifindex;
1389 * Join a multicast group.
1390 * the group is specified by a class D multicast address 224.0.0.0/8
1391 * in the in_addr structure passed in as a parameter.
1394 join_mcast_group(struct sock *sk, struct in_addr *addr, struct net_device *dev)
1396 struct ip_mreqn mreq;
1399 memset(&mreq, 0, sizeof(mreq));
1400 memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
1402 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1405 mreq.imr_ifindex = dev->ifindex;
1408 ret = ip_mc_join_group(sk, &mreq);
1414 #ifdef CONFIG_IP_VS_IPV6
1415 static int join_mcast_group6(struct sock *sk, struct in6_addr *addr,
1416 struct net_device *dev)
1420 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1424 ret = ipv6_sock_mc_join(sk, dev->ifindex, addr);
1431 static int bind_mcastif_addr(struct socket *sock, struct net_device *dev)
1434 struct sockaddr_in sin;
1436 addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
1438 pr_err("You probably need to specify IP address on "
1439 "multicast interface.\n");
1441 IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
1444 /* Now bind the socket with the address of multicast interface */
1445 sin.sin_family = AF_INET;
1446 sin.sin_addr.s_addr = addr;
1449 return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
1452 static void get_mcast_sockaddr(union ipvs_sockaddr *sa, int *salen,
1453 struct ipvs_sync_daemon_cfg *c, int id)
1455 if (AF_INET6 == c->mcast_af) {
1456 sa->in6 = (struct sockaddr_in6) {
1457 .sin6_family = AF_INET6,
1458 .sin6_port = htons(c->mcast_port + id),
1460 sa->in6.sin6_addr = c->mcast_group.in6;
1461 *salen = sizeof(sa->in6);
1463 sa->in = (struct sockaddr_in) {
1464 .sin_family = AF_INET,
1465 .sin_port = htons(c->mcast_port + id),
1467 sa->in.sin_addr = c->mcast_group.in;
1468 *salen = sizeof(sa->in);
1473 * Set up sending multicast socket over UDP
1475 static int make_send_sock(struct netns_ipvs *ipvs, int id,
1476 struct net_device *dev, struct socket **sock_ret)
1478 /* multicast addr */
1479 union ipvs_sockaddr mcast_addr;
1480 struct socket *sock;
1483 /* First create a socket */
1484 result = sock_create_kern(ipvs->net, ipvs->mcfg.mcast_af, SOCK_DGRAM,
1485 IPPROTO_UDP, &sock);
1487 pr_err("Error during creation of socket; terminating\n");
1491 result = set_mcast_if(sock->sk, dev);
1493 pr_err("Error setting outbound mcast interface\n");
1497 set_mcast_loop(sock->sk, 0);
1498 set_mcast_ttl(sock->sk, ipvs->mcfg.mcast_ttl);
1499 /* Allow fragmentation if MTU changes */
1500 set_mcast_pmtudisc(sock->sk, IP_PMTUDISC_DONT);
1501 result = sysctl_sync_sock_size(ipvs);
1503 set_sock_size(sock->sk, 1, result);
1505 if (AF_INET == ipvs->mcfg.mcast_af)
1506 result = bind_mcastif_addr(sock, dev);
1510 pr_err("Error binding address of the mcast interface\n");
1514 get_mcast_sockaddr(&mcast_addr, &salen, &ipvs->mcfg, id);
1515 result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr,
1518 pr_err("Error connecting to the multicast addr\n");
1530 * Set up receiving multicast socket over UDP
1532 static int make_receive_sock(struct netns_ipvs *ipvs, int id,
1533 struct net_device *dev, struct socket **sock_ret)
1535 /* multicast addr */
1536 union ipvs_sockaddr mcast_addr;
1537 struct socket *sock;
1540 /* First create a socket */
1541 result = sock_create_kern(ipvs->net, ipvs->bcfg.mcast_af, SOCK_DGRAM,
1542 IPPROTO_UDP, &sock);
1544 pr_err("Error during creation of socket; terminating\n");
1548 /* it is equivalent to the REUSEADDR option in user-space */
1549 sock->sk->sk_reuse = SK_CAN_REUSE;
1550 result = sysctl_sync_sock_size(ipvs);
1552 set_sock_size(sock->sk, 0, result);
1554 get_mcast_sockaddr(&mcast_addr, &salen, &ipvs->bcfg, id);
1555 sock->sk->sk_bound_dev_if = dev->ifindex;
1556 result = sock->ops->bind(sock, (struct sockaddr *)&mcast_addr, salen);
1558 pr_err("Error binding to the multicast addr\n");
1562 /* join the multicast group */
1563 #ifdef CONFIG_IP_VS_IPV6
1564 if (ipvs->bcfg.mcast_af == AF_INET6)
1565 result = join_mcast_group6(sock->sk, &mcast_addr.in6.sin6_addr,
1569 result = join_mcast_group(sock->sk, &mcast_addr.in.sin_addr,
1572 pr_err("Error joining to the multicast group\n");
1584 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
1586 struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
1591 iov.iov_base = (void *)buffer;
1592 iov.iov_len = length;
1594 len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
1601 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
1606 msize = ntohs(msg->size);
1608 ret = ip_vs_send_async(sock, (char *)msg, msize);
1609 if (ret >= 0 || ret == -EAGAIN)
1611 pr_err("ip_vs_send_async error %d\n", ret);
1616 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
1618 struct msghdr msg = {NULL,};
1619 struct kvec iov = {buffer, buflen};
1624 /* Receive a packet */
1625 iov_iter_kvec(&msg.msg_iter, READ | ITER_KVEC, &iov, 1, buflen);
1626 len = sock_recvmsg(sock, &msg, MSG_DONTWAIT);
1634 /* Wakeup the master thread for sending */
1635 static void master_wakeup_work_handler(struct work_struct *work)
1637 struct ipvs_master_sync_state *ms =
1638 container_of(work, struct ipvs_master_sync_state,
1639 master_wakeup_work.work);
1640 struct netns_ipvs *ipvs = ms->ipvs;
1642 spin_lock_bh(&ipvs->sync_lock);
1643 if (ms->sync_queue_len &&
1644 ms->sync_queue_delay < IPVS_SYNC_WAKEUP_RATE) {
1645 ms->sync_queue_delay = IPVS_SYNC_WAKEUP_RATE;
1646 wake_up_process(ms->master_thread);
1648 spin_unlock_bh(&ipvs->sync_lock);
1651 /* Get next buffer to send */
1652 static inline struct ip_vs_sync_buff *
1653 next_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
1655 struct ip_vs_sync_buff *sb;
1657 sb = sb_dequeue(ipvs, ms);
1660 /* Do not delay entries in buffer for more than 2 seconds */
1661 return get_curr_sync_buff(ipvs, ms, IPVS_SYNC_FLUSH_TIME);
1664 static int sync_thread_master(void *data)
1666 struct ip_vs_sync_thread_data *tinfo = data;
1667 struct netns_ipvs *ipvs = tinfo->ipvs;
1668 struct ipvs_master_sync_state *ms = &ipvs->ms[tinfo->id];
1669 struct sock *sk = tinfo->sock->sk;
1670 struct ip_vs_sync_buff *sb;
1672 pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
1673 "syncid = %d, id = %d\n",
1674 ipvs->mcfg.mcast_ifn, ipvs->mcfg.syncid, tinfo->id);
1677 sb = next_sync_buff(ipvs, ms);
1678 if (unlikely(kthread_should_stop()))
1681 schedule_timeout(IPVS_SYNC_CHECK_PERIOD);
1684 while (ip_vs_send_sync_msg(tinfo->sock, sb->mesg) < 0) {
1685 /* (Ab)use interruptible sleep to avoid increasing
1688 __wait_event_interruptible(*sk_sleep(sk),
1689 sock_writeable(sk) ||
1690 kthread_should_stop());
1691 if (unlikely(kthread_should_stop()))
1694 ip_vs_sync_buff_release(sb);
1698 __set_current_state(TASK_RUNNING);
1700 ip_vs_sync_buff_release(sb);
1702 /* clean up the sync_buff queue */
1703 while ((sb = sb_dequeue(ipvs, ms)))
1704 ip_vs_sync_buff_release(sb);
1705 __set_current_state(TASK_RUNNING);
1707 /* clean up the current sync_buff */
1708 sb = get_curr_sync_buff(ipvs, ms, 0);
1710 ip_vs_sync_buff_release(sb);
1712 /* release the sending multicast socket */
1713 sock_release(tinfo->sock);
1720 static int sync_thread_backup(void *data)
1722 struct ip_vs_sync_thread_data *tinfo = data;
1723 struct netns_ipvs *ipvs = tinfo->ipvs;
1726 pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
1727 "syncid = %d, id = %d\n",
1728 ipvs->bcfg.mcast_ifn, ipvs->bcfg.syncid, tinfo->id);
1730 while (!kthread_should_stop()) {
1731 wait_event_interruptible(*sk_sleep(tinfo->sock->sk),
1732 !skb_queue_empty(&tinfo->sock->sk->sk_receive_queue)
1733 || kthread_should_stop());
1735 /* do we have data now? */
1736 while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) {
1737 len = ip_vs_receive(tinfo->sock, tinfo->buf,
1738 ipvs->bcfg.sync_maxlen);
1741 pr_err("receiving message error\n");
1745 ip_vs_process_message(ipvs, tinfo->buf, len);
1749 /* release the sending multicast socket */
1750 sock_release(tinfo->sock);
1758 int start_sync_thread(struct netns_ipvs *ipvs, struct ipvs_sync_daemon_cfg *c,
1761 struct ip_vs_sync_thread_data *tinfo = NULL;
1762 struct task_struct **array = NULL, *task;
1763 struct net_device *dev;
1765 int (*threadfn)(void *data);
1766 int id = 0, count, hlen;
1767 int result = -ENOMEM;
1770 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1771 IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %zd bytes\n",
1772 sizeof(struct ip_vs_sync_conn_v0));
1774 /* Do not hold one mutex and then to block on another */
1777 if (mutex_trylock(&ipvs->sync_mutex))
1780 mutex_lock(&ipvs->sync_mutex);
1783 mutex_unlock(&ipvs->sync_mutex);
1786 if (!ipvs->sync_state) {
1787 count = clamp(sysctl_sync_ports(ipvs), 1, IPVS_SYNC_PORTS_MAX);
1788 ipvs->threads_mask = count - 1;
1790 count = ipvs->threads_mask + 1;
1792 if (c->mcast_af == AF_UNSPEC) {
1793 c->mcast_af = AF_INET;
1794 c->mcast_group.ip = cpu_to_be32(IP_VS_SYNC_GROUP);
1797 c->mcast_port = IP_VS_SYNC_PORT;
1801 dev = __dev_get_by_name(ipvs->net, c->mcast_ifn);
1803 pr_err("Unknown mcast interface: %s\n", c->mcast_ifn);
1807 hlen = (AF_INET6 == c->mcast_af) ?
1808 sizeof(struct ipv6hdr) + sizeof(struct udphdr) :
1809 sizeof(struct iphdr) + sizeof(struct udphdr);
1810 mtu = (state == IP_VS_STATE_BACKUP) ?
1811 clamp(dev->mtu, 1500U, 65535U) : 1500U;
1812 min_mtu = (state == IP_VS_STATE_BACKUP) ? 1024 : 1;
1815 c->sync_maxlen = clamp_t(unsigned int,
1816 c->sync_maxlen, min_mtu,
1819 c->sync_maxlen = mtu - hlen;
1821 if (state == IP_VS_STATE_MASTER) {
1827 name = "ipvs-m:%d:%d";
1828 threadfn = sync_thread_master;
1829 } else if (state == IP_VS_STATE_BACKUP) {
1831 if (ipvs->backup_threads)
1835 name = "ipvs-b:%d:%d";
1836 threadfn = sync_thread_backup;
1842 if (state == IP_VS_STATE_MASTER) {
1843 struct ipvs_master_sync_state *ms;
1846 ipvs->ms = kcalloc(count, sizeof(ipvs->ms[0]), GFP_KERNEL);
1850 for (id = 0; id < count; id++, ms++) {
1851 INIT_LIST_HEAD(&ms->sync_queue);
1852 ms->sync_queue_len = 0;
1853 ms->sync_queue_delay = 0;
1854 INIT_DELAYED_WORK(&ms->master_wakeup_work,
1855 master_wakeup_work_handler);
1859 array = kcalloc(count, sizeof(struct task_struct *),
1866 for (id = 0; id < count; id++) {
1868 tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL);
1873 if (state == IP_VS_STATE_BACKUP) {
1874 tinfo->buf = kmalloc(ipvs->bcfg.sync_maxlen,
1882 if (state == IP_VS_STATE_MASTER)
1883 result = make_send_sock(ipvs, id, dev, &tinfo->sock);
1885 result = make_receive_sock(ipvs, id, dev, &tinfo->sock);
1889 task = kthread_run(threadfn, tinfo, name, ipvs->gen, id);
1891 result = PTR_ERR(task);
1895 if (state == IP_VS_STATE_MASTER)
1896 ipvs->ms[id].master_thread = task;
1901 /* mark as active */
1903 if (state == IP_VS_STATE_BACKUP)
1904 ipvs->backup_threads = array;
1905 spin_lock_bh(&ipvs->sync_buff_lock);
1906 ipvs->sync_state |= state;
1907 spin_unlock_bh(&ipvs->sync_buff_lock);
1909 mutex_unlock(&ipvs->sync_mutex);
1912 /* increase the module use count */
1913 ip_vs_use_count_inc();
1918 /* We do not need RTNL lock anymore, release it here so that
1919 * sock_release below and in the kthreads can use rtnl_lock
1920 * to leave the mcast group.
1924 while (count-- > 0) {
1925 if (state == IP_VS_STATE_MASTER)
1926 kthread_stop(ipvs->ms[count].master_thread);
1928 kthread_stop(array[count]);
1930 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
1934 mutex_unlock(&ipvs->sync_mutex);
1937 sock_release(tinfo->sock);
1945 mutex_unlock(&ipvs->sync_mutex);
1951 int stop_sync_thread(struct netns_ipvs *ipvs, int state)
1953 struct task_struct **array;
1957 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1959 if (state == IP_VS_STATE_MASTER) {
1964 * The lock synchronizes with sb_queue_tail(), so that we don't
1965 * add sync buffers to the queue, when we are already in
1966 * progress of stopping the master sync daemon.
1969 spin_lock_bh(&ipvs->sync_buff_lock);
1970 spin_lock(&ipvs->sync_lock);
1971 ipvs->sync_state &= ~IP_VS_STATE_MASTER;
1972 spin_unlock(&ipvs->sync_lock);
1973 spin_unlock_bh(&ipvs->sync_buff_lock);
1976 for (id = ipvs->threads_mask; id >= 0; id--) {
1977 struct ipvs_master_sync_state *ms = &ipvs->ms[id];
1980 pr_info("stopping master sync thread %d ...\n",
1981 task_pid_nr(ms->master_thread));
1982 cancel_delayed_work_sync(&ms->master_wakeup_work);
1983 ret = kthread_stop(ms->master_thread);
1989 } else if (state == IP_VS_STATE_BACKUP) {
1990 if (!ipvs->backup_threads)
1993 ipvs->sync_state &= ~IP_VS_STATE_BACKUP;
1994 array = ipvs->backup_threads;
1996 for (id = ipvs->threads_mask; id >= 0; id--) {
1999 pr_info("stopping backup sync thread %d ...\n",
2000 task_pid_nr(array[id]));
2001 ret = kthread_stop(array[id]);
2006 ipvs->backup_threads = NULL;
2009 /* decrease the module use count */
2010 ip_vs_use_count_dec();
2016 * Initialize data struct for each netns
2018 int __net_init ip_vs_sync_net_init(struct netns_ipvs *ipvs)
2020 __mutex_init(&ipvs->sync_mutex, "ipvs->sync_mutex", &__ipvs_sync_key);
2021 spin_lock_init(&ipvs->sync_lock);
2022 spin_lock_init(&ipvs->sync_buff_lock);
2026 void ip_vs_sync_net_cleanup(struct netns_ipvs *ipvs)
2030 mutex_lock(&ipvs->sync_mutex);
2031 retc = stop_sync_thread(ipvs, IP_VS_STATE_MASTER);
2032 if (retc && retc != -ESRCH)
2033 pr_err("Failed to stop Master Daemon\n");
2035 retc = stop_sync_thread(ipvs, IP_VS_STATE_BACKUP);
2036 if (retc && retc != -ESRCH)
2037 pr_err("Failed to stop Backup Daemon\n");
2038 mutex_unlock(&ipvs->sync_mutex);