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 task_struct *task;
199 struct netns_ipvs *ipvs;
205 /* Version 0 definition of packet sizes */
206 #define SIMPLE_CONN_SIZE (sizeof(struct ip_vs_sync_conn_v0))
207 #define FULL_CONN_SIZE \
208 (sizeof(struct ip_vs_sync_conn_v0) + sizeof(struct ip_vs_sync_conn_options))
212 The master mulitcasts messages (Datagrams) to the backup load balancers
213 in the following format.
216 Note, first byte should be Zero, so ver 0 receivers will drop the packet.
219 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
220 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
221 | 0 | SyncID | Size |
222 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
223 | Count Conns | Version | Reserved, set to Zero |
224 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
226 | IPVS Sync Connection (1) |
227 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
231 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
233 | IPVS Sync Connection (n) |
234 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
238 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
239 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
240 | Count Conns | SyncID | Size |
241 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
242 | IPVS Sync Connection (1) |
245 #define SYNC_MESG_HEADER_LEN 4
246 #define MAX_CONNS_PER_SYNCBUFF 255 /* nr_conns in ip_vs_sync_mesg is 8 bit */
248 /* Version 0 header */
249 struct ip_vs_sync_mesg_v0 {
254 /* ip_vs_sync_conn entries start here */
257 /* Version 1 header */
258 struct ip_vs_sync_mesg {
259 __u8 reserved; /* must be zero */
263 __s8 version; /* SYNC_PROTO_VER */
265 /* ip_vs_sync_conn entries start here */
268 union ipvs_sockaddr {
269 struct sockaddr_in in;
270 struct sockaddr_in6 in6;
273 struct ip_vs_sync_buff {
274 struct list_head list;
275 unsigned long firstuse;
277 /* pointers for the message data */
278 struct ip_vs_sync_mesg *mesg;
284 * Copy of struct ip_vs_seq
285 * From unaligned network order to aligned host order
287 static void ntoh_seq(struct ip_vs_seq *no, struct ip_vs_seq *ho)
289 memset(ho, 0, sizeof(*ho));
290 ho->init_seq = get_unaligned_be32(&no->init_seq);
291 ho->delta = get_unaligned_be32(&no->delta);
292 ho->previous_delta = get_unaligned_be32(&no->previous_delta);
296 * Copy of struct ip_vs_seq
297 * From Aligned host order to unaligned network order
299 static void hton_seq(struct ip_vs_seq *ho, struct ip_vs_seq *no)
301 put_unaligned_be32(ho->init_seq, &no->init_seq);
302 put_unaligned_be32(ho->delta, &no->delta);
303 put_unaligned_be32(ho->previous_delta, &no->previous_delta);
306 static inline struct ip_vs_sync_buff *
307 sb_dequeue(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
309 struct ip_vs_sync_buff *sb;
311 spin_lock_bh(&ipvs->sync_lock);
312 if (list_empty(&ms->sync_queue)) {
314 __set_current_state(TASK_INTERRUPTIBLE);
316 sb = list_entry(ms->sync_queue.next, struct ip_vs_sync_buff,
319 ms->sync_queue_len--;
320 if (!ms->sync_queue_len)
321 ms->sync_queue_delay = 0;
323 spin_unlock_bh(&ipvs->sync_lock);
329 * Create a new sync buffer for Version 1 proto.
331 static inline struct ip_vs_sync_buff *
332 ip_vs_sync_buff_create(struct netns_ipvs *ipvs, unsigned int len)
334 struct ip_vs_sync_buff *sb;
336 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
339 len = max_t(unsigned int, len + sizeof(struct ip_vs_sync_mesg),
340 ipvs->mcfg.sync_maxlen);
341 sb->mesg = kmalloc(len, GFP_ATOMIC);
346 sb->mesg->reserved = 0; /* old nr_conns i.e. must be zero now */
347 sb->mesg->version = SYNC_PROTO_VER;
348 sb->mesg->syncid = ipvs->mcfg.syncid;
349 sb->mesg->size = htons(sizeof(struct ip_vs_sync_mesg));
350 sb->mesg->nr_conns = 0;
352 sb->head = (unsigned char *)sb->mesg + sizeof(struct ip_vs_sync_mesg);
353 sb->end = (unsigned char *)sb->mesg + len;
355 sb->firstuse = jiffies;
359 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
365 static inline void sb_queue_tail(struct netns_ipvs *ipvs,
366 struct ipvs_master_sync_state *ms)
368 struct ip_vs_sync_buff *sb = ms->sync_buff;
370 spin_lock(&ipvs->sync_lock);
371 if (ipvs->sync_state & IP_VS_STATE_MASTER &&
372 ms->sync_queue_len < sysctl_sync_qlen_max(ipvs)) {
373 if (!ms->sync_queue_len)
374 schedule_delayed_work(&ms->master_wakeup_work,
375 max(IPVS_SYNC_SEND_DELAY, 1));
376 ms->sync_queue_len++;
377 list_add_tail(&sb->list, &ms->sync_queue);
378 if ((++ms->sync_queue_delay) == IPVS_SYNC_WAKEUP_RATE) {
379 int id = (int)(ms - ipvs->ms);
381 wake_up_process(ipvs->master_tinfo[id].task);
384 ip_vs_sync_buff_release(sb);
385 spin_unlock(&ipvs->sync_lock);
389 * Get the current sync buffer if it has been created for more
390 * than the specified time or the specified time is zero.
392 static inline struct ip_vs_sync_buff *
393 get_curr_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms,
396 struct ip_vs_sync_buff *sb;
398 spin_lock_bh(&ipvs->sync_buff_lock);
400 if (sb && time_after_eq(jiffies - sb->firstuse, time)) {
401 ms->sync_buff = NULL;
402 __set_current_state(TASK_RUNNING);
405 spin_unlock_bh(&ipvs->sync_buff_lock);
410 select_master_thread_id(struct netns_ipvs *ipvs, struct ip_vs_conn *cp)
412 return ((long) cp >> (1 + ilog2(sizeof(*cp)))) & ipvs->threads_mask;
416 * Create a new sync buffer for Version 0 proto.
418 static inline struct ip_vs_sync_buff *
419 ip_vs_sync_buff_create_v0(struct netns_ipvs *ipvs, unsigned int len)
421 struct ip_vs_sync_buff *sb;
422 struct ip_vs_sync_mesg_v0 *mesg;
424 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
427 len = max_t(unsigned int, len + sizeof(struct ip_vs_sync_mesg_v0),
428 ipvs->mcfg.sync_maxlen);
429 sb->mesg = kmalloc(len, GFP_ATOMIC);
434 mesg = (struct ip_vs_sync_mesg_v0 *)sb->mesg;
436 mesg->syncid = ipvs->mcfg.syncid;
437 mesg->size = htons(sizeof(struct ip_vs_sync_mesg_v0));
438 sb->head = (unsigned char *)mesg + sizeof(struct ip_vs_sync_mesg_v0);
439 sb->end = (unsigned char *)mesg + len;
440 sb->firstuse = jiffies;
444 /* Check if connection is controlled by persistence */
445 static inline bool in_persistence(struct ip_vs_conn *cp)
447 for (cp = cp->control; cp; cp = cp->control) {
448 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
454 /* Check if conn should be synced.
455 * pkts: conn packets, use sysctl_sync_threshold to avoid packet check
456 * - (1) sync_refresh_period: reduce sync rate. Additionally, retry
457 * sync_retries times with period of sync_refresh_period/8
458 * - (2) if both sync_refresh_period and sync_period are 0 send sync only
459 * for state changes or only once when pkts matches sync_threshold
460 * - (3) templates: rate can be reduced only with sync_refresh_period or
463 static int ip_vs_sync_conn_needed(struct netns_ipvs *ipvs,
464 struct ip_vs_conn *cp, int pkts)
466 unsigned long orig = READ_ONCE(cp->sync_endtime);
467 unsigned long now = jiffies;
468 unsigned long n = (now + cp->timeout) & ~3UL;
469 unsigned int sync_refresh_period;
473 /* Check if we sync in current state */
474 if (unlikely(cp->flags & IP_VS_CONN_F_TEMPLATE))
476 else if (unlikely(sysctl_sync_persist_mode(ipvs) && in_persistence(cp)))
478 else if (likely(cp->protocol == IPPROTO_TCP)) {
479 if (!((1 << cp->state) &
480 ((1 << IP_VS_TCP_S_ESTABLISHED) |
481 (1 << IP_VS_TCP_S_FIN_WAIT) |
482 (1 << IP_VS_TCP_S_CLOSE) |
483 (1 << IP_VS_TCP_S_CLOSE_WAIT) |
484 (1 << IP_VS_TCP_S_TIME_WAIT))))
486 force = cp->state != cp->old_state;
487 if (force && cp->state != IP_VS_TCP_S_ESTABLISHED)
489 } else if (unlikely(cp->protocol == IPPROTO_SCTP)) {
490 if (!((1 << cp->state) &
491 ((1 << IP_VS_SCTP_S_ESTABLISHED) |
492 (1 << IP_VS_SCTP_S_SHUTDOWN_SENT) |
493 (1 << IP_VS_SCTP_S_SHUTDOWN_RECEIVED) |
494 (1 << IP_VS_SCTP_S_SHUTDOWN_ACK_SENT) |
495 (1 << IP_VS_SCTP_S_CLOSED))))
497 force = cp->state != cp->old_state;
498 if (force && cp->state != IP_VS_SCTP_S_ESTABLISHED)
501 /* UDP or another protocol with single state */
505 sync_refresh_period = sysctl_sync_refresh_period(ipvs);
506 if (sync_refresh_period > 0) {
507 long diff = n - orig;
508 long min_diff = max(cp->timeout >> 1, 10UL * HZ);
510 /* Avoid sync if difference is below sync_refresh_period
511 * and below the half timeout.
513 if (abs(diff) < min_t(long, sync_refresh_period, min_diff)) {
514 int retries = orig & 3;
516 if (retries >= sysctl_sync_retries(ipvs))
518 if (time_before(now, orig - cp->timeout +
519 (sync_refresh_period >> 3)))
524 sync_period = sysctl_sync_period(ipvs);
525 if (sync_period > 0) {
526 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE) &&
527 pkts % sync_period != sysctl_sync_threshold(ipvs))
529 } else if (!sync_refresh_period &&
530 pkts != sysctl_sync_threshold(ipvs))
534 cp->old_state = cp->state;
535 n = cmpxchg(&cp->sync_endtime, orig, n);
536 return n == orig || force;
540 * Version 0 , could be switched in by sys_ctl.
541 * Add an ip_vs_conn information into the current sync_buff.
543 static void ip_vs_sync_conn_v0(struct netns_ipvs *ipvs, struct ip_vs_conn *cp,
546 struct ip_vs_sync_mesg_v0 *m;
547 struct ip_vs_sync_conn_v0 *s;
548 struct ip_vs_sync_buff *buff;
549 struct ipvs_master_sync_state *ms;
553 if (unlikely(cp->af != AF_INET))
555 /* Do not sync ONE PACKET */
556 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
559 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
562 spin_lock_bh(&ipvs->sync_buff_lock);
563 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
564 spin_unlock_bh(&ipvs->sync_buff_lock);
568 id = select_master_thread_id(ipvs, cp);
570 buff = ms->sync_buff;
571 len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
574 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
575 /* Send buffer if it is for v1 */
576 if (buff->head + len > buff->end || !m->nr_conns) {
577 sb_queue_tail(ipvs, ms);
578 ms->sync_buff = NULL;
583 buff = ip_vs_sync_buff_create_v0(ipvs, len);
585 spin_unlock_bh(&ipvs->sync_buff_lock);
586 pr_err("ip_vs_sync_buff_create failed.\n");
589 ms->sync_buff = buff;
592 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
593 s = (struct ip_vs_sync_conn_v0 *) buff->head;
597 s->protocol = cp->protocol;
598 s->cport = cp->cport;
599 s->vport = cp->vport;
600 s->dport = cp->dport;
601 s->caddr = cp->caddr.ip;
602 s->vaddr = cp->vaddr.ip;
603 s->daddr = cp->daddr.ip;
604 s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
605 s->state = htons(cp->state);
606 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
607 struct ip_vs_sync_conn_options *opt =
608 (struct ip_vs_sync_conn_options *)&s[1];
609 memcpy(opt, &cp->in_seq, sizeof(*opt));
613 m->size = htons(ntohs(m->size) + len);
615 spin_unlock_bh(&ipvs->sync_buff_lock);
617 /* synchronize its controller if it has */
620 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
621 pkts = atomic_add_return(1, &cp->in_pkts);
623 pkts = sysctl_sync_threshold(ipvs);
624 ip_vs_sync_conn(ipvs, cp, pkts);
629 * Add an ip_vs_conn information into the current sync_buff.
630 * Called by ip_vs_in.
631 * Sending Version 1 messages
633 void ip_vs_sync_conn(struct netns_ipvs *ipvs, struct ip_vs_conn *cp, int pkts)
635 struct ip_vs_sync_mesg *m;
636 union ip_vs_sync_conn *s;
637 struct ip_vs_sync_buff *buff;
638 struct ipvs_master_sync_state *ms;
641 unsigned int len, pe_name_len, pad;
643 /* Handle old version of the protocol */
644 if (sysctl_sync_ver(ipvs) == 0) {
645 ip_vs_sync_conn_v0(ipvs, cp, pkts);
648 /* Do not sync ONE PACKET */
649 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
652 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
657 if (cp->pe_data_len) {
658 if (!cp->pe_data || !cp->dest) {
659 IP_VS_ERR_RL("SYNC, connection pe_data invalid\n");
662 pe_name_len = strnlen(cp->pe->name, IP_VS_PENAME_MAXLEN);
665 spin_lock_bh(&ipvs->sync_buff_lock);
666 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
667 spin_unlock_bh(&ipvs->sync_buff_lock);
671 id = select_master_thread_id(ipvs, cp);
674 #ifdef CONFIG_IP_VS_IPV6
675 if (cp->af == AF_INET6)
676 len = sizeof(struct ip_vs_sync_v6);
679 len = sizeof(struct ip_vs_sync_v4);
681 if (cp->flags & IP_VS_CONN_F_SEQ_MASK)
682 len += sizeof(struct ip_vs_sync_conn_options) + 2;
685 len += cp->pe_data_len + 2; /* + Param hdr field */
687 len += pe_name_len + 2;
689 /* check if there is a space for this one */
691 buff = ms->sync_buff;
694 pad = (4 - (size_t) buff->head) & 3;
695 /* Send buffer if it is for v0 */
696 if (buff->head + len + pad > buff->end || m->reserved) {
697 sb_queue_tail(ipvs, ms);
698 ms->sync_buff = NULL;
705 buff = ip_vs_sync_buff_create(ipvs, len);
707 spin_unlock_bh(&ipvs->sync_buff_lock);
708 pr_err("ip_vs_sync_buff_create failed.\n");
711 ms->sync_buff = buff;
716 buff->head += pad + len;
717 m->size = htons(ntohs(m->size) + pad + len);
718 /* Add ev. padding from prev. sync_conn */
722 s = (union ip_vs_sync_conn *)p;
724 /* Set message type & copy members */
725 s->v4.type = (cp->af == AF_INET6 ? STYPE_F_INET6 : 0);
726 s->v4.ver_size = htons(len & SVER_MASK); /* Version 0 */
727 s->v4.flags = htonl(cp->flags & ~IP_VS_CONN_F_HASHED);
728 s->v4.state = htons(cp->state);
729 s->v4.protocol = cp->protocol;
730 s->v4.cport = cp->cport;
731 s->v4.vport = cp->vport;
732 s->v4.dport = cp->dport;
733 s->v4.fwmark = htonl(cp->fwmark);
734 s->v4.timeout = htonl(cp->timeout / HZ);
737 #ifdef CONFIG_IP_VS_IPV6
738 if (cp->af == AF_INET6) {
739 p += sizeof(struct ip_vs_sync_v6);
740 s->v6.caddr = cp->caddr.in6;
741 s->v6.vaddr = cp->vaddr.in6;
742 s->v6.daddr = cp->daddr.in6;
746 p += sizeof(struct ip_vs_sync_v4); /* options ptr */
747 s->v4.caddr = cp->caddr.ip;
748 s->v4.vaddr = cp->vaddr.ip;
749 s->v4.daddr = cp->daddr.ip;
751 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
752 *(p++) = IPVS_OPT_SEQ_DATA;
753 *(p++) = sizeof(struct ip_vs_sync_conn_options);
754 hton_seq((struct ip_vs_seq *)p, &cp->in_seq);
755 p += sizeof(struct ip_vs_seq);
756 hton_seq((struct ip_vs_seq *)p, &cp->out_seq);
757 p += sizeof(struct ip_vs_seq);
760 if (cp->pe_data_len && cp->pe_data) {
761 *(p++) = IPVS_OPT_PE_DATA;
762 *(p++) = cp->pe_data_len;
763 memcpy(p, cp->pe_data, cp->pe_data_len);
764 p += cp->pe_data_len;
767 *(p++) = IPVS_OPT_PE_NAME;
768 *(p++) = pe_name_len;
769 memcpy(p, cp->pe->name, pe_name_len);
774 spin_unlock_bh(&ipvs->sync_buff_lock);
777 /* synchronize its controller if it has */
781 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
782 pkts = atomic_add_return(1, &cp->in_pkts);
784 pkts = sysctl_sync_threshold(ipvs);
789 * fill_param used by version 1
792 ip_vs_conn_fill_param_sync(struct netns_ipvs *ipvs, int af, union ip_vs_sync_conn *sc,
793 struct ip_vs_conn_param *p,
794 __u8 *pe_data, unsigned int pe_data_len,
795 __u8 *pe_name, unsigned int pe_name_len)
797 #ifdef CONFIG_IP_VS_IPV6
799 ip_vs_conn_fill_param(ipvs, af, sc->v6.protocol,
800 (const union nf_inet_addr *)&sc->v6.caddr,
802 (const union nf_inet_addr *)&sc->v6.vaddr,
806 ip_vs_conn_fill_param(ipvs, af, sc->v4.protocol,
807 (const union nf_inet_addr *)&sc->v4.caddr,
809 (const union nf_inet_addr *)&sc->v4.vaddr,
814 char buff[IP_VS_PENAME_MAXLEN+1];
816 memcpy(buff, pe_name, pe_name_len);
818 p->pe = __ip_vs_pe_getbyname(buff);
820 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n",
825 IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n");
829 p->pe_data = kmemdup(pe_data, pe_data_len, GFP_ATOMIC);
831 module_put(p->pe->module);
834 p->pe_data_len = pe_data_len;
840 * Connection Add / Update.
841 * Common for version 0 and 1 reception of backup sync_conns.
845 static void ip_vs_proc_conn(struct netns_ipvs *ipvs, struct ip_vs_conn_param *param,
846 unsigned int flags, unsigned int state,
847 unsigned int protocol, unsigned int type,
848 const union nf_inet_addr *daddr, __be16 dport,
849 unsigned long timeout, __u32 fwmark,
850 struct ip_vs_sync_conn_options *opt)
852 struct ip_vs_dest *dest;
853 struct ip_vs_conn *cp;
855 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
856 cp = ip_vs_conn_in_get(param);
857 if (cp && ((cp->dport != dport) ||
858 !ip_vs_addr_equal(cp->daf, &cp->daddr, daddr))) {
859 if (!(flags & IP_VS_CONN_F_INACTIVE)) {
860 ip_vs_conn_expire_now(cp);
861 __ip_vs_conn_put(cp);
864 /* This is the expiration message for the
865 * connection that was already replaced, so we
868 __ip_vs_conn_put(cp);
869 kfree(param->pe_data);
874 cp = ip_vs_ct_in_get(param);
879 kfree(param->pe_data);
882 spin_lock_bh(&cp->lock);
883 if ((cp->flags ^ flags) & IP_VS_CONN_F_INACTIVE &&
884 !(flags & IP_VS_CONN_F_TEMPLATE) && dest) {
885 if (flags & IP_VS_CONN_F_INACTIVE) {
886 atomic_dec(&dest->activeconns);
887 atomic_inc(&dest->inactconns);
889 atomic_inc(&dest->activeconns);
890 atomic_dec(&dest->inactconns);
893 flags &= IP_VS_CONN_F_BACKUP_UPD_MASK;
894 flags |= cp->flags & ~IP_VS_CONN_F_BACKUP_UPD_MASK;
896 spin_unlock_bh(&cp->lock);
898 ip_vs_try_bind_dest(cp);
901 * Find the appropriate destination for the connection.
902 * If it is not found the connection will remain unbound
906 /* This function is only invoked by the synchronization
907 * code. We do not currently support heterogeneous pools
908 * with synchronization, so we can make the assumption that
909 * the svc_af is the same as the dest_af
911 dest = ip_vs_find_dest(ipvs, type, type, daddr, dport,
912 param->vaddr, param->vport, protocol,
915 cp = ip_vs_conn_new(param, type, daddr, dport, flags, dest,
919 kfree(param->pe_data);
920 IP_VS_DBG(2, "BACKUP, add new conn. failed\n");
923 if (!(flags & IP_VS_CONN_F_TEMPLATE))
924 kfree(param->pe_data);
928 cp->in_seq = opt->in_seq;
929 cp->out_seq = opt->out_seq;
931 atomic_set(&cp->in_pkts, sysctl_sync_threshold(ipvs));
933 cp->old_state = cp->state;
935 * For Ver 0 messages style
936 * - Not possible to recover the right timeout for templates
937 * - can not find the right fwmark
938 * virtual service. If needed, we can do it for
939 * non-fwmark persistent services.
940 * Ver 1 messages style.
944 if (timeout > MAX_SCHEDULE_TIMEOUT / HZ)
945 timeout = MAX_SCHEDULE_TIMEOUT / HZ;
946 cp->timeout = timeout*HZ;
948 struct ip_vs_proto_data *pd;
950 pd = ip_vs_proto_data_get(ipvs, protocol);
951 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pd && pd->timeout_table)
952 cp->timeout = pd->timeout_table[state];
954 cp->timeout = (3*60*HZ);
960 * Process received multicast message for Version 0
962 static void ip_vs_process_message_v0(struct netns_ipvs *ipvs, const char *buffer,
965 struct ip_vs_sync_mesg_v0 *m = (struct ip_vs_sync_mesg_v0 *)buffer;
966 struct ip_vs_sync_conn_v0 *s;
967 struct ip_vs_sync_conn_options *opt;
968 struct ip_vs_protocol *pp;
969 struct ip_vs_conn_param param;
973 p = (char *)buffer + sizeof(struct ip_vs_sync_mesg_v0);
974 for (i=0; i<m->nr_conns; i++) {
975 unsigned int flags, state;
977 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
978 IP_VS_ERR_RL("BACKUP v0, bogus conn\n");
981 s = (struct ip_vs_sync_conn_v0 *) p;
982 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
983 flags &= ~IP_VS_CONN_F_HASHED;
984 if (flags & IP_VS_CONN_F_SEQ_MASK) {
985 opt = (struct ip_vs_sync_conn_options *)&s[1];
987 if (p > buffer+buflen) {
988 IP_VS_ERR_RL("BACKUP v0, Dropping buffer bogus conn options\n");
993 p += SIMPLE_CONN_SIZE;
996 state = ntohs(s->state);
997 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
998 pp = ip_vs_proto_get(s->protocol);
1000 IP_VS_DBG(2, "BACKUP v0, Unsupported protocol %u\n",
1004 if (state >= pp->num_states) {
1005 IP_VS_DBG(2, "BACKUP v0, Invalid %s state %u\n",
1010 if (state >= IP_VS_CTPL_S_LAST)
1011 IP_VS_DBG(7, "BACKUP v0, Invalid tpl state %u\n",
1015 ip_vs_conn_fill_param(ipvs, AF_INET, s->protocol,
1016 (const union nf_inet_addr *)&s->caddr,
1018 (const union nf_inet_addr *)&s->vaddr,
1021 /* Send timeout as Zero */
1022 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->protocol, AF_INET,
1023 (union nf_inet_addr *)&s->daddr, s->dport,
1031 static inline int ip_vs_proc_seqopt(__u8 *p, unsigned int plen,
1033 struct ip_vs_sync_conn_options *opt)
1035 struct ip_vs_sync_conn_options *topt;
1037 topt = (struct ip_vs_sync_conn_options *)p;
1039 if (plen != sizeof(struct ip_vs_sync_conn_options)) {
1040 IP_VS_DBG(2, "BACKUP, bogus conn options length\n");
1043 if (*opt_flags & IPVS_OPT_F_SEQ_DATA) {
1044 IP_VS_DBG(2, "BACKUP, conn options found twice\n");
1047 ntoh_seq(&topt->in_seq, &opt->in_seq);
1048 ntoh_seq(&topt->out_seq, &opt->out_seq);
1049 *opt_flags |= IPVS_OPT_F_SEQ_DATA;
1053 static int ip_vs_proc_str(__u8 *p, unsigned int plen, unsigned int *data_len,
1054 __u8 **data, unsigned int maxlen,
1055 __u32 *opt_flags, __u32 flag)
1057 if (plen > maxlen) {
1058 IP_VS_DBG(2, "BACKUP, bogus par.data len > %d\n", maxlen);
1061 if (*opt_flags & flag) {
1062 IP_VS_DBG(2, "BACKUP, Par.data found twice 0x%x\n", flag);
1071 * Process a Version 1 sync. connection
1073 static inline int ip_vs_proc_sync_conn(struct netns_ipvs *ipvs, __u8 *p, __u8 *msg_end)
1075 struct ip_vs_sync_conn_options opt;
1076 union ip_vs_sync_conn *s;
1077 struct ip_vs_protocol *pp;
1078 struct ip_vs_conn_param param;
1080 unsigned int af, state, pe_data_len=0, pe_name_len=0;
1081 __u8 *pe_data=NULL, *pe_name=NULL;
1085 s = (union ip_vs_sync_conn *) p;
1087 if (s->v6.type & STYPE_F_INET6) {
1088 #ifdef CONFIG_IP_VS_IPV6
1090 p += sizeof(struct ip_vs_sync_v6);
1092 IP_VS_DBG(3,"BACKUP, IPv6 msg received, and IPVS is not compiled for IPv6\n");
1096 } else if (!s->v4.type) {
1098 p += sizeof(struct ip_vs_sync_v4);
1105 /* Process optional params check Type & Len. */
1106 while (p < msg_end) {
1115 if (!plen || ((p + plen) > msg_end))
1117 /* Handle seq option p = param data */
1118 switch (ptype & ~IPVS_OPT_F_PARAM) {
1119 case IPVS_OPT_SEQ_DATA:
1120 if (ip_vs_proc_seqopt(p, plen, &opt_flags, &opt))
1124 case IPVS_OPT_PE_DATA:
1125 if (ip_vs_proc_str(p, plen, &pe_data_len, &pe_data,
1126 IP_VS_PEDATA_MAXLEN, &opt_flags,
1127 IPVS_OPT_F_PE_DATA))
1131 case IPVS_OPT_PE_NAME:
1132 if (ip_vs_proc_str(p, plen,&pe_name_len, &pe_name,
1133 IP_VS_PENAME_MAXLEN, &opt_flags,
1134 IPVS_OPT_F_PE_NAME))
1139 /* Param data mandatory ? */
1140 if (!(ptype & IPVS_OPT_F_PARAM)) {
1141 IP_VS_DBG(3, "BACKUP, Unknown mandatory param %d found\n",
1142 ptype & ~IPVS_OPT_F_PARAM);
1147 p += plen; /* Next option */
1150 /* Get flags and Mask off unsupported */
1151 flags = ntohl(s->v4.flags) & IP_VS_CONN_F_BACKUP_MASK;
1152 flags |= IP_VS_CONN_F_SYNC;
1153 state = ntohs(s->v4.state);
1155 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
1156 pp = ip_vs_proto_get(s->v4.protocol);
1158 IP_VS_DBG(3,"BACKUP, Unsupported protocol %u\n",
1163 if (state >= pp->num_states) {
1164 IP_VS_DBG(3, "BACKUP, Invalid %s state %u\n",
1170 if (state >= IP_VS_CTPL_S_LAST)
1171 IP_VS_DBG(7, "BACKUP, Invalid tpl state %u\n",
1174 if (ip_vs_conn_fill_param_sync(ipvs, af, s, ¶m, pe_data,
1175 pe_data_len, pe_name, pe_name_len)) {
1179 /* If only IPv4, just silent skip IPv6 */
1181 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->v4.protocol, af,
1182 (union nf_inet_addr *)&s->v4.daddr, s->v4.dport,
1183 ntohl(s->v4.timeout), ntohl(s->v4.fwmark),
1184 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1186 #ifdef CONFIG_IP_VS_IPV6
1188 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->v6.protocol, af,
1189 (union nf_inet_addr *)&s->v6.daddr, s->v6.dport,
1190 ntohl(s->v6.timeout), ntohl(s->v6.fwmark),
1191 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1194 ip_vs_pe_put(param.pe);
1198 IP_VS_DBG(2, "BACKUP, Single msg dropped err:%d\n", retc);
1203 * Process received multicast message and create the corresponding
1204 * ip_vs_conn entries.
1205 * Handles Version 0 & 1
1207 static void ip_vs_process_message(struct netns_ipvs *ipvs, __u8 *buffer,
1208 const size_t buflen)
1210 struct ip_vs_sync_mesg *m2 = (struct ip_vs_sync_mesg *)buffer;
1214 if (buflen < sizeof(struct ip_vs_sync_mesg_v0)) {
1215 IP_VS_DBG(2, "BACKUP, message header too short\n");
1219 if (buflen != ntohs(m2->size)) {
1220 IP_VS_DBG(2, "BACKUP, bogus message size\n");
1223 /* SyncID sanity check */
1224 if (ipvs->bcfg.syncid != 0 && m2->syncid != ipvs->bcfg.syncid) {
1225 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2->syncid);
1228 /* Handle version 1 message */
1229 if ((m2->version == SYNC_PROTO_VER) && (m2->reserved == 0)
1230 && (m2->spare == 0)) {
1232 msg_end = buffer + sizeof(struct ip_vs_sync_mesg);
1233 nr_conns = m2->nr_conns;
1235 for (i=0; i<nr_conns; i++) {
1236 union ip_vs_sync_conn *s;
1241 if (p + sizeof(s->v4) > buffer+buflen) {
1242 IP_VS_ERR_RL("BACKUP, Dropping buffer, to small\n");
1245 s = (union ip_vs_sync_conn *)p;
1246 size = ntohs(s->v4.ver_size) & SVER_MASK;
1248 /* Basic sanity checks */
1249 if (msg_end > buffer+buflen) {
1250 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n");
1253 if (ntohs(s->v4.ver_size) >> SVER_SHIFT) {
1254 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n",
1255 ntohs(s->v4.ver_size) >> SVER_SHIFT);
1258 /* Process a single sync_conn */
1259 retc = ip_vs_proc_sync_conn(ipvs, p, msg_end);
1261 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n",
1265 /* Make sure we have 32 bit alignment */
1266 msg_end = p + ((size + 3) & ~3);
1269 /* Old type of message */
1270 ip_vs_process_message_v0(ipvs, buffer, buflen);
1277 * Setup sndbuf (mode=1) or rcvbuf (mode=0)
1279 static void set_sock_size(struct sock *sk, int mode, int val)
1281 /* setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &val, sizeof(val)); */
1282 /* setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)); */
1285 val = clamp_t(int, val, (SOCK_MIN_SNDBUF + 1) / 2,
1287 sk->sk_sndbuf = val * 2;
1288 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1290 val = clamp_t(int, val, (SOCK_MIN_RCVBUF + 1) / 2,
1292 sk->sk_rcvbuf = val * 2;
1293 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1299 * Setup loopback of outgoing multicasts on a sending socket
1301 static void set_mcast_loop(struct sock *sk, u_char loop)
1303 struct inet_sock *inet = inet_sk(sk);
1305 /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
1307 inet->mc_loop = loop ? 1 : 0;
1308 #ifdef CONFIG_IP_VS_IPV6
1309 if (sk->sk_family == AF_INET6) {
1310 struct ipv6_pinfo *np = inet6_sk(sk);
1312 /* IPV6_MULTICAST_LOOP */
1313 np->mc_loop = loop ? 1 : 0;
1320 * Specify TTL for outgoing multicasts on a sending socket
1322 static void set_mcast_ttl(struct sock *sk, u_char ttl)
1324 struct inet_sock *inet = inet_sk(sk);
1326 /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
1329 #ifdef CONFIG_IP_VS_IPV6
1330 if (sk->sk_family == AF_INET6) {
1331 struct ipv6_pinfo *np = inet6_sk(sk);
1333 /* IPV6_MULTICAST_HOPS */
1334 np->mcast_hops = ttl;
1340 /* Control fragmentation of messages */
1341 static void set_mcast_pmtudisc(struct sock *sk, int val)
1343 struct inet_sock *inet = inet_sk(sk);
1345 /* setsockopt(sock, SOL_IP, IP_MTU_DISCOVER, &val, sizeof(val)); */
1347 inet->pmtudisc = val;
1348 #ifdef CONFIG_IP_VS_IPV6
1349 if (sk->sk_family == AF_INET6) {
1350 struct ipv6_pinfo *np = inet6_sk(sk);
1352 /* IPV6_MTU_DISCOVER */
1360 * Specifiy default interface for outgoing multicasts
1362 static int set_mcast_if(struct sock *sk, struct net_device *dev)
1364 struct inet_sock *inet = inet_sk(sk);
1366 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1370 inet->mc_index = dev->ifindex;
1371 /* inet->mc_addr = 0; */
1372 #ifdef CONFIG_IP_VS_IPV6
1373 if (sk->sk_family == AF_INET6) {
1374 struct ipv6_pinfo *np = inet6_sk(sk);
1376 /* IPV6_MULTICAST_IF */
1377 np->mcast_oif = dev->ifindex;
1387 * Join a multicast group.
1388 * the group is specified by a class D multicast address 224.0.0.0/8
1389 * in the in_addr structure passed in as a parameter.
1392 join_mcast_group(struct sock *sk, struct in_addr *addr, struct net_device *dev)
1394 struct ip_mreqn mreq;
1397 memset(&mreq, 0, sizeof(mreq));
1398 memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
1400 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1403 mreq.imr_ifindex = dev->ifindex;
1406 ret = ip_mc_join_group(sk, &mreq);
1412 #ifdef CONFIG_IP_VS_IPV6
1413 static int join_mcast_group6(struct sock *sk, struct in6_addr *addr,
1414 struct net_device *dev)
1418 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1422 ret = ipv6_sock_mc_join(sk, dev->ifindex, addr);
1429 static int bind_mcastif_addr(struct socket *sock, struct net_device *dev)
1432 struct sockaddr_in sin;
1434 addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
1436 pr_err("You probably need to specify IP address on "
1437 "multicast interface.\n");
1439 IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
1442 /* Now bind the socket with the address of multicast interface */
1443 sin.sin_family = AF_INET;
1444 sin.sin_addr.s_addr = addr;
1447 return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
1450 static void get_mcast_sockaddr(union ipvs_sockaddr *sa, int *salen,
1451 struct ipvs_sync_daemon_cfg *c, int id)
1453 if (AF_INET6 == c->mcast_af) {
1454 sa->in6 = (struct sockaddr_in6) {
1455 .sin6_family = AF_INET6,
1456 .sin6_port = htons(c->mcast_port + id),
1458 sa->in6.sin6_addr = c->mcast_group.in6;
1459 *salen = sizeof(sa->in6);
1461 sa->in = (struct sockaddr_in) {
1462 .sin_family = AF_INET,
1463 .sin_port = htons(c->mcast_port + id),
1465 sa->in.sin_addr = c->mcast_group.in;
1466 *salen = sizeof(sa->in);
1471 * Set up sending multicast socket over UDP
1473 static int make_send_sock(struct netns_ipvs *ipvs, int id,
1474 struct net_device *dev, struct socket **sock_ret)
1476 /* multicast addr */
1477 union ipvs_sockaddr mcast_addr;
1478 struct socket *sock;
1481 /* First create a socket */
1482 result = sock_create_kern(ipvs->net, ipvs->mcfg.mcast_af, SOCK_DGRAM,
1483 IPPROTO_UDP, &sock);
1485 pr_err("Error during creation of socket; terminating\n");
1489 result = set_mcast_if(sock->sk, dev);
1491 pr_err("Error setting outbound mcast interface\n");
1495 set_mcast_loop(sock->sk, 0);
1496 set_mcast_ttl(sock->sk, ipvs->mcfg.mcast_ttl);
1497 /* Allow fragmentation if MTU changes */
1498 set_mcast_pmtudisc(sock->sk, IP_PMTUDISC_DONT);
1499 result = sysctl_sync_sock_size(ipvs);
1501 set_sock_size(sock->sk, 1, result);
1503 if (AF_INET == ipvs->mcfg.mcast_af)
1504 result = bind_mcastif_addr(sock, dev);
1508 pr_err("Error binding address of the mcast interface\n");
1512 get_mcast_sockaddr(&mcast_addr, &salen, &ipvs->mcfg, id);
1513 result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr,
1516 pr_err("Error connecting to the multicast addr\n");
1528 * Set up receiving multicast socket over UDP
1530 static int make_receive_sock(struct netns_ipvs *ipvs, int id,
1531 struct net_device *dev, struct socket **sock_ret)
1533 /* multicast addr */
1534 union ipvs_sockaddr mcast_addr;
1535 struct socket *sock;
1538 /* First create a socket */
1539 result = sock_create_kern(ipvs->net, ipvs->bcfg.mcast_af, SOCK_DGRAM,
1540 IPPROTO_UDP, &sock);
1542 pr_err("Error during creation of socket; terminating\n");
1546 /* it is equivalent to the REUSEADDR option in user-space */
1547 sock->sk->sk_reuse = SK_CAN_REUSE;
1548 result = sysctl_sync_sock_size(ipvs);
1550 set_sock_size(sock->sk, 0, result);
1552 get_mcast_sockaddr(&mcast_addr, &salen, &ipvs->bcfg, id);
1553 sock->sk->sk_bound_dev_if = dev->ifindex;
1554 result = sock->ops->bind(sock, (struct sockaddr *)&mcast_addr, salen);
1556 pr_err("Error binding to the multicast addr\n");
1560 /* join the multicast group */
1561 #ifdef CONFIG_IP_VS_IPV6
1562 if (ipvs->bcfg.mcast_af == AF_INET6)
1563 result = join_mcast_group6(sock->sk, &mcast_addr.in6.sin6_addr,
1567 result = join_mcast_group(sock->sk, &mcast_addr.in.sin_addr,
1570 pr_err("Error joining to the multicast group\n");
1582 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
1584 struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
1589 iov.iov_base = (void *)buffer;
1590 iov.iov_len = length;
1592 len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
1599 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
1604 msize = ntohs(msg->size);
1606 ret = ip_vs_send_async(sock, (char *)msg, msize);
1607 if (ret >= 0 || ret == -EAGAIN)
1609 pr_err("ip_vs_send_async error %d\n", ret);
1614 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
1616 struct msghdr msg = {NULL,};
1617 struct kvec iov = {buffer, buflen};
1622 /* Receive a packet */
1623 iov_iter_kvec(&msg.msg_iter, READ | ITER_KVEC, &iov, 1, buflen);
1624 len = sock_recvmsg(sock, &msg, MSG_DONTWAIT);
1632 /* Wakeup the master thread for sending */
1633 static void master_wakeup_work_handler(struct work_struct *work)
1635 struct ipvs_master_sync_state *ms =
1636 container_of(work, struct ipvs_master_sync_state,
1637 master_wakeup_work.work);
1638 struct netns_ipvs *ipvs = ms->ipvs;
1640 spin_lock_bh(&ipvs->sync_lock);
1641 if (ms->sync_queue_len &&
1642 ms->sync_queue_delay < IPVS_SYNC_WAKEUP_RATE) {
1643 int id = (int)(ms - ipvs->ms);
1645 ms->sync_queue_delay = IPVS_SYNC_WAKEUP_RATE;
1646 wake_up_process(ipvs->master_tinfo[id].task);
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);
1716 static int sync_thread_backup(void *data)
1718 struct ip_vs_sync_thread_data *tinfo = data;
1719 struct netns_ipvs *ipvs = tinfo->ipvs;
1720 struct sock *sk = tinfo->sock->sk;
1721 struct udp_sock *up = udp_sk(sk);
1724 pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
1725 "syncid = %d, id = %d\n",
1726 ipvs->bcfg.mcast_ifn, ipvs->bcfg.syncid, tinfo->id);
1728 while (!kthread_should_stop()) {
1729 wait_event_interruptible(*sk_sleep(sk),
1730 !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
1731 !skb_queue_empty_lockless(&up->reader_queue) ||
1732 kthread_should_stop());
1734 /* do we have data now? */
1735 while (!skb_queue_empty_lockless(&sk->sk_receive_queue) ||
1736 !skb_queue_empty_lockless(&up->reader_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);
1753 int start_sync_thread(struct netns_ipvs *ipvs, struct ipvs_sync_daemon_cfg *c,
1756 struct ip_vs_sync_thread_data *ti = NULL, *tinfo;
1757 struct task_struct *task;
1758 struct net_device *dev;
1760 int (*threadfn)(void *data);
1761 int id = 0, count, hlen;
1762 int result = -ENOMEM;
1765 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1766 IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %zd bytes\n",
1767 sizeof(struct ip_vs_sync_conn_v0));
1769 /* increase the module use count */
1770 if (!ip_vs_use_count_inc())
1771 return -ENOPROTOOPT;
1773 /* Do not hold one mutex and then to block on another */
1776 if (mutex_trylock(&ipvs->sync_mutex))
1779 mutex_lock(&ipvs->sync_mutex);
1782 mutex_unlock(&ipvs->sync_mutex);
1785 if (!ipvs->sync_state) {
1786 count = clamp(sysctl_sync_ports(ipvs), 1, IPVS_SYNC_PORTS_MAX);
1787 ipvs->threads_mask = count - 1;
1789 count = ipvs->threads_mask + 1;
1791 if (c->mcast_af == AF_UNSPEC) {
1792 c->mcast_af = AF_INET;
1793 c->mcast_group.ip = cpu_to_be32(IP_VS_SYNC_GROUP);
1796 c->mcast_port = IP_VS_SYNC_PORT;
1800 dev = __dev_get_by_name(ipvs->net, c->mcast_ifn);
1802 pr_err("Unknown mcast interface: %s\n", c->mcast_ifn);
1806 hlen = (AF_INET6 == c->mcast_af) ?
1807 sizeof(struct ipv6hdr) + sizeof(struct udphdr) :
1808 sizeof(struct iphdr) + sizeof(struct udphdr);
1809 mtu = (state == IP_VS_STATE_BACKUP) ?
1810 clamp(dev->mtu, 1500U, 65535U) : 1500U;
1811 min_mtu = (state == IP_VS_STATE_BACKUP) ? 1024 : 1;
1814 c->sync_maxlen = clamp_t(unsigned int,
1815 c->sync_maxlen, min_mtu,
1818 c->sync_maxlen = mtu - hlen;
1820 if (state == IP_VS_STATE_MASTER) {
1826 name = "ipvs-m:%d:%d";
1827 threadfn = sync_thread_master;
1828 } else if (state == IP_VS_STATE_BACKUP) {
1830 if (ipvs->backup_tinfo)
1834 name = "ipvs-b:%d:%d";
1835 threadfn = sync_thread_backup;
1841 if (state == IP_VS_STATE_MASTER) {
1842 struct ipvs_master_sync_state *ms;
1845 ipvs->ms = kcalloc(count, sizeof(ipvs->ms[0]), GFP_KERNEL);
1849 for (id = 0; id < count; id++, ms++) {
1850 INIT_LIST_HEAD(&ms->sync_queue);
1851 ms->sync_queue_len = 0;
1852 ms->sync_queue_delay = 0;
1853 INIT_DELAYED_WORK(&ms->master_wakeup_work,
1854 master_wakeup_work_handler);
1859 ti = kcalloc(count, sizeof(struct ip_vs_sync_thread_data),
1864 for (id = 0; id < count; id++) {
1867 if (state == IP_VS_STATE_BACKUP) {
1869 tinfo->buf = kmalloc(ipvs->bcfg.sync_maxlen,
1875 if (state == IP_VS_STATE_MASTER)
1876 result = make_send_sock(ipvs, id, dev, &tinfo->sock);
1878 result = make_receive_sock(ipvs, id, dev, &tinfo->sock);
1882 task = kthread_run(threadfn, tinfo, name, ipvs->gen, id);
1884 result = PTR_ERR(task);
1890 /* mark as active */
1892 if (state == IP_VS_STATE_MASTER)
1893 ipvs->master_tinfo = ti;
1895 ipvs->backup_tinfo = ti;
1896 spin_lock_bh(&ipvs->sync_buff_lock);
1897 ipvs->sync_state |= state;
1898 spin_unlock_bh(&ipvs->sync_buff_lock);
1900 mutex_unlock(&ipvs->sync_mutex);
1906 /* We do not need RTNL lock anymore, release it here so that
1907 * sock_release below can use rtnl_lock to leave the mcast group.
1910 id = min(id, count - 1);
1912 for (tinfo = ti + id; tinfo >= ti; tinfo--) {
1914 kthread_stop(tinfo->task);
1917 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
1921 mutex_unlock(&ipvs->sync_mutex);
1923 /* No more mutexes, release socks */
1925 for (tinfo = ti + id; tinfo >= ti; tinfo--) {
1927 sock_release(tinfo->sock);
1933 /* decrease the module use count */
1934 ip_vs_use_count_dec();
1938 mutex_unlock(&ipvs->sync_mutex);
1941 /* decrease the module use count */
1942 ip_vs_use_count_dec();
1947 int stop_sync_thread(struct netns_ipvs *ipvs, int state)
1949 struct ip_vs_sync_thread_data *ti, *tinfo;
1953 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1955 mutex_lock(&ipvs->sync_mutex);
1956 if (state == IP_VS_STATE_MASTER) {
1960 ti = ipvs->master_tinfo;
1963 * The lock synchronizes with sb_queue_tail(), so that we don't
1964 * add sync buffers to the queue, when we are already in
1965 * progress of stopping the master sync daemon.
1968 spin_lock_bh(&ipvs->sync_buff_lock);
1969 spin_lock(&ipvs->sync_lock);
1970 ipvs->sync_state &= ~IP_VS_STATE_MASTER;
1971 spin_unlock(&ipvs->sync_lock);
1972 spin_unlock_bh(&ipvs->sync_buff_lock);
1975 for (id = ipvs->threads_mask; id >= 0; id--) {
1976 struct ipvs_master_sync_state *ms = &ipvs->ms[id];
1980 pr_info("stopping master sync thread %d ...\n",
1981 task_pid_nr(tinfo->task));
1982 cancel_delayed_work_sync(&ms->master_wakeup_work);
1983 ret = kthread_stop(tinfo->task);
1989 ipvs->master_tinfo = NULL;
1990 } else if (state == IP_VS_STATE_BACKUP) {
1992 if (!ipvs->backup_tinfo)
1994 ti = ipvs->backup_tinfo;
1996 ipvs->sync_state &= ~IP_VS_STATE_BACKUP;
1998 for (id = ipvs->threads_mask; id >= 0; id--) {
2002 pr_info("stopping backup sync thread %d ...\n",
2003 task_pid_nr(tinfo->task));
2004 ret = kthread_stop(tinfo->task);
2008 ipvs->backup_tinfo = NULL;
2012 id = ipvs->threads_mask;
2013 mutex_unlock(&ipvs->sync_mutex);
2015 /* No more mutexes, release socks */
2016 for (tinfo = ti + id; tinfo >= ti; tinfo--) {
2018 sock_release(tinfo->sock);
2023 /* decrease the module use count */
2024 ip_vs_use_count_dec();
2028 mutex_unlock(&ipvs->sync_mutex);
2033 * Initialize data struct for each netns
2035 int __net_init ip_vs_sync_net_init(struct netns_ipvs *ipvs)
2037 __mutex_init(&ipvs->sync_mutex, "ipvs->sync_mutex", &__ipvs_sync_key);
2038 spin_lock_init(&ipvs->sync_lock);
2039 spin_lock_init(&ipvs->sync_buff_lock);
2043 void ip_vs_sync_net_cleanup(struct netns_ipvs *ipvs)
2047 retc = stop_sync_thread(ipvs, IP_VS_STATE_MASTER);
2048 if (retc && retc != -ESRCH)
2049 pr_err("Failed to stop Master Daemon\n");
2051 retc = stop_sync_thread(ipvs, IP_VS_STATE_BACKUP);
2052 if (retc && retc != -ESRCH)
2053 pr_err("Failed to stop Backup Daemon\n");