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 /* Version 0 header */
246 struct ip_vs_sync_mesg_v0 {
251 /* ip_vs_sync_conn entries start here */
254 /* Version 1 header */
255 struct ip_vs_sync_mesg {
256 __u8 reserved; /* must be zero */
260 __s8 version; /* SYNC_PROTO_VER */
262 /* ip_vs_sync_conn entries start here */
265 union ipvs_sockaddr {
266 struct sockaddr_in in;
267 struct sockaddr_in6 in6;
270 struct ip_vs_sync_buff {
271 struct list_head list;
272 unsigned long firstuse;
274 /* pointers for the message data */
275 struct ip_vs_sync_mesg *mesg;
281 * Copy of struct ip_vs_seq
282 * From unaligned network order to aligned host order
284 static void ntoh_seq(struct ip_vs_seq *no, struct ip_vs_seq *ho)
286 memset(ho, 0, sizeof(*ho));
287 ho->init_seq = get_unaligned_be32(&no->init_seq);
288 ho->delta = get_unaligned_be32(&no->delta);
289 ho->previous_delta = get_unaligned_be32(&no->previous_delta);
293 * Copy of struct ip_vs_seq
294 * From Aligned host order to unaligned network order
296 static void hton_seq(struct ip_vs_seq *ho, struct ip_vs_seq *no)
298 put_unaligned_be32(ho->init_seq, &no->init_seq);
299 put_unaligned_be32(ho->delta, &no->delta);
300 put_unaligned_be32(ho->previous_delta, &no->previous_delta);
303 static inline struct ip_vs_sync_buff *
304 sb_dequeue(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
306 struct ip_vs_sync_buff *sb;
308 spin_lock_bh(&ipvs->sync_lock);
309 if (list_empty(&ms->sync_queue)) {
311 __set_current_state(TASK_INTERRUPTIBLE);
313 sb = list_entry(ms->sync_queue.next, struct ip_vs_sync_buff,
316 ms->sync_queue_len--;
317 if (!ms->sync_queue_len)
318 ms->sync_queue_delay = 0;
320 spin_unlock_bh(&ipvs->sync_lock);
326 * Create a new sync buffer for Version 1 proto.
328 static inline struct ip_vs_sync_buff *
329 ip_vs_sync_buff_create(struct netns_ipvs *ipvs, unsigned int len)
331 struct ip_vs_sync_buff *sb;
333 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
336 len = max_t(unsigned int, len + sizeof(struct ip_vs_sync_mesg),
337 ipvs->mcfg.sync_maxlen);
338 sb->mesg = kmalloc(len, GFP_ATOMIC);
343 sb->mesg->reserved = 0; /* old nr_conns i.e. must be zero now */
344 sb->mesg->version = SYNC_PROTO_VER;
345 sb->mesg->syncid = ipvs->mcfg.syncid;
346 sb->mesg->size = htons(sizeof(struct ip_vs_sync_mesg));
347 sb->mesg->nr_conns = 0;
349 sb->head = (unsigned char *)sb->mesg + sizeof(struct ip_vs_sync_mesg);
350 sb->end = (unsigned char *)sb->mesg + len;
352 sb->firstuse = jiffies;
356 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
362 static inline void sb_queue_tail(struct netns_ipvs *ipvs,
363 struct ipvs_master_sync_state *ms)
365 struct ip_vs_sync_buff *sb = ms->sync_buff;
367 spin_lock(&ipvs->sync_lock);
368 if (ipvs->sync_state & IP_VS_STATE_MASTER &&
369 ms->sync_queue_len < sysctl_sync_qlen_max(ipvs)) {
370 if (!ms->sync_queue_len)
371 schedule_delayed_work(&ms->master_wakeup_work,
372 max(IPVS_SYNC_SEND_DELAY, 1));
373 ms->sync_queue_len++;
374 list_add_tail(&sb->list, &ms->sync_queue);
375 if ((++ms->sync_queue_delay) == IPVS_SYNC_WAKEUP_RATE) {
376 int id = (int)(ms - ipvs->ms);
378 wake_up_process(ipvs->master_tinfo[id].task);
381 ip_vs_sync_buff_release(sb);
382 spin_unlock(&ipvs->sync_lock);
386 * Get the current sync buffer if it has been created for more
387 * than the specified time or the specified time is zero.
389 static inline struct ip_vs_sync_buff *
390 get_curr_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms,
393 struct ip_vs_sync_buff *sb;
395 spin_lock_bh(&ipvs->sync_buff_lock);
397 if (sb && time_after_eq(jiffies - sb->firstuse, time)) {
398 ms->sync_buff = NULL;
399 __set_current_state(TASK_RUNNING);
402 spin_unlock_bh(&ipvs->sync_buff_lock);
407 select_master_thread_id(struct netns_ipvs *ipvs, struct ip_vs_conn *cp)
409 return ((long) cp >> (1 + ilog2(sizeof(*cp)))) & ipvs->threads_mask;
413 * Create a new sync buffer for Version 0 proto.
415 static inline struct ip_vs_sync_buff *
416 ip_vs_sync_buff_create_v0(struct netns_ipvs *ipvs, unsigned int len)
418 struct ip_vs_sync_buff *sb;
419 struct ip_vs_sync_mesg_v0 *mesg;
421 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
424 len = max_t(unsigned int, len + sizeof(struct ip_vs_sync_mesg_v0),
425 ipvs->mcfg.sync_maxlen);
426 sb->mesg = kmalloc(len, GFP_ATOMIC);
431 mesg = (struct ip_vs_sync_mesg_v0 *)sb->mesg;
433 mesg->syncid = ipvs->mcfg.syncid;
434 mesg->size = htons(sizeof(struct ip_vs_sync_mesg_v0));
435 sb->head = (unsigned char *)mesg + sizeof(struct ip_vs_sync_mesg_v0);
436 sb->end = (unsigned char *)mesg + len;
437 sb->firstuse = jiffies;
441 /* Check if connection is controlled by persistence */
442 static inline bool in_persistence(struct ip_vs_conn *cp)
444 for (cp = cp->control; cp; cp = cp->control) {
445 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
451 /* Check if conn should be synced.
452 * pkts: conn packets, use sysctl_sync_threshold to avoid packet check
453 * - (1) sync_refresh_period: reduce sync rate. Additionally, retry
454 * sync_retries times with period of sync_refresh_period/8
455 * - (2) if both sync_refresh_period and sync_period are 0 send sync only
456 * for state changes or only once when pkts matches sync_threshold
457 * - (3) templates: rate can be reduced only with sync_refresh_period or
460 static int ip_vs_sync_conn_needed(struct netns_ipvs *ipvs,
461 struct ip_vs_conn *cp, int pkts)
463 unsigned long orig = READ_ONCE(cp->sync_endtime);
464 unsigned long now = jiffies;
465 unsigned long n = (now + cp->timeout) & ~3UL;
466 unsigned int sync_refresh_period;
470 /* Check if we sync in current state */
471 if (unlikely(cp->flags & IP_VS_CONN_F_TEMPLATE))
473 else if (unlikely(sysctl_sync_persist_mode(ipvs) && in_persistence(cp)))
475 else if (likely(cp->protocol == IPPROTO_TCP)) {
476 if (!((1 << cp->state) &
477 ((1 << IP_VS_TCP_S_ESTABLISHED) |
478 (1 << IP_VS_TCP_S_FIN_WAIT) |
479 (1 << IP_VS_TCP_S_CLOSE) |
480 (1 << IP_VS_TCP_S_CLOSE_WAIT) |
481 (1 << IP_VS_TCP_S_TIME_WAIT))))
483 force = cp->state != cp->old_state;
484 if (force && cp->state != IP_VS_TCP_S_ESTABLISHED)
486 } else if (unlikely(cp->protocol == IPPROTO_SCTP)) {
487 if (!((1 << cp->state) &
488 ((1 << IP_VS_SCTP_S_ESTABLISHED) |
489 (1 << IP_VS_SCTP_S_SHUTDOWN_SENT) |
490 (1 << IP_VS_SCTP_S_SHUTDOWN_RECEIVED) |
491 (1 << IP_VS_SCTP_S_SHUTDOWN_ACK_SENT) |
492 (1 << IP_VS_SCTP_S_CLOSED))))
494 force = cp->state != cp->old_state;
495 if (force && cp->state != IP_VS_SCTP_S_ESTABLISHED)
498 /* UDP or another protocol with single state */
502 sync_refresh_period = sysctl_sync_refresh_period(ipvs);
503 if (sync_refresh_period > 0) {
504 long diff = n - orig;
505 long min_diff = max(cp->timeout >> 1, 10UL * HZ);
507 /* Avoid sync if difference is below sync_refresh_period
508 * and below the half timeout.
510 if (abs(diff) < min_t(long, sync_refresh_period, min_diff)) {
511 int retries = orig & 3;
513 if (retries >= sysctl_sync_retries(ipvs))
515 if (time_before(now, orig - cp->timeout +
516 (sync_refresh_period >> 3)))
521 sync_period = sysctl_sync_period(ipvs);
522 if (sync_period > 0) {
523 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE) &&
524 pkts % sync_period != sysctl_sync_threshold(ipvs))
526 } else if (!sync_refresh_period &&
527 pkts != sysctl_sync_threshold(ipvs))
531 cp->old_state = cp->state;
532 n = cmpxchg(&cp->sync_endtime, orig, n);
533 return n == orig || force;
537 * Version 0 , could be switched in by sys_ctl.
538 * Add an ip_vs_conn information into the current sync_buff.
540 static void ip_vs_sync_conn_v0(struct netns_ipvs *ipvs, struct ip_vs_conn *cp,
543 struct ip_vs_sync_mesg_v0 *m;
544 struct ip_vs_sync_conn_v0 *s;
545 struct ip_vs_sync_buff *buff;
546 struct ipvs_master_sync_state *ms;
550 if (unlikely(cp->af != AF_INET))
552 /* Do not sync ONE PACKET */
553 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
556 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
559 spin_lock_bh(&ipvs->sync_buff_lock);
560 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
561 spin_unlock_bh(&ipvs->sync_buff_lock);
565 id = select_master_thread_id(ipvs, cp);
567 buff = ms->sync_buff;
568 len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
571 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
572 /* Send buffer if it is for v1 */
573 if (buff->head + len > buff->end || !m->nr_conns) {
574 sb_queue_tail(ipvs, ms);
575 ms->sync_buff = NULL;
580 buff = ip_vs_sync_buff_create_v0(ipvs, len);
582 spin_unlock_bh(&ipvs->sync_buff_lock);
583 pr_err("ip_vs_sync_buff_create failed.\n");
586 ms->sync_buff = buff;
589 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
590 s = (struct ip_vs_sync_conn_v0 *) buff->head;
594 s->protocol = cp->protocol;
595 s->cport = cp->cport;
596 s->vport = cp->vport;
597 s->dport = cp->dport;
598 s->caddr = cp->caddr.ip;
599 s->vaddr = cp->vaddr.ip;
600 s->daddr = cp->daddr.ip;
601 s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
602 s->state = htons(cp->state);
603 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
604 struct ip_vs_sync_conn_options *opt =
605 (struct ip_vs_sync_conn_options *)&s[1];
606 memcpy(opt, &cp->sync_conn_opt, sizeof(*opt));
610 m->size = htons(ntohs(m->size) + len);
612 spin_unlock_bh(&ipvs->sync_buff_lock);
614 /* synchronize its controller if it has */
617 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
618 pkts = atomic_inc_return(&cp->in_pkts);
620 pkts = sysctl_sync_threshold(ipvs);
621 ip_vs_sync_conn(ipvs, cp, pkts);
626 * Add an ip_vs_conn information into the current sync_buff.
627 * Called by ip_vs_in.
628 * Sending Version 1 messages
630 void ip_vs_sync_conn(struct netns_ipvs *ipvs, struct ip_vs_conn *cp, int pkts)
632 struct ip_vs_sync_mesg *m;
633 union ip_vs_sync_conn *s;
634 struct ip_vs_sync_buff *buff;
635 struct ipvs_master_sync_state *ms;
638 unsigned int len, pe_name_len, pad;
640 /* Handle old version of the protocol */
641 if (sysctl_sync_ver(ipvs) == 0) {
642 ip_vs_sync_conn_v0(ipvs, cp, pkts);
645 /* Do not sync ONE PACKET */
646 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
649 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
654 if (cp->pe_data_len) {
655 if (!cp->pe_data || !cp->dest) {
656 IP_VS_ERR_RL("SYNC, connection pe_data invalid\n");
659 pe_name_len = strnlen(cp->pe->name, IP_VS_PENAME_MAXLEN);
662 spin_lock_bh(&ipvs->sync_buff_lock);
663 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
664 spin_unlock_bh(&ipvs->sync_buff_lock);
668 id = select_master_thread_id(ipvs, cp);
671 #ifdef CONFIG_IP_VS_IPV6
672 if (cp->af == AF_INET6)
673 len = sizeof(struct ip_vs_sync_v6);
676 len = sizeof(struct ip_vs_sync_v4);
678 if (cp->flags & IP_VS_CONN_F_SEQ_MASK)
679 len += sizeof(struct ip_vs_sync_conn_options) + 2;
682 len += cp->pe_data_len + 2; /* + Param hdr field */
684 len += pe_name_len + 2;
686 /* check if there is a space for this one */
688 buff = ms->sync_buff;
691 pad = (4 - (size_t) buff->head) & 3;
692 /* Send buffer if it is for v0 */
693 if (buff->head + len + pad > buff->end || m->reserved) {
694 sb_queue_tail(ipvs, ms);
695 ms->sync_buff = NULL;
702 buff = ip_vs_sync_buff_create(ipvs, len);
704 spin_unlock_bh(&ipvs->sync_buff_lock);
705 pr_err("ip_vs_sync_buff_create failed.\n");
708 ms->sync_buff = buff;
713 buff->head += pad + len;
714 m->size = htons(ntohs(m->size) + pad + len);
715 /* Add ev. padding from prev. sync_conn */
719 s = (union ip_vs_sync_conn *)p;
721 /* Set message type & copy members */
722 s->v4.type = (cp->af == AF_INET6 ? STYPE_F_INET6 : 0);
723 s->v4.ver_size = htons(len & SVER_MASK); /* Version 0 */
724 s->v4.flags = htonl(cp->flags & ~IP_VS_CONN_F_HASHED);
725 s->v4.state = htons(cp->state);
726 s->v4.protocol = cp->protocol;
727 s->v4.cport = cp->cport;
728 s->v4.vport = cp->vport;
729 s->v4.dport = cp->dport;
730 s->v4.fwmark = htonl(cp->fwmark);
731 s->v4.timeout = htonl(cp->timeout / HZ);
734 #ifdef CONFIG_IP_VS_IPV6
735 if (cp->af == AF_INET6) {
736 p += sizeof(struct ip_vs_sync_v6);
737 s->v6.caddr = cp->caddr.in6;
738 s->v6.vaddr = cp->vaddr.in6;
739 s->v6.daddr = cp->daddr.in6;
743 p += sizeof(struct ip_vs_sync_v4); /* options ptr */
744 s->v4.caddr = cp->caddr.ip;
745 s->v4.vaddr = cp->vaddr.ip;
746 s->v4.daddr = cp->daddr.ip;
748 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
749 *(p++) = IPVS_OPT_SEQ_DATA;
750 *(p++) = sizeof(struct ip_vs_sync_conn_options);
751 hton_seq((struct ip_vs_seq *)p, &cp->in_seq);
752 p += sizeof(struct ip_vs_seq);
753 hton_seq((struct ip_vs_seq *)p, &cp->out_seq);
754 p += sizeof(struct ip_vs_seq);
757 if (cp->pe_data_len && cp->pe_data) {
758 *(p++) = IPVS_OPT_PE_DATA;
759 *(p++) = cp->pe_data_len;
760 memcpy(p, cp->pe_data, cp->pe_data_len);
761 p += cp->pe_data_len;
764 *(p++) = IPVS_OPT_PE_NAME;
765 *(p++) = pe_name_len;
766 memcpy(p, cp->pe->name, pe_name_len);
771 spin_unlock_bh(&ipvs->sync_buff_lock);
774 /* synchronize its controller if it has */
778 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
779 pkts = atomic_inc_return(&cp->in_pkts);
781 pkts = sysctl_sync_threshold(ipvs);
786 * fill_param used by version 1
789 ip_vs_conn_fill_param_sync(struct netns_ipvs *ipvs, int af, union ip_vs_sync_conn *sc,
790 struct ip_vs_conn_param *p,
791 __u8 *pe_data, unsigned int pe_data_len,
792 __u8 *pe_name, unsigned int pe_name_len)
794 #ifdef CONFIG_IP_VS_IPV6
796 ip_vs_conn_fill_param(ipvs, af, sc->v6.protocol,
797 (const union nf_inet_addr *)&sc->v6.caddr,
799 (const union nf_inet_addr *)&sc->v6.vaddr,
803 ip_vs_conn_fill_param(ipvs, af, sc->v4.protocol,
804 (const union nf_inet_addr *)&sc->v4.caddr,
806 (const union nf_inet_addr *)&sc->v4.vaddr,
811 char buff[IP_VS_PENAME_MAXLEN+1];
813 memcpy(buff, pe_name, pe_name_len);
815 p->pe = __ip_vs_pe_getbyname(buff);
817 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n",
822 IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n");
826 p->pe_data = kmemdup(pe_data, pe_data_len, GFP_ATOMIC);
828 module_put(p->pe->module);
831 p->pe_data_len = pe_data_len;
837 * Connection Add / Update.
838 * Common for version 0 and 1 reception of backup sync_conns.
842 static void ip_vs_proc_conn(struct netns_ipvs *ipvs, struct ip_vs_conn_param *param,
843 unsigned int flags, unsigned int state,
844 unsigned int protocol, unsigned int type,
845 const union nf_inet_addr *daddr, __be16 dport,
846 unsigned long timeout, __u32 fwmark,
847 struct ip_vs_sync_conn_options *opt)
849 struct ip_vs_dest *dest;
850 struct ip_vs_conn *cp;
852 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
853 cp = ip_vs_conn_in_get(param);
854 if (cp && ((cp->dport != dport) ||
855 !ip_vs_addr_equal(cp->daf, &cp->daddr, daddr))) {
856 if (!(flags & IP_VS_CONN_F_INACTIVE)) {
857 ip_vs_conn_expire_now(cp);
858 __ip_vs_conn_put(cp);
861 /* This is the expiration message for the
862 * connection that was already replaced, so we
865 __ip_vs_conn_put(cp);
866 kfree(param->pe_data);
871 cp = ip_vs_ct_in_get(param);
876 kfree(param->pe_data);
879 spin_lock_bh(&cp->lock);
880 if ((cp->flags ^ flags) & IP_VS_CONN_F_INACTIVE &&
881 !(flags & IP_VS_CONN_F_TEMPLATE) && dest) {
882 if (flags & IP_VS_CONN_F_INACTIVE) {
883 atomic_dec(&dest->activeconns);
884 atomic_inc(&dest->inactconns);
886 atomic_inc(&dest->activeconns);
887 atomic_dec(&dest->inactconns);
890 flags &= IP_VS_CONN_F_BACKUP_UPD_MASK;
891 flags |= cp->flags & ~IP_VS_CONN_F_BACKUP_UPD_MASK;
893 spin_unlock_bh(&cp->lock);
895 ip_vs_try_bind_dest(cp);
898 * Find the appropriate destination for the connection.
899 * If it is not found the connection will remain unbound
903 /* This function is only invoked by the synchronization
904 * code. We do not currently support heterogeneous pools
905 * with synchronization, so we can make the assumption that
906 * the svc_af is the same as the dest_af
908 dest = ip_vs_find_dest(ipvs, type, type, daddr, dport,
909 param->vaddr, param->vport, protocol,
912 cp = ip_vs_conn_new(param, type, daddr, dport, flags, dest,
916 kfree(param->pe_data);
917 IP_VS_DBG(2, "BACKUP, add new conn. failed\n");
920 if (!(flags & IP_VS_CONN_F_TEMPLATE))
921 kfree(param->pe_data);
925 cp->in_seq = opt->in_seq;
926 cp->out_seq = opt->out_seq;
928 atomic_set(&cp->in_pkts, sysctl_sync_threshold(ipvs));
930 cp->old_state = cp->state;
932 * For Ver 0 messages style
933 * - Not possible to recover the right timeout for templates
934 * - can not find the right fwmark
935 * virtual service. If needed, we can do it for
936 * non-fwmark persistent services.
937 * Ver 1 messages style.
941 if (timeout > MAX_SCHEDULE_TIMEOUT / HZ)
942 timeout = MAX_SCHEDULE_TIMEOUT / HZ;
943 cp->timeout = timeout*HZ;
945 struct ip_vs_proto_data *pd;
947 pd = ip_vs_proto_data_get(ipvs, protocol);
948 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pd && pd->timeout_table)
949 cp->timeout = pd->timeout_table[state];
951 cp->timeout = (3*60*HZ);
957 * Process received multicast message for Version 0
959 static void ip_vs_process_message_v0(struct netns_ipvs *ipvs, const char *buffer,
962 struct ip_vs_sync_mesg_v0 *m = (struct ip_vs_sync_mesg_v0 *)buffer;
963 struct ip_vs_sync_conn_v0 *s;
964 struct ip_vs_sync_conn_options *opt;
965 struct ip_vs_protocol *pp;
966 struct ip_vs_conn_param param;
970 p = (char *)buffer + sizeof(struct ip_vs_sync_mesg_v0);
971 for (i=0; i<m->nr_conns; i++) {
972 unsigned int flags, state;
974 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
975 IP_VS_ERR_RL("BACKUP v0, bogus conn\n");
978 s = (struct ip_vs_sync_conn_v0 *) p;
979 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
980 flags &= ~IP_VS_CONN_F_HASHED;
981 if (flags & IP_VS_CONN_F_SEQ_MASK) {
982 opt = (struct ip_vs_sync_conn_options *)&s[1];
984 if (p > buffer+buflen) {
985 IP_VS_ERR_RL("BACKUP v0, Dropping buffer bogus conn options\n");
990 p += SIMPLE_CONN_SIZE;
993 state = ntohs(s->state);
994 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
995 pp = ip_vs_proto_get(s->protocol);
997 IP_VS_DBG(2, "BACKUP v0, Unsupported protocol %u\n",
1001 if (state >= pp->num_states) {
1002 IP_VS_DBG(2, "BACKUP v0, Invalid %s state %u\n",
1007 if (state >= IP_VS_CTPL_S_LAST)
1008 IP_VS_DBG(7, "BACKUP v0, Invalid tpl state %u\n",
1012 ip_vs_conn_fill_param(ipvs, AF_INET, s->protocol,
1013 (const union nf_inet_addr *)&s->caddr,
1015 (const union nf_inet_addr *)&s->vaddr,
1018 /* Send timeout as Zero */
1019 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->protocol, AF_INET,
1020 (union nf_inet_addr *)&s->daddr, s->dport,
1028 static inline int ip_vs_proc_seqopt(__u8 *p, unsigned int plen,
1030 struct ip_vs_sync_conn_options *opt)
1032 struct ip_vs_sync_conn_options *topt;
1034 topt = (struct ip_vs_sync_conn_options *)p;
1036 if (plen != sizeof(struct ip_vs_sync_conn_options)) {
1037 IP_VS_DBG(2, "BACKUP, bogus conn options length\n");
1040 if (*opt_flags & IPVS_OPT_F_SEQ_DATA) {
1041 IP_VS_DBG(2, "BACKUP, conn options found twice\n");
1044 ntoh_seq(&topt->in_seq, &opt->in_seq);
1045 ntoh_seq(&topt->out_seq, &opt->out_seq);
1046 *opt_flags |= IPVS_OPT_F_SEQ_DATA;
1050 static int ip_vs_proc_str(__u8 *p, unsigned int plen, unsigned int *data_len,
1051 __u8 **data, unsigned int maxlen,
1052 __u32 *opt_flags, __u32 flag)
1054 if (plen > maxlen) {
1055 IP_VS_DBG(2, "BACKUP, bogus par.data len > %d\n", maxlen);
1058 if (*opt_flags & flag) {
1059 IP_VS_DBG(2, "BACKUP, Par.data found twice 0x%x\n", flag);
1068 * Process a Version 1 sync. connection
1070 static inline int ip_vs_proc_sync_conn(struct netns_ipvs *ipvs, __u8 *p, __u8 *msg_end)
1072 struct ip_vs_sync_conn_options opt;
1073 union ip_vs_sync_conn *s;
1074 struct ip_vs_protocol *pp;
1075 struct ip_vs_conn_param param;
1077 unsigned int af, state, pe_data_len=0, pe_name_len=0;
1078 __u8 *pe_data=NULL, *pe_name=NULL;
1082 s = (union ip_vs_sync_conn *) p;
1084 if (s->v6.type & STYPE_F_INET6) {
1085 #ifdef CONFIG_IP_VS_IPV6
1087 p += sizeof(struct ip_vs_sync_v6);
1089 IP_VS_DBG(3,"BACKUP, IPv6 msg received, and IPVS is not compiled for IPv6\n");
1093 } else if (!s->v4.type) {
1095 p += sizeof(struct ip_vs_sync_v4);
1102 /* Process optional params check Type & Len. */
1103 while (p < msg_end) {
1112 if (!plen || ((p + plen) > msg_end))
1114 /* Handle seq option p = param data */
1115 switch (ptype & ~IPVS_OPT_F_PARAM) {
1116 case IPVS_OPT_SEQ_DATA:
1117 if (ip_vs_proc_seqopt(p, plen, &opt_flags, &opt))
1121 case IPVS_OPT_PE_DATA:
1122 if (ip_vs_proc_str(p, plen, &pe_data_len, &pe_data,
1123 IP_VS_PEDATA_MAXLEN, &opt_flags,
1124 IPVS_OPT_F_PE_DATA))
1128 case IPVS_OPT_PE_NAME:
1129 if (ip_vs_proc_str(p, plen,&pe_name_len, &pe_name,
1130 IP_VS_PENAME_MAXLEN, &opt_flags,
1131 IPVS_OPT_F_PE_NAME))
1136 /* Param data mandatory ? */
1137 if (!(ptype & IPVS_OPT_F_PARAM)) {
1138 IP_VS_DBG(3, "BACKUP, Unknown mandatory param %d found\n",
1139 ptype & ~IPVS_OPT_F_PARAM);
1144 p += plen; /* Next option */
1147 /* Get flags and Mask off unsupported */
1148 flags = ntohl(s->v4.flags) & IP_VS_CONN_F_BACKUP_MASK;
1149 flags |= IP_VS_CONN_F_SYNC;
1150 state = ntohs(s->v4.state);
1152 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
1153 pp = ip_vs_proto_get(s->v4.protocol);
1155 IP_VS_DBG(3,"BACKUP, Unsupported protocol %u\n",
1160 if (state >= pp->num_states) {
1161 IP_VS_DBG(3, "BACKUP, Invalid %s state %u\n",
1167 if (state >= IP_VS_CTPL_S_LAST)
1168 IP_VS_DBG(7, "BACKUP, Invalid tpl state %u\n",
1171 if (ip_vs_conn_fill_param_sync(ipvs, af, s, ¶m, pe_data,
1172 pe_data_len, pe_name, pe_name_len)) {
1176 /* If only IPv4, just silent skip IPv6 */
1178 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->v4.protocol, af,
1179 (union nf_inet_addr *)&s->v4.daddr, s->v4.dport,
1180 ntohl(s->v4.timeout), ntohl(s->v4.fwmark),
1181 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1183 #ifdef CONFIG_IP_VS_IPV6
1185 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->v6.protocol, af,
1186 (union nf_inet_addr *)&s->v6.daddr, s->v6.dport,
1187 ntohl(s->v6.timeout), ntohl(s->v6.fwmark),
1188 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1191 ip_vs_pe_put(param.pe);
1195 IP_VS_DBG(2, "BACKUP, Single msg dropped err:%d\n", retc);
1200 * Process received multicast message and create the corresponding
1201 * ip_vs_conn entries.
1202 * Handles Version 0 & 1
1204 static void ip_vs_process_message(struct netns_ipvs *ipvs, __u8 *buffer,
1205 const size_t buflen)
1207 struct ip_vs_sync_mesg *m2 = (struct ip_vs_sync_mesg *)buffer;
1211 if (buflen < sizeof(struct ip_vs_sync_mesg_v0)) {
1212 IP_VS_DBG(2, "BACKUP, message header too short\n");
1216 if (buflen != ntohs(m2->size)) {
1217 IP_VS_DBG(2, "BACKUP, bogus message size\n");
1220 /* SyncID sanity check */
1221 if (ipvs->bcfg.syncid != 0 && m2->syncid != ipvs->bcfg.syncid) {
1222 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2->syncid);
1225 /* Handle version 1 message */
1226 if ((m2->version == SYNC_PROTO_VER) && (m2->reserved == 0)
1227 && (m2->spare == 0)) {
1229 msg_end = buffer + sizeof(struct ip_vs_sync_mesg);
1230 nr_conns = m2->nr_conns;
1232 for (i=0; i<nr_conns; i++) {
1233 union ip_vs_sync_conn *s;
1238 if (p + sizeof(s->v4) > buffer+buflen) {
1239 IP_VS_ERR_RL("BACKUP, Dropping buffer, too small\n");
1242 s = (union ip_vs_sync_conn *)p;
1243 size = ntohs(s->v4.ver_size) & SVER_MASK;
1245 /* Basic sanity checks */
1246 if (msg_end > buffer+buflen) {
1247 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n");
1250 if (ntohs(s->v4.ver_size) >> SVER_SHIFT) {
1251 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n",
1252 ntohs(s->v4.ver_size) >> SVER_SHIFT);
1255 /* Process a single sync_conn */
1256 retc = ip_vs_proc_sync_conn(ipvs, p, msg_end);
1258 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n",
1262 /* Make sure we have 32 bit alignment */
1263 msg_end = p + ((size + 3) & ~3);
1266 /* Old type of message */
1267 ip_vs_process_message_v0(ipvs, buffer, buflen);
1274 * Setup sndbuf (mode=1) or rcvbuf (mode=0)
1276 static void set_sock_size(struct sock *sk, int mode, int val)
1278 /* setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &val, sizeof(val)); */
1279 /* setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)); */
1282 val = clamp_t(int, val, (SOCK_MIN_SNDBUF + 1) / 2,
1283 READ_ONCE(sysctl_wmem_max));
1284 sk->sk_sndbuf = val * 2;
1285 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1287 val = clamp_t(int, val, (SOCK_MIN_RCVBUF + 1) / 2,
1288 READ_ONCE(sysctl_rmem_max));
1289 sk->sk_rcvbuf = val * 2;
1290 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1296 * Setup loopback of outgoing multicasts on a sending socket
1298 static void set_mcast_loop(struct sock *sk, u_char loop)
1300 /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
1301 inet_assign_bit(MC_LOOP, sk, loop);
1302 #ifdef CONFIG_IP_VS_IPV6
1303 if (READ_ONCE(sk->sk_family) == AF_INET6) {
1304 /* IPV6_MULTICAST_LOOP */
1305 inet6_assign_bit(MC6_LOOP, sk, loop);
1311 * Specify TTL for outgoing multicasts on a sending socket
1313 static void set_mcast_ttl(struct sock *sk, u_char ttl)
1315 struct inet_sock *inet = inet_sk(sk);
1317 /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
1319 WRITE_ONCE(inet->mc_ttl, ttl);
1320 #ifdef CONFIG_IP_VS_IPV6
1321 if (sk->sk_family == AF_INET6) {
1322 struct ipv6_pinfo *np = inet6_sk(sk);
1324 /* IPV6_MULTICAST_HOPS */
1325 WRITE_ONCE(np->mcast_hops, ttl);
1331 /* Control fragmentation of messages */
1332 static void set_mcast_pmtudisc(struct sock *sk, int val)
1334 struct inet_sock *inet = inet_sk(sk);
1336 /* setsockopt(sock, SOL_IP, IP_MTU_DISCOVER, &val, sizeof(val)); */
1338 WRITE_ONCE(inet->pmtudisc, val);
1339 #ifdef CONFIG_IP_VS_IPV6
1340 if (sk->sk_family == AF_INET6) {
1341 struct ipv6_pinfo *np = inet6_sk(sk);
1343 /* IPV6_MTU_DISCOVER */
1344 WRITE_ONCE(np->pmtudisc, val);
1351 * Specifiy default interface for outgoing multicasts
1353 static int set_mcast_if(struct sock *sk, struct net_device *dev)
1355 struct inet_sock *inet = inet_sk(sk);
1357 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1361 inet->mc_index = dev->ifindex;
1362 /* inet->mc_addr = 0; */
1363 #ifdef CONFIG_IP_VS_IPV6
1364 if (sk->sk_family == AF_INET6) {
1365 struct ipv6_pinfo *np = inet6_sk(sk);
1367 /* IPV6_MULTICAST_IF */
1368 WRITE_ONCE(np->mcast_oif, dev->ifindex);
1378 * Join a multicast group.
1379 * the group is specified by a class D multicast address 224.0.0.0/8
1380 * in the in_addr structure passed in as a parameter.
1383 join_mcast_group(struct sock *sk, struct in_addr *addr, struct net_device *dev)
1385 struct ip_mreqn mreq;
1388 memset(&mreq, 0, sizeof(mreq));
1389 memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
1391 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1394 mreq.imr_ifindex = dev->ifindex;
1397 ret = ip_mc_join_group(sk, &mreq);
1403 #ifdef CONFIG_IP_VS_IPV6
1404 static int join_mcast_group6(struct sock *sk, struct in6_addr *addr,
1405 struct net_device *dev)
1409 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1413 ret = ipv6_sock_mc_join(sk, dev->ifindex, addr);
1420 static int bind_mcastif_addr(struct socket *sock, struct net_device *dev)
1423 struct sockaddr_in sin;
1425 addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
1427 pr_err("You probably need to specify IP address on "
1428 "multicast interface.\n");
1430 IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
1433 /* Now bind the socket with the address of multicast interface */
1434 sin.sin_family = AF_INET;
1435 sin.sin_addr.s_addr = addr;
1438 return kernel_bind(sock, (struct sockaddr *)&sin, sizeof(sin));
1441 static void get_mcast_sockaddr(union ipvs_sockaddr *sa, int *salen,
1442 struct ipvs_sync_daemon_cfg *c, int id)
1444 if (AF_INET6 == c->mcast_af) {
1445 sa->in6 = (struct sockaddr_in6) {
1446 .sin6_family = AF_INET6,
1447 .sin6_port = htons(c->mcast_port + id),
1449 sa->in6.sin6_addr = c->mcast_group.in6;
1450 *salen = sizeof(sa->in6);
1452 sa->in = (struct sockaddr_in) {
1453 .sin_family = AF_INET,
1454 .sin_port = htons(c->mcast_port + id),
1456 sa->in.sin_addr = c->mcast_group.in;
1457 *salen = sizeof(sa->in);
1462 * Set up sending multicast socket over UDP
1464 static int make_send_sock(struct netns_ipvs *ipvs, int id,
1465 struct net_device *dev, struct socket **sock_ret)
1467 /* multicast addr */
1468 union ipvs_sockaddr mcast_addr;
1469 struct socket *sock;
1472 /* First create a socket */
1473 result = sock_create_kern(ipvs->net, ipvs->mcfg.mcast_af, SOCK_DGRAM,
1474 IPPROTO_UDP, &sock);
1476 pr_err("Error during creation of socket; terminating\n");
1480 result = set_mcast_if(sock->sk, dev);
1482 pr_err("Error setting outbound mcast interface\n");
1486 set_mcast_loop(sock->sk, 0);
1487 set_mcast_ttl(sock->sk, ipvs->mcfg.mcast_ttl);
1488 /* Allow fragmentation if MTU changes */
1489 set_mcast_pmtudisc(sock->sk, IP_PMTUDISC_DONT);
1490 result = sysctl_sync_sock_size(ipvs);
1492 set_sock_size(sock->sk, 1, result);
1494 if (AF_INET == ipvs->mcfg.mcast_af)
1495 result = bind_mcastif_addr(sock, dev);
1499 pr_err("Error binding address of the mcast interface\n");
1503 get_mcast_sockaddr(&mcast_addr, &salen, &ipvs->mcfg, id);
1504 result = kernel_connect(sock, (struct sockaddr *)&mcast_addr,
1507 pr_err("Error connecting to the multicast addr\n");
1519 * Set up receiving multicast socket over UDP
1521 static int make_receive_sock(struct netns_ipvs *ipvs, int id,
1522 struct net_device *dev, struct socket **sock_ret)
1524 /* multicast addr */
1525 union ipvs_sockaddr mcast_addr;
1526 struct socket *sock;
1529 /* First create a socket */
1530 result = sock_create_kern(ipvs->net, ipvs->bcfg.mcast_af, SOCK_DGRAM,
1531 IPPROTO_UDP, &sock);
1533 pr_err("Error during creation of socket; terminating\n");
1537 /* it is equivalent to the REUSEADDR option in user-space */
1538 sock->sk->sk_reuse = SK_CAN_REUSE;
1539 result = sysctl_sync_sock_size(ipvs);
1541 set_sock_size(sock->sk, 0, result);
1543 get_mcast_sockaddr(&mcast_addr, &salen, &ipvs->bcfg, id);
1544 sock->sk->sk_bound_dev_if = dev->ifindex;
1545 result = kernel_bind(sock, (struct sockaddr *)&mcast_addr, salen);
1547 pr_err("Error binding to the multicast addr\n");
1551 /* join the multicast group */
1552 #ifdef CONFIG_IP_VS_IPV6
1553 if (ipvs->bcfg.mcast_af == AF_INET6)
1554 result = join_mcast_group6(sock->sk, &mcast_addr.in6.sin6_addr,
1558 result = join_mcast_group(sock->sk, &mcast_addr.in.sin_addr,
1561 pr_err("Error joining to the multicast group\n");
1573 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
1575 struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
1579 iov.iov_base = (void *)buffer;
1580 iov.iov_len = length;
1582 len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
1588 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
1593 msize = ntohs(msg->size);
1595 ret = ip_vs_send_async(sock, (char *)msg, msize);
1596 if (ret >= 0 || ret == -EAGAIN)
1598 pr_err("ip_vs_send_async error %d\n", ret);
1603 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
1605 struct msghdr msg = {NULL,};
1606 struct kvec iov = {buffer, buflen};
1609 /* Receive a packet */
1610 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &iov, 1, buflen);
1611 len = sock_recvmsg(sock, &msg, MSG_DONTWAIT);
1618 /* Wakeup the master thread for sending */
1619 static void master_wakeup_work_handler(struct work_struct *work)
1621 struct ipvs_master_sync_state *ms =
1622 container_of(work, struct ipvs_master_sync_state,
1623 master_wakeup_work.work);
1624 struct netns_ipvs *ipvs = ms->ipvs;
1626 spin_lock_bh(&ipvs->sync_lock);
1627 if (ms->sync_queue_len &&
1628 ms->sync_queue_delay < IPVS_SYNC_WAKEUP_RATE) {
1629 int id = (int)(ms - ipvs->ms);
1631 ms->sync_queue_delay = IPVS_SYNC_WAKEUP_RATE;
1632 wake_up_process(ipvs->master_tinfo[id].task);
1634 spin_unlock_bh(&ipvs->sync_lock);
1637 /* Get next buffer to send */
1638 static inline struct ip_vs_sync_buff *
1639 next_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
1641 struct ip_vs_sync_buff *sb;
1643 sb = sb_dequeue(ipvs, ms);
1646 /* Do not delay entries in buffer for more than 2 seconds */
1647 return get_curr_sync_buff(ipvs, ms, IPVS_SYNC_FLUSH_TIME);
1650 static int sync_thread_master(void *data)
1652 struct ip_vs_sync_thread_data *tinfo = data;
1653 struct netns_ipvs *ipvs = tinfo->ipvs;
1654 struct ipvs_master_sync_state *ms = &ipvs->ms[tinfo->id];
1655 struct sock *sk = tinfo->sock->sk;
1656 struct ip_vs_sync_buff *sb;
1658 pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
1659 "syncid = %d, id = %d\n",
1660 ipvs->mcfg.mcast_ifn, ipvs->mcfg.syncid, tinfo->id);
1663 sb = next_sync_buff(ipvs, ms);
1664 if (unlikely(kthread_should_stop()))
1667 schedule_timeout(IPVS_SYNC_CHECK_PERIOD);
1670 while (ip_vs_send_sync_msg(tinfo->sock, sb->mesg) < 0) {
1671 /* (Ab)use interruptible sleep to avoid increasing
1674 __wait_event_interruptible(*sk_sleep(sk),
1675 sock_writeable(sk) ||
1676 kthread_should_stop());
1677 if (unlikely(kthread_should_stop()))
1680 ip_vs_sync_buff_release(sb);
1684 __set_current_state(TASK_RUNNING);
1686 ip_vs_sync_buff_release(sb);
1688 /* clean up the sync_buff queue */
1689 while ((sb = sb_dequeue(ipvs, ms)))
1690 ip_vs_sync_buff_release(sb);
1691 __set_current_state(TASK_RUNNING);
1693 /* clean up the current sync_buff */
1694 sb = get_curr_sync_buff(ipvs, ms, 0);
1696 ip_vs_sync_buff_release(sb);
1702 static int sync_thread_backup(void *data)
1704 struct ip_vs_sync_thread_data *tinfo = data;
1705 struct netns_ipvs *ipvs = tinfo->ipvs;
1706 struct sock *sk = tinfo->sock->sk;
1707 struct udp_sock *up = udp_sk(sk);
1710 pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
1711 "syncid = %d, id = %d\n",
1712 ipvs->bcfg.mcast_ifn, ipvs->bcfg.syncid, tinfo->id);
1714 while (!kthread_should_stop()) {
1715 wait_event_interruptible(*sk_sleep(sk),
1716 !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
1717 !skb_queue_empty_lockless(&up->reader_queue) ||
1718 kthread_should_stop());
1720 /* do we have data now? */
1721 while (!skb_queue_empty_lockless(&sk->sk_receive_queue) ||
1722 !skb_queue_empty_lockless(&up->reader_queue)) {
1723 len = ip_vs_receive(tinfo->sock, tinfo->buf,
1724 ipvs->bcfg.sync_maxlen);
1727 pr_err("receiving message error\n");
1731 ip_vs_process_message(ipvs, tinfo->buf, len);
1739 int start_sync_thread(struct netns_ipvs *ipvs, struct ipvs_sync_daemon_cfg *c,
1742 struct ip_vs_sync_thread_data *ti = NULL, *tinfo;
1743 struct task_struct *task;
1744 struct net_device *dev;
1746 int (*threadfn)(void *data);
1747 int id = 0, count, hlen;
1748 int result = -ENOMEM;
1751 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1752 IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %zd bytes\n",
1753 sizeof(struct ip_vs_sync_conn_v0));
1755 /* increase the module use count */
1756 if (!ip_vs_use_count_inc())
1757 return -ENOPROTOOPT;
1759 /* Do not hold one mutex and then to block on another */
1762 if (mutex_trylock(&ipvs->sync_mutex))
1765 mutex_lock(&ipvs->sync_mutex);
1768 mutex_unlock(&ipvs->sync_mutex);
1771 if (!ipvs->sync_state) {
1772 count = clamp(sysctl_sync_ports(ipvs), 1, IPVS_SYNC_PORTS_MAX);
1773 ipvs->threads_mask = count - 1;
1775 count = ipvs->threads_mask + 1;
1777 if (c->mcast_af == AF_UNSPEC) {
1778 c->mcast_af = AF_INET;
1779 c->mcast_group.ip = cpu_to_be32(IP_VS_SYNC_GROUP);
1782 c->mcast_port = IP_VS_SYNC_PORT;
1786 dev = __dev_get_by_name(ipvs->net, c->mcast_ifn);
1788 pr_err("Unknown mcast interface: %s\n", c->mcast_ifn);
1792 hlen = (AF_INET6 == c->mcast_af) ?
1793 sizeof(struct ipv6hdr) + sizeof(struct udphdr) :
1794 sizeof(struct iphdr) + sizeof(struct udphdr);
1795 mtu = (state == IP_VS_STATE_BACKUP) ?
1796 clamp(dev->mtu, 1500U, 65535U) : 1500U;
1797 min_mtu = (state == IP_VS_STATE_BACKUP) ? 1024 : 1;
1800 c->sync_maxlen = clamp_t(unsigned int,
1801 c->sync_maxlen, min_mtu,
1804 c->sync_maxlen = mtu - hlen;
1806 if (state == IP_VS_STATE_MASTER) {
1812 name = "ipvs-m:%d:%d";
1813 threadfn = sync_thread_master;
1814 } else if (state == IP_VS_STATE_BACKUP) {
1816 if (ipvs->backup_tinfo)
1820 name = "ipvs-b:%d:%d";
1821 threadfn = sync_thread_backup;
1827 if (state == IP_VS_STATE_MASTER) {
1828 struct ipvs_master_sync_state *ms;
1831 ipvs->ms = kcalloc(count, sizeof(ipvs->ms[0]), GFP_KERNEL);
1835 for (id = 0; id < count; id++, ms++) {
1836 INIT_LIST_HEAD(&ms->sync_queue);
1837 ms->sync_queue_len = 0;
1838 ms->sync_queue_delay = 0;
1839 INIT_DELAYED_WORK(&ms->master_wakeup_work,
1840 master_wakeup_work_handler);
1845 ti = kcalloc(count, sizeof(struct ip_vs_sync_thread_data),
1850 for (id = 0; id < count; id++) {
1853 if (state == IP_VS_STATE_BACKUP) {
1855 tinfo->buf = kmalloc(ipvs->bcfg.sync_maxlen,
1861 if (state == IP_VS_STATE_MASTER)
1862 result = make_send_sock(ipvs, id, dev, &tinfo->sock);
1864 result = make_receive_sock(ipvs, id, dev, &tinfo->sock);
1868 task = kthread_run(threadfn, tinfo, name, ipvs->gen, id);
1870 result = PTR_ERR(task);
1876 /* mark as active */
1878 if (state == IP_VS_STATE_MASTER)
1879 ipvs->master_tinfo = ti;
1881 ipvs->backup_tinfo = ti;
1882 spin_lock_bh(&ipvs->sync_buff_lock);
1883 ipvs->sync_state |= state;
1884 spin_unlock_bh(&ipvs->sync_buff_lock);
1886 mutex_unlock(&ipvs->sync_mutex);
1892 /* We do not need RTNL lock anymore, release it here so that
1893 * sock_release below can use rtnl_lock to leave the mcast group.
1896 id = min(id, count - 1);
1898 for (tinfo = ti + id; tinfo >= ti; tinfo--) {
1900 kthread_stop(tinfo->task);
1903 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
1907 mutex_unlock(&ipvs->sync_mutex);
1909 /* No more mutexes, release socks */
1911 for (tinfo = ti + id; tinfo >= ti; tinfo--) {
1913 sock_release(tinfo->sock);
1919 /* decrease the module use count */
1920 ip_vs_use_count_dec();
1924 mutex_unlock(&ipvs->sync_mutex);
1927 /* decrease the module use count */
1928 ip_vs_use_count_dec();
1933 int stop_sync_thread(struct netns_ipvs *ipvs, int state)
1935 struct ip_vs_sync_thread_data *ti, *tinfo;
1939 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1941 mutex_lock(&ipvs->sync_mutex);
1942 if (state == IP_VS_STATE_MASTER) {
1946 ti = ipvs->master_tinfo;
1949 * The lock synchronizes with sb_queue_tail(), so that we don't
1950 * add sync buffers to the queue, when we are already in
1951 * progress of stopping the master sync daemon.
1954 spin_lock_bh(&ipvs->sync_buff_lock);
1955 spin_lock(&ipvs->sync_lock);
1956 ipvs->sync_state &= ~IP_VS_STATE_MASTER;
1957 spin_unlock(&ipvs->sync_lock);
1958 spin_unlock_bh(&ipvs->sync_buff_lock);
1961 for (id = ipvs->threads_mask; id >= 0; id--) {
1962 struct ipvs_master_sync_state *ms = &ipvs->ms[id];
1966 pr_info("stopping master sync thread %d ...\n",
1967 task_pid_nr(tinfo->task));
1968 cancel_delayed_work_sync(&ms->master_wakeup_work);
1969 ret = kthread_stop(tinfo->task);
1975 ipvs->master_tinfo = NULL;
1976 } else if (state == IP_VS_STATE_BACKUP) {
1978 if (!ipvs->backup_tinfo)
1980 ti = ipvs->backup_tinfo;
1982 ipvs->sync_state &= ~IP_VS_STATE_BACKUP;
1984 for (id = ipvs->threads_mask; id >= 0; id--) {
1988 pr_info("stopping backup sync thread %d ...\n",
1989 task_pid_nr(tinfo->task));
1990 ret = kthread_stop(tinfo->task);
1994 ipvs->backup_tinfo = NULL;
1998 id = ipvs->threads_mask;
1999 mutex_unlock(&ipvs->sync_mutex);
2001 /* No more mutexes, release socks */
2002 for (tinfo = ti + id; tinfo >= ti; tinfo--) {
2004 sock_release(tinfo->sock);
2009 /* decrease the module use count */
2010 ip_vs_use_count_dec();
2014 mutex_unlock(&ipvs->sync_mutex);
2019 * Initialize data struct for each netns
2021 int __net_init ip_vs_sync_net_init(struct netns_ipvs *ipvs)
2023 __mutex_init(&ipvs->sync_mutex, "ipvs->sync_mutex", &__ipvs_sync_key);
2024 spin_lock_init(&ipvs->sync_lock);
2025 spin_lock_init(&ipvs->sync_buff_lock);
2029 void ip_vs_sync_net_cleanup(struct netns_ipvs *ipvs)
2033 retc = stop_sync_thread(ipvs, IP_VS_STATE_MASTER);
2034 if (retc && retc != -ESRCH)
2035 pr_err("Failed to stop Master Daemon\n");
2037 retc = stop_sync_thread(ipvs, IP_VS_STATE_BACKUP);
2038 if (retc && retc != -ESRCH)
2039 pr_err("Failed to stop Backup Daemon\n");