projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma
[linux-2.6-block.git] / net / netfilter / ipvs / ip_vs_sync.c
CommitLineData
1da177e4
LT		 1/*
2 * IPVS An implementation of the IP virtual server support for the
3 * LINUX operating system. IPVS is now implemented as a module
4 * over the NetFilter framework. IPVS can be used to build a
5 * high-performance and highly available server based on a
6 * cluster of servers.
7 *
b880c1f0
HS		 8 * Version 1, is capable of handling both version 0 and 1 messages.
9 * Version 0 is the plain old format.
10 * Note Version 0 receivers will just drop Ver 1 messages.
11 * Version 1 is capable of handle IPv6, Persistence data,
12 * time-outs, and firewall marks.
13 * In ver.1 "ip_vs_sync_conn_options" will be sent in netw. order.
14 * Ver. 0 can be turned on by sysctl -w net.ipv4.vs.sync_version=0
15 *
16 * Definitions Message: is a complete datagram
17 * Sync_conn: is a part of a Message
18 * Param Data is an option to a Sync_conn.
19 *
1da177e4
LT		 20 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
21 *
22 * ip_vs_sync: sync connection info from master load balancer to backups
23 * through multicast
24 *
25 * Changes:
26 * Alexandre Cassen : Added master & backup support at a time.
27 * Alexandre Cassen : Added SyncID support for incoming sync
28 * messages filtering.
29 * Justin Ossevoort : Fix endian problem on sync message size.
b880c1f0
HS		 30 * Hans Schillstrom : Added Version 1: i.e. IPv6,
31 * Persistence support, fwmark and time-out.
1da177e4
LT		 32 */
33
9aada7ac
HE		 34#define KMSG_COMPONENT "IPVS"
35#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
36
1da177e4
LT		 37#include <linux/module.h>
38#include <linux/slab.h>
14c85021 39#include <linux/inetdevice.h>
1da177e4
LT		 40#include <linux/net.h>
41#include <linux/completion.h>
42#include <linux/delay.h>
43#include <linux/skbuff.h>
44#include <linux/in.h>
45#include <linux/igmp.h> /* for ip_mc_join_group */
14c85021 46#include <linux/udp.h>
e6dd731c 47#include <linux/err.h>
998e7a76 48#include <linux/kthread.h>
ba6fd850 49#include <linux/wait.h>
e6f225eb 50#include <linux/kernel.h>
1da177e4 51
fe5e7a1e
HS		 52#include <asm/unaligned.h> /* Used for ntoh_seq and hton_seq */
53
1da177e4
LT		 54#include <net/ip.h>
55#include <net/sock.h>
1da177e4
LT		 56
57#include <net/ip_vs.h>
58
59#define IP_VS_SYNC_GROUP 0xe0000051 /* multicast addr - 224.0.0.81 */
60#define IP_VS_SYNC_PORT 8848 /* multicast port */
61
2981bc9a 62#define SYNC_PROTO_VER 1 /* Protocol version in header */
1da177e4 63
ae1d48b2 64static struct lock_class_key __ipvs_sync_key;
1da177e4
LT		 65/*
66 * IPVS sync connection entry
2981bc9a 67 * Version 0, i.e. original version.
1da177e4 68 */
2981bc9a 69struct ip_vs_sync_conn_v0 {
1da177e4
LT		 70 __u8 reserved;
71
72 /* Protocol, addresses and port numbers */
73 __u8 protocol; /* Which protocol (TCP/UDP) */
014d730d
AV		 74 __be16 cport;
75 __be16 vport;
76 __be16 dport;
77 __be32 caddr; /* client address */
78 __be32 vaddr; /* virtual address */
79 __be32 daddr; /* destination address */
1da177e4
LT		 80
81 /* Flags and state transition */
014d730d
AV		 82 __be16 flags; /* status flags */
83 __be16 state; /* state info */
1da177e4
LT		 84
85 /* The sequence options start here */
86};
87
88struct ip_vs_sync_conn_options {
89 struct ip_vs_seq in_seq; /* incoming seq. struct */
90 struct ip_vs_seq out_seq; /* outgoing seq. struct */
91};
92
2981bc9a
HS		 93/*
94 Sync Connection format (sync_conn)
95
96 0 1 2 3
97 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
98 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
99 | Type | Protocol | Ver. | Size |
100 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
101 | Flags |
102 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
103 | State | cport |
104 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
105 | vport | dport |
106 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
107 | fwmark |
108 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
109 | timeout (in sec.) |
110 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
111 | ... |
112 | IP-Addresses (v4 or v6) |
113 | ... |
114 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
115 Optional Parameters.
116 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
117 | Param. Type | Param. Length | Param. data |
118 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
119 | ... |
120 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
121 | | Param Type | Param. Length |
122 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
123 | Param data |
124 | Last Param data should be padded for 32 bit alignment |
125 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
126*/
127
128/*
129 * Type 0, IPv4 sync connection format
130 */
131struct ip_vs_sync_v4 {
132 __u8 type;
133 __u8 protocol; /* Which protocol (TCP/UDP) */
134 __be16 ver_size; /* Version msb 4 bits */
135 /* Flags and state transition */
136 __be32 flags; /* status flags */
137 __be16 state; /* state info */
138 /* Protocol, addresses and port numbers */
139 __be16 cport;
140 __be16 vport;
141 __be16 dport;
142 __be32 fwmark; /* Firewall mark from skb */
143 __be32 timeout; /* cp timeout */
144 __be32 caddr; /* client address */
145 __be32 vaddr; /* virtual address */
146 __be32 daddr; /* destination address */
147 /* The sequence options start here */
148 /* PE data padded to 32bit alignment after seq. options */
149};
150/*
151 * Type 2 messages IPv6
152 */
153struct ip_vs_sync_v6 {
154 __u8 type;
155 __u8 protocol; /* Which protocol (TCP/UDP) */
156 __be16 ver_size; /* Version msb 4 bits */
157 /* Flags and state transition */
158 __be32 flags; /* status flags */
159 __be16 state; /* state info */
160 /* Protocol, addresses and port numbers */
161 __be16 cport;
162 __be16 vport;
163 __be16 dport;
164 __be32 fwmark; /* Firewall mark from skb */
165 __be32 timeout; /* cp timeout */
166 struct in6_addr caddr; /* client address */
167 struct in6_addr vaddr; /* virtual address */
168 struct in6_addr daddr; /* destination address */
169 /* The sequence options start here */
170 /* PE data padded to 32bit alignment after seq. options */
171};
172
173union ip_vs_sync_conn {
174 struct ip_vs_sync_v4 v4;
175 struct ip_vs_sync_v6 v6;
176};
177
178/* Bits in Type field in above */
179#define STYPE_INET6 0
180#define STYPE_F_INET6 (1 << STYPE_INET6)
181
182#define SVER_SHIFT 12 /* Shift to get version */
183#define SVER_MASK 0x0fff /* Mask to strip version */
184
185#define IPVS_OPT_SEQ_DATA 1
186#define IPVS_OPT_PE_DATA 2
187#define IPVS_OPT_PE_NAME 3
188#define IPVS_OPT_PARAM 7
189
190#define IPVS_OPT_F_SEQ_DATA (1 << (IPVS_OPT_SEQ_DATA-1))
191#define IPVS_OPT_F_PE_DATA (1 << (IPVS_OPT_PE_DATA-1))
192#define IPVS_OPT_F_PE_NAME (1 << (IPVS_OPT_PE_NAME-1))
193#define IPVS_OPT_F_PARAM (1 << (IPVS_OPT_PARAM-1))
194
cc0191ae 195struct ip_vs_sync_thread_data {
f131315f 196 struct net *net;
998e7a76
SW		 197 struct socket *sock;
198 char *buf;
f73181c8 199 int id;
cc0191ae
NH		 200};
201
2981bc9a
HS		 202/* Version 0 definition of packet sizes */
203#define SIMPLE_CONN_SIZE (sizeof(struct ip_vs_sync_conn_v0))
1da177e4 204#define FULL_CONN_SIZE \
2981bc9a 205(sizeof(struct ip_vs_sync_conn_v0) + sizeof(struct ip_vs_sync_conn_options))
1da177e4
LT		 206
207
208/*
2981bc9a
HS		 209 The master mulitcasts messages (Datagrams) to the backup load balancers
210 in the following format.
211
212 Version 1:
213 Note, first byte should be Zero, so ver 0 receivers will drop the packet.
1da177e4
LT		 214
215 0 1 2 3
216 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
217 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2981bc9a
HS		 218 | 0 | SyncID | Size |
219 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
220 | Count Conns | Version | Reserved, set to Zero |
1da177e4
LT		 221 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
222 | |
223 | IPVS Sync Connection (1) |
224 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
225 | . |
2981bc9a 226 ~ . ~
1da177e4
LT		 227 | . |
228 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
229 | |
230 | IPVS Sync Connection (n) |
231 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2981bc9a
HS		 232
233 Version 0 Header
234 0 1 2 3
235 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
236 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
237 | Count Conns | SyncID | Size |
238 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
239 | IPVS Sync Connection (1) |
1da177e4
LT		 240*/
241
242#define SYNC_MESG_HEADER_LEN 4
e6f225eb 243#define MAX_CONNS_PER_SYNCBUFF 255 /* nr_conns in ip_vs_sync_mesg is 8 bit */
1da177e4 244
2981bc9a 245/* Version 0 header */
986a0757 246struct ip_vs_sync_mesg_v0 {
1da177e4
LT		 247 __u8 nr_conns;
248 __u8 syncid;
38561437 249 __be16 size;
1da177e4
LT		 250
251 /* ip_vs_sync_conn entries start here */
252};
253
2981bc9a 254/* Version 1 header */
986a0757 255struct ip_vs_sync_mesg {
2981bc9a
HS		 256 __u8 reserved; /* must be zero */
257 __u8 syncid;
38561437 258 __be16 size;
2981bc9a
HS		 259 __u8 nr_conns;
260 __s8 version; /* SYNC_PROTO_VER */
261 __u16 spare;
262 /* ip_vs_sync_conn entries start here */
263};
264
1da177e4
LT		 265struct ip_vs_sync_buff {
266 struct list_head list;
267 unsigned long firstuse;
268
269 /* pointers for the message data */
270 struct ip_vs_sync_mesg *mesg;
271 unsigned char *head;
272 unsigned char *end;
273};
274
fe5e7a1e
HS		 275/*
276 * Copy of struct ip_vs_seq
277 * From unaligned network order to aligned host order
278 */
279static void ntoh_seq(struct ip_vs_seq *no, struct ip_vs_seq *ho)
280{
281 ho->init_seq = get_unaligned_be32(&no->init_seq);
282 ho->delta = get_unaligned_be32(&no->delta);
283 ho->previous_delta = get_unaligned_be32(&no->previous_delta);
284}
1da177e4 285
986a0757
HS		 286/*
287 * Copy of struct ip_vs_seq
288 * From Aligned host order to unaligned network order
289 */
290static void hton_seq(struct ip_vs_seq *ho, struct ip_vs_seq *no)
291{
292 put_unaligned_be32(ho->init_seq, &no->init_seq);
293 put_unaligned_be32(ho->delta, &no->delta);
294 put_unaligned_be32(ho->previous_delta, &no->previous_delta);
295}
296
f73181c8
PNA		 297static inline struct ip_vs_sync_buff *
298sb_dequeue(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
1da177e4
LT		 299{
300 struct ip_vs_sync_buff *sb;
301
f131315f 302 spin_lock_bh(&ipvs->sync_lock);
f73181c8 303 if (list_empty(&ms->sync_queue)) {
1da177e4 304 sb = NULL;
1c003b15 305 __set_current_state(TASK_INTERRUPTIBLE);
1da177e4 306 } else {
f73181c8 307 sb = list_entry(ms->sync_queue.next, struct ip_vs_sync_buff,
1da177e4
LT		 308 list);
309 list_del(&sb->list);
f73181c8
PNA		 310 ms->sync_queue_len--;
311 if (!ms->sync_queue_len)
312 ms->sync_queue_delay = 0;
1da177e4 313 }
f131315f 314 spin_unlock_bh(&ipvs->sync_lock);
1da177e4
LT		 315
316 return sb;
317}
318
986a0757
HS		 319/*
320 * Create a new sync buffer for Version 1 proto.
321 */
f131315f
HS		 322static inline struct ip_vs_sync_buff *
323ip_vs_sync_buff_create(struct netns_ipvs *ipvs)
1da177e4
LT		 324{
325 struct ip_vs_sync_buff *sb;
326
327 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
328 return NULL;
329
f131315f
HS		 330 sb->mesg = kmalloc(ipvs->send_mesg_maxlen, GFP_ATOMIC);
331 if (!sb->mesg) {
1da177e4
LT		 332 kfree(sb);
333 return NULL;
334 }
f73181c8 335 sb->mesg->reserved = 0; /* old nr_conns i.e. must be zero now */
986a0757 336 sb->mesg->version = SYNC_PROTO_VER;
f131315f 337 sb->mesg->syncid = ipvs->master_syncid;
38561437 338 sb->mesg->size = htons(sizeof(struct ip_vs_sync_mesg));
986a0757
HS		 339 sb->mesg->nr_conns = 0;
340 sb->mesg->spare = 0;
341 sb->head = (unsigned char *)sb->mesg + sizeof(struct ip_vs_sync_mesg);
f131315f 342 sb->end = (unsigned char *)sb->mesg + ipvs->send_mesg_maxlen;
986a0757 343
1da177e4
LT		 344 sb->firstuse = jiffies;
345 return sb;
346}
347
348static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
349{
350 kfree(sb->mesg);
351 kfree(sb);
352}
353
f73181c8
PNA		 354static inline void sb_queue_tail(struct netns_ipvs *ipvs,
355 struct ipvs_master_sync_state *ms)
998e7a76 356{
f73181c8 357 struct ip_vs_sync_buff *sb = ms->sync_buff;
f131315f
HS		 358
359 spin_lock(&ipvs->sync_lock);
1c003b15 360 if (ipvs->sync_state & IP_VS_STATE_MASTER &&
f73181c8
PNA		 361 ms->sync_queue_len < sysctl_sync_qlen_max(ipvs)) {
362 if (!ms->sync_queue_len)
363 schedule_delayed_work(&ms->master_wakeup_work,
1c003b15 364 max(IPVS_SYNC_SEND_DELAY, 1));
f73181c8
PNA		 365 ms->sync_queue_len++;
366 list_add_tail(&sb->list, &ms->sync_queue);
367 if ((++ms->sync_queue_delay) == IPVS_SYNC_WAKEUP_RATE)
368 wake_up_process(ms->master_thread);
1c003b15 369 } else
998e7a76 370 ip_vs_sync_buff_release(sb);
f131315f 371 spin_unlock(&ipvs->sync_lock);
998e7a76
SW		 372}
373
1da177e4
LT		 374/*
375 * Get the current sync buffer if it has been created for more
376 * than the specified time or the specified time is zero.
377 */
378static inline struct ip_vs_sync_buff *
f73181c8
PNA		 379get_curr_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms,
380 unsigned long time)
1da177e4
LT		 381{
382 struct ip_vs_sync_buff *sb;
383
f131315f 384 spin_lock_bh(&ipvs->sync_buff_lock);
f73181c8
PNA		 385 sb = ms->sync_buff;
386 if (sb && time_after_eq(jiffies - sb->firstuse, time)) {
387 ms->sync_buff = NULL;
1c003b15 388 __set_current_state(TASK_RUNNING);
1da177e4
LT		 389 } else
390 sb = NULL;
f131315f 391 spin_unlock_bh(&ipvs->sync_buff_lock);
1da177e4
LT		 392 return sb;
393}
394
f73181c8
PNA		 395static inline int
396select_master_thread_id(struct netns_ipvs *ipvs, struct ip_vs_conn *cp)
f131315f 397{
f73181c8 398 return ((long) cp >> (1 + ilog2(sizeof(*cp)))) & ipvs->threads_mask;
b880c1f0
HS		 399}
400
401/*
402 * Create a new sync buffer for Version 0 proto.
403 */
f131315f
HS		 404static inline struct ip_vs_sync_buff *
405ip_vs_sync_buff_create_v0(struct netns_ipvs *ipvs)
b880c1f0
HS		 406{
407 struct ip_vs_sync_buff *sb;
408 struct ip_vs_sync_mesg_v0 *mesg;
409
410 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
411 return NULL;
412
f131315f
HS		 413 sb->mesg = kmalloc(ipvs->send_mesg_maxlen, GFP_ATOMIC);
414 if (!sb->mesg) {
b880c1f0
HS		 415 kfree(sb);
416 return NULL;
417 }
418 mesg = (struct ip_vs_sync_mesg_v0 *)sb->mesg;
419 mesg->nr_conns = 0;
f131315f 420 mesg->syncid = ipvs->master_syncid;
38561437 421 mesg->size = htons(sizeof(struct ip_vs_sync_mesg_v0));
f131315f
HS		 422 sb->head = (unsigned char *)mesg + sizeof(struct ip_vs_sync_mesg_v0);
423 sb->end = (unsigned char *)mesg + ipvs->send_mesg_maxlen;
b880c1f0
HS		 424 sb->firstuse = jiffies;
425 return sb;
426}
427
4d0c875d
JA		 428/* Check if connection is controlled by persistence */
429static inline bool in_persistence(struct ip_vs_conn *cp)
430{
431 for (cp = cp->control; cp; cp = cp->control) {
432 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
433 return true;
434 }
435 return false;
436}
437
749c42b6
JA		 438/* Check if conn should be synced.
439 * pkts: conn packets, use sysctl_sync_threshold to avoid packet check
440 * - (1) sync_refresh_period: reduce sync rate. Additionally, retry
441 * sync_retries times with period of sync_refresh_period/8
442 * - (2) if both sync_refresh_period and sync_period are 0 send sync only
443 * for state changes or only once when pkts matches sync_threshold
444 * - (3) templates: rate can be reduced only with sync_refresh_period or
445 * with (2)
446 */
447static int ip_vs_sync_conn_needed(struct netns_ipvs *ipvs,
448 struct ip_vs_conn *cp, int pkts)
449{
450 unsigned long orig = ACCESS_ONCE(cp->sync_endtime);
451 unsigned long now = jiffies;
452 unsigned long n = (now + cp->timeout) & ~3UL;
453 unsigned int sync_refresh_period;
454 int sync_period;
455 int force;
456
457 /* Check if we sync in current state */
458 if (unlikely(cp->flags & IP_VS_CONN_F_TEMPLATE))
459 force = 0;
4d0c875d
JA		 460 else if (unlikely(sysctl_sync_persist_mode(ipvs) && in_persistence(cp)))
461 return 0;
749c42b6
JA		 462 else if (likely(cp->protocol == IPPROTO_TCP)) {
463 if (!((1 << cp->state) &
464 ((1 << IP_VS_TCP_S_ESTABLISHED) |
465 (1 << IP_VS_TCP_S_FIN_WAIT) |
466 (1 << IP_VS_TCP_S_CLOSE) |
467 (1 << IP_VS_TCP_S_CLOSE_WAIT) |
468 (1 << IP_VS_TCP_S_TIME_WAIT))))
469 return 0;
470 force = cp->state != cp->old_state;
471 if (force && cp->state != IP_VS_TCP_S_ESTABLISHED)
472 goto set;
473 } else if (unlikely(cp->protocol == IPPROTO_SCTP)) {
474 if (!((1 << cp->state) &
475 ((1 << IP_VS_SCTP_S_ESTABLISHED) |
61e7c420
JA		 476 (1 << IP_VS_SCTP_S_SHUTDOWN_SENT) |
477 (1 << IP_VS_SCTP_S_SHUTDOWN_RECEIVED) |
478 (1 << IP_VS_SCTP_S_SHUTDOWN_ACK_SENT) |
479 (1 << IP_VS_SCTP_S_CLOSED))))
749c42b6
JA		 480 return 0;
481 force = cp->state != cp->old_state;
482 if (force && cp->state != IP_VS_SCTP_S_ESTABLISHED)
483 goto set;
484 } else {
485 /* UDP or another protocol with single state */
486 force = 0;
487 }
488
489 sync_refresh_period = sysctl_sync_refresh_period(ipvs);
490 if (sync_refresh_period > 0) {
491 long diff = n - orig;
492 long min_diff = max(cp->timeout >> 1, 10UL * HZ);
493
494 /* Avoid sync if difference is below sync_refresh_period
495 * and below the half timeout.
496 */
497 if (abs(diff) < min_t(long, sync_refresh_period, min_diff)) {
498 int retries = orig & 3;
499
500 if (retries >= sysctl_sync_retries(ipvs))
501 return 0;
502 if (time_before(now, orig - cp->timeout +
503 (sync_refresh_period >> 3)))
504 return 0;
505 n |= retries + 1;
506 }
507 }
508 sync_period = sysctl_sync_period(ipvs);
509 if (sync_period > 0) {
510 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE) &&
511 pkts % sync_period != sysctl_sync_threshold(ipvs))
512 return 0;
513 } else if (sync_refresh_period <= 0 &&
514 pkts != sysctl_sync_threshold(ipvs))
515 return 0;
516
517set:
518 cp->old_state = cp->state;
519 n = cmpxchg(&cp->sync_endtime, orig, n);
520 return n == orig || force;
521}
522
b880c1f0
HS		 523/*
524 * Version 0 , could be switched in by sys_ctl.
525 * Add an ip_vs_conn information into the current sync_buff.
526 */
749c42b6
JA		 527static void ip_vs_sync_conn_v0(struct net *net, struct ip_vs_conn *cp,
528 int pkts)
b880c1f0 529{
f131315f 530 struct netns_ipvs *ipvs = net_ipvs(net);
b880c1f0
HS		 531 struct ip_vs_sync_mesg_v0 *m;
532 struct ip_vs_sync_conn_v0 *s;
f73181c8
PNA		 533 struct ip_vs_sync_buff *buff;
534 struct ipvs_master_sync_state *ms;
535 int id;
b880c1f0
HS		 536 int len;
537
538 if (unlikely(cp->af != AF_INET))
539 return;
540 /* Do not sync ONE PACKET */
541 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
542 return;
543
749c42b6
JA		 544 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
545 return;
546
ac69269a 547 spin_lock_bh(&ipvs->sync_buff_lock);
f73181c8 548 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
ac69269a 549 spin_unlock_bh(&ipvs->sync_buff_lock);
f73181c8
PNA		 550 return;
551 }
552
553 id = select_master_thread_id(ipvs, cp);
554 ms = &ipvs->ms[id];
555 buff = ms->sync_buff;
556 if (buff) {
557 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
558 /* Send buffer if it is for v1 */
559 if (!m->nr_conns) {
560 sb_queue_tail(ipvs, ms);
561 ms->sync_buff = NULL;
562 buff = NULL;
563 }
564 }
565 if (!buff) {
566 buff = ip_vs_sync_buff_create_v0(ipvs);
567 if (!buff) {
ac69269a 568 spin_unlock_bh(&ipvs->sync_buff_lock);
b880c1f0
HS		 569 pr_err("ip_vs_sync_buff_create failed.\n");
570 return;
571 }
f73181c8 572 ms->sync_buff = buff;
b880c1f0
HS		 573 }
574
575 len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
576 SIMPLE_CONN_SIZE;
f73181c8
PNA		 577 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
578 s = (struct ip_vs_sync_conn_v0 *) buff->head;
b880c1f0
HS		 579
580 /* copy members */
581 s->reserved = 0;
582 s->protocol = cp->protocol;
583 s->cport = cp->cport;
584 s->vport = cp->vport;
585 s->dport = cp->dport;
586 s->caddr = cp->caddr.ip;
587 s->vaddr = cp->vaddr.ip;
588 s->daddr = cp->daddr.ip;
589 s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
590 s->state = htons(cp->state);
591 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
592 struct ip_vs_sync_conn_options *opt =
593 (struct ip_vs_sync_conn_options *)&s[1];
594 memcpy(opt, &cp->in_seq, sizeof(*opt));
595 }
596
597 m->nr_conns++;
38561437 598 m->size = htons(ntohs(m->size) + len);
f73181c8 599 buff->head += len;
b880c1f0
HS		 600
601 /* check if there is a space for next one */
f73181c8
PNA		 602 if (buff->head + FULL_CONN_SIZE > buff->end) {
603 sb_queue_tail(ipvs, ms);
604 ms->sync_buff = NULL;
b880c1f0 605 }
ac69269a 606 spin_unlock_bh(&ipvs->sync_buff_lock);
b880c1f0
HS		 607
608 /* synchronize its controller if it has */
749c42b6
JA		 609 cp = cp->control;
610 if (cp) {
611 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
612 pkts = atomic_add_return(1, &cp->in_pkts);
613 else
614 pkts = sysctl_sync_threshold(ipvs);
56184858 615 ip_vs_sync_conn(net, cp, pkts);
749c42b6 616 }
b880c1f0
HS		 617}
618
1da177e4
LT		 619/*
620 * Add an ip_vs_conn information into the current sync_buff.
621 * Called by ip_vs_in.
986a0757 622 * Sending Version 1 messages
1da177e4 623 */
749c42b6 624void ip_vs_sync_conn(struct net *net, struct ip_vs_conn *cp, int pkts)
1da177e4 625{
f131315f 626 struct netns_ipvs *ipvs = net_ipvs(net);
1da177e4 627 struct ip_vs_sync_mesg *m;
986a0757 628 union ip_vs_sync_conn *s;
f73181c8
PNA		 629 struct ip_vs_sync_buff *buff;
630 struct ipvs_master_sync_state *ms;
631 int id;
986a0757
HS		 632 __u8 *p;
633 unsigned int len, pe_name_len, pad;
634
b880c1f0 635 /* Handle old version of the protocol */
7532e8d4 636 if (sysctl_sync_ver(ipvs) == 0) {
749c42b6 637 ip_vs_sync_conn_v0(net, cp, pkts);
b880c1f0
HS		 638 return;
639 }
986a0757
HS		 640 /* Do not sync ONE PACKET */
641 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
642 goto control;
643sloop:
749c42b6
JA		 644 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
645 goto control;
646
986a0757
HS		 647 /* Sanity checks */
648 pe_name_len = 0;
649 if (cp->pe_data_len) {
650 if (!cp->pe_data || !cp->dest) {
651 IP_VS_ERR_RL("SYNC, connection pe_data invalid\n");
652 return;
653 }
654 pe_name_len = strnlen(cp->pe->name, IP_VS_PENAME_MAXLEN);
655 }
1da177e4 656
ac69269a 657 spin_lock_bh(&ipvs->sync_buff_lock);
f73181c8 658 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
ac69269a 659 spin_unlock_bh(&ipvs->sync_buff_lock);
f73181c8
PNA		 660 return;
661 }
662
663 id = select_master_thread_id(ipvs, cp);
664 ms = &ipvs->ms[id];
986a0757
HS		 665
666#ifdef CONFIG_IP_VS_IPV6
667 if (cp->af == AF_INET6)
668 len = sizeof(struct ip_vs_sync_v6);
669 else
670#endif
671 len = sizeof(struct ip_vs_sync_v4);
672
673 if (cp->flags & IP_VS_CONN_F_SEQ_MASK)
674 len += sizeof(struct ip_vs_sync_conn_options) + 2;
675
676 if (cp->pe_data_len)
677 len += cp->pe_data_len + 2; /* + Param hdr field */
678 if (pe_name_len)
679 len += pe_name_len + 2;
680
681 /* check if there is a space for this one */
682 pad = 0;
f73181c8
PNA		 683 buff = ms->sync_buff;
684 if (buff) {
685 m = buff->mesg;
686 pad = (4 - (size_t) buff->head) & 3;
687 /* Send buffer if it is for v0 */
688 if (buff->head + len + pad > buff->end || m->reserved) {
689 sb_queue_tail(ipvs, ms);
690 ms->sync_buff = NULL;
691 buff = NULL;
986a0757
HS		 692 pad = 0;
693 }
694 }
695
f73181c8
PNA		 696 if (!buff) {
697 buff = ip_vs_sync_buff_create(ipvs);
698 if (!buff) {
ac69269a 699 spin_unlock_bh(&ipvs->sync_buff_lock);
1e3e238e 700 pr_err("ip_vs_sync_buff_create failed.\n");
1da177e4
LT		 701 return;
702 }
f73181c8
PNA		 703 ms->sync_buff = buff;
704 m = buff->mesg;
1da177e4
LT		 705 }
706
f73181c8
PNA		 707 p = buff->head;
708 buff->head += pad + len;
38561437 709 m->size = htons(ntohs(m->size) + pad + len);
986a0757
HS		 710 /* Add ev. padding from prev. sync_conn */
711 while (pad--)
712 *(p++) = 0;
713
714 s = (union ip_vs_sync_conn *)p;
715
716 /* Set message type & copy members */
717 s->v4.type = (cp->af == AF_INET6 ? STYPE_F_INET6 : 0);
718 s->v4.ver_size = htons(len & SVER_MASK); /* Version 0 */
719 s->v4.flags = htonl(cp->flags & ~IP_VS_CONN_F_HASHED);
720 s->v4.state = htons(cp->state);
721 s->v4.protocol = cp->protocol;
722 s->v4.cport = cp->cport;
723 s->v4.vport = cp->vport;
724 s->v4.dport = cp->dport;
725 s->v4.fwmark = htonl(cp->fwmark);
726 s->v4.timeout = htonl(cp->timeout / HZ);
1da177e4 727 m->nr_conns++;
1da177e4 728
986a0757
HS		 729#ifdef CONFIG_IP_VS_IPV6
730 if (cp->af == AF_INET6) {
731 p += sizeof(struct ip_vs_sync_v6);
4e3fd7a0
AD		 732 s->v6.caddr = cp->caddr.in6;
733 s->v6.vaddr = cp->vaddr.in6;
734 s->v6.daddr = cp->daddr.in6;
986a0757
HS		 735 } else
736#endif
737 {
738 p += sizeof(struct ip_vs_sync_v4); /* options ptr */
739 s->v4.caddr = cp->caddr.ip;
740 s->v4.vaddr = cp->vaddr.ip;
741 s->v4.daddr = cp->daddr.ip;
742 }
743 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
744 *(p++) = IPVS_OPT_SEQ_DATA;
745 *(p++) = sizeof(struct ip_vs_sync_conn_options);
746 hton_seq((struct ip_vs_seq *)p, &cp->in_seq);
747 p += sizeof(struct ip_vs_seq);
748 hton_seq((struct ip_vs_seq *)p, &cp->out_seq);
749 p += sizeof(struct ip_vs_seq);
1da177e4 750 }
986a0757
HS		 751 /* Handle pe data */
752 if (cp->pe_data_len && cp->pe_data) {
753 *(p++) = IPVS_OPT_PE_DATA;
754 *(p++) = cp->pe_data_len;
755 memcpy(p, cp->pe_data, cp->pe_data_len);
756 p += cp->pe_data_len;
757 if (pe_name_len) {
758 /* Add PE_NAME */
759 *(p++) = IPVS_OPT_PE_NAME;
760 *(p++) = pe_name_len;
761 memcpy(p, cp->pe->name, pe_name_len);
762 p += pe_name_len;
763 }
764 }
765
ac69269a 766 spin_unlock_bh(&ipvs->sync_buff_lock);
1da177e4 767
986a0757 768control:
1da177e4 769 /* synchronize its controller if it has */
986a0757
HS		 770 cp = cp->control;
771 if (!cp)
772 return;
749c42b6
JA		 773 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
774 pkts = atomic_add_return(1, &cp->in_pkts);
775 else
776 pkts = sysctl_sync_threshold(ipvs);
986a0757 777 goto sloop;
1da177e4
LT		 778}
779
fe5e7a1e
HS		 780/*
781 * fill_param used by version 1
782 */
85999283 783static inline int
6e67e586 784ip_vs_conn_fill_param_sync(struct net *net, int af, union ip_vs_sync_conn *sc,
fe5e7a1e
HS		 785 struct ip_vs_conn_param *p,
786 __u8 *pe_data, unsigned int pe_data_len,
787 __u8 *pe_name, unsigned int pe_name_len)
85999283 788{
fe5e7a1e
HS		 789#ifdef CONFIG_IP_VS_IPV6
790 if (af == AF_INET6)
6e67e586 791 ip_vs_conn_fill_param(net, af, sc->v6.protocol,
fe5e7a1e
HS		 792 (const union nf_inet_addr *)&sc->v6.caddr,
793 sc->v6.cport,
794 (const union nf_inet_addr *)&sc->v6.vaddr,
795 sc->v6.vport, p);
796 else
797#endif
6e67e586 798 ip_vs_conn_fill_param(net, af, sc->v4.protocol,
fe5e7a1e
HS		 799 (const union nf_inet_addr *)&sc->v4.caddr,
800 sc->v4.cport,
801 (const union nf_inet_addr *)&sc->v4.vaddr,
802 sc->v4.vport, p);
803 /* Handle pe data */
804 if (pe_data_len) {
805 if (pe_name_len) {
806 char buff[IP_VS_PENAME_MAXLEN+1];
807
808 memcpy(buff, pe_name, pe_name_len);
809 buff[pe_name_len]=0;
810 p->pe = __ip_vs_pe_getbyname(buff);
811 if (!p->pe) {
f131315f
HS		 812 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n",
813 buff);
fe5e7a1e
HS		 814 return 1;
815 }
816 } else {
817 IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n");
818 return 1;
819 }
820
6060c74a 821 p->pe_data = kmemdup(pe_data, pe_data_len, GFP_ATOMIC);
fe5e7a1e 822 if (!p->pe_data) {
982f4051 823 module_put(p->pe->module);
fe5e7a1e
HS		 824 return -ENOMEM;
825 }
fe5e7a1e
HS		 826 p->pe_data_len = pe_data_len;
827 }
85999283
SH		 828 return 0;
829}
1da177e4
LT		 830
831/*
fe5e7a1e
HS		 832 * Connection Add / Update.
833 * Common for version 0 and 1 reception of backup sync_conns.
834 * Param: ...
835 * timeout is in sec.
1da177e4 836 */
9bbac6a9
HS		 837static void ip_vs_proc_conn(struct net *net, struct ip_vs_conn_param *param,
838 unsigned int flags, unsigned int state,
839 unsigned int protocol, unsigned int type,
fe5e7a1e
HS		 840 const union nf_inet_addr *daddr, __be16 dport,
841 unsigned long timeout, __u32 fwmark,
9bbac6a9 842 struct ip_vs_sync_conn_options *opt)
fe5e7a1e
HS		 843{
844 struct ip_vs_dest *dest;
845 struct ip_vs_conn *cp;
a0840e2e 846 struct netns_ipvs *ipvs = net_ipvs(net);
fe5e7a1e 847
d752c364 848 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
fe5e7a1e 849 cp = ip_vs_conn_in_get(param);
d752c364
MRL		 850 if (cp && ((cp->dport != dport) ||
851 !ip_vs_addr_equal(cp->daf, &cp->daddr, daddr))) {
852 if (!(flags & IP_VS_CONN_F_INACTIVE)) {
853 ip_vs_conn_expire_now(cp);
854 __ip_vs_conn_put(cp);
855 cp = NULL;
856 } else {
857 /* This is the expiration message for the
858 * connection that was already replaced, so we
859 * just ignore it.
860 */
861 __ip_vs_conn_put(cp);
862 kfree(param->pe_data);
863 return;
864 }
865 }
866 } else {
fe5e7a1e 867 cp = ip_vs_ct_in_get(param);
d752c364 868 }
fe5e7a1e 869
cdcc5e90
JA		 870 if (cp) {
871 /* Free pe_data */
fe5e7a1e 872 kfree(param->pe_data);
cdcc5e90
JA		 873
874 dest = cp->dest;
ac69269a 875 spin_lock_bh(&cp->lock);
cdcc5e90
JA		 876 if ((cp->flags ^ flags) & IP_VS_CONN_F_INACTIVE &&
877 !(flags & IP_VS_CONN_F_TEMPLATE) && dest) {
878 if (flags & IP_VS_CONN_F_INACTIVE) {
879 atomic_dec(&dest->activeconns);
880 atomic_inc(&dest->inactconns);
881 } else {
882 atomic_inc(&dest->activeconns);
883 atomic_dec(&dest->inactconns);
884 }
885 }
886 flags &= IP_VS_CONN_F_BACKUP_UPD_MASK;
887 flags |= cp->flags & ~IP_VS_CONN_F_BACKUP_UPD_MASK;
888 cp->flags = flags;
ac69269a 889 spin_unlock_bh(&cp->lock);
413c2d04
JA		 890 if (!dest)
891 ip_vs_try_bind_dest(cp);
cdcc5e90 892 } else {
fe5e7a1e
HS		 893 /*
894 * Find the appropriate destination for the connection.
895 * If it is not found the connection will remain unbound
896 * but still handled.
897 */
413c2d04 898 rcu_read_lock();
655eef10
AG		 899 /* This function is only invoked by the synchronization
900 * code. We do not currently support heterogeneous pools
901 * with synchronization, so we can make the assumption that
902 * the svc_af is the same as the dest_af
903 */
904 dest = ip_vs_find_dest(net, type, type, daddr, dport,
905 param->vaddr, param->vport, protocol,
906 fwmark, flags);
fe5e7a1e 907
ba38528a
AG		 908 cp = ip_vs_conn_new(param, type, daddr, dport, flags, dest,
909 fwmark);
413c2d04 910 rcu_read_unlock();
fe5e7a1e 911 if (!cp) {
b734427a 912 kfree(param->pe_data);
fe5e7a1e
HS		 913 IP_VS_DBG(2, "BACKUP, add new conn. failed\n");
914 return;
915 }
528c943f
JA		 916 if (!(flags & IP_VS_CONN_F_TEMPLATE))
917 kfree(param->pe_data);
fe5e7a1e
HS		 918 }
919
920 if (opt)
921 memcpy(&cp->in_seq, opt, sizeof(*opt));
59e0350e 922 atomic_set(&cp->in_pkts, sysctl_sync_threshold(ipvs));
fe5e7a1e
HS		 923 cp->state = state;
924 cp->old_state = cp->state;
925 /*
926 * For Ver 0 messages style
927 * - Not possible to recover the right timeout for templates
928 * - can not find the right fwmark
929 * virtual service. If needed, we can do it for
930 * non-fwmark persistent services.
931 * Ver 1 messages style.
932 * - No problem.
933 */
934 if (timeout) {
935 if (timeout > MAX_SCHEDULE_TIMEOUT / HZ)
936 timeout = MAX_SCHEDULE_TIMEOUT / HZ;
937 cp->timeout = timeout*HZ;
9bbac6a9
HS		 938 } else {
939 struct ip_vs_proto_data *pd;
940
941 pd = ip_vs_proto_data_get(net, protocol);
942 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pd && pd->timeout_table)
943 cp->timeout = pd->timeout_table[state];
944 else
945 cp->timeout = (3*60*HZ);
946 }
fe5e7a1e
HS		 947 ip_vs_conn_put(cp);
948}
949
950/*
951 * Process received multicast message for Version 0
952 */
9bbac6a9
HS		 953static void ip_vs_process_message_v0(struct net *net, const char *buffer,
954 const size_t buflen)
1da177e4 955{
986a0757 956 struct ip_vs_sync_mesg_v0 *m = (struct ip_vs_sync_mesg_v0 *)buffer;
2981bc9a 957 struct ip_vs_sync_conn_v0 *s;
1da177e4 958 struct ip_vs_sync_conn_options *opt;
5c81833c 959 struct ip_vs_protocol *pp;
f11017ec 960 struct ip_vs_conn_param param;
1da177e4
LT		 961 char *p;
962 int i;
963
986a0757 964 p = (char *)buffer + sizeof(struct ip_vs_sync_mesg_v0);
1da177e4 965 for (i=0; i<m->nr_conns; i++) {
95c96174 966 unsigned int flags, state;
87375ab4 967
2ad17def 968 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
fe5e7a1e 969 IP_VS_ERR_RL("BACKUP v0, bogus conn\n");
2ad17def
JA		 970 return;
971 }
2981bc9a 972 s = (struct ip_vs_sync_conn_v0 *) p;
7a4fbb1f 973 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
2ad17def
JA		 974 flags &= ~IP_VS_CONN_F_HASHED;
975 if (flags & IP_VS_CONN_F_SEQ_MASK) {
976 opt = (struct ip_vs_sync_conn_options *)&s[1];
977 p += FULL_CONN_SIZE;
978 if (p > buffer+buflen) {
fe5e7a1e 979 IP_VS_ERR_RL("BACKUP v0, Dropping buffer bogus conn options\n");
2ad17def
JA		 980 return;
981 }
982 } else {
983 opt = NULL;
984 p += SIMPLE_CONN_SIZE;
985 }
986
b209639e 987 state = ntohs(s->state);
2ad17def
JA		 988 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
989 pp = ip_vs_proto_get(s->protocol);
990 if (!pp) {
fe5e7a1e 991 IP_VS_DBG(2, "BACKUP v0, Unsupported protocol %u\n",
2ad17def
JA		 992 s->protocol);
993 continue;
994 }
995 if (state >= pp->num_states) {
fe5e7a1e 996 IP_VS_DBG(2, "BACKUP v0, Invalid %s state %u\n",
2ad17def
JA		 997 pp->name, state);
998 continue;
999 }
1000 } else {
1001 /* protocol in templates is not used for state/timeout */
2ad17def 1002 if (state > 0) {
fe5e7a1e 1003 IP_VS_DBG(2, "BACKUP v0, Invalid template state %u\n",
2ad17def
JA		 1004 state);
1005 state = 0;
1006 }
1007 }
1008
6e67e586 1009 ip_vs_conn_fill_param(net, AF_INET, s->protocol,
fe5e7a1e
HS		 1010 (const union nf_inet_addr *)&s->caddr,
1011 s->cport,
1012 (const union nf_inet_addr *)&s->vaddr,
1013 s->vport, &param);
1014
1015 /* Send timeout as Zero */
9bbac6a9 1016 ip_vs_proc_conn(net, &param, flags, state, s->protocol, AF_INET,
fe5e7a1e 1017 (union nf_inet_addr *)&s->daddr, s->dport,
9bbac6a9 1018 0, 0, opt);
fe5e7a1e
HS		 1019 }
1020}
1021
1022/*
1023 * Handle options
1024 */
1025static inline int ip_vs_proc_seqopt(__u8 *p, unsigned int plen,
1026 __u32 *opt_flags,
1027 struct ip_vs_sync_conn_options *opt)
1028{
1029 struct ip_vs_sync_conn_options *topt;
1030
1031 topt = (struct ip_vs_sync_conn_options *)p;
1032
1033 if (plen != sizeof(struct ip_vs_sync_conn_options)) {
1034 IP_VS_DBG(2, "BACKUP, bogus conn options length\n");
1035 return -EINVAL;
1036 }
1037 if (*opt_flags & IPVS_OPT_F_SEQ_DATA) {
1038 IP_VS_DBG(2, "BACKUP, conn options found twice\n");
1039 return -EINVAL;
1040 }
1041 ntoh_seq(&topt->in_seq, &opt->in_seq);
1042 ntoh_seq(&topt->out_seq, &opt->out_seq);
1043 *opt_flags |= IPVS_OPT_F_SEQ_DATA;
1044 return 0;
1045}
1046
1047static int ip_vs_proc_str(__u8 *p, unsigned int plen, unsigned int *data_len,
1048 __u8 **data, unsigned int maxlen,
1049 __u32 *opt_flags, __u32 flag)
1050{
1051 if (plen > maxlen) {
1052 IP_VS_DBG(2, "BACKUP, bogus par.data len > %d\n", maxlen);
1053 return -EINVAL;
1054 }
1055 if (*opt_flags & flag) {
1056 IP_VS_DBG(2, "BACKUP, Par.data found twice 0x%x\n", flag);
1057 return -EINVAL;
1058 }
1059 *data_len = plen;
1060 *data = p;
1061 *opt_flags |= flag;
1062 return 0;
1063}
1064/*
1065 * Process a Version 1 sync. connection
1066 */
9bbac6a9 1067static inline int ip_vs_proc_sync_conn(struct net *net, __u8 *p, __u8 *msg_end)
fe5e7a1e
HS		 1068{
1069 struct ip_vs_sync_conn_options opt;
1070 union ip_vs_sync_conn *s;
1071 struct ip_vs_protocol *pp;
1072 struct ip_vs_conn_param param;
1073 __u32 flags;
1074 unsigned int af, state, pe_data_len=0, pe_name_len=0;
1075 __u8 *pe_data=NULL, *pe_name=NULL;
1076 __u32 opt_flags=0;
1077 int retc=0;
1078
1079 s = (union ip_vs_sync_conn *) p;
1080
1081 if (s->v6.type & STYPE_F_INET6) {
1082#ifdef CONFIG_IP_VS_IPV6
1083 af = AF_INET6;
1084 p += sizeof(struct ip_vs_sync_v6);
1085#else
1086 IP_VS_DBG(3,"BACKUP, IPv6 msg received, and IPVS is not compiled for IPv6\n");
1087 retc = 10;
1088 goto out;
1089#endif
1090 } else if (!s->v4.type) {
1091 af = AF_INET;
1092 p += sizeof(struct ip_vs_sync_v4);
1093 } else {
1094 return -10;
1095 }
1096 if (p > msg_end)
1097 return -20;
1098
1099 /* Process optional params check Type & Len. */
1100 while (p < msg_end) {
1101 int ptype;
1102 int plen;
1103
1104 if (p+2 > msg_end)
1105 return -30;
1106 ptype = *(p++);
1107 plen = *(p++);
1108
1109 if (!plen || ((p + plen) > msg_end))
1110 return -40;
1111 /* Handle seq option p = param data */
1112 switch (ptype & ~IPVS_OPT_F_PARAM) {
1113 case IPVS_OPT_SEQ_DATA:
1114 if (ip_vs_proc_seqopt(p, plen, &opt_flags, &opt))
1115 return -50;
1116 break;
1117
1118 case IPVS_OPT_PE_DATA:
1119 if (ip_vs_proc_str(p, plen, &pe_data_len, &pe_data,
1120 IP_VS_PEDATA_MAXLEN, &opt_flags,
1121 IPVS_OPT_F_PE_DATA))
1122 return -60;
1123 break;
1124
1125 case IPVS_OPT_PE_NAME:
1126 if (ip_vs_proc_str(p, plen,&pe_name_len, &pe_name,
1127 IP_VS_PENAME_MAXLEN, &opt_flags,
1128 IPVS_OPT_F_PE_NAME))
1129 return -70;
1130 break;
1131
1132 default:
1133 /* Param data mandatory ? */
1134 if (!(ptype & IPVS_OPT_F_PARAM)) {
1135 IP_VS_DBG(3, "BACKUP, Unknown mandatory param %d found\n",
1136 ptype & ~IPVS_OPT_F_PARAM);
1137 retc = 20;
1138 goto out;
1139 }
f11017ec 1140 }
fe5e7a1e
HS		 1141 p += plen; /* Next option */
1142 }
1143
1144 /* Get flags and Mask off unsupported */
1145 flags = ntohl(s->v4.flags) & IP_VS_CONN_F_BACKUP_MASK;
1146 flags |= IP_VS_CONN_F_SYNC;
1147 state = ntohs(s->v4.state);
1148
1149 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
1150 pp = ip_vs_proto_get(s->v4.protocol);
1151 if (!pp) {
1152 IP_VS_DBG(3,"BACKUP, Unsupported protocol %u\n",
1153 s->v4.protocol);
1154 retc = 30;
1155 goto out;
1156 }
1157 if (state >= pp->num_states) {
1158 IP_VS_DBG(3, "BACKUP, Invalid %s state %u\n",
1159 pp->name, state);
1160 retc = 40;
1161 goto out;
1162 }
1163 } else {
1164 /* protocol in templates is not used for state/timeout */
fe5e7a1e
HS		 1165 if (state > 0) {
1166 IP_VS_DBG(3, "BACKUP, Invalid template state %u\n",
1167 state);
1168 state = 0;
1169 }
1170 }
6e67e586
HS		 1171 if (ip_vs_conn_fill_param_sync(net, af, s, &param, pe_data,
1172 pe_data_len, pe_name, pe_name_len)) {
fe5e7a1e
HS		 1173 retc = 50;
1174 goto out;
1175 }
1176 /* If only IPv4, just silent skip IPv6 */
1177 if (af == AF_INET)
9bbac6a9 1178 ip_vs_proc_conn(net, &param, flags, state, s->v4.protocol, af,
fe5e7a1e
HS		 1179 (union nf_inet_addr *)&s->v4.daddr, s->v4.dport,
1180 ntohl(s->v4.timeout), ntohl(s->v4.fwmark),
9bbac6a9
HS		 1181 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1182 );
fe5e7a1e
HS		 1183#ifdef CONFIG_IP_VS_IPV6
1184 else
9bbac6a9 1185 ip_vs_proc_conn(net, &param, flags, state, s->v6.protocol, af,
fe5e7a1e
HS		 1186 (union nf_inet_addr *)&s->v6.daddr, s->v6.dport,
1187 ntohl(s->v6.timeout), ntohl(s->v6.fwmark),
9bbac6a9
HS		 1188 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1189 );
fe5e7a1e 1190#endif
528c943f 1191 ip_vs_pe_put(param.pe);
fe5e7a1e
HS		 1192 return 0;
1193 /* Error exit */
1194out:
1195 IP_VS_DBG(2, "BACKUP, Single msg dropped err:%d\n", retc);
1196 return retc;
1197
1198}
1199/*
1200 * Process received multicast message and create the corresponding
1201 * ip_vs_conn entries.
1202 * Handles Version 0 & 1
1203 */
9bbac6a9
HS		 1204static void ip_vs_process_message(struct net *net, __u8 *buffer,
1205 const size_t buflen)
fe5e7a1e 1206{
f131315f 1207 struct netns_ipvs *ipvs = net_ipvs(net);
986a0757 1208 struct ip_vs_sync_mesg *m2 = (struct ip_vs_sync_mesg *)buffer;
fe5e7a1e 1209 __u8 *p, *msg_end;
986a0757 1210 int i, nr_conns;
fe5e7a1e 1211
986a0757 1212 if (buflen < sizeof(struct ip_vs_sync_mesg_v0)) {
fe5e7a1e
HS		 1213 IP_VS_DBG(2, "BACKUP, message header too short\n");
1214 return;
1215 }
fe5e7a1e 1216
38561437 1217 if (buflen != ntohs(m2->size)) {
fe5e7a1e
HS		 1218 IP_VS_DBG(2, "BACKUP, bogus message size\n");
1219 return;
1220 }
1221 /* SyncID sanity check */
f131315f 1222 if (ipvs->backup_syncid != 0 && m2->syncid != ipvs->backup_syncid) {
fe5e7a1e
HS		 1223 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2->syncid);
1224 return;
1225 }
1226 /* Handle version 1 message */
1227 if ((m2->version == SYNC_PROTO_VER) && (m2->reserved == 0)
1228 && (m2->spare == 0)) {
1229
986a0757 1230 msg_end = buffer + sizeof(struct ip_vs_sync_mesg);
fe5e7a1e
HS		 1231 nr_conns = m2->nr_conns;
1232
1233 for (i=0; i<nr_conns; i++) {
1234 union ip_vs_sync_conn *s;
95c96174 1235 unsigned int size;
fe5e7a1e
HS		 1236 int retc;
1237
1238 p = msg_end;
1239 if (p + sizeof(s->v4) > buffer+buflen) {
1240 IP_VS_ERR_RL("BACKUP, Dropping buffer, to small\n");
1241 return;
b209639e 1242 }
fe5e7a1e
HS		 1243 s = (union ip_vs_sync_conn *)p;
1244 size = ntohs(s->v4.ver_size) & SVER_MASK;
1245 msg_end = p + size;
1246 /* Basic sanity checks */
1247 if (msg_end > buffer+buflen) {
1248 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n");
1da177e4
LT		 1249 return;
1250 }
fe5e7a1e
HS		 1251 if (ntohs(s->v4.ver_size) >> SVER_SHIFT) {
1252 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n",
1253 ntohs(s->v4.ver_size) >> SVER_SHIFT);
1254 return;
b209639e 1255 }
fe5e7a1e 1256 /* Process a single sync_conn */
9bbac6a9
HS		 1257 retc = ip_vs_proc_sync_conn(net, p, msg_end);
1258 if (retc < 0) {
fe5e7a1e
HS		 1259 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n",
1260 retc);
1261 return;
2906f66a 1262 }
fe5e7a1e
HS		 1263 /* Make sure we have 32 bit alignment */
1264 msg_end = p + ((size + 3) & ~3);
b209639e 1265 }
fe5e7a1e
HS		 1266 } else {
1267 /* Old type of message */
9bbac6a9 1268 ip_vs_process_message_v0(net, buffer, buflen);
fe5e7a1e 1269 return;
1da177e4
LT		 1270 }
1271}
1272
1273
1c003b15
PNA		 1274/*
1275 * Setup sndbuf (mode=1) or rcvbuf (mode=0)
1276 */
1277static void set_sock_size(struct sock *sk, int mode, int val)
1278{
1279 /* setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &val, sizeof(val)); */
1280 /* setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)); */
1281 lock_sock(sk);
1282 if (mode) {
1283 val = clamp_t(int, val, (SOCK_MIN_SNDBUF + 1) / 2,
1284 sysctl_wmem_max);
1285 sk->sk_sndbuf = val * 2;
1286 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1287 } else {
1288 val = clamp_t(int, val, (SOCK_MIN_RCVBUF + 1) / 2,
1289 sysctl_rmem_max);
1290 sk->sk_rcvbuf = val * 2;
1291 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1292 }
1293 release_sock(sk);
1294}
1295
1da177e4
LT		 1296/*
1297 * Setup loopback of outgoing multicasts on a sending socket
1298 */
1299static void set_mcast_loop(struct sock *sk, u_char loop)
1300{
1301 struct inet_sock *inet = inet_sk(sk);
1302
1303 /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
1304 lock_sock(sk);
1305 inet->mc_loop = loop ? 1 : 0;
1306 release_sock(sk);
1307}
1308
1309/*
1310 * Specify TTL for outgoing multicasts on a sending socket
1311 */
1312static void set_mcast_ttl(struct sock *sk, u_char ttl)
1313{
1314 struct inet_sock *inet = inet_sk(sk);
1315
1316 /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
1317 lock_sock(sk);
1318 inet->mc_ttl = ttl;
1319 release_sock(sk);
1320}
1321
1322/*
1323 * Specifiy default interface for outgoing multicasts
1324 */
1325static int set_mcast_if(struct sock *sk, char *ifname)
1326{
1327 struct net_device *dev;
1328 struct inet_sock *inet = inet_sk(sk);
f131315f 1329 struct net *net = sock_net(sk);
1da177e4 1330
f131315f
HS		 1331 dev = __dev_get_by_name(net, ifname);
1332 if (!dev)
1da177e4
LT		 1333 return -ENODEV;
1334
1335 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1336 return -EINVAL;
1337
1338 lock_sock(sk);
1339 inet->mc_index = dev->ifindex;
1340 /* inet->mc_addr = 0; */
1341 release_sock(sk);
1342
1343 return 0;
1344}
1345
1346
1347/*
1348 * Set the maximum length of sync message according to the
1349 * specified interface's MTU.
1350 */
f131315f 1351static int set_sync_mesg_maxlen(struct net *net, int sync_state)
1da177e4 1352{
f131315f 1353 struct netns_ipvs *ipvs = net_ipvs(net);
1da177e4
LT		 1354 struct net_device *dev;
1355 int num;
1356
1357 if (sync_state == IP_VS_STATE_MASTER) {
f131315f
HS		 1358 dev = __dev_get_by_name(net, ipvs->master_mcast_ifn);
1359 if (!dev)
1da177e4
LT		 1360 return -ENODEV;
1361
1362 num = (dev->mtu - sizeof(struct iphdr) -
1363 sizeof(struct udphdr) -
1364 SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE;
f131315f 1365 ipvs->send_mesg_maxlen = SYNC_MESG_HEADER_LEN +
e6f225eb 1366 SIMPLE_CONN_SIZE * min(num, MAX_CONNS_PER_SYNCBUFF);
1da177e4 1367 IP_VS_DBG(7, "setting the maximum length of sync sending "
f131315f 1368 "message %d.\n", ipvs->send_mesg_maxlen);
1da177e4 1369 } else if (sync_state == IP_VS_STATE_BACKUP) {
f131315f
HS		 1370 dev = __dev_get_by_name(net, ipvs->backup_mcast_ifn);
1371 if (!dev)
1da177e4
LT		 1372 return -ENODEV;
1373
f131315f 1374 ipvs->recv_mesg_maxlen = dev->mtu -
1da177e4
LT		 1375 sizeof(struct iphdr) - sizeof(struct udphdr);
1376 IP_VS_DBG(7, "setting the maximum length of sync receiving "
f131315f 1377 "message %d.\n", ipvs->recv_mesg_maxlen);
1da177e4
LT		 1378 }
1379
1380 return 0;
1381}
1382
1383
1384/*
1385 * Join a multicast group.
1386 * the group is specified by a class D multicast address 224.0.0.0/8
1387 * in the in_addr structure passed in as a parameter.
1388 */
1389static int
1390join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
1391{
f131315f 1392 struct net *net = sock_net(sk);
1da177e4
LT		 1393 struct ip_mreqn mreq;
1394 struct net_device *dev;
1395 int ret;
1396
1397 memset(&mreq, 0, sizeof(mreq));
1398 memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
1399
f131315f
HS		 1400 dev = __dev_get_by_name(net, ifname);
1401 if (!dev)
1da177e4
LT		 1402 return -ENODEV;
1403 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1404 return -EINVAL;
1405
1406 mreq.imr_ifindex = dev->ifindex;
1407
54ff9ef3 1408 rtnl_lock();
1da177e4
LT		 1409 lock_sock(sk);
1410 ret = ip_mc_join_group(sk, &mreq);
1411 release_sock(sk);
54ff9ef3 1412 rtnl_unlock();
1da177e4
LT		 1413
1414 return ret;
1415}
1416
1417
1418static int bind_mcastif_addr(struct socket *sock, char *ifname)
1419{
f131315f 1420 struct net *net = sock_net(sock->sk);
1da177e4 1421 struct net_device *dev;
a61ced5d 1422 __be32 addr;
1da177e4
LT		 1423 struct sockaddr_in sin;
1424
f131315f
HS		 1425 dev = __dev_get_by_name(net, ifname);
1426 if (!dev)
1da177e4
LT		 1427 return -ENODEV;
1428
1429 addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
1430 if (!addr)
1e3e238e
HE		 1431 pr_err("You probably need to specify IP address on "
1432 "multicast interface.\n");
1da177e4 1433
14d5e834
HH		 1434 IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
1435 ifname, &addr);
1da177e4
LT		 1436
1437 /* Now bind the socket with the address of multicast interface */
1438 sin.sin_family = AF_INET;
1439 sin.sin_addr.s_addr = addr;
1440 sin.sin_port = 0;
1441
1442 return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
1443}
1444
1445/*
1446 * Set up sending multicast socket over UDP
1447 */
f73181c8 1448static struct socket *make_send_sock(struct net *net, int id)
1da177e4 1449{
f131315f 1450 struct netns_ipvs *ipvs = net_ipvs(net);
f73181c8
PNA		 1451 /* multicast addr */
1452 struct sockaddr_in mcast_addr = {
1453 .sin_family = AF_INET,
1454 .sin_port = cpu_to_be16(IP_VS_SYNC_PORT + id),
1455 .sin_addr.s_addr = cpu_to_be32(IP_VS_SYNC_GROUP),
1456 };
1da177e4 1457 struct socket *sock;
e6dd731c 1458 int result;
1da177e4 1459
26abe143
EB		 1460 /* First create a socket */
1461 result = sock_create_kern(net, PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
e6dd731c 1462 if (result < 0) {
1e3e238e 1463 pr_err("Error during creation of socket; terminating\n");
e6dd731c 1464 return ERR_PTR(result);
1da177e4 1465 }
f131315f 1466 result = set_mcast_if(sock->sk, ipvs->master_mcast_ifn);
e6dd731c 1467 if (result < 0) {
1e3e238e 1468 pr_err("Error setting outbound mcast interface\n");
1da177e4
LT		 1469 goto error;
1470 }
1471
1472 set_mcast_loop(sock->sk, 0);
1473 set_mcast_ttl(sock->sk, 1);
1c003b15
PNA		 1474 result = sysctl_sync_sock_size(ipvs);
1475 if (result > 0)
1476 set_sock_size(sock->sk, 1, result);
1da177e4 1477
f131315f 1478 result = bind_mcastif_addr(sock, ipvs->master_mcast_ifn);
e6dd731c 1479 if (result < 0) {
1e3e238e 1480 pr_err("Error binding address of the mcast interface\n");
1da177e4
LT		 1481 goto error;
1482 }
1483
e6dd731c
SW		 1484 result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr,
1485 sizeof(struct sockaddr), 0);
1486 if (result < 0) {
1e3e238e 1487 pr_err("Error connecting to the multicast addr\n");
1da177e4
LT		 1488 goto error;
1489 }
1490
1491 return sock;
1492
1ae132b0 1493error:
26abe143 1494 sock_release(sock);
e6dd731c 1495 return ERR_PTR(result);
1da177e4
LT		 1496}
1497
1498
1499/*
1500 * Set up receiving multicast socket over UDP
1501 */
f73181c8 1502static struct socket *make_receive_sock(struct net *net, int id)
1da177e4 1503{
f131315f 1504 struct netns_ipvs *ipvs = net_ipvs(net);
f73181c8
PNA		 1505 /* multicast addr */
1506 struct sockaddr_in mcast_addr = {
1507 .sin_family = AF_INET,
1508 .sin_port = cpu_to_be16(IP_VS_SYNC_PORT + id),
1509 .sin_addr.s_addr = cpu_to_be32(IP_VS_SYNC_GROUP),
1510 };
1da177e4 1511 struct socket *sock;
e6dd731c 1512 int result;
1da177e4
LT		 1513
1514 /* First create a socket */
26abe143 1515 result = sock_create_kern(net, PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
e6dd731c 1516 if (result < 0) {
1e3e238e 1517 pr_err("Error during creation of socket; terminating\n");
e6dd731c 1518 return ERR_PTR(result);
1da177e4 1519 }
1da177e4 1520 /* it is equivalent to the REUSEADDR option in user-space */
4a17fd52 1521 sock->sk->sk_reuse = SK_CAN_REUSE;
1c003b15
PNA		 1522 result = sysctl_sync_sock_size(ipvs);
1523 if (result > 0)
1524 set_sock_size(sock->sk, 0, result);
1da177e4 1525
e6dd731c
SW		 1526 result = sock->ops->bind(sock, (struct sockaddr *) &mcast_addr,
1527 sizeof(struct sockaddr));
1528 if (result < 0) {
1e3e238e 1529 pr_err("Error binding to the multicast addr\n");
1da177e4
LT		 1530 goto error;
1531 }
1532
1533 /* join the multicast group */
e6dd731c
SW		 1534 result = join_mcast_group(sock->sk,
1535 (struct in_addr *) &mcast_addr.sin_addr,
f131315f 1536 ipvs->backup_mcast_ifn);
e6dd731c 1537 if (result < 0) {
1e3e238e 1538 pr_err("Error joining to the multicast group\n");
1da177e4
LT		 1539 goto error;
1540 }
1541
1542 return sock;
1543
1ae132b0 1544error:
26abe143 1545 sock_release(sock);
e6dd731c 1546 return ERR_PTR(result);
1da177e4
LT		 1547}
1548
1549
1550static int
1551ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
1552{
1553 struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
1554 struct kvec iov;
1555 int len;
1556
1557 EnterFunction(7);
1558 iov.iov_base = (void *)buffer;
1559 iov.iov_len = length;
1560
1561 len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
1562
1563 LeaveFunction(7);
1564 return len;
1565}
1566
1c003b15 1567static int
1da177e4
LT		 1568ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
1569{
1570 int msize;
1c003b15 1571 int ret;
1da177e4 1572
38561437 1573 msize = ntohs(msg->size);
1da177e4 1574
1c003b15
PNA		 1575 ret = ip_vs_send_async(sock, (char *)msg, msize);
1576 if (ret >= 0 || ret == -EAGAIN)
1577 return ret;
1578 pr_err("ip_vs_send_async error %d\n", ret);
1579 return 0;
1da177e4
LT		 1580}
1581
1582static int
1583ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
1584{
1585 struct msghdr msg = {NULL,};
1586 struct kvec iov;
1587 int len;
1588
1589 EnterFunction(7);
1590
1591 /* Receive a packet */
1592 iov.iov_base = buffer;
1593 iov.iov_len = (size_t)buflen;
1594
f73181c8 1595 len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, MSG_DONTWAIT);
1da177e4
LT		 1596
1597 if (len < 0)
f73181c8 1598 return len;
1da177e4
LT		 1599
1600 LeaveFunction(7);
1601 return len;
1602}
1603
1c003b15
PNA		 1604/* Wakeup the master thread for sending */
1605static void master_wakeup_work_handler(struct work_struct *work)
1606{
f73181c8
PNA		 1607 struct ipvs_master_sync_state *ms =
1608 container_of(work, struct ipvs_master_sync_state,
1609 master_wakeup_work.work);
1610 struct netns_ipvs *ipvs = ms->ipvs;
1c003b15
PNA		 1611
1612 spin_lock_bh(&ipvs->sync_lock);
f73181c8
PNA		 1613 if (ms->sync_queue_len &&
1614 ms->sync_queue_delay < IPVS_SYNC_WAKEUP_RATE) {
1615 ms->sync_queue_delay = IPVS_SYNC_WAKEUP_RATE;
1616 wake_up_process(ms->master_thread);
1c003b15
PNA		 1617 }
1618 spin_unlock_bh(&ipvs->sync_lock);
1619}
1620
1621/* Get next buffer to send */
1622static inline struct ip_vs_sync_buff *
f73181c8 1623next_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
1c003b15
PNA		 1624{
1625 struct ip_vs_sync_buff *sb;
1626
f73181c8 1627 sb = sb_dequeue(ipvs, ms);
1c003b15
PNA		 1628 if (sb)
1629 return sb;
1630 /* Do not delay entries in buffer for more than 2 seconds */
f73181c8 1631 return get_curr_sync_buff(ipvs, ms, IPVS_SYNC_FLUSH_TIME);
1c003b15 1632}
1da177e4 1633
998e7a76 1634static int sync_thread_master(void *data)
1da177e4 1635{
998e7a76 1636 struct ip_vs_sync_thread_data *tinfo = data;
f131315f 1637 struct netns_ipvs *ipvs = net_ipvs(tinfo->net);
f73181c8 1638 struct ipvs_master_sync_state *ms = &ipvs->ms[tinfo->id];
1c003b15 1639 struct sock *sk = tinfo->sock->sk;
1da177e4
LT		 1640 struct ip_vs_sync_buff *sb;
1641
1e3e238e 1642 pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
f73181c8
PNA		 1643 "syncid = %d, id = %d\n",
1644 ipvs->master_mcast_ifn, ipvs->master_syncid, tinfo->id);
1da177e4 1645
1c003b15 1646 for (;;) {
f73181c8 1647 sb = next_sync_buff(ipvs, ms);
1c003b15
PNA		 1648 if (unlikely(kthread_should_stop()))
1649 break;
1650 if (!sb) {
1651 schedule_timeout(IPVS_SYNC_CHECK_PERIOD);
1652 continue;
1da177e4 1653 }
1c003b15 1654 while (ip_vs_send_sync_msg(tinfo->sock, sb->mesg) < 0) {
9dcbe1b8
GU		 1655 /* (Ab)use interruptible sleep to avoid increasing
1656 * the load avg.
1657 */
1658 __wait_event_interruptible(*sk_sleep(sk),
1c003b15 1659 sock_writeable(sk) ||
35a2af94 1660 kthread_should_stop());
1c003b15
PNA		 1661 if (unlikely(kthread_should_stop()))
1662 goto done;
1da177e4 1663 }
1c003b15 1664 ip_vs_sync_buff_release(sb);
1da177e4
LT		 1665 }
1666
1c003b15
PNA		 1667done:
1668 __set_current_state(TASK_RUNNING);
1669 if (sb)
1670 ip_vs_sync_buff_release(sb);
1671
1da177e4 1672 /* clean up the sync_buff queue */
f73181c8 1673 while ((sb = sb_dequeue(ipvs, ms)))
1da177e4 1674 ip_vs_sync_buff_release(sb);
1c003b15 1675 __set_current_state(TASK_RUNNING);
1da177e4
LT		 1676
1677 /* clean up the current sync_buff */
f73181c8 1678 sb = get_curr_sync_buff(ipvs, ms, 0);
f131315f 1679 if (sb)
1da177e4 1680 ip_vs_sync_buff_release(sb);
1da177e4
LT		 1681
1682 /* release the sending multicast socket */
26abe143 1683 sock_release(tinfo->sock);
998e7a76
SW		 1684 kfree(tinfo);
1685
1686 return 0;
1da177e4
LT		 1687}
1688
1689
998e7a76 1690static int sync_thread_backup(void *data)
1da177e4 1691{
998e7a76 1692 struct ip_vs_sync_thread_data *tinfo = data;
f131315f 1693 struct netns_ipvs *ipvs = net_ipvs(tinfo->net);
1da177e4
LT		 1694 int len;
1695
1e3e238e 1696 pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
f73181c8
PNA		 1697 "syncid = %d, id = %d\n",
1698 ipvs->backup_mcast_ifn, ipvs->backup_syncid, tinfo->id);
1da177e4 1699
998e7a76 1700 while (!kthread_should_stop()) {
aa395145 1701 wait_event_interruptible(*sk_sleep(tinfo->sock->sk),
ba6fd850
SW		 1702 !skb_queue_empty(&tinfo->sock->sk->sk_receive_queue)
1703 || kthread_should_stop());
1704
998e7a76
SW		 1705 /* do we have data now? */
1706 while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) {
1707 len = ip_vs_receive(tinfo->sock, tinfo->buf,
f131315f 1708 ipvs->recv_mesg_maxlen);
998e7a76 1709 if (len <= 0) {
f73181c8
PNA		 1710 if (len != -EAGAIN)
1711 pr_err("receiving message error\n");
1da177e4
LT		 1712 break;
1713 }
998e7a76 1714
f131315f 1715 ip_vs_process_message(tinfo->net, tinfo->buf, len);
1da177e4 1716 }
1da177e4
LT		 1717 }
1718
1719 /* release the sending multicast socket */
26abe143 1720 sock_release(tinfo->sock);
998e7a76
SW		 1721 kfree(tinfo->buf);
1722 kfree(tinfo);
1da177e4 1723
998e7a76 1724 return 0;
1da177e4
LT		 1725}
1726
1727
f131315f 1728int start_sync_thread(struct net *net, int state, char *mcast_ifn, __u8 syncid)
1da177e4 1729{
998e7a76 1730 struct ip_vs_sync_thread_data *tinfo;
f73181c8 1731 struct task_struct **array = NULL, *task;
998e7a76 1732 struct socket *sock;
f131315f 1733 struct netns_ipvs *ipvs = net_ipvs(net);
f73181c8 1734 char *name;
998e7a76 1735 int (*threadfn)(void *data);
f73181c8 1736 int id, count;
998e7a76 1737 int result = -ENOMEM;
1da177e4 1738
1e3e238e 1739 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
998e7a76 1740 IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n",
2981bc9a 1741 sizeof(struct ip_vs_sync_conn_v0));
1da177e4 1742
f73181c8
PNA		 1743 if (!ipvs->sync_state) {
1744 count = clamp(sysctl_sync_ports(ipvs), 1, IPVS_SYNC_PORTS_MAX);
1745 ipvs->threads_mask = count - 1;
1746 } else
1747 count = ipvs->threads_mask + 1;
ae1d48b2 1748
998e7a76 1749 if (state == IP_VS_STATE_MASTER) {
f73181c8 1750 if (ipvs->ms)
998e7a76 1751 return -EEXIST;
1da177e4 1752
f131315f
HS		 1753 strlcpy(ipvs->master_mcast_ifn, mcast_ifn,
1754 sizeof(ipvs->master_mcast_ifn));
1755 ipvs->master_syncid = syncid;
f73181c8 1756 name = "ipvs-m:%d:%d";
998e7a76 1757 threadfn = sync_thread_master;
998e7a76 1758 } else if (state == IP_VS_STATE_BACKUP) {
f73181c8 1759 if (ipvs->backup_threads)
998e7a76
SW		 1760 return -EEXIST;
1761
f131315f
HS		 1762 strlcpy(ipvs->backup_mcast_ifn, mcast_ifn,
1763 sizeof(ipvs->backup_mcast_ifn));
1764 ipvs->backup_syncid = syncid;
f73181c8 1765 name = "ipvs-b:%d:%d";
998e7a76 1766 threadfn = sync_thread_backup;
1da177e4 1767 } else {
1da177e4
LT		 1768 return -EINVAL;
1769 }
1770
f73181c8
PNA		 1771 if (state == IP_VS_STATE_MASTER) {
1772 struct ipvs_master_sync_state *ms;
1da177e4 1773
f73181c8
PNA		 1774 ipvs->ms = kzalloc(count * sizeof(ipvs->ms[0]), GFP_KERNEL);
1775 if (!ipvs->ms)
1776 goto out;
1777 ms = ipvs->ms;
1778 for (id = 0; id < count; id++, ms++) {
1779 INIT_LIST_HEAD(&ms->sync_queue);
1780 ms->sync_queue_len = 0;
1781 ms->sync_queue_delay = 0;
1782 INIT_DELAYED_WORK(&ms->master_wakeup_work,
1783 master_wakeup_work_handler);
1784 ms->ipvs = ipvs;
1785 }
1786 } else {
1787 array = kzalloc(count * sizeof(struct task_struct *),
1788 GFP_KERNEL);
1789 if (!array)
1790 goto out;
998e7a76 1791 }
f73181c8 1792 set_sync_mesg_maxlen(net, state);
1da177e4 1793
f73181c8
PNA		 1794 tinfo = NULL;
1795 for (id = 0; id < count; id++) {
1796 if (state == IP_VS_STATE_MASTER)
1797 sock = make_send_sock(net, id);
1798 else
1799 sock = make_receive_sock(net, id);
1800 if (IS_ERR(sock)) {
1801 result = PTR_ERR(sock);
1802 goto outtinfo;
1803 }
1804 tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL);
1805 if (!tinfo)
1806 goto outsocket;
1807 tinfo->net = net;
1808 tinfo->sock = sock;
1809 if (state == IP_VS_STATE_BACKUP) {
1810 tinfo->buf = kmalloc(ipvs->recv_mesg_maxlen,
1811 GFP_KERNEL);
1812 if (!tinfo->buf)
1813 goto outtinfo;
b425df4c
DC		 1814 } else {
1815 tinfo->buf = NULL;
f73181c8
PNA		 1816 }
1817 tinfo->id = id;
1da177e4 1818
f73181c8
PNA		 1819 task = kthread_run(threadfn, tinfo, name, ipvs->gen, id);
1820 if (IS_ERR(task)) {
1821 result = PTR_ERR(task);
1822 goto outtinfo;
1823 }
1824 tinfo = NULL;
1825 if (state == IP_VS_STATE_MASTER)
1826 ipvs->ms[id].master_thread = task;
1827 else
1828 array[id] = task;
998e7a76 1829 }
1da177e4 1830
998e7a76 1831 /* mark as active */
f73181c8
PNA		 1832
1833 if (state == IP_VS_STATE_BACKUP)
1834 ipvs->backup_threads = array;
1835 spin_lock_bh(&ipvs->sync_buff_lock);
f131315f 1836 ipvs->sync_state |= state;
f73181c8 1837 spin_unlock_bh(&ipvs->sync_buff_lock);
1da177e4 1838
998e7a76
SW		 1839 /* increase the module use count */
1840 ip_vs_use_count_inc();
1da177e4
LT		 1841
1842 return 0;
1da177e4 1843
998e7a76 1844outsocket:
26abe143 1845 sock_release(sock);
f73181c8
PNA		 1846
1847outtinfo:
1848 if (tinfo) {
26abe143 1849 sock_release(tinfo->sock);
f73181c8
PNA		 1850 kfree(tinfo->buf);
1851 kfree(tinfo);
1852 }
1853 count = id;
1854 while (count-- > 0) {
1855 if (state == IP_VS_STATE_MASTER)
1856 kthread_stop(ipvs->ms[count].master_thread);
1857 else
1858 kthread_stop(array[count]);
1859 }
1860 kfree(array);
1861
998e7a76 1862out:
f73181c8
PNA		 1863 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
1864 kfree(ipvs->ms);
1865 ipvs->ms = NULL;
1866 }
998e7a76 1867 return result;
1da177e4
LT		 1868}
1869
1870
f131315f 1871int stop_sync_thread(struct net *net, int state)
1da177e4 1872{
f131315f 1873 struct netns_ipvs *ipvs = net_ipvs(net);
f73181c8
PNA		 1874 struct task_struct **array;
1875 int id;
1ae132b0 1876 int retc = -EINVAL;
f131315f 1877
1e3e238e 1878 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1da177e4 1879
1da177e4 1880 if (state == IP_VS_STATE_MASTER) {
f73181c8 1881 if (!ipvs->ms)
998e7a76 1882 return -ESRCH;
cc0191ae 1883
998e7a76
SW		 1884 /*
1885 * The lock synchronizes with sb_queue_tail(), so that we don't
1886 * add sync buffers to the queue, when we are already in
1887 * progress of stopping the master sync daemon.
1888 */
1da177e4 1889
f73181c8
PNA		 1890 spin_lock_bh(&ipvs->sync_buff_lock);
1891 spin_lock(&ipvs->sync_lock);
f131315f 1892 ipvs->sync_state &= ~IP_VS_STATE_MASTER;
f73181c8
PNA		 1893 spin_unlock(&ipvs->sync_lock);
1894 spin_unlock_bh(&ipvs->sync_buff_lock);
1895
1896 retc = 0;
1897 for (id = ipvs->threads_mask; id >= 0; id--) {
1898 struct ipvs_master_sync_state *ms = &ipvs->ms[id];
1899 int ret;
1900
1901 pr_info("stopping master sync thread %d ...\n",
1902 task_pid_nr(ms->master_thread));
1903 cancel_delayed_work_sync(&ms->master_wakeup_work);
1904 ret = kthread_stop(ms->master_thread);
1905 if (retc >= 0)
1906 retc = ret;
1907 }
1908 kfree(ipvs->ms);
1909 ipvs->ms = NULL;
998e7a76 1910 } else if (state == IP_VS_STATE_BACKUP) {
f73181c8 1911 if (!ipvs->backup_threads)
998e7a76
SW		 1912 return -ESRCH;
1913
f131315f 1914 ipvs->sync_state &= ~IP_VS_STATE_BACKUP;
f73181c8
PNA		 1915 array = ipvs->backup_threads;
1916 retc = 0;
1917 for (id = ipvs->threads_mask; id >= 0; id--) {
1918 int ret;
1919
1920 pr_info("stopping backup sync thread %d ...\n",
1921 task_pid_nr(array[id]));
1922 ret = kthread_stop(array[id]);
1923 if (retc >= 0)
1924 retc = ret;
1925 }
1926 kfree(array);
1927 ipvs->backup_threads = NULL;
998e7a76 1928 }
1da177e4 1929
998e7a76
SW		 1930 /* decrease the module use count */
1931 ip_vs_use_count_dec();
1da177e4 1932
1ae132b0 1933 return retc;
1da177e4 1934}
61b1ab45
HS		 1935
1936/*
1937 * Initialize data struct for each netns
1938 */
503cf15a 1939int __net_init ip_vs_sync_net_init(struct net *net)
61b1ab45 1940{
f131315f
HS		 1941 struct netns_ipvs *ipvs = net_ipvs(net);
1942
ae1d48b2 1943 __mutex_init(&ipvs->sync_mutex, "ipvs->sync_mutex", &__ipvs_sync_key);
f131315f
HS		 1944 spin_lock_init(&ipvs->sync_lock);
1945 spin_lock_init(&ipvs->sync_buff_lock);
61b1ab45
HS		 1946 return 0;
1947}
1948
503cf15a 1949void ip_vs_sync_net_cleanup(struct net *net)
61b1ab45 1950{
1ae132b0 1951 int retc;
ae1d48b2 1952 struct netns_ipvs *ipvs = net_ipvs(net);
1ae132b0 1953
ae1d48b2 1954 mutex_lock(&ipvs->sync_mutex);
1ae132b0
HS		 1955 retc = stop_sync_thread(net, IP_VS_STATE_MASTER);
1956 if (retc && retc != -ESRCH)
1957 pr_err("Failed to stop Master Daemon\n");
1958
1959 retc = stop_sync_thread(net, IP_VS_STATE_BACKUP);
1960 if (retc && retc != -ESRCH)
1961 pr_err("Failed to stop Backup Daemon\n");
ae1d48b2 1962 mutex_unlock(&ipvs->sync_mutex);
61b1ab45 1963}
Linux 6.x block layer and io_uring tree(s)