projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branches 'acpi-ec', 'acpi-video', 'acpi-button' and 'acpi-thermal'
[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 {
37b68e6d 196 struct netns_ipvs *ipvs;
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
d3328817
JA		 265union ipvs_sockaddr {
266 struct sockaddr_in in;
267 struct sockaddr_in6 in6;
268};
269
1da177e4
LT		 270struct ip_vs_sync_buff {
271 struct list_head list;
272 unsigned long firstuse;
273
274 /* pointers for the message data */
275 struct ip_vs_sync_mesg *mesg;
276 unsigned char *head;
277 unsigned char *end;
278};
279
fe5e7a1e
HS		 280/*
281 * Copy of struct ip_vs_seq
282 * From unaligned network order to aligned host order
283 */
284static void ntoh_seq(struct ip_vs_seq *no, struct ip_vs_seq *ho)
285{
286 ho->init_seq = get_unaligned_be32(&no->init_seq);
287 ho->delta = get_unaligned_be32(&no->delta);
288 ho->previous_delta = get_unaligned_be32(&no->previous_delta);
289}
1da177e4 290
986a0757
HS		 291/*
292 * Copy of struct ip_vs_seq
293 * From Aligned host order to unaligned network order
294 */
295static void hton_seq(struct ip_vs_seq *ho, struct ip_vs_seq *no)
296{
297 put_unaligned_be32(ho->init_seq, &no->init_seq);
298 put_unaligned_be32(ho->delta, &no->delta);
299 put_unaligned_be32(ho->previous_delta, &no->previous_delta);
300}
301
f73181c8
PNA		 302static inline struct ip_vs_sync_buff *
303sb_dequeue(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
1da177e4
LT		 304{
305 struct ip_vs_sync_buff *sb;
306
f131315f 307 spin_lock_bh(&ipvs->sync_lock);
f73181c8 308 if (list_empty(&ms->sync_queue)) {
1da177e4 309 sb = NULL;
1c003b15 310 __set_current_state(TASK_INTERRUPTIBLE);
1da177e4 311 } else {
f73181c8 312 sb = list_entry(ms->sync_queue.next, struct ip_vs_sync_buff,
1da177e4
LT		 313 list);
314 list_del(&sb->list);
f73181c8
PNA		 315 ms->sync_queue_len--;
316 if (!ms->sync_queue_len)
317 ms->sync_queue_delay = 0;
1da177e4 318 }
f131315f 319 spin_unlock_bh(&ipvs->sync_lock);
1da177e4
LT		 320
321 return sb;
322}
323
986a0757
HS		 324/*
325 * Create a new sync buffer for Version 1 proto.
326 */
f131315f 327static inline struct ip_vs_sync_buff *
e4ff6751 328ip_vs_sync_buff_create(struct netns_ipvs *ipvs, unsigned int len)
1da177e4
LT		 329{
330 struct ip_vs_sync_buff *sb;
331
332 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
333 return NULL;
334
e4ff6751
JA		 335 len = max_t(unsigned int, len + sizeof(struct ip_vs_sync_mesg),
336 ipvs->mcfg.sync_maxlen);
337 sb->mesg = kmalloc(len, GFP_ATOMIC);
f131315f 338 if (!sb->mesg) {
1da177e4
LT		 339 kfree(sb);
340 return NULL;
341 }
f73181c8 342 sb->mesg->reserved = 0; /* old nr_conns i.e. must be zero now */
986a0757 343 sb->mesg->version = SYNC_PROTO_VER;
e4ff6751 344 sb->mesg->syncid = ipvs->mcfg.syncid;
38561437 345 sb->mesg->size = htons(sizeof(struct ip_vs_sync_mesg));
986a0757
HS		 346 sb->mesg->nr_conns = 0;
347 sb->mesg->spare = 0;
348 sb->head = (unsigned char *)sb->mesg + sizeof(struct ip_vs_sync_mesg);
e4ff6751 349 sb->end = (unsigned char *)sb->mesg + len;
986a0757 350
1da177e4
LT		 351 sb->firstuse = jiffies;
352 return sb;
353}
354
355static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
356{
357 kfree(sb->mesg);
358 kfree(sb);
359}
360
f73181c8
PNA		 361static inline void sb_queue_tail(struct netns_ipvs *ipvs,
362 struct ipvs_master_sync_state *ms)
998e7a76 363{
f73181c8 364 struct ip_vs_sync_buff *sb = ms->sync_buff;
f131315f
HS		 365
366 spin_lock(&ipvs->sync_lock);
1c003b15 367 if (ipvs->sync_state & IP_VS_STATE_MASTER &&
f73181c8
PNA		 368 ms->sync_queue_len < sysctl_sync_qlen_max(ipvs)) {
369 if (!ms->sync_queue_len)
370 schedule_delayed_work(&ms->master_wakeup_work,
1c003b15 371 max(IPVS_SYNC_SEND_DELAY, 1));
f73181c8
PNA		 372 ms->sync_queue_len++;
373 list_add_tail(&sb->list, &ms->sync_queue);
374 if ((++ms->sync_queue_delay) == IPVS_SYNC_WAKEUP_RATE)
375 wake_up_process(ms->master_thread);
1c003b15 376 } else
998e7a76 377 ip_vs_sync_buff_release(sb);
f131315f 378 spin_unlock(&ipvs->sync_lock);
998e7a76
SW		 379}
380
1da177e4
LT		 381/*
382 * Get the current sync buffer if it has been created for more
383 * than the specified time or the specified time is zero.
384 */
385static inline struct ip_vs_sync_buff *
f73181c8
PNA		 386get_curr_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms,
387 unsigned long time)
1da177e4
LT		 388{
389 struct ip_vs_sync_buff *sb;
390
f131315f 391 spin_lock_bh(&ipvs->sync_buff_lock);
f73181c8
PNA		 392 sb = ms->sync_buff;
393 if (sb && time_after_eq(jiffies - sb->firstuse, time)) {
394 ms->sync_buff = NULL;
1c003b15 395 __set_current_state(TASK_RUNNING);
1da177e4
LT		 396 } else
397 sb = NULL;
f131315f 398 spin_unlock_bh(&ipvs->sync_buff_lock);
1da177e4
LT		 399 return sb;
400}
401
f73181c8
PNA		 402static inline int
403select_master_thread_id(struct netns_ipvs *ipvs, struct ip_vs_conn *cp)
f131315f 404{
f73181c8 405 return ((long) cp >> (1 + ilog2(sizeof(*cp)))) & ipvs->threads_mask;
b880c1f0
HS		 406}
407
408/*
409 * Create a new sync buffer for Version 0 proto.
410 */
f131315f 411static inline struct ip_vs_sync_buff *
e4ff6751 412ip_vs_sync_buff_create_v0(struct netns_ipvs *ipvs, unsigned int len)
b880c1f0
HS		 413{
414 struct ip_vs_sync_buff *sb;
415 struct ip_vs_sync_mesg_v0 *mesg;
416
417 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
418 return NULL;
419
e4ff6751
JA		 420 len = max_t(unsigned int, len + sizeof(struct ip_vs_sync_mesg_v0),
421 ipvs->mcfg.sync_maxlen);
422 sb->mesg = kmalloc(len, GFP_ATOMIC);
f131315f 423 if (!sb->mesg) {
b880c1f0
HS		 424 kfree(sb);
425 return NULL;
426 }
427 mesg = (struct ip_vs_sync_mesg_v0 *)sb->mesg;
428 mesg->nr_conns = 0;
e4ff6751 429 mesg->syncid = ipvs->mcfg.syncid;
38561437 430 mesg->size = htons(sizeof(struct ip_vs_sync_mesg_v0));
f131315f 431 sb->head = (unsigned char *)mesg + sizeof(struct ip_vs_sync_mesg_v0);
e4ff6751 432 sb->end = (unsigned char *)mesg + len;
b880c1f0
HS		 433 sb->firstuse = jiffies;
434 return sb;
435}
436
4d0c875d
JA		 437/* Check if connection is controlled by persistence */
438static inline bool in_persistence(struct ip_vs_conn *cp)
439{
440 for (cp = cp->control; cp; cp = cp->control) {
441 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
442 return true;
443 }
444 return false;
445}
446
749c42b6
JA		 447/* Check if conn should be synced.
448 * pkts: conn packets, use sysctl_sync_threshold to avoid packet check
449 * - (1) sync_refresh_period: reduce sync rate. Additionally, retry
450 * sync_retries times with period of sync_refresh_period/8
451 * - (2) if both sync_refresh_period and sync_period are 0 send sync only
452 * for state changes or only once when pkts matches sync_threshold
453 * - (3) templates: rate can be reduced only with sync_refresh_period or
454 * with (2)
455 */
456static int ip_vs_sync_conn_needed(struct netns_ipvs *ipvs,
457 struct ip_vs_conn *cp, int pkts)
458{
459 unsigned long orig = ACCESS_ONCE(cp->sync_endtime);
460 unsigned long now = jiffies;
461 unsigned long n = (now + cp->timeout) & ~3UL;
462 unsigned int sync_refresh_period;
463 int sync_period;
464 int force;
465
466 /* Check if we sync in current state */
467 if (unlikely(cp->flags & IP_VS_CONN_F_TEMPLATE))
468 force = 0;
4d0c875d
JA		 469 else if (unlikely(sysctl_sync_persist_mode(ipvs) && in_persistence(cp)))
470 return 0;
749c42b6
JA		 471 else if (likely(cp->protocol == IPPROTO_TCP)) {
472 if (!((1 << cp->state) &
473 ((1 << IP_VS_TCP_S_ESTABLISHED) |
474 (1 << IP_VS_TCP_S_FIN_WAIT) |
475 (1 << IP_VS_TCP_S_CLOSE) |
476 (1 << IP_VS_TCP_S_CLOSE_WAIT) |
477 (1 << IP_VS_TCP_S_TIME_WAIT))))
478 return 0;
479 force = cp->state != cp->old_state;
480 if (force && cp->state != IP_VS_TCP_S_ESTABLISHED)
481 goto set;
482 } else if (unlikely(cp->protocol == IPPROTO_SCTP)) {
483 if (!((1 << cp->state) &
484 ((1 << IP_VS_SCTP_S_ESTABLISHED) |
61e7c420
JA		 485 (1 << IP_VS_SCTP_S_SHUTDOWN_SENT) |
486 (1 << IP_VS_SCTP_S_SHUTDOWN_RECEIVED) |
487 (1 << IP_VS_SCTP_S_SHUTDOWN_ACK_SENT) |
488 (1 << IP_VS_SCTP_S_CLOSED))))
749c42b6
JA		 489 return 0;
490 force = cp->state != cp->old_state;
491 if (force && cp->state != IP_VS_SCTP_S_ESTABLISHED)
492 goto set;
493 } else {
494 /* UDP or another protocol with single state */
495 force = 0;
496 }
497
498 sync_refresh_period = sysctl_sync_refresh_period(ipvs);
499 if (sync_refresh_period > 0) {
500 long diff = n - orig;
501 long min_diff = max(cp->timeout >> 1, 10UL * HZ);
502
503 /* Avoid sync if difference is below sync_refresh_period
504 * and below the half timeout.
505 */
506 if (abs(diff) < min_t(long, sync_refresh_period, min_diff)) {
507 int retries = orig & 3;
508
509 if (retries >= sysctl_sync_retries(ipvs))
510 return 0;
511 if (time_before(now, orig - cp->timeout +
512 (sync_refresh_period >> 3)))
513 return 0;
514 n |= retries + 1;
515 }
516 }
517 sync_period = sysctl_sync_period(ipvs);
518 if (sync_period > 0) {
519 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE) &&
520 pkts % sync_period != sysctl_sync_threshold(ipvs))
521 return 0;
522 } else if (sync_refresh_period <= 0 &&
523 pkts != sysctl_sync_threshold(ipvs))
524 return 0;
525
526set:
527 cp->old_state = cp->state;
528 n = cmpxchg(&cp->sync_endtime, orig, n);
529 return n == orig || force;
530}
531
b880c1f0
HS		 532/*
533 * Version 0 , could be switched in by sys_ctl.
534 * Add an ip_vs_conn information into the current sync_buff.
535 */
72e9481e 536static void ip_vs_sync_conn_v0(struct netns_ipvs *ipvs, struct ip_vs_conn *cp,
749c42b6 537 int pkts)
b880c1f0
HS		 538{
539 struct ip_vs_sync_mesg_v0 *m;
540 struct ip_vs_sync_conn_v0 *s;
f73181c8
PNA		 541 struct ip_vs_sync_buff *buff;
542 struct ipvs_master_sync_state *ms;
543 int id;
e4ff6751 544 unsigned int len;
b880c1f0
HS		 545
546 if (unlikely(cp->af != AF_INET))
547 return;
548 /* Do not sync ONE PACKET */
549 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
550 return;
551
749c42b6
JA		 552 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
553 return;
554
ac69269a 555 spin_lock_bh(&ipvs->sync_buff_lock);
f73181c8 556 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
ac69269a 557 spin_unlock_bh(&ipvs->sync_buff_lock);
f73181c8
PNA		 558 return;
559 }
560
561 id = select_master_thread_id(ipvs, cp);
562 ms = &ipvs->ms[id];
563 buff = ms->sync_buff;
e4ff6751
JA		 564 len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
565 SIMPLE_CONN_SIZE;
f73181c8
PNA		 566 if (buff) {
567 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
568 /* Send buffer if it is for v1 */
e4ff6751 569 if (buff->head + len > buff->end || !m->nr_conns) {
f73181c8
PNA		 570 sb_queue_tail(ipvs, ms);
571 ms->sync_buff = NULL;
572 buff = NULL;
573 }
574 }
575 if (!buff) {
e4ff6751 576 buff = ip_vs_sync_buff_create_v0(ipvs, len);
f73181c8 577 if (!buff) {
ac69269a 578 spin_unlock_bh(&ipvs->sync_buff_lock);
b880c1f0
HS		 579 pr_err("ip_vs_sync_buff_create failed.\n");
580 return;
581 }
f73181c8 582 ms->sync_buff = buff;
b880c1f0
HS		 583 }
584
f73181c8
PNA		 585 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
586 s = (struct ip_vs_sync_conn_v0 *) buff->head;
b880c1f0
HS		 587
588 /* copy members */
589 s->reserved = 0;
590 s->protocol = cp->protocol;
591 s->cport = cp->cport;
592 s->vport = cp->vport;
593 s->dport = cp->dport;
594 s->caddr = cp->caddr.ip;
595 s->vaddr = cp->vaddr.ip;
596 s->daddr = cp->daddr.ip;
597 s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
598 s->state = htons(cp->state);
599 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
600 struct ip_vs_sync_conn_options *opt =
601 (struct ip_vs_sync_conn_options *)&s[1];
602 memcpy(opt, &cp->in_seq, sizeof(*opt));
603 }
604
605 m->nr_conns++;
38561437 606 m->size = htons(ntohs(m->size) + len);
f73181c8 607 buff->head += len;
ac69269a 608 spin_unlock_bh(&ipvs->sync_buff_lock);
b880c1f0
HS		 609
610 /* synchronize its controller if it has */
749c42b6
JA		 611 cp = cp->control;
612 if (cp) {
613 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
614 pkts = atomic_add_return(1, &cp->in_pkts);
615 else
616 pkts = sysctl_sync_threshold(ipvs);
b61a8c1a 617 ip_vs_sync_conn(ipvs, cp, pkts);
749c42b6 618 }
b880c1f0
HS		 619}
620
1da177e4
LT		 621/*
622 * Add an ip_vs_conn information into the current sync_buff.
623 * Called by ip_vs_in.
986a0757 624 * Sending Version 1 messages
1da177e4 625 */
b61a8c1a 626void ip_vs_sync_conn(struct netns_ipvs *ipvs, struct ip_vs_conn *cp, int pkts)
1da177e4
LT		 627{
628 struct ip_vs_sync_mesg *m;
986a0757 629 union ip_vs_sync_conn *s;
f73181c8
PNA		 630 struct ip_vs_sync_buff *buff;
631 struct ipvs_master_sync_state *ms;
632 int id;
986a0757
HS		 633 __u8 *p;
634 unsigned int len, pe_name_len, pad;
635
b880c1f0 636 /* Handle old version of the protocol */
7532e8d4 637 if (sysctl_sync_ver(ipvs) == 0) {
72e9481e 638 ip_vs_sync_conn_v0(ipvs, cp, pkts);
b880c1f0
HS		 639 return;
640 }
986a0757
HS		 641 /* Do not sync ONE PACKET */
642 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
643 goto control;
644sloop:
749c42b6
JA		 645 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
646 goto control;
647
986a0757
HS		 648 /* Sanity checks */
649 pe_name_len = 0;
650 if (cp->pe_data_len) {
651 if (!cp->pe_data || !cp->dest) {
652 IP_VS_ERR_RL("SYNC, connection pe_data invalid\n");
653 return;
654 }
655 pe_name_len = strnlen(cp->pe->name, IP_VS_PENAME_MAXLEN);
656 }
1da177e4 657
ac69269a 658 spin_lock_bh(&ipvs->sync_buff_lock);
f73181c8 659 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
ac69269a 660 spin_unlock_bh(&ipvs->sync_buff_lock);
f73181c8
PNA		 661 return;
662 }
663
664 id = select_master_thread_id(ipvs, cp);
665 ms = &ipvs->ms[id];
986a0757
HS		 666
667#ifdef CONFIG_IP_VS_IPV6
668 if (cp->af == AF_INET6)
669 len = sizeof(struct ip_vs_sync_v6);
670 else
671#endif
672 len = sizeof(struct ip_vs_sync_v4);
673
674 if (cp->flags & IP_VS_CONN_F_SEQ_MASK)
675 len += sizeof(struct ip_vs_sync_conn_options) + 2;
676
677 if (cp->pe_data_len)
678 len += cp->pe_data_len + 2; /* + Param hdr field */
679 if (pe_name_len)
680 len += pe_name_len + 2;
681
682 /* check if there is a space for this one */
683 pad = 0;
f73181c8
PNA		 684 buff = ms->sync_buff;
685 if (buff) {
686 m = buff->mesg;
687 pad = (4 - (size_t) buff->head) & 3;
688 /* Send buffer if it is for v0 */
689 if (buff->head + len + pad > buff->end || m->reserved) {
690 sb_queue_tail(ipvs, ms);
691 ms->sync_buff = NULL;
692 buff = NULL;
986a0757
HS		 693 pad = 0;
694 }
695 }
696
f73181c8 697 if (!buff) {
e4ff6751 698 buff = ip_vs_sync_buff_create(ipvs, len);
f73181c8 699 if (!buff) {
ac69269a 700 spin_unlock_bh(&ipvs->sync_buff_lock);
1e3e238e 701 pr_err("ip_vs_sync_buff_create failed.\n");
1da177e4
LT		 702 return;
703 }
f73181c8
PNA		 704 ms->sync_buff = buff;
705 m = buff->mesg;
1da177e4
LT		 706 }
707
f73181c8
PNA		 708 p = buff->head;
709 buff->head += pad + len;
38561437 710 m->size = htons(ntohs(m->size) + pad + len);
986a0757
HS		 711 /* Add ev. padding from prev. sync_conn */
712 while (pad--)
713 *(p++) = 0;
714
715 s = (union ip_vs_sync_conn *)p;
716
717 /* Set message type & copy members */
718 s->v4.type = (cp->af == AF_INET6 ? STYPE_F_INET6 : 0);
719 s->v4.ver_size = htons(len & SVER_MASK); /* Version 0 */
720 s->v4.flags = htonl(cp->flags & ~IP_VS_CONN_F_HASHED);
721 s->v4.state = htons(cp->state);
722 s->v4.protocol = cp->protocol;
723 s->v4.cport = cp->cport;
724 s->v4.vport = cp->vport;
725 s->v4.dport = cp->dport;
726 s->v4.fwmark = htonl(cp->fwmark);
727 s->v4.timeout = htonl(cp->timeout / HZ);
1da177e4 728 m->nr_conns++;
1da177e4 729
986a0757
HS		 730#ifdef CONFIG_IP_VS_IPV6
731 if (cp->af == AF_INET6) {
732 p += sizeof(struct ip_vs_sync_v6);
4e3fd7a0
AD		 733 s->v6.caddr = cp->caddr.in6;
734 s->v6.vaddr = cp->vaddr.in6;
735 s->v6.daddr = cp->daddr.in6;
986a0757
HS		 736 } else
737#endif
738 {
739 p += sizeof(struct ip_vs_sync_v4); /* options ptr */
740 s->v4.caddr = cp->caddr.ip;
741 s->v4.vaddr = cp->vaddr.ip;
742 s->v4.daddr = cp->daddr.ip;
743 }
744 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
745 *(p++) = IPVS_OPT_SEQ_DATA;
746 *(p++) = sizeof(struct ip_vs_sync_conn_options);
747 hton_seq((struct ip_vs_seq *)p, &cp->in_seq);
748 p += sizeof(struct ip_vs_seq);
749 hton_seq((struct ip_vs_seq *)p, &cp->out_seq);
750 p += sizeof(struct ip_vs_seq);
1da177e4 751 }
986a0757
HS		 752 /* Handle pe data */
753 if (cp->pe_data_len && cp->pe_data) {
754 *(p++) = IPVS_OPT_PE_DATA;
755 *(p++) = cp->pe_data_len;
756 memcpy(p, cp->pe_data, cp->pe_data_len);
757 p += cp->pe_data_len;
758 if (pe_name_len) {
759 /* Add PE_NAME */
760 *(p++) = IPVS_OPT_PE_NAME;
761 *(p++) = pe_name_len;
762 memcpy(p, cp->pe->name, pe_name_len);
763 p += pe_name_len;
764 }
765 }
766
ac69269a 767 spin_unlock_bh(&ipvs->sync_buff_lock);
1da177e4 768
986a0757 769control:
1da177e4 770 /* synchronize its controller if it has */
986a0757
HS		 771 cp = cp->control;
772 if (!cp)
773 return;
749c42b6
JA		 774 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
775 pkts = atomic_add_return(1, &cp->in_pkts);
776 else
777 pkts = sysctl_sync_threshold(ipvs);
986a0757 778 goto sloop;
1da177e4
LT		 779}
780
fe5e7a1e
HS		 781/*
782 * fill_param used by version 1
783 */
85999283 784static inline int
19913dec 785ip_vs_conn_fill_param_sync(struct netns_ipvs *ipvs, int af, union ip_vs_sync_conn *sc,
fe5e7a1e
HS		 786 struct ip_vs_conn_param *p,
787 __u8 *pe_data, unsigned int pe_data_len,
788 __u8 *pe_name, unsigned int pe_name_len)
85999283 789{
fe5e7a1e
HS		 790#ifdef CONFIG_IP_VS_IPV6
791 if (af == AF_INET6)
19913dec 792 ip_vs_conn_fill_param(ipvs, af, sc->v6.protocol,
fe5e7a1e
HS		 793 (const union nf_inet_addr *)&sc->v6.caddr,
794 sc->v6.cport,
795 (const union nf_inet_addr *)&sc->v6.vaddr,
796 sc->v6.vport, p);
797 else
798#endif
19913dec 799 ip_vs_conn_fill_param(ipvs, af, sc->v4.protocol,
fe5e7a1e
HS		 800 (const union nf_inet_addr *)&sc->v4.caddr,
801 sc->v4.cport,
802 (const union nf_inet_addr *)&sc->v4.vaddr,
803 sc->v4.vport, p);
804 /* Handle pe data */
805 if (pe_data_len) {
806 if (pe_name_len) {
807 char buff[IP_VS_PENAME_MAXLEN+1];
808
809 memcpy(buff, pe_name, pe_name_len);
810 buff[pe_name_len]=0;
811 p->pe = __ip_vs_pe_getbyname(buff);
812 if (!p->pe) {
f131315f
HS		 813 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n",
814 buff);
fe5e7a1e
HS		 815 return 1;
816 }
817 } else {
818 IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n");
819 return 1;
820 }
821
6060c74a 822 p->pe_data = kmemdup(pe_data, pe_data_len, GFP_ATOMIC);
fe5e7a1e 823 if (!p->pe_data) {
982f4051 824 module_put(p->pe->module);
fe5e7a1e
HS		 825 return -ENOMEM;
826 }
fe5e7a1e
HS		 827 p->pe_data_len = pe_data_len;
828 }
85999283
SH		 829 return 0;
830}
1da177e4
LT		 831
832/*
fe5e7a1e
HS		 833 * Connection Add / Update.
834 * Common for version 0 and 1 reception of backup sync_conns.
835 * Param: ...
836 * timeout is in sec.
1da177e4 837 */
4f30665b 838static void ip_vs_proc_conn(struct netns_ipvs *ipvs, struct ip_vs_conn_param *param,
9bbac6a9
HS		 839 unsigned int flags, unsigned int state,
840 unsigned int protocol, unsigned int type,
fe5e7a1e
HS		 841 const union nf_inet_addr *daddr, __be16 dport,
842 unsigned long timeout, __u32 fwmark,
9bbac6a9 843 struct ip_vs_sync_conn_options *opt)
fe5e7a1e
HS		 844{
845 struct ip_vs_dest *dest;
846 struct ip_vs_conn *cp;
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 */
dc2add6f 904 dest = ip_vs_find_dest(ipvs, type, type, daddr, dport,
655eef10
AG		 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
18d6ade6 941 pd = ip_vs_proto_data_get(ipvs, protocol);
9bbac6a9
HS		 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 */
19913dec 953static void ip_vs_process_message_v0(struct netns_ipvs *ipvs, const char *buffer,
9bbac6a9 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
19913dec 1009 ip_vs_conn_fill_param(ipvs, 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 */
4f30665b 1016 ip_vs_proc_conn(ipvs, &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 */
1fc12004 1067static inline int ip_vs_proc_sync_conn(struct netns_ipvs *ipvs, __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 }
4f30665b 1171 if (ip_vs_conn_fill_param_sync(ipvs, af, s, &param, pe_data,
6e67e586 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)
4f30665b 1178 ip_vs_proc_conn(ipvs, &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
4f30665b 1185 ip_vs_proc_conn(ipvs, &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 */
7d537f3a 1204static void ip_vs_process_message(struct netns_ipvs *ipvs, __u8 *buffer,
9bbac6a9 1205 const size_t buflen)
fe5e7a1e 1206{
986a0757 1207 struct ip_vs_sync_mesg *m2 = (struct ip_vs_sync_mesg *)buffer;
fe5e7a1e 1208 __u8 *p, *msg_end;
986a0757 1209 int i, nr_conns;
fe5e7a1e 1210
986a0757 1211 if (buflen < sizeof(struct ip_vs_sync_mesg_v0)) {
fe5e7a1e
HS		 1212 IP_VS_DBG(2, "BACKUP, message header too short\n");
1213 return;
1214 }
fe5e7a1e 1215
38561437 1216 if (buflen != ntohs(m2->size)) {
fe5e7a1e
HS		 1217 IP_VS_DBG(2, "BACKUP, bogus message size\n");
1218 return;
1219 }
1220 /* SyncID sanity check */
e4ff6751 1221 if (ipvs->bcfg.syncid != 0 && m2->syncid != ipvs->bcfg.syncid) {
fe5e7a1e
HS		 1222 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2->syncid);
1223 return;
1224 }
1225 /* Handle version 1 message */
1226 if ((m2->version == SYNC_PROTO_VER) && (m2->reserved == 0)
1227 && (m2->spare == 0)) {
1228
986a0757 1229 msg_end = buffer + sizeof(struct ip_vs_sync_mesg);
fe5e7a1e
HS		 1230 nr_conns = m2->nr_conns;
1231
1232 for (i=0; i<nr_conns; i++) {
1233 union ip_vs_sync_conn *s;
95c96174 1234 unsigned int size;
fe5e7a1e
HS		 1235 int retc;
1236
1237 p = msg_end;
1238 if (p + sizeof(s->v4) > buffer+buflen) {
1239 IP_VS_ERR_RL("BACKUP, Dropping buffer, to small\n");
1240 return;
b209639e 1241 }
fe5e7a1e
HS		 1242 s = (union ip_vs_sync_conn *)p;
1243 size = ntohs(s->v4.ver_size) & SVER_MASK;
1244 msg_end = p + size;
1245 /* Basic sanity checks */
1246 if (msg_end > buffer+buflen) {
1247 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n");
1da177e4
LT		 1248 return;
1249 }
fe5e7a1e
HS		 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);
1253 return;
b209639e 1254 }
fe5e7a1e 1255 /* Process a single sync_conn */
1fc12004 1256 retc = ip_vs_proc_sync_conn(ipvs, p, msg_end);
9bbac6a9 1257 if (retc < 0) {
fe5e7a1e
HS		 1258 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n",
1259 retc);
1260 return;
2906f66a 1261 }
fe5e7a1e
HS		 1262 /* Make sure we have 32 bit alignment */
1263 msg_end = p + ((size + 3) & ~3);
b209639e 1264 }
fe5e7a1e
HS		 1265 } else {
1266 /* Old type of message */
19913dec 1267 ip_vs_process_message_v0(ipvs, buffer, buflen);
fe5e7a1e 1268 return;
1da177e4
LT		 1269 }
1270}
1271
1272
1c003b15
PNA		 1273/*
1274 * Setup sndbuf (mode=1) or rcvbuf (mode=0)
1275 */
1276static void set_sock_size(struct sock *sk, int mode, int val)
1277{
1278 /* setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &val, sizeof(val)); */
1279 /* setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)); */
1280 lock_sock(sk);
1281 if (mode) {
1282 val = clamp_t(int, val, (SOCK_MIN_SNDBUF + 1) / 2,
1283 sysctl_wmem_max);
1284 sk->sk_sndbuf = val * 2;
1285 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1286 } else {
1287 val = clamp_t(int, val, (SOCK_MIN_RCVBUF + 1) / 2,
1288 sysctl_rmem_max);
1289 sk->sk_rcvbuf = val * 2;
1290 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1291 }
1292 release_sock(sk);
1293}
1294
1da177e4
LT		 1295/*
1296 * Setup loopback of outgoing multicasts on a sending socket
1297 */
1298static void set_mcast_loop(struct sock *sk, u_char loop)
1299{
1300 struct inet_sock *inet = inet_sk(sk);
1301
1302 /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
1303 lock_sock(sk);
1304 inet->mc_loop = loop ? 1 : 0;
d3328817
JA		 1305#ifdef CONFIG_IP_VS_IPV6
1306 if (sk->sk_family == AF_INET6) {
1307 struct ipv6_pinfo *np = inet6_sk(sk);
1308
1309 /* IPV6_MULTICAST_LOOP */
1310 np->mc_loop = loop ? 1 : 0;
1311 }
1312#endif
1da177e4
LT		 1313 release_sock(sk);
1314}
1315
1316/*
1317 * Specify TTL for outgoing multicasts on a sending socket
1318 */
1319static void set_mcast_ttl(struct sock *sk, u_char ttl)
1320{
1321 struct inet_sock *inet = inet_sk(sk);
1322
1323 /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
1324 lock_sock(sk);
1325 inet->mc_ttl = ttl;
d3328817
JA		 1326#ifdef CONFIG_IP_VS_IPV6
1327 if (sk->sk_family == AF_INET6) {
1328 struct ipv6_pinfo *np = inet6_sk(sk);
1329
1330 /* IPV6_MULTICAST_HOPS */
1331 np->mcast_hops = ttl;
1332 }
1333#endif
1da177e4
LT		 1334 release_sock(sk);
1335}
1336
e4ff6751
JA		 1337/* Control fragmentation of messages */
1338static void set_mcast_pmtudisc(struct sock *sk, int val)
1339{
1340 struct inet_sock *inet = inet_sk(sk);
1341
1342 /* setsockopt(sock, SOL_IP, IP_MTU_DISCOVER, &val, sizeof(val)); */
1343 lock_sock(sk);
1344 inet->pmtudisc = val;
d3328817
JA		 1345#ifdef CONFIG_IP_VS_IPV6
1346 if (sk->sk_family == AF_INET6) {
1347 struct ipv6_pinfo *np = inet6_sk(sk);
1348
1349 /* IPV6_MTU_DISCOVER */
1350 np->pmtudisc = val;
1351 }
1352#endif
e4ff6751
JA		 1353 release_sock(sk);
1354}
1355
1da177e4
LT		 1356/*
1357 * Specifiy default interface for outgoing multicasts
1358 */
1359static int set_mcast_if(struct sock *sk, char *ifname)
1360{
1361 struct net_device *dev;
1362 struct inet_sock *inet = inet_sk(sk);
f131315f 1363 struct net *net = sock_net(sk);
1da177e4 1364
f131315f
HS		 1365 dev = __dev_get_by_name(net, ifname);
1366 if (!dev)
1da177e4
LT		 1367 return -ENODEV;
1368
1369 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1370 return -EINVAL;
1371
1372 lock_sock(sk);
1373 inet->mc_index = dev->ifindex;
1374 /* inet->mc_addr = 0; */
d3328817
JA		 1375#ifdef CONFIG_IP_VS_IPV6
1376 if (sk->sk_family == AF_INET6) {
1377 struct ipv6_pinfo *np = inet6_sk(sk);
1378
1379 /* IPV6_MULTICAST_IF */
1380 np->mcast_oif = dev->ifindex;
1381 }
1382#endif
1da177e4
LT		 1383 release_sock(sk);
1384
1385 return 0;
1386}
1387
1388
1da177e4
LT		 1389/*
1390 * Join a multicast group.
1391 * the group is specified by a class D multicast address 224.0.0.0/8
1392 * in the in_addr structure passed in as a parameter.
1393 */
1394static int
1395join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
1396{
f131315f 1397 struct net *net = sock_net(sk);
1da177e4
LT		 1398 struct ip_mreqn mreq;
1399 struct net_device *dev;
1400 int ret;
1401
1402 memset(&mreq, 0, sizeof(mreq));
1403 memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
1404
f131315f
HS		 1405 dev = __dev_get_by_name(net, ifname);
1406 if (!dev)
1da177e4
LT		 1407 return -ENODEV;
1408 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1409 return -EINVAL;
1410
1411 mreq.imr_ifindex = dev->ifindex;
1412
1413 lock_sock(sk);
1414 ret = ip_mc_join_group(sk, &mreq);
1415 release_sock(sk);
1416
1417 return ret;
1418}
1419
d3328817
JA		 1420#ifdef CONFIG_IP_VS_IPV6
1421static int join_mcast_group6(struct sock *sk, struct in6_addr *addr,
1422 char *ifname)
1423{
1424 struct net *net = sock_net(sk);
1425 struct net_device *dev;
1426 int ret;
1427
1428 dev = __dev_get_by_name(net, ifname);
1429 if (!dev)
1430 return -ENODEV;
1431 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1432 return -EINVAL;
1433
1434 lock_sock(sk);
1435 ret = ipv6_sock_mc_join(sk, dev->ifindex, addr);
1436 release_sock(sk);
1437
1438 return ret;
1439}
1440#endif
1da177e4
LT		 1441
1442static int bind_mcastif_addr(struct socket *sock, char *ifname)
1443{
f131315f 1444 struct net *net = sock_net(sock->sk);
1da177e4 1445 struct net_device *dev;
a61ced5d 1446 __be32 addr;
1da177e4
LT		 1447 struct sockaddr_in sin;
1448
f131315f
HS		 1449 dev = __dev_get_by_name(net, ifname);
1450 if (!dev)
1da177e4
LT		 1451 return -ENODEV;
1452
1453 addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
1454 if (!addr)
1e3e238e
HE		 1455 pr_err("You probably need to specify IP address on "
1456 "multicast interface.\n");
1da177e4 1457
14d5e834
HH		 1458 IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
1459 ifname, &addr);
1da177e4
LT		 1460
1461 /* Now bind the socket with the address of multicast interface */
1462 sin.sin_family = AF_INET;
1463 sin.sin_addr.s_addr = addr;
1464 sin.sin_port = 0;
1465
1466 return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
1467}
1468
d3328817
JA		 1469static void get_mcast_sockaddr(union ipvs_sockaddr *sa, int *salen,
1470 struct ipvs_sync_daemon_cfg *c, int id)
1471{
1472 if (AF_INET6 == c->mcast_af) {
1473 sa->in6 = (struct sockaddr_in6) {
1474 .sin6_family = AF_INET6,
1475 .sin6_port = htons(c->mcast_port + id),
1476 };
1477 sa->in6.sin6_addr = c->mcast_group.in6;
1478 *salen = sizeof(sa->in6);
1479 } else {
1480 sa->in = (struct sockaddr_in) {
1481 .sin_family = AF_INET,
1482 .sin_port = htons(c->mcast_port + id),
1483 };
1484 sa->in.sin_addr = c->mcast_group.in;
1485 *salen = sizeof(sa->in);
1486 }
1487}
1488
1da177e4
LT		 1489/*
1490 * Set up sending multicast socket over UDP
1491 */
68c76b6a 1492static struct socket *make_send_sock(struct netns_ipvs *ipvs, int id)
1da177e4 1493{
f73181c8 1494 /* multicast addr */
d3328817 1495 union ipvs_sockaddr mcast_addr;
1da177e4 1496 struct socket *sock;
d3328817 1497 int result, salen;
1da177e4 1498
26abe143 1499 /* First create a socket */
68c76b6a 1500 result = sock_create_kern(ipvs->net, ipvs->mcfg.mcast_af, SOCK_DGRAM,
d3328817 1501 IPPROTO_UDP, &sock);
e6dd731c 1502 if (result < 0) {
1e3e238e 1503 pr_err("Error during creation of socket; terminating\n");
e6dd731c 1504 return ERR_PTR(result);
1da177e4 1505 }
e4ff6751 1506 result = set_mcast_if(sock->sk, ipvs->mcfg.mcast_ifn);
e6dd731c 1507 if (result < 0) {
1e3e238e 1508 pr_err("Error setting outbound mcast interface\n");
1da177e4
LT		 1509 goto error;
1510 }
1511
1512 set_mcast_loop(sock->sk, 0);
d3328817 1513 set_mcast_ttl(sock->sk, ipvs->mcfg.mcast_ttl);
e4ff6751
JA		 1514 /* Allow fragmentation if MTU changes */
1515 set_mcast_pmtudisc(sock->sk, IP_PMTUDISC_DONT);
1c003b15
PNA		 1516 result = sysctl_sync_sock_size(ipvs);
1517 if (result > 0)
1518 set_sock_size(sock->sk, 1, result);
1da177e4 1519
d3328817
JA		 1520 if (AF_INET == ipvs->mcfg.mcast_af)
1521 result = bind_mcastif_addr(sock, ipvs->mcfg.mcast_ifn);
1522 else
1523 result = 0;
e6dd731c 1524 if (result < 0) {
1e3e238e 1525 pr_err("Error binding address of the mcast interface\n");
1da177e4
LT		 1526 goto error;
1527 }
1528
d3328817 1529 get_mcast_sockaddr(&mcast_addr, &salen, &ipvs->mcfg, id);
e6dd731c 1530 result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr,
d3328817 1531 salen, 0);
e6dd731c 1532 if (result < 0) {
1e3e238e 1533 pr_err("Error connecting to the multicast addr\n");
1da177e4
LT		 1534 goto error;
1535 }
1536
1537 return sock;
1538
1ae132b0 1539error:
26abe143 1540 sock_release(sock);
e6dd731c 1541 return ERR_PTR(result);
1da177e4
LT		 1542}
1543
1544
1545/*
1546 * Set up receiving multicast socket over UDP
1547 */
3777ed68
QA		 1548static struct socket *make_receive_sock(struct netns_ipvs *ipvs, int id,
1549 int ifindex)
1da177e4 1550{
f73181c8 1551 /* multicast addr */
d3328817 1552 union ipvs_sockaddr mcast_addr;
1da177e4 1553 struct socket *sock;
d3328817 1554 int result, salen;
1da177e4
LT		 1555
1556 /* First create a socket */
fd124e2f 1557 result = sock_create_kern(ipvs->net, ipvs->bcfg.mcast_af, SOCK_DGRAM,
d3328817 1558 IPPROTO_UDP, &sock);
e6dd731c 1559 if (result < 0) {
1e3e238e 1560 pr_err("Error during creation of socket; terminating\n");
e6dd731c 1561 return ERR_PTR(result);
1da177e4 1562 }
1da177e4 1563 /* it is equivalent to the REUSEADDR option in user-space */
4a17fd52 1564 sock->sk->sk_reuse = SK_CAN_REUSE;
1c003b15
PNA		 1565 result = sysctl_sync_sock_size(ipvs);
1566 if (result > 0)
1567 set_sock_size(sock->sk, 0, result);
1da177e4 1568
d3328817 1569 get_mcast_sockaddr(&mcast_addr, &salen, &ipvs->bcfg, id);
3777ed68 1570 sock->sk->sk_bound_dev_if = ifindex;
d3328817 1571 result = sock->ops->bind(sock, (struct sockaddr *)&mcast_addr, salen);
e6dd731c 1572 if (result < 0) {
1e3e238e 1573 pr_err("Error binding to the multicast addr\n");
1da177e4
LT		 1574 goto error;
1575 }
1576
1577 /* join the multicast group */
d3328817
JA		 1578#ifdef CONFIG_IP_VS_IPV6
1579 if (ipvs->bcfg.mcast_af == AF_INET6)
1580 result = join_mcast_group6(sock->sk, &mcast_addr.in6.sin6_addr,
1581 ipvs->bcfg.mcast_ifn);
1582 else
1583#endif
1584 result = join_mcast_group(sock->sk, &mcast_addr.in.sin_addr,
1585 ipvs->bcfg.mcast_ifn);
e6dd731c 1586 if (result < 0) {
1e3e238e 1587 pr_err("Error joining to the multicast group\n");
1da177e4
LT		 1588 goto error;
1589 }
1590
1591 return sock;
1592
1ae132b0 1593error:
26abe143 1594 sock_release(sock);
e6dd731c 1595 return ERR_PTR(result);
1da177e4
LT		 1596}
1597
1598
1599static int
1600ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
1601{
1602 struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
1603 struct kvec iov;
1604 int len;
1605
1606 EnterFunction(7);
1607 iov.iov_base = (void *)buffer;
1608 iov.iov_len = length;
1609
1610 len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
1611
1612 LeaveFunction(7);
1613 return len;
1614}
1615
1c003b15 1616static int
1da177e4
LT		 1617ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
1618{
1619 int msize;
1c003b15 1620 int ret;
1da177e4 1621
38561437 1622 msize = ntohs(msg->size);
1da177e4 1623
1c003b15
PNA		 1624 ret = ip_vs_send_async(sock, (char *)msg, msize);
1625 if (ret >= 0 || ret == -EAGAIN)
1626 return ret;
1627 pr_err("ip_vs_send_async error %d\n", ret);
1628 return 0;
1da177e4
LT		 1629}
1630
1631static int
1632ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
1633{
1634 struct msghdr msg = {NULL,};
1635 struct kvec iov;
1636 int len;
1637
1638 EnterFunction(7);
1639
1640 /* Receive a packet */
1641 iov.iov_base = buffer;
1642 iov.iov_len = (size_t)buflen;
1643
f73181c8 1644 len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, MSG_DONTWAIT);
1da177e4
LT		 1645
1646 if (len < 0)
f73181c8 1647 return len;
1da177e4
LT		 1648
1649 LeaveFunction(7);
1650 return len;
1651}
1652
1c003b15
PNA		 1653/* Wakeup the master thread for sending */
1654static void master_wakeup_work_handler(struct work_struct *work)
1655{
f73181c8
PNA		 1656 struct ipvs_master_sync_state *ms =
1657 container_of(work, struct ipvs_master_sync_state,
1658 master_wakeup_work.work);
1659 struct netns_ipvs *ipvs = ms->ipvs;
1c003b15
PNA		 1660
1661 spin_lock_bh(&ipvs->sync_lock);
f73181c8
PNA		 1662 if (ms->sync_queue_len &&
1663 ms->sync_queue_delay < IPVS_SYNC_WAKEUP_RATE) {
1664 ms->sync_queue_delay = IPVS_SYNC_WAKEUP_RATE;
1665 wake_up_process(ms->master_thread);
1c003b15
PNA		 1666 }
1667 spin_unlock_bh(&ipvs->sync_lock);
1668}
1669
1670/* Get next buffer to send */
1671static inline struct ip_vs_sync_buff *
f73181c8 1672next_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
1c003b15
PNA		 1673{
1674 struct ip_vs_sync_buff *sb;
1675
f73181c8 1676 sb = sb_dequeue(ipvs, ms);
1c003b15
PNA		 1677 if (sb)
1678 return sb;
1679 /* Do not delay entries in buffer for more than 2 seconds */
f73181c8 1680 return get_curr_sync_buff(ipvs, ms, IPVS_SYNC_FLUSH_TIME);
1c003b15 1681}
1da177e4 1682
998e7a76 1683static int sync_thread_master(void *data)
1da177e4 1684{
998e7a76 1685 struct ip_vs_sync_thread_data *tinfo = data;
37b68e6d 1686 struct netns_ipvs *ipvs = tinfo->ipvs;
f73181c8 1687 struct ipvs_master_sync_state *ms = &ipvs->ms[tinfo->id];
1c003b15 1688 struct sock *sk = tinfo->sock->sk;
1da177e4
LT		 1689 struct ip_vs_sync_buff *sb;
1690
1e3e238e 1691 pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
f73181c8 1692 "syncid = %d, id = %d\n",
e4ff6751 1693 ipvs->mcfg.mcast_ifn, ipvs->mcfg.syncid, tinfo->id);
1da177e4 1694
1c003b15 1695 for (;;) {
f73181c8 1696 sb = next_sync_buff(ipvs, ms);
1c003b15
PNA		 1697 if (unlikely(kthread_should_stop()))
1698 break;
1699 if (!sb) {
1700 schedule_timeout(IPVS_SYNC_CHECK_PERIOD);
1701 continue;
1da177e4 1702 }
1c003b15 1703 while (ip_vs_send_sync_msg(tinfo->sock, sb->mesg) < 0) {
9dcbe1b8
GU		 1704 /* (Ab)use interruptible sleep to avoid increasing
1705 * the load avg.
1706 */
1707 __wait_event_interruptible(*sk_sleep(sk),
1c003b15 1708 sock_writeable(sk) ||
35a2af94 1709 kthread_should_stop());
1c003b15
PNA		 1710 if (unlikely(kthread_should_stop()))
1711 goto done;
1da177e4 1712 }
1c003b15 1713 ip_vs_sync_buff_release(sb);
1da177e4
LT		 1714 }
1715
1c003b15
PNA		 1716done:
1717 __set_current_state(TASK_RUNNING);
1718 if (sb)
1719 ip_vs_sync_buff_release(sb);
1720
1da177e4 1721 /* clean up the sync_buff queue */
f73181c8 1722 while ((sb = sb_dequeue(ipvs, ms)))
1da177e4 1723 ip_vs_sync_buff_release(sb);
1c003b15 1724 __set_current_state(TASK_RUNNING);
1da177e4
LT		 1725
1726 /* clean up the current sync_buff */
f73181c8 1727 sb = get_curr_sync_buff(ipvs, ms, 0);
f131315f 1728 if (sb)
1da177e4 1729 ip_vs_sync_buff_release(sb);
1da177e4
LT		 1730
1731 /* release the sending multicast socket */
26abe143 1732 sock_release(tinfo->sock);
998e7a76
SW		 1733 kfree(tinfo);
1734
1735 return 0;
1da177e4
LT		 1736}
1737
1738
998e7a76 1739static int sync_thread_backup(void *data)
1da177e4 1740{
998e7a76 1741 struct ip_vs_sync_thread_data *tinfo = data;
37b68e6d 1742 struct netns_ipvs *ipvs = tinfo->ipvs;
1da177e4
LT		 1743 int len;
1744
1e3e238e 1745 pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
f73181c8 1746 "syncid = %d, id = %d\n",
e4ff6751 1747 ipvs->bcfg.mcast_ifn, ipvs->bcfg.syncid, tinfo->id);
1da177e4 1748
998e7a76 1749 while (!kthread_should_stop()) {
aa395145 1750 wait_event_interruptible(*sk_sleep(tinfo->sock->sk),
ba6fd850
SW		 1751 !skb_queue_empty(&tinfo->sock->sk->sk_receive_queue)
1752 || kthread_should_stop());
1753
998e7a76
SW		 1754 /* do we have data now? */
1755 while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) {
1756 len = ip_vs_receive(tinfo->sock, tinfo->buf,
e4ff6751 1757 ipvs->bcfg.sync_maxlen);
998e7a76 1758 if (len <= 0) {
f73181c8
PNA		 1759 if (len != -EAGAIN)
1760 pr_err("receiving message error\n");
1da177e4
LT		 1761 break;
1762 }
998e7a76 1763
7d537f3a 1764 ip_vs_process_message(ipvs, tinfo->buf, len);
1da177e4 1765 }
1da177e4
LT		 1766 }
1767
1768 /* release the sending multicast socket */
26abe143 1769 sock_release(tinfo->sock);
998e7a76
SW		 1770 kfree(tinfo->buf);
1771 kfree(tinfo);
1da177e4 1772
998e7a76 1773 return 0;
1da177e4
LT		 1774}
1775
1776
6ac121d7 1777int start_sync_thread(struct netns_ipvs *ipvs, struct ipvs_sync_daemon_cfg *c,
e4ff6751 1778 int state)
1da177e4 1779{
998e7a76 1780 struct ip_vs_sync_thread_data *tinfo;
f73181c8 1781 struct task_struct **array = NULL, *task;
998e7a76 1782 struct socket *sock;
e4ff6751 1783 struct net_device *dev;
f73181c8 1784 char *name;
998e7a76 1785 int (*threadfn)(void *data);
e4ff6751 1786 int id, count, hlen;
998e7a76 1787 int result = -ENOMEM;
e4ff6751 1788 u16 mtu, min_mtu;
1da177e4 1789
1e3e238e 1790 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
998e7a76 1791 IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n",
2981bc9a 1792 sizeof(struct ip_vs_sync_conn_v0));
1da177e4 1793
f73181c8
PNA		 1794 if (!ipvs->sync_state) {
1795 count = clamp(sysctl_sync_ports(ipvs), 1, IPVS_SYNC_PORTS_MAX);
1796 ipvs->threads_mask = count - 1;
1797 } else
1798 count = ipvs->threads_mask + 1;
ae1d48b2 1799
d3328817
JA		 1800 if (c->mcast_af == AF_UNSPEC) {
1801 c->mcast_af = AF_INET;
1802 c->mcast_group.ip = cpu_to_be32(IP_VS_SYNC_GROUP);
1803 }
1804 if (!c->mcast_port)
1805 c->mcast_port = IP_VS_SYNC_PORT;
1806 if (!c->mcast_ttl)
1807 c->mcast_ttl = 1;
1808
6ac121d7 1809 dev = __dev_get_by_name(ipvs->net, c->mcast_ifn);
e4ff6751
JA		 1810 if (!dev) {
1811 pr_err("Unknown mcast interface: %s\n", c->mcast_ifn);
1812 return -ENODEV;
1813 }
d3328817
JA		 1814 hlen = (AF_INET6 == c->mcast_af) ?
1815 sizeof(struct ipv6hdr) + sizeof(struct udphdr) :
1816 sizeof(struct iphdr) + sizeof(struct udphdr);
e4ff6751
JA		 1817 mtu = (state == IP_VS_STATE_BACKUP) ?
1818 clamp(dev->mtu, 1500U, 65535U) : 1500U;
1819 min_mtu = (state == IP_VS_STATE_BACKUP) ? 1024 : 1;
1820
1821 if (c->sync_maxlen)
1822 c->sync_maxlen = clamp_t(unsigned int,
1823 c->sync_maxlen, min_mtu,
1824 65535 - hlen);
1825 else
1826 c->sync_maxlen = mtu - hlen;
1827
998e7a76 1828 if (state == IP_VS_STATE_MASTER) {
f73181c8 1829 if (ipvs->ms)
998e7a76 1830 return -EEXIST;
1da177e4 1831
e4ff6751 1832 ipvs->mcfg = *c;
f73181c8 1833 name = "ipvs-m:%d:%d";
998e7a76 1834 threadfn = sync_thread_master;
998e7a76 1835 } else if (state == IP_VS_STATE_BACKUP) {
f73181c8 1836 if (ipvs->backup_threads)
998e7a76
SW		 1837 return -EEXIST;
1838
e4ff6751 1839 ipvs->bcfg = *c;
f73181c8 1840 name = "ipvs-b:%d:%d";
998e7a76 1841 threadfn = sync_thread_backup;
1da177e4 1842 } else {
1da177e4
LT		 1843 return -EINVAL;
1844 }
1845
f73181c8
PNA		 1846 if (state == IP_VS_STATE_MASTER) {
1847 struct ipvs_master_sync_state *ms;
1da177e4 1848
f73181c8
PNA		 1849 ipvs->ms = kzalloc(count * sizeof(ipvs->ms[0]), GFP_KERNEL);
1850 if (!ipvs->ms)
1851 goto out;
1852 ms = ipvs->ms;
1853 for (id = 0; id < count; id++, ms++) {
1854 INIT_LIST_HEAD(&ms->sync_queue);
1855 ms->sync_queue_len = 0;
1856 ms->sync_queue_delay = 0;
1857 INIT_DELAYED_WORK(&ms->master_wakeup_work,
1858 master_wakeup_work_handler);
1859 ms->ipvs = ipvs;
1860 }
1861 } else {
1862 array = kzalloc(count * sizeof(struct task_struct *),
1863 GFP_KERNEL);
1864 if (!array)
1865 goto out;
998e7a76 1866 }
1da177e4 1867
f73181c8
PNA		 1868 tinfo = NULL;
1869 for (id = 0; id < count; id++) {
1870 if (state == IP_VS_STATE_MASTER)
68c76b6a 1871 sock = make_send_sock(ipvs, id);
f73181c8 1872 else
3777ed68 1873 sock = make_receive_sock(ipvs, id, dev->ifindex);
f73181c8
PNA		 1874 if (IS_ERR(sock)) {
1875 result = PTR_ERR(sock);
1876 goto outtinfo;
1877 }
1878 tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL);
1879 if (!tinfo)
1880 goto outsocket;
37b68e6d 1881 tinfo->ipvs = ipvs;
f73181c8
PNA		 1882 tinfo->sock = sock;
1883 if (state == IP_VS_STATE_BACKUP) {
e4ff6751 1884 tinfo->buf = kmalloc(ipvs->bcfg.sync_maxlen,
f73181c8
PNA		 1885 GFP_KERNEL);
1886 if (!tinfo->buf)
1887 goto outtinfo;
b425df4c
DC		 1888 } else {
1889 tinfo->buf = NULL;
f73181c8
PNA		 1890 }
1891 tinfo->id = id;
1da177e4 1892
f73181c8
PNA		 1893 task = kthread_run(threadfn, tinfo, name, ipvs->gen, id);
1894 if (IS_ERR(task)) {
1895 result = PTR_ERR(task);
1896 goto outtinfo;
1897 }
1898 tinfo = NULL;
1899 if (state == IP_VS_STATE_MASTER)
1900 ipvs->ms[id].master_thread = task;
1901 else
1902 array[id] = task;
998e7a76 1903 }
1da177e4 1904
998e7a76 1905 /* mark as active */
f73181c8
PNA		 1906
1907 if (state == IP_VS_STATE_BACKUP)
1908 ipvs->backup_threads = array;
1909 spin_lock_bh(&ipvs->sync_buff_lock);
f131315f 1910 ipvs->sync_state |= state;
f73181c8 1911 spin_unlock_bh(&ipvs->sync_buff_lock);
1da177e4 1912
998e7a76
SW		 1913 /* increase the module use count */
1914 ip_vs_use_count_inc();
1da177e4
LT		 1915
1916 return 0;
1da177e4 1917
998e7a76 1918outsocket:
26abe143 1919 sock_release(sock);
f73181c8
PNA		 1920
1921outtinfo:
1922 if (tinfo) {
26abe143 1923 sock_release(tinfo->sock);
f73181c8
PNA		 1924 kfree(tinfo->buf);
1925 kfree(tinfo);
1926 }
1927 count = id;
1928 while (count-- > 0) {
1929 if (state == IP_VS_STATE_MASTER)
1930 kthread_stop(ipvs->ms[count].master_thread);
1931 else
1932 kthread_stop(array[count]);
1933 }
1934 kfree(array);
1935
998e7a76 1936out:
f73181c8
PNA		 1937 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
1938 kfree(ipvs->ms);
1939 ipvs->ms = NULL;
1940 }
998e7a76 1941 return result;
1da177e4
LT		 1942}
1943
1944
b3cf3cbf 1945int stop_sync_thread(struct netns_ipvs *ipvs, int state)
1da177e4 1946{
f73181c8
PNA		 1947 struct task_struct **array;
1948 int id;
1ae132b0 1949 int retc = -EINVAL;
f131315f 1950
1e3e238e 1951 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1da177e4 1952
1da177e4 1953 if (state == IP_VS_STATE_MASTER) {
f73181c8 1954 if (!ipvs->ms)
998e7a76 1955 return -ESRCH;
cc0191ae 1956
998e7a76
SW		 1957 /*
1958 * The lock synchronizes with sb_queue_tail(), so that we don't
1959 * add sync buffers to the queue, when we are already in
1960 * progress of stopping the master sync daemon.
1961 */
1da177e4 1962
f73181c8
PNA		 1963 spin_lock_bh(&ipvs->sync_buff_lock);
1964 spin_lock(&ipvs->sync_lock);
f131315f 1965 ipvs->sync_state &= ~IP_VS_STATE_MASTER;
f73181c8
PNA		 1966 spin_unlock(&ipvs->sync_lock);
1967 spin_unlock_bh(&ipvs->sync_buff_lock);
1968
1969 retc = 0;
1970 for (id = ipvs->threads_mask; id >= 0; id--) {
1971 struct ipvs_master_sync_state *ms = &ipvs->ms[id];
1972 int ret;
1973
1974 pr_info("stopping master sync thread %d ...\n",
1975 task_pid_nr(ms->master_thread));
1976 cancel_delayed_work_sync(&ms->master_wakeup_work);
1977 ret = kthread_stop(ms->master_thread);
1978 if (retc >= 0)
1979 retc = ret;
1980 }
1981 kfree(ipvs->ms);
1982 ipvs->ms = NULL;
998e7a76 1983 } else if (state == IP_VS_STATE_BACKUP) {
f73181c8 1984 if (!ipvs->backup_threads)
998e7a76
SW		 1985 return -ESRCH;
1986
f131315f 1987 ipvs->sync_state &= ~IP_VS_STATE_BACKUP;
f73181c8
PNA		 1988 array = ipvs->backup_threads;
1989 retc = 0;
1990 for (id = ipvs->threads_mask; id >= 0; id--) {
1991 int ret;
1992
1993 pr_info("stopping backup sync thread %d ...\n",
1994 task_pid_nr(array[id]));
1995 ret = kthread_stop(array[id]);
1996 if (retc >= 0)
1997 retc = ret;
1998 }
1999 kfree(array);
2000 ipvs->backup_threads = NULL;
998e7a76 2001 }
1da177e4 2002
998e7a76
SW		 2003 /* decrease the module use count */
2004 ip_vs_use_count_dec();
1da177e4 2005
1ae132b0 2006 return retc;
1da177e4 2007}
61b1ab45
HS		 2008
2009/*
2010 * Initialize data struct for each netns
2011 */
802cb437 2012int __net_init ip_vs_sync_net_init(struct netns_ipvs *ipvs)
61b1ab45 2013{
ae1d48b2 2014 __mutex_init(&ipvs->sync_mutex, "ipvs->sync_mutex", &__ipvs_sync_key);
f131315f
HS		 2015 spin_lock_init(&ipvs->sync_lock);
2016 spin_lock_init(&ipvs->sync_buff_lock);
61b1ab45
HS		 2017 return 0;
2018}
2019
ebea1f7c 2020void ip_vs_sync_net_cleanup(struct netns_ipvs *ipvs)
61b1ab45 2021{
1ae132b0
HS		 2022 int retc;
2023
ae1d48b2 2024 mutex_lock(&ipvs->sync_mutex);
b3cf3cbf 2025 retc = stop_sync_thread(ipvs, IP_VS_STATE_MASTER);
1ae132b0
HS		 2026 if (retc && retc != -ESRCH)
2027 pr_err("Failed to stop Master Daemon\n");
2028
b3cf3cbf 2029 retc = stop_sync_thread(ipvs, IP_VS_STATE_BACKUP);
1ae132b0
HS		 2030 if (retc && retc != -ESRCH)
2031 pr_err("Failed to stop Backup Daemon\n");
ae1d48b2 2032 mutex_unlock(&ipvs->sync_mutex);
61b1ab45 2033}
Linux 6.x block layer and io_uring tree(s)