Commit | Line | Data |
---|---|---|
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 | * | |
8 | * Version: $Id: ip_vs_sync.c,v 1.13 2003/06/08 09:31:19 wensong Exp $ | |
9 | * | |
10 | * Authors: Wensong Zhang <wensong@linuxvirtualserver.org> | |
11 | * | |
12 | * ip_vs_sync: sync connection info from master load balancer to backups | |
13 | * through multicast | |
14 | * | |
15 | * Changes: | |
16 | * Alexandre Cassen : Added master & backup support at a time. | |
17 | * Alexandre Cassen : Added SyncID support for incoming sync | |
18 | * messages filtering. | |
19 | * Justin Ossevoort : Fix endian problem on sync message size. | |
20 | */ | |
21 | ||
22 | #include <linux/module.h> | |
23 | #include <linux/slab.h> | |
14c85021 | 24 | #include <linux/inetdevice.h> |
1da177e4 LT |
25 | #include <linux/net.h> |
26 | #include <linux/completion.h> | |
27 | #include <linux/delay.h> | |
28 | #include <linux/skbuff.h> | |
29 | #include <linux/in.h> | |
30 | #include <linux/igmp.h> /* for ip_mc_join_group */ | |
14c85021 | 31 | #include <linux/udp.h> |
1da177e4 LT |
32 | |
33 | #include <net/ip.h> | |
34 | #include <net/sock.h> | |
35 | #include <asm/uaccess.h> /* for get_fs and set_fs */ | |
36 | ||
37 | #include <net/ip_vs.h> | |
38 | ||
39 | #define IP_VS_SYNC_GROUP 0xe0000051 /* multicast addr - 224.0.0.81 */ | |
40 | #define IP_VS_SYNC_PORT 8848 /* multicast port */ | |
41 | ||
42 | ||
43 | /* | |
44 | * IPVS sync connection entry | |
45 | */ | |
46 | struct ip_vs_sync_conn { | |
47 | __u8 reserved; | |
48 | ||
49 | /* Protocol, addresses and port numbers */ | |
50 | __u8 protocol; /* Which protocol (TCP/UDP) */ | |
014d730d AV |
51 | __be16 cport; |
52 | __be16 vport; | |
53 | __be16 dport; | |
54 | __be32 caddr; /* client address */ | |
55 | __be32 vaddr; /* virtual address */ | |
56 | __be32 daddr; /* destination address */ | |
1da177e4 LT |
57 | |
58 | /* Flags and state transition */ | |
014d730d AV |
59 | __be16 flags; /* status flags */ |
60 | __be16 state; /* state info */ | |
1da177e4 LT |
61 | |
62 | /* The sequence options start here */ | |
63 | }; | |
64 | ||
65 | struct ip_vs_sync_conn_options { | |
66 | struct ip_vs_seq in_seq; /* incoming seq. struct */ | |
67 | struct ip_vs_seq out_seq; /* outgoing seq. struct */ | |
68 | }; | |
69 | ||
cc0191ae NH |
70 | struct ip_vs_sync_thread_data { |
71 | struct completion *startup; | |
72 | int state; | |
73 | }; | |
74 | ||
1da177e4 LT |
75 | #define IP_VS_SYNC_CONN_TIMEOUT (3*60*HZ) |
76 | #define SIMPLE_CONN_SIZE (sizeof(struct ip_vs_sync_conn)) | |
77 | #define FULL_CONN_SIZE \ | |
78 | (sizeof(struct ip_vs_sync_conn) + sizeof(struct ip_vs_sync_conn_options)) | |
79 | ||
80 | ||
81 | /* | |
82 | The master mulitcasts messages to the backup load balancers in the | |
83 | following format. | |
84 | ||
85 | 0 1 2 3 | |
86 | 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 | |
87 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
88 | | Count Conns | SyncID | Size | | |
89 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
90 | | | | |
91 | | IPVS Sync Connection (1) | | |
92 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
93 | | . | | |
94 | | . | | |
95 | | . | | |
96 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
97 | | | | |
98 | | IPVS Sync Connection (n) | | |
99 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
100 | */ | |
101 | ||
102 | #define SYNC_MESG_HEADER_LEN 4 | |
103 | ||
104 | struct ip_vs_sync_mesg { | |
105 | __u8 nr_conns; | |
106 | __u8 syncid; | |
107 | __u16 size; | |
108 | ||
109 | /* ip_vs_sync_conn entries start here */ | |
110 | }; | |
111 | ||
112 | /* the maximum length of sync (sending/receiving) message */ | |
113 | static int sync_send_mesg_maxlen; | |
114 | static int sync_recv_mesg_maxlen; | |
115 | ||
116 | struct ip_vs_sync_buff { | |
117 | struct list_head list; | |
118 | unsigned long firstuse; | |
119 | ||
120 | /* pointers for the message data */ | |
121 | struct ip_vs_sync_mesg *mesg; | |
122 | unsigned char *head; | |
123 | unsigned char *end; | |
124 | }; | |
125 | ||
126 | ||
127 | /* the sync_buff list head and the lock */ | |
128 | static LIST_HEAD(ip_vs_sync_queue); | |
129 | static DEFINE_SPINLOCK(ip_vs_sync_lock); | |
130 | ||
131 | /* current sync_buff for accepting new conn entries */ | |
132 | static struct ip_vs_sync_buff *curr_sb = NULL; | |
133 | static DEFINE_SPINLOCK(curr_sb_lock); | |
134 | ||
135 | /* ipvs sync daemon state */ | |
136 | volatile int ip_vs_sync_state = IP_VS_STATE_NONE; | |
137 | volatile int ip_vs_master_syncid = 0; | |
138 | volatile int ip_vs_backup_syncid = 0; | |
139 | ||
140 | /* multicast interface name */ | |
141 | char ip_vs_master_mcast_ifn[IP_VS_IFNAME_MAXLEN]; | |
142 | char ip_vs_backup_mcast_ifn[IP_VS_IFNAME_MAXLEN]; | |
143 | ||
144 | /* multicast addr */ | |
145 | static struct sockaddr_in mcast_addr; | |
146 | ||
147 | ||
148 | static inline void sb_queue_tail(struct ip_vs_sync_buff *sb) | |
149 | { | |
150 | spin_lock(&ip_vs_sync_lock); | |
151 | list_add_tail(&sb->list, &ip_vs_sync_queue); | |
152 | spin_unlock(&ip_vs_sync_lock); | |
153 | } | |
154 | ||
155 | static inline struct ip_vs_sync_buff * sb_dequeue(void) | |
156 | { | |
157 | struct ip_vs_sync_buff *sb; | |
158 | ||
159 | spin_lock_bh(&ip_vs_sync_lock); | |
160 | if (list_empty(&ip_vs_sync_queue)) { | |
161 | sb = NULL; | |
162 | } else { | |
163 | sb = list_entry(ip_vs_sync_queue.next, | |
164 | struct ip_vs_sync_buff, | |
165 | list); | |
166 | list_del(&sb->list); | |
167 | } | |
168 | spin_unlock_bh(&ip_vs_sync_lock); | |
169 | ||
170 | return sb; | |
171 | } | |
172 | ||
173 | static inline struct ip_vs_sync_buff * ip_vs_sync_buff_create(void) | |
174 | { | |
175 | struct ip_vs_sync_buff *sb; | |
176 | ||
177 | if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC))) | |
178 | return NULL; | |
179 | ||
180 | if (!(sb->mesg=kmalloc(sync_send_mesg_maxlen, GFP_ATOMIC))) { | |
181 | kfree(sb); | |
182 | return NULL; | |
183 | } | |
184 | sb->mesg->nr_conns = 0; | |
185 | sb->mesg->syncid = ip_vs_master_syncid; | |
186 | sb->mesg->size = 4; | |
187 | sb->head = (unsigned char *)sb->mesg + 4; | |
188 | sb->end = (unsigned char *)sb->mesg + sync_send_mesg_maxlen; | |
189 | sb->firstuse = jiffies; | |
190 | return sb; | |
191 | } | |
192 | ||
193 | static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb) | |
194 | { | |
195 | kfree(sb->mesg); | |
196 | kfree(sb); | |
197 | } | |
198 | ||
199 | /* | |
200 | * Get the current sync buffer if it has been created for more | |
201 | * than the specified time or the specified time is zero. | |
202 | */ | |
203 | static inline struct ip_vs_sync_buff * | |
204 | get_curr_sync_buff(unsigned long time) | |
205 | { | |
206 | struct ip_vs_sync_buff *sb; | |
207 | ||
208 | spin_lock_bh(&curr_sb_lock); | |
209 | if (curr_sb && (time == 0 || | |
210 | time_before(jiffies - curr_sb->firstuse, time))) { | |
211 | sb = curr_sb; | |
212 | curr_sb = NULL; | |
213 | } else | |
214 | sb = NULL; | |
215 | spin_unlock_bh(&curr_sb_lock); | |
216 | return sb; | |
217 | } | |
218 | ||
219 | ||
220 | /* | |
221 | * Add an ip_vs_conn information into the current sync_buff. | |
222 | * Called by ip_vs_in. | |
223 | */ | |
224 | void ip_vs_sync_conn(struct ip_vs_conn *cp) | |
225 | { | |
226 | struct ip_vs_sync_mesg *m; | |
227 | struct ip_vs_sync_conn *s; | |
228 | int len; | |
229 | ||
230 | spin_lock(&curr_sb_lock); | |
231 | if (!curr_sb) { | |
232 | if (!(curr_sb=ip_vs_sync_buff_create())) { | |
233 | spin_unlock(&curr_sb_lock); | |
234 | IP_VS_ERR("ip_vs_sync_buff_create failed.\n"); | |
235 | return; | |
236 | } | |
237 | } | |
238 | ||
239 | len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE : | |
240 | SIMPLE_CONN_SIZE; | |
241 | m = curr_sb->mesg; | |
242 | s = (struct ip_vs_sync_conn *)curr_sb->head; | |
243 | ||
244 | /* copy members */ | |
245 | s->protocol = cp->protocol; | |
246 | s->cport = cp->cport; | |
247 | s->vport = cp->vport; | |
248 | s->dport = cp->dport; | |
249 | s->caddr = cp->caddr; | |
250 | s->vaddr = cp->vaddr; | |
251 | s->daddr = cp->daddr; | |
252 | s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED); | |
253 | s->state = htons(cp->state); | |
254 | if (cp->flags & IP_VS_CONN_F_SEQ_MASK) { | |
255 | struct ip_vs_sync_conn_options *opt = | |
256 | (struct ip_vs_sync_conn_options *)&s[1]; | |
257 | memcpy(opt, &cp->in_seq, sizeof(*opt)); | |
258 | } | |
259 | ||
260 | m->nr_conns++; | |
261 | m->size += len; | |
262 | curr_sb->head += len; | |
263 | ||
264 | /* check if there is a space for next one */ | |
265 | if (curr_sb->head+FULL_CONN_SIZE > curr_sb->end) { | |
266 | sb_queue_tail(curr_sb); | |
267 | curr_sb = NULL; | |
268 | } | |
269 | spin_unlock(&curr_sb_lock); | |
270 | ||
271 | /* synchronize its controller if it has */ | |
272 | if (cp->control) | |
273 | ip_vs_sync_conn(cp->control); | |
274 | } | |
275 | ||
276 | ||
277 | /* | |
278 | * Process received multicast message and create the corresponding | |
279 | * ip_vs_conn entries. | |
280 | */ | |
281 | static void ip_vs_process_message(const char *buffer, const size_t buflen) | |
282 | { | |
283 | struct ip_vs_sync_mesg *m = (struct ip_vs_sync_mesg *)buffer; | |
284 | struct ip_vs_sync_conn *s; | |
285 | struct ip_vs_sync_conn_options *opt; | |
286 | struct ip_vs_conn *cp; | |
287 | char *p; | |
288 | int i; | |
289 | ||
290 | /* Convert size back to host byte order */ | |
291 | m->size = ntohs(m->size); | |
292 | ||
293 | if (buflen != m->size) { | |
294 | IP_VS_ERR("bogus message\n"); | |
295 | return; | |
296 | } | |
297 | ||
298 | /* SyncID sanity check */ | |
299 | if (ip_vs_backup_syncid != 0 && m->syncid != ip_vs_backup_syncid) { | |
300 | IP_VS_DBG(7, "Ignoring incoming msg with syncid = %d\n", | |
301 | m->syncid); | |
302 | return; | |
303 | } | |
304 | ||
305 | p = (char *)buffer + sizeof(struct ip_vs_sync_mesg); | |
306 | for (i=0; i<m->nr_conns; i++) { | |
87375ab4 JA |
307 | unsigned flags; |
308 | ||
1da177e4 | 309 | s = (struct ip_vs_sync_conn *)p; |
87375ab4 JA |
310 | flags = ntohs(s->flags); |
311 | if (!(flags & IP_VS_CONN_F_TEMPLATE)) | |
312 | cp = ip_vs_conn_in_get(s->protocol, | |
313 | s->caddr, s->cport, | |
314 | s->vaddr, s->vport); | |
315 | else | |
316 | cp = ip_vs_ct_in_get(s->protocol, | |
317 | s->caddr, s->cport, | |
318 | s->vaddr, s->vport); | |
1da177e4 LT |
319 | if (!cp) { |
320 | cp = ip_vs_conn_new(s->protocol, | |
321 | s->caddr, s->cport, | |
322 | s->vaddr, s->vport, | |
323 | s->daddr, s->dport, | |
87375ab4 | 324 | flags, NULL); |
1da177e4 LT |
325 | if (!cp) { |
326 | IP_VS_ERR("ip_vs_conn_new failed\n"); | |
327 | return; | |
328 | } | |
329 | cp->state = ntohs(s->state); | |
330 | } else if (!cp->dest) { | |
331 | /* it is an entry created by the synchronization */ | |
332 | cp->state = ntohs(s->state); | |
87375ab4 | 333 | cp->flags = flags | IP_VS_CONN_F_HASHED; |
1da177e4 LT |
334 | } /* Note that we don't touch its state and flags |
335 | if it is a normal entry. */ | |
336 | ||
87375ab4 | 337 | if (flags & IP_VS_CONN_F_SEQ_MASK) { |
1da177e4 LT |
338 | opt = (struct ip_vs_sync_conn_options *)&s[1]; |
339 | memcpy(&cp->in_seq, opt, sizeof(*opt)); | |
340 | p += FULL_CONN_SIZE; | |
341 | } else | |
342 | p += SIMPLE_CONN_SIZE; | |
343 | ||
344 | atomic_set(&cp->in_pkts, sysctl_ip_vs_sync_threshold[0]); | |
345 | cp->timeout = IP_VS_SYNC_CONN_TIMEOUT; | |
346 | ip_vs_conn_put(cp); | |
347 | ||
348 | if (p > buffer+buflen) { | |
349 | IP_VS_ERR("bogus message\n"); | |
350 | return; | |
351 | } | |
352 | } | |
353 | } | |
354 | ||
355 | ||
356 | /* | |
357 | * Setup loopback of outgoing multicasts on a sending socket | |
358 | */ | |
359 | static void set_mcast_loop(struct sock *sk, u_char loop) | |
360 | { | |
361 | struct inet_sock *inet = inet_sk(sk); | |
362 | ||
363 | /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */ | |
364 | lock_sock(sk); | |
365 | inet->mc_loop = loop ? 1 : 0; | |
366 | release_sock(sk); | |
367 | } | |
368 | ||
369 | /* | |
370 | * Specify TTL for outgoing multicasts on a sending socket | |
371 | */ | |
372 | static void set_mcast_ttl(struct sock *sk, u_char ttl) | |
373 | { | |
374 | struct inet_sock *inet = inet_sk(sk); | |
375 | ||
376 | /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */ | |
377 | lock_sock(sk); | |
378 | inet->mc_ttl = ttl; | |
379 | release_sock(sk); | |
380 | } | |
381 | ||
382 | /* | |
383 | * Specifiy default interface for outgoing multicasts | |
384 | */ | |
385 | static int set_mcast_if(struct sock *sk, char *ifname) | |
386 | { | |
387 | struct net_device *dev; | |
388 | struct inet_sock *inet = inet_sk(sk); | |
389 | ||
881d966b | 390 | if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL) |
1da177e4 LT |
391 | return -ENODEV; |
392 | ||
393 | if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) | |
394 | return -EINVAL; | |
395 | ||
396 | lock_sock(sk); | |
397 | inet->mc_index = dev->ifindex; | |
398 | /* inet->mc_addr = 0; */ | |
399 | release_sock(sk); | |
400 | ||
401 | return 0; | |
402 | } | |
403 | ||
404 | ||
405 | /* | |
406 | * Set the maximum length of sync message according to the | |
407 | * specified interface's MTU. | |
408 | */ | |
409 | static int set_sync_mesg_maxlen(int sync_state) | |
410 | { | |
411 | struct net_device *dev; | |
412 | int num; | |
413 | ||
414 | if (sync_state == IP_VS_STATE_MASTER) { | |
881d966b | 415 | if ((dev = __dev_get_by_name(&init_net, ip_vs_master_mcast_ifn)) == NULL) |
1da177e4 LT |
416 | return -ENODEV; |
417 | ||
418 | num = (dev->mtu - sizeof(struct iphdr) - | |
419 | sizeof(struct udphdr) - | |
420 | SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE; | |
421 | sync_send_mesg_maxlen = | |
422 | SYNC_MESG_HEADER_LEN + SIMPLE_CONN_SIZE * num; | |
423 | IP_VS_DBG(7, "setting the maximum length of sync sending " | |
424 | "message %d.\n", sync_send_mesg_maxlen); | |
425 | } else if (sync_state == IP_VS_STATE_BACKUP) { | |
881d966b | 426 | if ((dev = __dev_get_by_name(&init_net, ip_vs_backup_mcast_ifn)) == NULL) |
1da177e4 LT |
427 | return -ENODEV; |
428 | ||
429 | sync_recv_mesg_maxlen = dev->mtu - | |
430 | sizeof(struct iphdr) - sizeof(struct udphdr); | |
431 | IP_VS_DBG(7, "setting the maximum length of sync receiving " | |
432 | "message %d.\n", sync_recv_mesg_maxlen); | |
433 | } | |
434 | ||
435 | return 0; | |
436 | } | |
437 | ||
438 | ||
439 | /* | |
440 | * Join a multicast group. | |
441 | * the group is specified by a class D multicast address 224.0.0.0/8 | |
442 | * in the in_addr structure passed in as a parameter. | |
443 | */ | |
444 | static int | |
445 | join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname) | |
446 | { | |
447 | struct ip_mreqn mreq; | |
448 | struct net_device *dev; | |
449 | int ret; | |
450 | ||
451 | memset(&mreq, 0, sizeof(mreq)); | |
452 | memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr)); | |
453 | ||
881d966b | 454 | if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL) |
1da177e4 LT |
455 | return -ENODEV; |
456 | if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) | |
457 | return -EINVAL; | |
458 | ||
459 | mreq.imr_ifindex = dev->ifindex; | |
460 | ||
461 | lock_sock(sk); | |
462 | ret = ip_mc_join_group(sk, &mreq); | |
463 | release_sock(sk); | |
464 | ||
465 | return ret; | |
466 | } | |
467 | ||
468 | ||
469 | static int bind_mcastif_addr(struct socket *sock, char *ifname) | |
470 | { | |
471 | struct net_device *dev; | |
a61ced5d | 472 | __be32 addr; |
1da177e4 LT |
473 | struct sockaddr_in sin; |
474 | ||
881d966b | 475 | if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL) |
1da177e4 LT |
476 | return -ENODEV; |
477 | ||
478 | addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE); | |
479 | if (!addr) | |
480 | IP_VS_ERR("You probably need to specify IP address on " | |
481 | "multicast interface.\n"); | |
482 | ||
483 | IP_VS_DBG(7, "binding socket with (%s) %u.%u.%u.%u\n", | |
484 | ifname, NIPQUAD(addr)); | |
485 | ||
486 | /* Now bind the socket with the address of multicast interface */ | |
487 | sin.sin_family = AF_INET; | |
488 | sin.sin_addr.s_addr = addr; | |
489 | sin.sin_port = 0; | |
490 | ||
491 | return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin)); | |
492 | } | |
493 | ||
494 | /* | |
495 | * Set up sending multicast socket over UDP | |
496 | */ | |
497 | static struct socket * make_send_sock(void) | |
498 | { | |
499 | struct socket *sock; | |
500 | ||
501 | /* First create a socket */ | |
502 | if (sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock) < 0) { | |
503 | IP_VS_ERR("Error during creation of socket; terminating\n"); | |
504 | return NULL; | |
505 | } | |
506 | ||
507 | if (set_mcast_if(sock->sk, ip_vs_master_mcast_ifn) < 0) { | |
508 | IP_VS_ERR("Error setting outbound mcast interface\n"); | |
509 | goto error; | |
510 | } | |
511 | ||
512 | set_mcast_loop(sock->sk, 0); | |
513 | set_mcast_ttl(sock->sk, 1); | |
514 | ||
515 | if (bind_mcastif_addr(sock, ip_vs_master_mcast_ifn) < 0) { | |
516 | IP_VS_ERR("Error binding address of the mcast interface\n"); | |
517 | goto error; | |
518 | } | |
519 | ||
520 | if (sock->ops->connect(sock, | |
521 | (struct sockaddr*)&mcast_addr, | |
522 | sizeof(struct sockaddr), 0) < 0) { | |
523 | IP_VS_ERR("Error connecting to the multicast addr\n"); | |
524 | goto error; | |
525 | } | |
526 | ||
527 | return sock; | |
528 | ||
529 | error: | |
530 | sock_release(sock); | |
531 | return NULL; | |
532 | } | |
533 | ||
534 | ||
535 | /* | |
536 | * Set up receiving multicast socket over UDP | |
537 | */ | |
538 | static struct socket * make_receive_sock(void) | |
539 | { | |
540 | struct socket *sock; | |
541 | ||
542 | /* First create a socket */ | |
543 | if (sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock) < 0) { | |
544 | IP_VS_ERR("Error during creation of socket; terminating\n"); | |
545 | return NULL; | |
546 | } | |
547 | ||
548 | /* it is equivalent to the REUSEADDR option in user-space */ | |
549 | sock->sk->sk_reuse = 1; | |
550 | ||
551 | if (sock->ops->bind(sock, | |
552 | (struct sockaddr*)&mcast_addr, | |
553 | sizeof(struct sockaddr)) < 0) { | |
554 | IP_VS_ERR("Error binding to the multicast addr\n"); | |
555 | goto error; | |
556 | } | |
557 | ||
558 | /* join the multicast group */ | |
559 | if (join_mcast_group(sock->sk, | |
560 | (struct in_addr*)&mcast_addr.sin_addr, | |
561 | ip_vs_backup_mcast_ifn) < 0) { | |
562 | IP_VS_ERR("Error joining to the multicast group\n"); | |
563 | goto error; | |
564 | } | |
565 | ||
566 | return sock; | |
567 | ||
568 | error: | |
569 | sock_release(sock); | |
570 | return NULL; | |
571 | } | |
572 | ||
573 | ||
574 | static int | |
575 | ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length) | |
576 | { | |
577 | struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL}; | |
578 | struct kvec iov; | |
579 | int len; | |
580 | ||
581 | EnterFunction(7); | |
582 | iov.iov_base = (void *)buffer; | |
583 | iov.iov_len = length; | |
584 | ||
585 | len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length)); | |
586 | ||
587 | LeaveFunction(7); | |
588 | return len; | |
589 | } | |
590 | ||
591 | static void | |
592 | ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg) | |
593 | { | |
594 | int msize; | |
595 | ||
596 | msize = msg->size; | |
597 | ||
598 | /* Put size in network byte order */ | |
599 | msg->size = htons(msg->size); | |
600 | ||
601 | if (ip_vs_send_async(sock, (char *)msg, msize) != msize) | |
602 | IP_VS_ERR("ip_vs_send_async error\n"); | |
603 | } | |
604 | ||
605 | static int | |
606 | ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen) | |
607 | { | |
608 | struct msghdr msg = {NULL,}; | |
609 | struct kvec iov; | |
610 | int len; | |
611 | ||
612 | EnterFunction(7); | |
613 | ||
614 | /* Receive a packet */ | |
615 | iov.iov_base = buffer; | |
616 | iov.iov_len = (size_t)buflen; | |
617 | ||
618 | len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, 0); | |
619 | ||
620 | if (len < 0) | |
621 | return -1; | |
622 | ||
623 | LeaveFunction(7); | |
624 | return len; | |
625 | } | |
626 | ||
627 | ||
628 | static DECLARE_WAIT_QUEUE_HEAD(sync_wait); | |
629 | static pid_t sync_master_pid = 0; | |
630 | static pid_t sync_backup_pid = 0; | |
631 | ||
632 | static DECLARE_WAIT_QUEUE_HEAD(stop_sync_wait); | |
633 | static int stop_master_sync = 0; | |
634 | static int stop_backup_sync = 0; | |
635 | ||
636 | static void sync_master_loop(void) | |
637 | { | |
638 | struct socket *sock; | |
639 | struct ip_vs_sync_buff *sb; | |
640 | ||
641 | /* create the sending multicast socket */ | |
642 | sock = make_send_sock(); | |
643 | if (!sock) | |
644 | return; | |
645 | ||
646 | IP_VS_INFO("sync thread started: state = MASTER, mcast_ifn = %s, " | |
647 | "syncid = %d\n", | |
648 | ip_vs_master_mcast_ifn, ip_vs_master_syncid); | |
649 | ||
650 | for (;;) { | |
651 | while ((sb=sb_dequeue())) { | |
652 | ip_vs_send_sync_msg(sock, sb->mesg); | |
653 | ip_vs_sync_buff_release(sb); | |
654 | } | |
655 | ||
656 | /* check if entries stay in curr_sb for 2 seconds */ | |
657 | if ((sb = get_curr_sync_buff(2*HZ))) { | |
658 | ip_vs_send_sync_msg(sock, sb->mesg); | |
659 | ip_vs_sync_buff_release(sb); | |
660 | } | |
661 | ||
662 | if (stop_master_sync) | |
663 | break; | |
664 | ||
89eaeb09 | 665 | msleep_interruptible(1000); |
1da177e4 LT |
666 | } |
667 | ||
668 | /* clean up the sync_buff queue */ | |
669 | while ((sb=sb_dequeue())) { | |
670 | ip_vs_sync_buff_release(sb); | |
671 | } | |
672 | ||
673 | /* clean up the current sync_buff */ | |
674 | if ((sb = get_curr_sync_buff(0))) { | |
675 | ip_vs_sync_buff_release(sb); | |
676 | } | |
677 | ||
678 | /* release the sending multicast socket */ | |
679 | sock_release(sock); | |
680 | } | |
681 | ||
682 | ||
683 | static void sync_backup_loop(void) | |
684 | { | |
685 | struct socket *sock; | |
686 | char *buf; | |
687 | int len; | |
688 | ||
689 | if (!(buf = kmalloc(sync_recv_mesg_maxlen, GFP_ATOMIC))) { | |
690 | IP_VS_ERR("sync_backup_loop: kmalloc error\n"); | |
691 | return; | |
692 | } | |
693 | ||
694 | /* create the receiving multicast socket */ | |
695 | sock = make_receive_sock(); | |
696 | if (!sock) | |
697 | goto out; | |
698 | ||
699 | IP_VS_INFO("sync thread started: state = BACKUP, mcast_ifn = %s, " | |
700 | "syncid = %d\n", | |
701 | ip_vs_backup_mcast_ifn, ip_vs_backup_syncid); | |
702 | ||
703 | for (;;) { | |
704 | /* do you have data now? */ | |
705 | while (!skb_queue_empty(&(sock->sk->sk_receive_queue))) { | |
706 | if ((len = | |
707 | ip_vs_receive(sock, buf, | |
708 | sync_recv_mesg_maxlen)) <= 0) { | |
709 | IP_VS_ERR("receiving message error\n"); | |
710 | break; | |
711 | } | |
712 | /* disable bottom half, because it accessed the data | |
713 | shared by softirq while getting/creating conns */ | |
714 | local_bh_disable(); | |
715 | ip_vs_process_message(buf, len); | |
716 | local_bh_enable(); | |
717 | } | |
718 | ||
719 | if (stop_backup_sync) | |
720 | break; | |
721 | ||
89eaeb09 | 722 | msleep_interruptible(1000); |
1da177e4 LT |
723 | } |
724 | ||
725 | /* release the sending multicast socket */ | |
726 | sock_release(sock); | |
727 | ||
728 | out: | |
729 | kfree(buf); | |
730 | } | |
731 | ||
732 | ||
733 | static void set_sync_pid(int sync_state, pid_t sync_pid) | |
734 | { | |
735 | if (sync_state == IP_VS_STATE_MASTER) | |
736 | sync_master_pid = sync_pid; | |
737 | else if (sync_state == IP_VS_STATE_BACKUP) | |
738 | sync_backup_pid = sync_pid; | |
739 | } | |
740 | ||
741 | static void set_stop_sync(int sync_state, int set) | |
742 | { | |
743 | if (sync_state == IP_VS_STATE_MASTER) | |
744 | stop_master_sync = set; | |
745 | else if (sync_state == IP_VS_STATE_BACKUP) | |
746 | stop_backup_sync = set; | |
747 | else { | |
748 | stop_master_sync = set; | |
749 | stop_backup_sync = set; | |
750 | } | |
751 | } | |
752 | ||
753 | static int sync_thread(void *startup) | |
754 | { | |
755 | DECLARE_WAITQUEUE(wait, current); | |
756 | mm_segment_t oldmm; | |
757 | int state; | |
758 | const char *name; | |
cc0191ae | 759 | struct ip_vs_sync_thread_data *tinfo = startup; |
1da177e4 LT |
760 | |
761 | /* increase the module use count */ | |
762 | ip_vs_use_count_inc(); | |
763 | ||
764 | if (ip_vs_sync_state & IP_VS_STATE_MASTER && !sync_master_pid) { | |
765 | state = IP_VS_STATE_MASTER; | |
766 | name = "ipvs_syncmaster"; | |
767 | } else if (ip_vs_sync_state & IP_VS_STATE_BACKUP && !sync_backup_pid) { | |
768 | state = IP_VS_STATE_BACKUP; | |
769 | name = "ipvs_syncbackup"; | |
770 | } else { | |
771 | IP_VS_BUG(); | |
772 | ip_vs_use_count_dec(); | |
773 | return -EINVAL; | |
774 | } | |
775 | ||
776 | daemonize(name); | |
777 | ||
778 | oldmm = get_fs(); | |
779 | set_fs(KERNEL_DS); | |
780 | ||
781 | /* Block all signals */ | |
782 | spin_lock_irq(¤t->sighand->siglock); | |
783 | siginitsetinv(¤t->blocked, 0); | |
784 | recalc_sigpending(); | |
785 | spin_unlock_irq(¤t->sighand->siglock); | |
786 | ||
787 | /* set the maximum length of sync message */ | |
788 | set_sync_mesg_maxlen(state); | |
789 | ||
790 | /* set up multicast address */ | |
791 | mcast_addr.sin_family = AF_INET; | |
792 | mcast_addr.sin_port = htons(IP_VS_SYNC_PORT); | |
793 | mcast_addr.sin_addr.s_addr = htonl(IP_VS_SYNC_GROUP); | |
794 | ||
795 | add_wait_queue(&sync_wait, &wait); | |
796 | ||
797 | set_sync_pid(state, current->pid); | |
cc0191ae NH |
798 | complete(tinfo->startup); |
799 | ||
800 | /* | |
801 | * once we call the completion queue above, we should | |
802 | * null out that reference, since its allocated on the | |
803 | * stack of the creating kernel thread | |
804 | */ | |
805 | tinfo->startup = NULL; | |
1da177e4 LT |
806 | |
807 | /* processing master/backup loop here */ | |
808 | if (state == IP_VS_STATE_MASTER) | |
809 | sync_master_loop(); | |
810 | else if (state == IP_VS_STATE_BACKUP) | |
811 | sync_backup_loop(); | |
812 | else IP_VS_BUG(); | |
813 | ||
814 | remove_wait_queue(&sync_wait, &wait); | |
815 | ||
816 | /* thread exits */ | |
cc0191ae NH |
817 | |
818 | /* | |
819 | * If we weren't explicitly stopped, then we | |
820 | * exited in error, and should undo our state | |
821 | */ | |
822 | if ((!stop_master_sync) && (!stop_backup_sync)) | |
823 | ip_vs_sync_state -= tinfo->state; | |
824 | ||
1da177e4 LT |
825 | set_sync_pid(state, 0); |
826 | IP_VS_INFO("sync thread stopped!\n"); | |
827 | ||
828 | set_fs(oldmm); | |
829 | ||
830 | /* decrease the module use count */ | |
831 | ip_vs_use_count_dec(); | |
832 | ||
833 | set_stop_sync(state, 0); | |
834 | wake_up(&stop_sync_wait); | |
835 | ||
cc0191ae NH |
836 | /* |
837 | * we need to free the structure that was allocated | |
838 | * for us in start_sync_thread | |
839 | */ | |
840 | kfree(tinfo); | |
1da177e4 LT |
841 | return 0; |
842 | } | |
843 | ||
844 | ||
845 | static int fork_sync_thread(void *startup) | |
846 | { | |
847 | pid_t pid; | |
848 | ||
849 | /* fork the sync thread here, then the parent process of the | |
850 | sync thread is the init process after this thread exits. */ | |
851 | repeat: | |
852 | if ((pid = kernel_thread(sync_thread, startup, 0)) < 0) { | |
853 | IP_VS_ERR("could not create sync_thread due to %d... " | |
854 | "retrying.\n", pid); | |
89eaeb09 | 855 | msleep_interruptible(1000); |
1da177e4 LT |
856 | goto repeat; |
857 | } | |
858 | ||
859 | return 0; | |
860 | } | |
861 | ||
862 | ||
863 | int start_sync_thread(int state, char *mcast_ifn, __u8 syncid) | |
864 | { | |
6e9a4738 | 865 | DECLARE_COMPLETION_ONSTACK(startup); |
1da177e4 | 866 | pid_t pid; |
cc0191ae | 867 | struct ip_vs_sync_thread_data *tinfo; |
1da177e4 LT |
868 | |
869 | if ((state == IP_VS_STATE_MASTER && sync_master_pid) || | |
870 | (state == IP_VS_STATE_BACKUP && sync_backup_pid)) | |
871 | return -EEXIST; | |
872 | ||
cc0191ae NH |
873 | /* |
874 | * Note that tinfo will be freed in sync_thread on exit | |
875 | */ | |
876 | tinfo = kmalloc(sizeof(struct ip_vs_sync_thread_data), GFP_KERNEL); | |
877 | if (!tinfo) | |
878 | return -ENOMEM; | |
879 | ||
1da177e4 LT |
880 | IP_VS_DBG(7, "%s: pid %d\n", __FUNCTION__, current->pid); |
881 | IP_VS_DBG(7, "Each ip_vs_sync_conn entry need %Zd bytes\n", | |
882 | sizeof(struct ip_vs_sync_conn)); | |
883 | ||
884 | ip_vs_sync_state |= state; | |
885 | if (state == IP_VS_STATE_MASTER) { | |
37004af3 SH |
886 | strlcpy(ip_vs_master_mcast_ifn, mcast_ifn, |
887 | sizeof(ip_vs_master_mcast_ifn)); | |
1da177e4 LT |
888 | ip_vs_master_syncid = syncid; |
889 | } else { | |
37004af3 SH |
890 | strlcpy(ip_vs_backup_mcast_ifn, mcast_ifn, |
891 | sizeof(ip_vs_backup_mcast_ifn)); | |
1da177e4 LT |
892 | ip_vs_backup_syncid = syncid; |
893 | } | |
894 | ||
cc0191ae NH |
895 | tinfo->state = state; |
896 | tinfo->startup = &startup; | |
897 | ||
1da177e4 | 898 | repeat: |
cc0191ae | 899 | if ((pid = kernel_thread(fork_sync_thread, tinfo, 0)) < 0) { |
1da177e4 LT |
900 | IP_VS_ERR("could not create fork_sync_thread due to %d... " |
901 | "retrying.\n", pid); | |
89eaeb09 | 902 | msleep_interruptible(1000); |
1da177e4 LT |
903 | goto repeat; |
904 | } | |
905 | ||
906 | wait_for_completion(&startup); | |
907 | ||
908 | return 0; | |
909 | } | |
910 | ||
911 | ||
912 | int stop_sync_thread(int state) | |
913 | { | |
914 | DECLARE_WAITQUEUE(wait, current); | |
915 | ||
916 | if ((state == IP_VS_STATE_MASTER && !sync_master_pid) || | |
917 | (state == IP_VS_STATE_BACKUP && !sync_backup_pid)) | |
918 | return -ESRCH; | |
919 | ||
920 | IP_VS_DBG(7, "%s: pid %d\n", __FUNCTION__, current->pid); | |
921 | IP_VS_INFO("stopping sync thread %d ...\n", | |
37004af3 SH |
922 | (state == IP_VS_STATE_MASTER) ? |
923 | sync_master_pid : sync_backup_pid); | |
1da177e4 LT |
924 | |
925 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
926 | add_wait_queue(&stop_sync_wait, &wait); | |
927 | set_stop_sync(state, 1); | |
928 | ip_vs_sync_state -= state; | |
929 | wake_up(&sync_wait); | |
930 | schedule(); | |
931 | __set_current_state(TASK_RUNNING); | |
932 | remove_wait_queue(&stop_sync_wait, &wait); | |
933 | ||
934 | /* Note: no need to reap the sync thread, because its parent | |
935 | process is the init process */ | |
936 | ||
937 | if ((state == IP_VS_STATE_MASTER && stop_master_sync) || | |
938 | (state == IP_VS_STATE_BACKUP && stop_backup_sync)) | |
939 | IP_VS_BUG(); | |
940 | ||
941 | return 0; | |
942 | } |