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 | ||
70 | #define IP_VS_SYNC_CONN_TIMEOUT (3*60*HZ) | |
71 | #define SIMPLE_CONN_SIZE (sizeof(struct ip_vs_sync_conn)) | |
72 | #define FULL_CONN_SIZE \ | |
73 | (sizeof(struct ip_vs_sync_conn) + sizeof(struct ip_vs_sync_conn_options)) | |
74 | ||
75 | ||
76 | /* | |
77 | The master mulitcasts messages to the backup load balancers in the | |
78 | following format. | |
79 | ||
80 | 0 1 2 3 | |
81 | 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 | |
82 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
83 | | Count Conns | SyncID | Size | | |
84 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
85 | | | | |
86 | | IPVS Sync Connection (1) | | |
87 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
88 | | . | | |
89 | | . | | |
90 | | . | | |
91 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
92 | | | | |
93 | | IPVS Sync Connection (n) | | |
94 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
95 | */ | |
96 | ||
97 | #define SYNC_MESG_HEADER_LEN 4 | |
98 | ||
99 | struct ip_vs_sync_mesg { | |
100 | __u8 nr_conns; | |
101 | __u8 syncid; | |
102 | __u16 size; | |
103 | ||
104 | /* ip_vs_sync_conn entries start here */ | |
105 | }; | |
106 | ||
107 | /* the maximum length of sync (sending/receiving) message */ | |
108 | static int sync_send_mesg_maxlen; | |
109 | static int sync_recv_mesg_maxlen; | |
110 | ||
111 | struct ip_vs_sync_buff { | |
112 | struct list_head list; | |
113 | unsigned long firstuse; | |
114 | ||
115 | /* pointers for the message data */ | |
116 | struct ip_vs_sync_mesg *mesg; | |
117 | unsigned char *head; | |
118 | unsigned char *end; | |
119 | }; | |
120 | ||
121 | ||
122 | /* the sync_buff list head and the lock */ | |
123 | static LIST_HEAD(ip_vs_sync_queue); | |
124 | static DEFINE_SPINLOCK(ip_vs_sync_lock); | |
125 | ||
126 | /* current sync_buff for accepting new conn entries */ | |
127 | static struct ip_vs_sync_buff *curr_sb = NULL; | |
128 | static DEFINE_SPINLOCK(curr_sb_lock); | |
129 | ||
130 | /* ipvs sync daemon state */ | |
131 | volatile int ip_vs_sync_state = IP_VS_STATE_NONE; | |
132 | volatile int ip_vs_master_syncid = 0; | |
133 | volatile int ip_vs_backup_syncid = 0; | |
134 | ||
135 | /* multicast interface name */ | |
136 | char ip_vs_master_mcast_ifn[IP_VS_IFNAME_MAXLEN]; | |
137 | char ip_vs_backup_mcast_ifn[IP_VS_IFNAME_MAXLEN]; | |
138 | ||
139 | /* multicast addr */ | |
140 | static struct sockaddr_in mcast_addr; | |
141 | ||
142 | ||
143 | static inline void sb_queue_tail(struct ip_vs_sync_buff *sb) | |
144 | { | |
145 | spin_lock(&ip_vs_sync_lock); | |
146 | list_add_tail(&sb->list, &ip_vs_sync_queue); | |
147 | spin_unlock(&ip_vs_sync_lock); | |
148 | } | |
149 | ||
150 | static inline struct ip_vs_sync_buff * sb_dequeue(void) | |
151 | { | |
152 | struct ip_vs_sync_buff *sb; | |
153 | ||
154 | spin_lock_bh(&ip_vs_sync_lock); | |
155 | if (list_empty(&ip_vs_sync_queue)) { | |
156 | sb = NULL; | |
157 | } else { | |
158 | sb = list_entry(ip_vs_sync_queue.next, | |
159 | struct ip_vs_sync_buff, | |
160 | list); | |
161 | list_del(&sb->list); | |
162 | } | |
163 | spin_unlock_bh(&ip_vs_sync_lock); | |
164 | ||
165 | return sb; | |
166 | } | |
167 | ||
168 | static inline struct ip_vs_sync_buff * ip_vs_sync_buff_create(void) | |
169 | { | |
170 | struct ip_vs_sync_buff *sb; | |
171 | ||
172 | if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC))) | |
173 | return NULL; | |
174 | ||
175 | if (!(sb->mesg=kmalloc(sync_send_mesg_maxlen, GFP_ATOMIC))) { | |
176 | kfree(sb); | |
177 | return NULL; | |
178 | } | |
179 | sb->mesg->nr_conns = 0; | |
180 | sb->mesg->syncid = ip_vs_master_syncid; | |
181 | sb->mesg->size = 4; | |
182 | sb->head = (unsigned char *)sb->mesg + 4; | |
183 | sb->end = (unsigned char *)sb->mesg + sync_send_mesg_maxlen; | |
184 | sb->firstuse = jiffies; | |
185 | return sb; | |
186 | } | |
187 | ||
188 | static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb) | |
189 | { | |
190 | kfree(sb->mesg); | |
191 | kfree(sb); | |
192 | } | |
193 | ||
194 | /* | |
195 | * Get the current sync buffer if it has been created for more | |
196 | * than the specified time or the specified time is zero. | |
197 | */ | |
198 | static inline struct ip_vs_sync_buff * | |
199 | get_curr_sync_buff(unsigned long time) | |
200 | { | |
201 | struct ip_vs_sync_buff *sb; | |
202 | ||
203 | spin_lock_bh(&curr_sb_lock); | |
204 | if (curr_sb && (time == 0 || | |
205 | time_before(jiffies - curr_sb->firstuse, time))) { | |
206 | sb = curr_sb; | |
207 | curr_sb = NULL; | |
208 | } else | |
209 | sb = NULL; | |
210 | spin_unlock_bh(&curr_sb_lock); | |
211 | return sb; | |
212 | } | |
213 | ||
214 | ||
215 | /* | |
216 | * Add an ip_vs_conn information into the current sync_buff. | |
217 | * Called by ip_vs_in. | |
218 | */ | |
219 | void ip_vs_sync_conn(struct ip_vs_conn *cp) | |
220 | { | |
221 | struct ip_vs_sync_mesg *m; | |
222 | struct ip_vs_sync_conn *s; | |
223 | int len; | |
224 | ||
225 | spin_lock(&curr_sb_lock); | |
226 | if (!curr_sb) { | |
227 | if (!(curr_sb=ip_vs_sync_buff_create())) { | |
228 | spin_unlock(&curr_sb_lock); | |
229 | IP_VS_ERR("ip_vs_sync_buff_create failed.\n"); | |
230 | return; | |
231 | } | |
232 | } | |
233 | ||
234 | len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE : | |
235 | SIMPLE_CONN_SIZE; | |
236 | m = curr_sb->mesg; | |
237 | s = (struct ip_vs_sync_conn *)curr_sb->head; | |
238 | ||
239 | /* copy members */ | |
240 | s->protocol = cp->protocol; | |
241 | s->cport = cp->cport; | |
242 | s->vport = cp->vport; | |
243 | s->dport = cp->dport; | |
244 | s->caddr = cp->caddr; | |
245 | s->vaddr = cp->vaddr; | |
246 | s->daddr = cp->daddr; | |
247 | s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED); | |
248 | s->state = htons(cp->state); | |
249 | if (cp->flags & IP_VS_CONN_F_SEQ_MASK) { | |
250 | struct ip_vs_sync_conn_options *opt = | |
251 | (struct ip_vs_sync_conn_options *)&s[1]; | |
252 | memcpy(opt, &cp->in_seq, sizeof(*opt)); | |
253 | } | |
254 | ||
255 | m->nr_conns++; | |
256 | m->size += len; | |
257 | curr_sb->head += len; | |
258 | ||
259 | /* check if there is a space for next one */ | |
260 | if (curr_sb->head+FULL_CONN_SIZE > curr_sb->end) { | |
261 | sb_queue_tail(curr_sb); | |
262 | curr_sb = NULL; | |
263 | } | |
264 | spin_unlock(&curr_sb_lock); | |
265 | ||
266 | /* synchronize its controller if it has */ | |
267 | if (cp->control) | |
268 | ip_vs_sync_conn(cp->control); | |
269 | } | |
270 | ||
271 | ||
272 | /* | |
273 | * Process received multicast message and create the corresponding | |
274 | * ip_vs_conn entries. | |
275 | */ | |
276 | static void ip_vs_process_message(const char *buffer, const size_t buflen) | |
277 | { | |
278 | struct ip_vs_sync_mesg *m = (struct ip_vs_sync_mesg *)buffer; | |
279 | struct ip_vs_sync_conn *s; | |
280 | struct ip_vs_sync_conn_options *opt; | |
281 | struct ip_vs_conn *cp; | |
282 | char *p; | |
283 | int i; | |
284 | ||
285 | /* Convert size back to host byte order */ | |
286 | m->size = ntohs(m->size); | |
287 | ||
288 | if (buflen != m->size) { | |
289 | IP_VS_ERR("bogus message\n"); | |
290 | return; | |
291 | } | |
292 | ||
293 | /* SyncID sanity check */ | |
294 | if (ip_vs_backup_syncid != 0 && m->syncid != ip_vs_backup_syncid) { | |
295 | IP_VS_DBG(7, "Ignoring incoming msg with syncid = %d\n", | |
296 | m->syncid); | |
297 | return; | |
298 | } | |
299 | ||
300 | p = (char *)buffer + sizeof(struct ip_vs_sync_mesg); | |
301 | for (i=0; i<m->nr_conns; i++) { | |
87375ab4 JA |
302 | unsigned flags; |
303 | ||
1da177e4 | 304 | s = (struct ip_vs_sync_conn *)p; |
87375ab4 JA |
305 | flags = ntohs(s->flags); |
306 | if (!(flags & IP_VS_CONN_F_TEMPLATE)) | |
307 | cp = ip_vs_conn_in_get(s->protocol, | |
308 | s->caddr, s->cport, | |
309 | s->vaddr, s->vport); | |
310 | else | |
311 | cp = ip_vs_ct_in_get(s->protocol, | |
312 | s->caddr, s->cport, | |
313 | s->vaddr, s->vport); | |
1da177e4 LT |
314 | if (!cp) { |
315 | cp = ip_vs_conn_new(s->protocol, | |
316 | s->caddr, s->cport, | |
317 | s->vaddr, s->vport, | |
318 | s->daddr, s->dport, | |
87375ab4 | 319 | flags, NULL); |
1da177e4 LT |
320 | if (!cp) { |
321 | IP_VS_ERR("ip_vs_conn_new failed\n"); | |
322 | return; | |
323 | } | |
324 | cp->state = ntohs(s->state); | |
325 | } else if (!cp->dest) { | |
326 | /* it is an entry created by the synchronization */ | |
327 | cp->state = ntohs(s->state); | |
87375ab4 | 328 | cp->flags = flags | IP_VS_CONN_F_HASHED; |
1da177e4 LT |
329 | } /* Note that we don't touch its state and flags |
330 | if it is a normal entry. */ | |
331 | ||
87375ab4 | 332 | if (flags & IP_VS_CONN_F_SEQ_MASK) { |
1da177e4 LT |
333 | opt = (struct ip_vs_sync_conn_options *)&s[1]; |
334 | memcpy(&cp->in_seq, opt, sizeof(*opt)); | |
335 | p += FULL_CONN_SIZE; | |
336 | } else | |
337 | p += SIMPLE_CONN_SIZE; | |
338 | ||
339 | atomic_set(&cp->in_pkts, sysctl_ip_vs_sync_threshold[0]); | |
340 | cp->timeout = IP_VS_SYNC_CONN_TIMEOUT; | |
341 | ip_vs_conn_put(cp); | |
342 | ||
343 | if (p > buffer+buflen) { | |
344 | IP_VS_ERR("bogus message\n"); | |
345 | return; | |
346 | } | |
347 | } | |
348 | } | |
349 | ||
350 | ||
351 | /* | |
352 | * Setup loopback of outgoing multicasts on a sending socket | |
353 | */ | |
354 | static void set_mcast_loop(struct sock *sk, u_char loop) | |
355 | { | |
356 | struct inet_sock *inet = inet_sk(sk); | |
357 | ||
358 | /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */ | |
359 | lock_sock(sk); | |
360 | inet->mc_loop = loop ? 1 : 0; | |
361 | release_sock(sk); | |
362 | } | |
363 | ||
364 | /* | |
365 | * Specify TTL for outgoing multicasts on a sending socket | |
366 | */ | |
367 | static void set_mcast_ttl(struct sock *sk, u_char ttl) | |
368 | { | |
369 | struct inet_sock *inet = inet_sk(sk); | |
370 | ||
371 | /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */ | |
372 | lock_sock(sk); | |
373 | inet->mc_ttl = ttl; | |
374 | release_sock(sk); | |
375 | } | |
376 | ||
377 | /* | |
378 | * Specifiy default interface for outgoing multicasts | |
379 | */ | |
380 | static int set_mcast_if(struct sock *sk, char *ifname) | |
381 | { | |
382 | struct net_device *dev; | |
383 | struct inet_sock *inet = inet_sk(sk); | |
384 | ||
385 | if ((dev = __dev_get_by_name(ifname)) == NULL) | |
386 | return -ENODEV; | |
387 | ||
388 | if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) | |
389 | return -EINVAL; | |
390 | ||
391 | lock_sock(sk); | |
392 | inet->mc_index = dev->ifindex; | |
393 | /* inet->mc_addr = 0; */ | |
394 | release_sock(sk); | |
395 | ||
396 | return 0; | |
397 | } | |
398 | ||
399 | ||
400 | /* | |
401 | * Set the maximum length of sync message according to the | |
402 | * specified interface's MTU. | |
403 | */ | |
404 | static int set_sync_mesg_maxlen(int sync_state) | |
405 | { | |
406 | struct net_device *dev; | |
407 | int num; | |
408 | ||
409 | if (sync_state == IP_VS_STATE_MASTER) { | |
410 | if ((dev = __dev_get_by_name(ip_vs_master_mcast_ifn)) == NULL) | |
411 | return -ENODEV; | |
412 | ||
413 | num = (dev->mtu - sizeof(struct iphdr) - | |
414 | sizeof(struct udphdr) - | |
415 | SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE; | |
416 | sync_send_mesg_maxlen = | |
417 | SYNC_MESG_HEADER_LEN + SIMPLE_CONN_SIZE * num; | |
418 | IP_VS_DBG(7, "setting the maximum length of sync sending " | |
419 | "message %d.\n", sync_send_mesg_maxlen); | |
420 | } else if (sync_state == IP_VS_STATE_BACKUP) { | |
421 | if ((dev = __dev_get_by_name(ip_vs_backup_mcast_ifn)) == NULL) | |
422 | return -ENODEV; | |
423 | ||
424 | sync_recv_mesg_maxlen = dev->mtu - | |
425 | sizeof(struct iphdr) - sizeof(struct udphdr); | |
426 | IP_VS_DBG(7, "setting the maximum length of sync receiving " | |
427 | "message %d.\n", sync_recv_mesg_maxlen); | |
428 | } | |
429 | ||
430 | return 0; | |
431 | } | |
432 | ||
433 | ||
434 | /* | |
435 | * Join a multicast group. | |
436 | * the group is specified by a class D multicast address 224.0.0.0/8 | |
437 | * in the in_addr structure passed in as a parameter. | |
438 | */ | |
439 | static int | |
440 | join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname) | |
441 | { | |
442 | struct ip_mreqn mreq; | |
443 | struct net_device *dev; | |
444 | int ret; | |
445 | ||
446 | memset(&mreq, 0, sizeof(mreq)); | |
447 | memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr)); | |
448 | ||
449 | if ((dev = __dev_get_by_name(ifname)) == NULL) | |
450 | return -ENODEV; | |
451 | if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) | |
452 | return -EINVAL; | |
453 | ||
454 | mreq.imr_ifindex = dev->ifindex; | |
455 | ||
456 | lock_sock(sk); | |
457 | ret = ip_mc_join_group(sk, &mreq); | |
458 | release_sock(sk); | |
459 | ||
460 | return ret; | |
461 | } | |
462 | ||
463 | ||
464 | static int bind_mcastif_addr(struct socket *sock, char *ifname) | |
465 | { | |
466 | struct net_device *dev; | |
a61ced5d | 467 | __be32 addr; |
1da177e4 LT |
468 | struct sockaddr_in sin; |
469 | ||
470 | if ((dev = __dev_get_by_name(ifname)) == NULL) | |
471 | return -ENODEV; | |
472 | ||
473 | addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE); | |
474 | if (!addr) | |
475 | IP_VS_ERR("You probably need to specify IP address on " | |
476 | "multicast interface.\n"); | |
477 | ||
478 | IP_VS_DBG(7, "binding socket with (%s) %u.%u.%u.%u\n", | |
479 | ifname, NIPQUAD(addr)); | |
480 | ||
481 | /* Now bind the socket with the address of multicast interface */ | |
482 | sin.sin_family = AF_INET; | |
483 | sin.sin_addr.s_addr = addr; | |
484 | sin.sin_port = 0; | |
485 | ||
486 | return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin)); | |
487 | } | |
488 | ||
489 | /* | |
490 | * Set up sending multicast socket over UDP | |
491 | */ | |
492 | static struct socket * make_send_sock(void) | |
493 | { | |
494 | struct socket *sock; | |
495 | ||
496 | /* First create a socket */ | |
497 | if (sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock) < 0) { | |
498 | IP_VS_ERR("Error during creation of socket; terminating\n"); | |
499 | return NULL; | |
500 | } | |
501 | ||
502 | if (set_mcast_if(sock->sk, ip_vs_master_mcast_ifn) < 0) { | |
503 | IP_VS_ERR("Error setting outbound mcast interface\n"); | |
504 | goto error; | |
505 | } | |
506 | ||
507 | set_mcast_loop(sock->sk, 0); | |
508 | set_mcast_ttl(sock->sk, 1); | |
509 | ||
510 | if (bind_mcastif_addr(sock, ip_vs_master_mcast_ifn) < 0) { | |
511 | IP_VS_ERR("Error binding address of the mcast interface\n"); | |
512 | goto error; | |
513 | } | |
514 | ||
515 | if (sock->ops->connect(sock, | |
516 | (struct sockaddr*)&mcast_addr, | |
517 | sizeof(struct sockaddr), 0) < 0) { | |
518 | IP_VS_ERR("Error connecting to the multicast addr\n"); | |
519 | goto error; | |
520 | } | |
521 | ||
522 | return sock; | |
523 | ||
524 | error: | |
525 | sock_release(sock); | |
526 | return NULL; | |
527 | } | |
528 | ||
529 | ||
530 | /* | |
531 | * Set up receiving multicast socket over UDP | |
532 | */ | |
533 | static struct socket * make_receive_sock(void) | |
534 | { | |
535 | struct socket *sock; | |
536 | ||
537 | /* First create a socket */ | |
538 | if (sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock) < 0) { | |
539 | IP_VS_ERR("Error during creation of socket; terminating\n"); | |
540 | return NULL; | |
541 | } | |
542 | ||
543 | /* it is equivalent to the REUSEADDR option in user-space */ | |
544 | sock->sk->sk_reuse = 1; | |
545 | ||
546 | if (sock->ops->bind(sock, | |
547 | (struct sockaddr*)&mcast_addr, | |
548 | sizeof(struct sockaddr)) < 0) { | |
549 | IP_VS_ERR("Error binding to the multicast addr\n"); | |
550 | goto error; | |
551 | } | |
552 | ||
553 | /* join the multicast group */ | |
554 | if (join_mcast_group(sock->sk, | |
555 | (struct in_addr*)&mcast_addr.sin_addr, | |
556 | ip_vs_backup_mcast_ifn) < 0) { | |
557 | IP_VS_ERR("Error joining to the multicast group\n"); | |
558 | goto error; | |
559 | } | |
560 | ||
561 | return sock; | |
562 | ||
563 | error: | |
564 | sock_release(sock); | |
565 | return NULL; | |
566 | } | |
567 | ||
568 | ||
569 | static int | |
570 | ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length) | |
571 | { | |
572 | struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL}; | |
573 | struct kvec iov; | |
574 | int len; | |
575 | ||
576 | EnterFunction(7); | |
577 | iov.iov_base = (void *)buffer; | |
578 | iov.iov_len = length; | |
579 | ||
580 | len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length)); | |
581 | ||
582 | LeaveFunction(7); | |
583 | return len; | |
584 | } | |
585 | ||
586 | static void | |
587 | ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg) | |
588 | { | |
589 | int msize; | |
590 | ||
591 | msize = msg->size; | |
592 | ||
593 | /* Put size in network byte order */ | |
594 | msg->size = htons(msg->size); | |
595 | ||
596 | if (ip_vs_send_async(sock, (char *)msg, msize) != msize) | |
597 | IP_VS_ERR("ip_vs_send_async error\n"); | |
598 | } | |
599 | ||
600 | static int | |
601 | ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen) | |
602 | { | |
603 | struct msghdr msg = {NULL,}; | |
604 | struct kvec iov; | |
605 | int len; | |
606 | ||
607 | EnterFunction(7); | |
608 | ||
609 | /* Receive a packet */ | |
610 | iov.iov_base = buffer; | |
611 | iov.iov_len = (size_t)buflen; | |
612 | ||
613 | len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, 0); | |
614 | ||
615 | if (len < 0) | |
616 | return -1; | |
617 | ||
618 | LeaveFunction(7); | |
619 | return len; | |
620 | } | |
621 | ||
622 | ||
623 | static DECLARE_WAIT_QUEUE_HEAD(sync_wait); | |
624 | static pid_t sync_master_pid = 0; | |
625 | static pid_t sync_backup_pid = 0; | |
626 | ||
627 | static DECLARE_WAIT_QUEUE_HEAD(stop_sync_wait); | |
628 | static int stop_master_sync = 0; | |
629 | static int stop_backup_sync = 0; | |
630 | ||
631 | static void sync_master_loop(void) | |
632 | { | |
633 | struct socket *sock; | |
634 | struct ip_vs_sync_buff *sb; | |
635 | ||
636 | /* create the sending multicast socket */ | |
637 | sock = make_send_sock(); | |
638 | if (!sock) | |
639 | return; | |
640 | ||
641 | IP_VS_INFO("sync thread started: state = MASTER, mcast_ifn = %s, " | |
642 | "syncid = %d\n", | |
643 | ip_vs_master_mcast_ifn, ip_vs_master_syncid); | |
644 | ||
645 | for (;;) { | |
646 | while ((sb=sb_dequeue())) { | |
647 | ip_vs_send_sync_msg(sock, sb->mesg); | |
648 | ip_vs_sync_buff_release(sb); | |
649 | } | |
650 | ||
651 | /* check if entries stay in curr_sb for 2 seconds */ | |
652 | if ((sb = get_curr_sync_buff(2*HZ))) { | |
653 | ip_vs_send_sync_msg(sock, sb->mesg); | |
654 | ip_vs_sync_buff_release(sb); | |
655 | } | |
656 | ||
657 | if (stop_master_sync) | |
658 | break; | |
659 | ||
660 | ssleep(1); | |
661 | } | |
662 | ||
663 | /* clean up the sync_buff queue */ | |
664 | while ((sb=sb_dequeue())) { | |
665 | ip_vs_sync_buff_release(sb); | |
666 | } | |
667 | ||
668 | /* clean up the current sync_buff */ | |
669 | if ((sb = get_curr_sync_buff(0))) { | |
670 | ip_vs_sync_buff_release(sb); | |
671 | } | |
672 | ||
673 | /* release the sending multicast socket */ | |
674 | sock_release(sock); | |
675 | } | |
676 | ||
677 | ||
678 | static void sync_backup_loop(void) | |
679 | { | |
680 | struct socket *sock; | |
681 | char *buf; | |
682 | int len; | |
683 | ||
684 | if (!(buf = kmalloc(sync_recv_mesg_maxlen, GFP_ATOMIC))) { | |
685 | IP_VS_ERR("sync_backup_loop: kmalloc error\n"); | |
686 | return; | |
687 | } | |
688 | ||
689 | /* create the receiving multicast socket */ | |
690 | sock = make_receive_sock(); | |
691 | if (!sock) | |
692 | goto out; | |
693 | ||
694 | IP_VS_INFO("sync thread started: state = BACKUP, mcast_ifn = %s, " | |
695 | "syncid = %d\n", | |
696 | ip_vs_backup_mcast_ifn, ip_vs_backup_syncid); | |
697 | ||
698 | for (;;) { | |
699 | /* do you have data now? */ | |
700 | while (!skb_queue_empty(&(sock->sk->sk_receive_queue))) { | |
701 | if ((len = | |
702 | ip_vs_receive(sock, buf, | |
703 | sync_recv_mesg_maxlen)) <= 0) { | |
704 | IP_VS_ERR("receiving message error\n"); | |
705 | break; | |
706 | } | |
707 | /* disable bottom half, because it accessed the data | |
708 | shared by softirq while getting/creating conns */ | |
709 | local_bh_disable(); | |
710 | ip_vs_process_message(buf, len); | |
711 | local_bh_enable(); | |
712 | } | |
713 | ||
714 | if (stop_backup_sync) | |
715 | break; | |
716 | ||
717 | ssleep(1); | |
718 | } | |
719 | ||
720 | /* release the sending multicast socket */ | |
721 | sock_release(sock); | |
722 | ||
723 | out: | |
724 | kfree(buf); | |
725 | } | |
726 | ||
727 | ||
728 | static void set_sync_pid(int sync_state, pid_t sync_pid) | |
729 | { | |
730 | if (sync_state == IP_VS_STATE_MASTER) | |
731 | sync_master_pid = sync_pid; | |
732 | else if (sync_state == IP_VS_STATE_BACKUP) | |
733 | sync_backup_pid = sync_pid; | |
734 | } | |
735 | ||
736 | static void set_stop_sync(int sync_state, int set) | |
737 | { | |
738 | if (sync_state == IP_VS_STATE_MASTER) | |
739 | stop_master_sync = set; | |
740 | else if (sync_state == IP_VS_STATE_BACKUP) | |
741 | stop_backup_sync = set; | |
742 | else { | |
743 | stop_master_sync = set; | |
744 | stop_backup_sync = set; | |
745 | } | |
746 | } | |
747 | ||
748 | static int sync_thread(void *startup) | |
749 | { | |
750 | DECLARE_WAITQUEUE(wait, current); | |
751 | mm_segment_t oldmm; | |
752 | int state; | |
753 | const char *name; | |
754 | ||
755 | /* increase the module use count */ | |
756 | ip_vs_use_count_inc(); | |
757 | ||
758 | if (ip_vs_sync_state & IP_VS_STATE_MASTER && !sync_master_pid) { | |
759 | state = IP_VS_STATE_MASTER; | |
760 | name = "ipvs_syncmaster"; | |
761 | } else if (ip_vs_sync_state & IP_VS_STATE_BACKUP && !sync_backup_pid) { | |
762 | state = IP_VS_STATE_BACKUP; | |
763 | name = "ipvs_syncbackup"; | |
764 | } else { | |
765 | IP_VS_BUG(); | |
766 | ip_vs_use_count_dec(); | |
767 | return -EINVAL; | |
768 | } | |
769 | ||
770 | daemonize(name); | |
771 | ||
772 | oldmm = get_fs(); | |
773 | set_fs(KERNEL_DS); | |
774 | ||
775 | /* Block all signals */ | |
776 | spin_lock_irq(¤t->sighand->siglock); | |
777 | siginitsetinv(¤t->blocked, 0); | |
778 | recalc_sigpending(); | |
779 | spin_unlock_irq(¤t->sighand->siglock); | |
780 | ||
781 | /* set the maximum length of sync message */ | |
782 | set_sync_mesg_maxlen(state); | |
783 | ||
784 | /* set up multicast address */ | |
785 | mcast_addr.sin_family = AF_INET; | |
786 | mcast_addr.sin_port = htons(IP_VS_SYNC_PORT); | |
787 | mcast_addr.sin_addr.s_addr = htonl(IP_VS_SYNC_GROUP); | |
788 | ||
789 | add_wait_queue(&sync_wait, &wait); | |
790 | ||
791 | set_sync_pid(state, current->pid); | |
792 | complete((struct completion *)startup); | |
793 | ||
794 | /* processing master/backup loop here */ | |
795 | if (state == IP_VS_STATE_MASTER) | |
796 | sync_master_loop(); | |
797 | else if (state == IP_VS_STATE_BACKUP) | |
798 | sync_backup_loop(); | |
799 | else IP_VS_BUG(); | |
800 | ||
801 | remove_wait_queue(&sync_wait, &wait); | |
802 | ||
803 | /* thread exits */ | |
804 | set_sync_pid(state, 0); | |
805 | IP_VS_INFO("sync thread stopped!\n"); | |
806 | ||
807 | set_fs(oldmm); | |
808 | ||
809 | /* decrease the module use count */ | |
810 | ip_vs_use_count_dec(); | |
811 | ||
812 | set_stop_sync(state, 0); | |
813 | wake_up(&stop_sync_wait); | |
814 | ||
815 | return 0; | |
816 | } | |
817 | ||
818 | ||
819 | static int fork_sync_thread(void *startup) | |
820 | { | |
821 | pid_t pid; | |
822 | ||
823 | /* fork the sync thread here, then the parent process of the | |
824 | sync thread is the init process after this thread exits. */ | |
825 | repeat: | |
826 | if ((pid = kernel_thread(sync_thread, startup, 0)) < 0) { | |
827 | IP_VS_ERR("could not create sync_thread due to %d... " | |
828 | "retrying.\n", pid); | |
829 | ssleep(1); | |
830 | goto repeat; | |
831 | } | |
832 | ||
833 | return 0; | |
834 | } | |
835 | ||
836 | ||
837 | int start_sync_thread(int state, char *mcast_ifn, __u8 syncid) | |
838 | { | |
839 | DECLARE_COMPLETION(startup); | |
840 | pid_t pid; | |
841 | ||
842 | if ((state == IP_VS_STATE_MASTER && sync_master_pid) || | |
843 | (state == IP_VS_STATE_BACKUP && sync_backup_pid)) | |
844 | return -EEXIST; | |
845 | ||
846 | IP_VS_DBG(7, "%s: pid %d\n", __FUNCTION__, current->pid); | |
847 | IP_VS_DBG(7, "Each ip_vs_sync_conn entry need %Zd bytes\n", | |
848 | sizeof(struct ip_vs_sync_conn)); | |
849 | ||
850 | ip_vs_sync_state |= state; | |
851 | if (state == IP_VS_STATE_MASTER) { | |
4da62fc7 | 852 | strlcpy(ip_vs_master_mcast_ifn, mcast_ifn, sizeof(ip_vs_master_mcast_ifn)); |
1da177e4 LT |
853 | ip_vs_master_syncid = syncid; |
854 | } else { | |
4da62fc7 | 855 | strlcpy(ip_vs_backup_mcast_ifn, mcast_ifn, sizeof(ip_vs_backup_mcast_ifn)); |
1da177e4 LT |
856 | ip_vs_backup_syncid = syncid; |
857 | } | |
858 | ||
859 | repeat: | |
860 | if ((pid = kernel_thread(fork_sync_thread, &startup, 0)) < 0) { | |
861 | IP_VS_ERR("could not create fork_sync_thread due to %d... " | |
862 | "retrying.\n", pid); | |
863 | ssleep(1); | |
864 | goto repeat; | |
865 | } | |
866 | ||
867 | wait_for_completion(&startup); | |
868 | ||
869 | return 0; | |
870 | } | |
871 | ||
872 | ||
873 | int stop_sync_thread(int state) | |
874 | { | |
875 | DECLARE_WAITQUEUE(wait, current); | |
876 | ||
877 | if ((state == IP_VS_STATE_MASTER && !sync_master_pid) || | |
878 | (state == IP_VS_STATE_BACKUP && !sync_backup_pid)) | |
879 | return -ESRCH; | |
880 | ||
881 | IP_VS_DBG(7, "%s: pid %d\n", __FUNCTION__, current->pid); | |
882 | IP_VS_INFO("stopping sync thread %d ...\n", | |
883 | (state == IP_VS_STATE_MASTER) ? sync_master_pid : sync_backup_pid); | |
884 | ||
885 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
886 | add_wait_queue(&stop_sync_wait, &wait); | |
887 | set_stop_sync(state, 1); | |
888 | ip_vs_sync_state -= state; | |
889 | wake_up(&sync_wait); | |
890 | schedule(); | |
891 | __set_current_state(TASK_RUNNING); | |
892 | remove_wait_queue(&stop_sync_wait, &wait); | |
893 | ||
894 | /* Note: no need to reap the sync thread, because its parent | |
895 | process is the init process */ | |
896 | ||
897 | if ((state == IP_VS_STATE_MASTER && stop_master_sync) || | |
898 | (state == IP_VS_STATE_BACKUP && stop_backup_sync)) | |
899 | IP_VS_BUG(); | |
900 | ||
901 | return 0; | |
902 | } |