]>
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> | |
24 | #include <linux/net.h> | |
25 | #include <linux/completion.h> | |
26 | #include <linux/delay.h> | |
27 | #include <linux/skbuff.h> | |
28 | #include <linux/in.h> | |
29 | #include <linux/igmp.h> /* for ip_mc_join_group */ | |
30 | ||
31 | #include <net/ip.h> | |
32 | #include <net/sock.h> | |
33 | #include <asm/uaccess.h> /* for get_fs and set_fs */ | |
34 | ||
35 | #include <net/ip_vs.h> | |
36 | ||
37 | #define IP_VS_SYNC_GROUP 0xe0000051 /* multicast addr - 224.0.0.81 */ | |
38 | #define IP_VS_SYNC_PORT 8848 /* multicast port */ | |
39 | ||
40 | ||
41 | /* | |
42 | * IPVS sync connection entry | |
43 | */ | |
44 | struct ip_vs_sync_conn { | |
45 | __u8 reserved; | |
46 | ||
47 | /* Protocol, addresses and port numbers */ | |
48 | __u8 protocol; /* Which protocol (TCP/UDP) */ | |
49 | __u16 cport; | |
50 | __u16 vport; | |
51 | __u16 dport; | |
52 | __u32 caddr; /* client address */ | |
53 | __u32 vaddr; /* virtual address */ | |
54 | __u32 daddr; /* destination address */ | |
55 | ||
56 | /* Flags and state transition */ | |
57 | __u16 flags; /* status flags */ | |
58 | __u16 state; /* state info */ | |
59 | ||
60 | /* The sequence options start here */ | |
61 | }; | |
62 | ||
63 | struct ip_vs_sync_conn_options { | |
64 | struct ip_vs_seq in_seq; /* incoming seq. struct */ | |
65 | struct ip_vs_seq out_seq; /* outgoing seq. struct */ | |
66 | }; | |
67 | ||
68 | #define IP_VS_SYNC_CONN_TIMEOUT (3*60*HZ) | |
69 | #define SIMPLE_CONN_SIZE (sizeof(struct ip_vs_sync_conn)) | |
70 | #define FULL_CONN_SIZE \ | |
71 | (sizeof(struct ip_vs_sync_conn) + sizeof(struct ip_vs_sync_conn_options)) | |
72 | ||
73 | ||
74 | /* | |
75 | The master mulitcasts messages to the backup load balancers in the | |
76 | following format. | |
77 | ||
78 | 0 1 2 3 | |
79 | 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 | |
80 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
81 | | Count Conns | SyncID | Size | | |
82 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
83 | | | | |
84 | | IPVS Sync Connection (1) | | |
85 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
86 | | . | | |
87 | | . | | |
88 | | . | | |
89 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
90 | | | | |
91 | | IPVS Sync Connection (n) | | |
92 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
93 | */ | |
94 | ||
95 | #define SYNC_MESG_HEADER_LEN 4 | |
96 | ||
97 | struct ip_vs_sync_mesg { | |
98 | __u8 nr_conns; | |
99 | __u8 syncid; | |
100 | __u16 size; | |
101 | ||
102 | /* ip_vs_sync_conn entries start here */ | |
103 | }; | |
104 | ||
105 | /* the maximum length of sync (sending/receiving) message */ | |
106 | static int sync_send_mesg_maxlen; | |
107 | static int sync_recv_mesg_maxlen; | |
108 | ||
109 | struct ip_vs_sync_buff { | |
110 | struct list_head list; | |
111 | unsigned long firstuse; | |
112 | ||
113 | /* pointers for the message data */ | |
114 | struct ip_vs_sync_mesg *mesg; | |
115 | unsigned char *head; | |
116 | unsigned char *end; | |
117 | }; | |
118 | ||
119 | ||
120 | /* the sync_buff list head and the lock */ | |
121 | static LIST_HEAD(ip_vs_sync_queue); | |
122 | static DEFINE_SPINLOCK(ip_vs_sync_lock); | |
123 | ||
124 | /* current sync_buff for accepting new conn entries */ | |
125 | static struct ip_vs_sync_buff *curr_sb = NULL; | |
126 | static DEFINE_SPINLOCK(curr_sb_lock); | |
127 | ||
128 | /* ipvs sync daemon state */ | |
129 | volatile int ip_vs_sync_state = IP_VS_STATE_NONE; | |
130 | volatile int ip_vs_master_syncid = 0; | |
131 | volatile int ip_vs_backup_syncid = 0; | |
132 | ||
133 | /* multicast interface name */ | |
134 | char ip_vs_master_mcast_ifn[IP_VS_IFNAME_MAXLEN]; | |
135 | char ip_vs_backup_mcast_ifn[IP_VS_IFNAME_MAXLEN]; | |
136 | ||
137 | /* multicast addr */ | |
138 | static struct sockaddr_in mcast_addr; | |
139 | ||
140 | ||
141 | static inline void sb_queue_tail(struct ip_vs_sync_buff *sb) | |
142 | { | |
143 | spin_lock(&ip_vs_sync_lock); | |
144 | list_add_tail(&sb->list, &ip_vs_sync_queue); | |
145 | spin_unlock(&ip_vs_sync_lock); | |
146 | } | |
147 | ||
148 | static inline struct ip_vs_sync_buff * sb_dequeue(void) | |
149 | { | |
150 | struct ip_vs_sync_buff *sb; | |
151 | ||
152 | spin_lock_bh(&ip_vs_sync_lock); | |
153 | if (list_empty(&ip_vs_sync_queue)) { | |
154 | sb = NULL; | |
155 | } else { | |
156 | sb = list_entry(ip_vs_sync_queue.next, | |
157 | struct ip_vs_sync_buff, | |
158 | list); | |
159 | list_del(&sb->list); | |
160 | } | |
161 | spin_unlock_bh(&ip_vs_sync_lock); | |
162 | ||
163 | return sb; | |
164 | } | |
165 | ||
166 | static inline struct ip_vs_sync_buff * ip_vs_sync_buff_create(void) | |
167 | { | |
168 | struct ip_vs_sync_buff *sb; | |
169 | ||
170 | if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC))) | |
171 | return NULL; | |
172 | ||
173 | if (!(sb->mesg=kmalloc(sync_send_mesg_maxlen, GFP_ATOMIC))) { | |
174 | kfree(sb); | |
175 | return NULL; | |
176 | } | |
177 | sb->mesg->nr_conns = 0; | |
178 | sb->mesg->syncid = ip_vs_master_syncid; | |
179 | sb->mesg->size = 4; | |
180 | sb->head = (unsigned char *)sb->mesg + 4; | |
181 | sb->end = (unsigned char *)sb->mesg + sync_send_mesg_maxlen; | |
182 | sb->firstuse = jiffies; | |
183 | return sb; | |
184 | } | |
185 | ||
186 | static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb) | |
187 | { | |
188 | kfree(sb->mesg); | |
189 | kfree(sb); | |
190 | } | |
191 | ||
192 | /* | |
193 | * Get the current sync buffer if it has been created for more | |
194 | * than the specified time or the specified time is zero. | |
195 | */ | |
196 | static inline struct ip_vs_sync_buff * | |
197 | get_curr_sync_buff(unsigned long time) | |
198 | { | |
199 | struct ip_vs_sync_buff *sb; | |
200 | ||
201 | spin_lock_bh(&curr_sb_lock); | |
202 | if (curr_sb && (time == 0 || | |
203 | time_before(jiffies - curr_sb->firstuse, time))) { | |
204 | sb = curr_sb; | |
205 | curr_sb = NULL; | |
206 | } else | |
207 | sb = NULL; | |
208 | spin_unlock_bh(&curr_sb_lock); | |
209 | return sb; | |
210 | } | |
211 | ||
212 | ||
213 | /* | |
214 | * Add an ip_vs_conn information into the current sync_buff. | |
215 | * Called by ip_vs_in. | |
216 | */ | |
217 | void ip_vs_sync_conn(struct ip_vs_conn *cp) | |
218 | { | |
219 | struct ip_vs_sync_mesg *m; | |
220 | struct ip_vs_sync_conn *s; | |
221 | int len; | |
222 | ||
223 | spin_lock(&curr_sb_lock); | |
224 | if (!curr_sb) { | |
225 | if (!(curr_sb=ip_vs_sync_buff_create())) { | |
226 | spin_unlock(&curr_sb_lock); | |
227 | IP_VS_ERR("ip_vs_sync_buff_create failed.\n"); | |
228 | return; | |
229 | } | |
230 | } | |
231 | ||
232 | len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE : | |
233 | SIMPLE_CONN_SIZE; | |
234 | m = curr_sb->mesg; | |
235 | s = (struct ip_vs_sync_conn *)curr_sb->head; | |
236 | ||
237 | /* copy members */ | |
238 | s->protocol = cp->protocol; | |
239 | s->cport = cp->cport; | |
240 | s->vport = cp->vport; | |
241 | s->dport = cp->dport; | |
242 | s->caddr = cp->caddr; | |
243 | s->vaddr = cp->vaddr; | |
244 | s->daddr = cp->daddr; | |
245 | s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED); | |
246 | s->state = htons(cp->state); | |
247 | if (cp->flags & IP_VS_CONN_F_SEQ_MASK) { | |
248 | struct ip_vs_sync_conn_options *opt = | |
249 | (struct ip_vs_sync_conn_options *)&s[1]; | |
250 | memcpy(opt, &cp->in_seq, sizeof(*opt)); | |
251 | } | |
252 | ||
253 | m->nr_conns++; | |
254 | m->size += len; | |
255 | curr_sb->head += len; | |
256 | ||
257 | /* check if there is a space for next one */ | |
258 | if (curr_sb->head+FULL_CONN_SIZE > curr_sb->end) { | |
259 | sb_queue_tail(curr_sb); | |
260 | curr_sb = NULL; | |
261 | } | |
262 | spin_unlock(&curr_sb_lock); | |
263 | ||
264 | /* synchronize its controller if it has */ | |
265 | if (cp->control) | |
266 | ip_vs_sync_conn(cp->control); | |
267 | } | |
268 | ||
269 | ||
270 | /* | |
271 | * Process received multicast message and create the corresponding | |
272 | * ip_vs_conn entries. | |
273 | */ | |
274 | static void ip_vs_process_message(const char *buffer, const size_t buflen) | |
275 | { | |
276 | struct ip_vs_sync_mesg *m = (struct ip_vs_sync_mesg *)buffer; | |
277 | struct ip_vs_sync_conn *s; | |
278 | struct ip_vs_sync_conn_options *opt; | |
279 | struct ip_vs_conn *cp; | |
280 | char *p; | |
281 | int i; | |
282 | ||
283 | /* Convert size back to host byte order */ | |
284 | m->size = ntohs(m->size); | |
285 | ||
286 | if (buflen != m->size) { | |
287 | IP_VS_ERR("bogus message\n"); | |
288 | return; | |
289 | } | |
290 | ||
291 | /* SyncID sanity check */ | |
292 | if (ip_vs_backup_syncid != 0 && m->syncid != ip_vs_backup_syncid) { | |
293 | IP_VS_DBG(7, "Ignoring incoming msg with syncid = %d\n", | |
294 | m->syncid); | |
295 | return; | |
296 | } | |
297 | ||
298 | p = (char *)buffer + sizeof(struct ip_vs_sync_mesg); | |
299 | for (i=0; i<m->nr_conns; i++) { | |
300 | s = (struct ip_vs_sync_conn *)p; | |
301 | cp = ip_vs_conn_in_get(s->protocol, | |
302 | s->caddr, s->cport, | |
303 | s->vaddr, s->vport); | |
304 | if (!cp) { | |
305 | cp = ip_vs_conn_new(s->protocol, | |
306 | s->caddr, s->cport, | |
307 | s->vaddr, s->vport, | |
308 | s->daddr, s->dport, | |
309 | ntohs(s->flags), NULL); | |
310 | if (!cp) { | |
311 | IP_VS_ERR("ip_vs_conn_new failed\n"); | |
312 | return; | |
313 | } | |
314 | cp->state = ntohs(s->state); | |
315 | } else if (!cp->dest) { | |
316 | /* it is an entry created by the synchronization */ | |
317 | cp->state = ntohs(s->state); | |
318 | cp->flags = ntohs(s->flags) | IP_VS_CONN_F_HASHED; | |
319 | } /* Note that we don't touch its state and flags | |
320 | if it is a normal entry. */ | |
321 | ||
322 | if (ntohs(s->flags) & IP_VS_CONN_F_SEQ_MASK) { | |
323 | opt = (struct ip_vs_sync_conn_options *)&s[1]; | |
324 | memcpy(&cp->in_seq, opt, sizeof(*opt)); | |
325 | p += FULL_CONN_SIZE; | |
326 | } else | |
327 | p += SIMPLE_CONN_SIZE; | |
328 | ||
329 | atomic_set(&cp->in_pkts, sysctl_ip_vs_sync_threshold[0]); | |
330 | cp->timeout = IP_VS_SYNC_CONN_TIMEOUT; | |
331 | ip_vs_conn_put(cp); | |
332 | ||
333 | if (p > buffer+buflen) { | |
334 | IP_VS_ERR("bogus message\n"); | |
335 | return; | |
336 | } | |
337 | } | |
338 | } | |
339 | ||
340 | ||
341 | /* | |
342 | * Setup loopback of outgoing multicasts on a sending socket | |
343 | */ | |
344 | static void set_mcast_loop(struct sock *sk, u_char loop) | |
345 | { | |
346 | struct inet_sock *inet = inet_sk(sk); | |
347 | ||
348 | /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */ | |
349 | lock_sock(sk); | |
350 | inet->mc_loop = loop ? 1 : 0; | |
351 | release_sock(sk); | |
352 | } | |
353 | ||
354 | /* | |
355 | * Specify TTL for outgoing multicasts on a sending socket | |
356 | */ | |
357 | static void set_mcast_ttl(struct sock *sk, u_char ttl) | |
358 | { | |
359 | struct inet_sock *inet = inet_sk(sk); | |
360 | ||
361 | /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */ | |
362 | lock_sock(sk); | |
363 | inet->mc_ttl = ttl; | |
364 | release_sock(sk); | |
365 | } | |
366 | ||
367 | /* | |
368 | * Specifiy default interface for outgoing multicasts | |
369 | */ | |
370 | static int set_mcast_if(struct sock *sk, char *ifname) | |
371 | { | |
372 | struct net_device *dev; | |
373 | struct inet_sock *inet = inet_sk(sk); | |
374 | ||
375 | if ((dev = __dev_get_by_name(ifname)) == NULL) | |
376 | return -ENODEV; | |
377 | ||
378 | if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) | |
379 | return -EINVAL; | |
380 | ||
381 | lock_sock(sk); | |
382 | inet->mc_index = dev->ifindex; | |
383 | /* inet->mc_addr = 0; */ | |
384 | release_sock(sk); | |
385 | ||
386 | return 0; | |
387 | } | |
388 | ||
389 | ||
390 | /* | |
391 | * Set the maximum length of sync message according to the | |
392 | * specified interface's MTU. | |
393 | */ | |
394 | static int set_sync_mesg_maxlen(int sync_state) | |
395 | { | |
396 | struct net_device *dev; | |
397 | int num; | |
398 | ||
399 | if (sync_state == IP_VS_STATE_MASTER) { | |
400 | if ((dev = __dev_get_by_name(ip_vs_master_mcast_ifn)) == NULL) | |
401 | return -ENODEV; | |
402 | ||
403 | num = (dev->mtu - sizeof(struct iphdr) - | |
404 | sizeof(struct udphdr) - | |
405 | SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE; | |
406 | sync_send_mesg_maxlen = | |
407 | SYNC_MESG_HEADER_LEN + SIMPLE_CONN_SIZE * num; | |
408 | IP_VS_DBG(7, "setting the maximum length of sync sending " | |
409 | "message %d.\n", sync_send_mesg_maxlen); | |
410 | } else if (sync_state == IP_VS_STATE_BACKUP) { | |
411 | if ((dev = __dev_get_by_name(ip_vs_backup_mcast_ifn)) == NULL) | |
412 | return -ENODEV; | |
413 | ||
414 | sync_recv_mesg_maxlen = dev->mtu - | |
415 | sizeof(struct iphdr) - sizeof(struct udphdr); | |
416 | IP_VS_DBG(7, "setting the maximum length of sync receiving " | |
417 | "message %d.\n", sync_recv_mesg_maxlen); | |
418 | } | |
419 | ||
420 | return 0; | |
421 | } | |
422 | ||
423 | ||
424 | /* | |
425 | * Join a multicast group. | |
426 | * the group is specified by a class D multicast address 224.0.0.0/8 | |
427 | * in the in_addr structure passed in as a parameter. | |
428 | */ | |
429 | static int | |
430 | join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname) | |
431 | { | |
432 | struct ip_mreqn mreq; | |
433 | struct net_device *dev; | |
434 | int ret; | |
435 | ||
436 | memset(&mreq, 0, sizeof(mreq)); | |
437 | memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr)); | |
438 | ||
439 | if ((dev = __dev_get_by_name(ifname)) == NULL) | |
440 | return -ENODEV; | |
441 | if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) | |
442 | return -EINVAL; | |
443 | ||
444 | mreq.imr_ifindex = dev->ifindex; | |
445 | ||
446 | lock_sock(sk); | |
447 | ret = ip_mc_join_group(sk, &mreq); | |
448 | release_sock(sk); | |
449 | ||
450 | return ret; | |
451 | } | |
452 | ||
453 | ||
454 | static int bind_mcastif_addr(struct socket *sock, char *ifname) | |
455 | { | |
456 | struct net_device *dev; | |
457 | u32 addr; | |
458 | struct sockaddr_in sin; | |
459 | ||
460 | if ((dev = __dev_get_by_name(ifname)) == NULL) | |
461 | return -ENODEV; | |
462 | ||
463 | addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE); | |
464 | if (!addr) | |
465 | IP_VS_ERR("You probably need to specify IP address on " | |
466 | "multicast interface.\n"); | |
467 | ||
468 | IP_VS_DBG(7, "binding socket with (%s) %u.%u.%u.%u\n", | |
469 | ifname, NIPQUAD(addr)); | |
470 | ||
471 | /* Now bind the socket with the address of multicast interface */ | |
472 | sin.sin_family = AF_INET; | |
473 | sin.sin_addr.s_addr = addr; | |
474 | sin.sin_port = 0; | |
475 | ||
476 | return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin)); | |
477 | } | |
478 | ||
479 | /* | |
480 | * Set up sending multicast socket over UDP | |
481 | */ | |
482 | static struct socket * make_send_sock(void) | |
483 | { | |
484 | struct socket *sock; | |
485 | ||
486 | /* First create a socket */ | |
487 | if (sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock) < 0) { | |
488 | IP_VS_ERR("Error during creation of socket; terminating\n"); | |
489 | return NULL; | |
490 | } | |
491 | ||
492 | if (set_mcast_if(sock->sk, ip_vs_master_mcast_ifn) < 0) { | |
493 | IP_VS_ERR("Error setting outbound mcast interface\n"); | |
494 | goto error; | |
495 | } | |
496 | ||
497 | set_mcast_loop(sock->sk, 0); | |
498 | set_mcast_ttl(sock->sk, 1); | |
499 | ||
500 | if (bind_mcastif_addr(sock, ip_vs_master_mcast_ifn) < 0) { | |
501 | IP_VS_ERR("Error binding address of the mcast interface\n"); | |
502 | goto error; | |
503 | } | |
504 | ||
505 | if (sock->ops->connect(sock, | |
506 | (struct sockaddr*)&mcast_addr, | |
507 | sizeof(struct sockaddr), 0) < 0) { | |
508 | IP_VS_ERR("Error connecting to the multicast addr\n"); | |
509 | goto error; | |
510 | } | |
511 | ||
512 | return sock; | |
513 | ||
514 | error: | |
515 | sock_release(sock); | |
516 | return NULL; | |
517 | } | |
518 | ||
519 | ||
520 | /* | |
521 | * Set up receiving multicast socket over UDP | |
522 | */ | |
523 | static struct socket * make_receive_sock(void) | |
524 | { | |
525 | struct socket *sock; | |
526 | ||
527 | /* First create a socket */ | |
528 | if (sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock) < 0) { | |
529 | IP_VS_ERR("Error during creation of socket; terminating\n"); | |
530 | return NULL; | |
531 | } | |
532 | ||
533 | /* it is equivalent to the REUSEADDR option in user-space */ | |
534 | sock->sk->sk_reuse = 1; | |
535 | ||
536 | if (sock->ops->bind(sock, | |
537 | (struct sockaddr*)&mcast_addr, | |
538 | sizeof(struct sockaddr)) < 0) { | |
539 | IP_VS_ERR("Error binding to the multicast addr\n"); | |
540 | goto error; | |
541 | } | |
542 | ||
543 | /* join the multicast group */ | |
544 | if (join_mcast_group(sock->sk, | |
545 | (struct in_addr*)&mcast_addr.sin_addr, | |
546 | ip_vs_backup_mcast_ifn) < 0) { | |
547 | IP_VS_ERR("Error joining to the multicast group\n"); | |
548 | goto error; | |
549 | } | |
550 | ||
551 | return sock; | |
552 | ||
553 | error: | |
554 | sock_release(sock); | |
555 | return NULL; | |
556 | } | |
557 | ||
558 | ||
559 | static int | |
560 | ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length) | |
561 | { | |
562 | struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL}; | |
563 | struct kvec iov; | |
564 | int len; | |
565 | ||
566 | EnterFunction(7); | |
567 | iov.iov_base = (void *)buffer; | |
568 | iov.iov_len = length; | |
569 | ||
570 | len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length)); | |
571 | ||
572 | LeaveFunction(7); | |
573 | return len; | |
574 | } | |
575 | ||
576 | static void | |
577 | ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg) | |
578 | { | |
579 | int msize; | |
580 | ||
581 | msize = msg->size; | |
582 | ||
583 | /* Put size in network byte order */ | |
584 | msg->size = htons(msg->size); | |
585 | ||
586 | if (ip_vs_send_async(sock, (char *)msg, msize) != msize) | |
587 | IP_VS_ERR("ip_vs_send_async error\n"); | |
588 | } | |
589 | ||
590 | static int | |
591 | ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen) | |
592 | { | |
593 | struct msghdr msg = {NULL,}; | |
594 | struct kvec iov; | |
595 | int len; | |
596 | ||
597 | EnterFunction(7); | |
598 | ||
599 | /* Receive a packet */ | |
600 | iov.iov_base = buffer; | |
601 | iov.iov_len = (size_t)buflen; | |
602 | ||
603 | len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, 0); | |
604 | ||
605 | if (len < 0) | |
606 | return -1; | |
607 | ||
608 | LeaveFunction(7); | |
609 | return len; | |
610 | } | |
611 | ||
612 | ||
613 | static DECLARE_WAIT_QUEUE_HEAD(sync_wait); | |
614 | static pid_t sync_master_pid = 0; | |
615 | static pid_t sync_backup_pid = 0; | |
616 | ||
617 | static DECLARE_WAIT_QUEUE_HEAD(stop_sync_wait); | |
618 | static int stop_master_sync = 0; | |
619 | static int stop_backup_sync = 0; | |
620 | ||
621 | static void sync_master_loop(void) | |
622 | { | |
623 | struct socket *sock; | |
624 | struct ip_vs_sync_buff *sb; | |
625 | ||
626 | /* create the sending multicast socket */ | |
627 | sock = make_send_sock(); | |
628 | if (!sock) | |
629 | return; | |
630 | ||
631 | IP_VS_INFO("sync thread started: state = MASTER, mcast_ifn = %s, " | |
632 | "syncid = %d\n", | |
633 | ip_vs_master_mcast_ifn, ip_vs_master_syncid); | |
634 | ||
635 | for (;;) { | |
636 | while ((sb=sb_dequeue())) { | |
637 | ip_vs_send_sync_msg(sock, sb->mesg); | |
638 | ip_vs_sync_buff_release(sb); | |
639 | } | |
640 | ||
641 | /* check if entries stay in curr_sb for 2 seconds */ | |
642 | if ((sb = get_curr_sync_buff(2*HZ))) { | |
643 | ip_vs_send_sync_msg(sock, sb->mesg); | |
644 | ip_vs_sync_buff_release(sb); | |
645 | } | |
646 | ||
647 | if (stop_master_sync) | |
648 | break; | |
649 | ||
650 | ssleep(1); | |
651 | } | |
652 | ||
653 | /* clean up the sync_buff queue */ | |
654 | while ((sb=sb_dequeue())) { | |
655 | ip_vs_sync_buff_release(sb); | |
656 | } | |
657 | ||
658 | /* clean up the current sync_buff */ | |
659 | if ((sb = get_curr_sync_buff(0))) { | |
660 | ip_vs_sync_buff_release(sb); | |
661 | } | |
662 | ||
663 | /* release the sending multicast socket */ | |
664 | sock_release(sock); | |
665 | } | |
666 | ||
667 | ||
668 | static void sync_backup_loop(void) | |
669 | { | |
670 | struct socket *sock; | |
671 | char *buf; | |
672 | int len; | |
673 | ||
674 | if (!(buf = kmalloc(sync_recv_mesg_maxlen, GFP_ATOMIC))) { | |
675 | IP_VS_ERR("sync_backup_loop: kmalloc error\n"); | |
676 | return; | |
677 | } | |
678 | ||
679 | /* create the receiving multicast socket */ | |
680 | sock = make_receive_sock(); | |
681 | if (!sock) | |
682 | goto out; | |
683 | ||
684 | IP_VS_INFO("sync thread started: state = BACKUP, mcast_ifn = %s, " | |
685 | "syncid = %d\n", | |
686 | ip_vs_backup_mcast_ifn, ip_vs_backup_syncid); | |
687 | ||
688 | for (;;) { | |
689 | /* do you have data now? */ | |
690 | while (!skb_queue_empty(&(sock->sk->sk_receive_queue))) { | |
691 | if ((len = | |
692 | ip_vs_receive(sock, buf, | |
693 | sync_recv_mesg_maxlen)) <= 0) { | |
694 | IP_VS_ERR("receiving message error\n"); | |
695 | break; | |
696 | } | |
697 | /* disable bottom half, because it accessed the data | |
698 | shared by softirq while getting/creating conns */ | |
699 | local_bh_disable(); | |
700 | ip_vs_process_message(buf, len); | |
701 | local_bh_enable(); | |
702 | } | |
703 | ||
704 | if (stop_backup_sync) | |
705 | break; | |
706 | ||
707 | ssleep(1); | |
708 | } | |
709 | ||
710 | /* release the sending multicast socket */ | |
711 | sock_release(sock); | |
712 | ||
713 | out: | |
714 | kfree(buf); | |
715 | } | |
716 | ||
717 | ||
718 | static void set_sync_pid(int sync_state, pid_t sync_pid) | |
719 | { | |
720 | if (sync_state == IP_VS_STATE_MASTER) | |
721 | sync_master_pid = sync_pid; | |
722 | else if (sync_state == IP_VS_STATE_BACKUP) | |
723 | sync_backup_pid = sync_pid; | |
724 | } | |
725 | ||
726 | static void set_stop_sync(int sync_state, int set) | |
727 | { | |
728 | if (sync_state == IP_VS_STATE_MASTER) | |
729 | stop_master_sync = set; | |
730 | else if (sync_state == IP_VS_STATE_BACKUP) | |
731 | stop_backup_sync = set; | |
732 | else { | |
733 | stop_master_sync = set; | |
734 | stop_backup_sync = set; | |
735 | } | |
736 | } | |
737 | ||
738 | static int sync_thread(void *startup) | |
739 | { | |
740 | DECLARE_WAITQUEUE(wait, current); | |
741 | mm_segment_t oldmm; | |
742 | int state; | |
743 | const char *name; | |
744 | ||
745 | /* increase the module use count */ | |
746 | ip_vs_use_count_inc(); | |
747 | ||
748 | if (ip_vs_sync_state & IP_VS_STATE_MASTER && !sync_master_pid) { | |
749 | state = IP_VS_STATE_MASTER; | |
750 | name = "ipvs_syncmaster"; | |
751 | } else if (ip_vs_sync_state & IP_VS_STATE_BACKUP && !sync_backup_pid) { | |
752 | state = IP_VS_STATE_BACKUP; | |
753 | name = "ipvs_syncbackup"; | |
754 | } else { | |
755 | IP_VS_BUG(); | |
756 | ip_vs_use_count_dec(); | |
757 | return -EINVAL; | |
758 | } | |
759 | ||
760 | daemonize(name); | |
761 | ||
762 | oldmm = get_fs(); | |
763 | set_fs(KERNEL_DS); | |
764 | ||
765 | /* Block all signals */ | |
766 | spin_lock_irq(¤t->sighand->siglock); | |
767 | siginitsetinv(¤t->blocked, 0); | |
768 | recalc_sigpending(); | |
769 | spin_unlock_irq(¤t->sighand->siglock); | |
770 | ||
771 | /* set the maximum length of sync message */ | |
772 | set_sync_mesg_maxlen(state); | |
773 | ||
774 | /* set up multicast address */ | |
775 | mcast_addr.sin_family = AF_INET; | |
776 | mcast_addr.sin_port = htons(IP_VS_SYNC_PORT); | |
777 | mcast_addr.sin_addr.s_addr = htonl(IP_VS_SYNC_GROUP); | |
778 | ||
779 | add_wait_queue(&sync_wait, &wait); | |
780 | ||
781 | set_sync_pid(state, current->pid); | |
782 | complete((struct completion *)startup); | |
783 | ||
784 | /* processing master/backup loop here */ | |
785 | if (state == IP_VS_STATE_MASTER) | |
786 | sync_master_loop(); | |
787 | else if (state == IP_VS_STATE_BACKUP) | |
788 | sync_backup_loop(); | |
789 | else IP_VS_BUG(); | |
790 | ||
791 | remove_wait_queue(&sync_wait, &wait); | |
792 | ||
793 | /* thread exits */ | |
794 | set_sync_pid(state, 0); | |
795 | IP_VS_INFO("sync thread stopped!\n"); | |
796 | ||
797 | set_fs(oldmm); | |
798 | ||
799 | /* decrease the module use count */ | |
800 | ip_vs_use_count_dec(); | |
801 | ||
802 | set_stop_sync(state, 0); | |
803 | wake_up(&stop_sync_wait); | |
804 | ||
805 | return 0; | |
806 | } | |
807 | ||
808 | ||
809 | static int fork_sync_thread(void *startup) | |
810 | { | |
811 | pid_t pid; | |
812 | ||
813 | /* fork the sync thread here, then the parent process of the | |
814 | sync thread is the init process after this thread exits. */ | |
815 | repeat: | |
816 | if ((pid = kernel_thread(sync_thread, startup, 0)) < 0) { | |
817 | IP_VS_ERR("could not create sync_thread due to %d... " | |
818 | "retrying.\n", pid); | |
819 | ssleep(1); | |
820 | goto repeat; | |
821 | } | |
822 | ||
823 | return 0; | |
824 | } | |
825 | ||
826 | ||
827 | int start_sync_thread(int state, char *mcast_ifn, __u8 syncid) | |
828 | { | |
829 | DECLARE_COMPLETION(startup); | |
830 | pid_t pid; | |
831 | ||
832 | if ((state == IP_VS_STATE_MASTER && sync_master_pid) || | |
833 | (state == IP_VS_STATE_BACKUP && sync_backup_pid)) | |
834 | return -EEXIST; | |
835 | ||
836 | IP_VS_DBG(7, "%s: pid %d\n", __FUNCTION__, current->pid); | |
837 | IP_VS_DBG(7, "Each ip_vs_sync_conn entry need %Zd bytes\n", | |
838 | sizeof(struct ip_vs_sync_conn)); | |
839 | ||
840 | ip_vs_sync_state |= state; | |
841 | if (state == IP_VS_STATE_MASTER) { | |
4da62fc7 | 842 | strlcpy(ip_vs_master_mcast_ifn, mcast_ifn, sizeof(ip_vs_master_mcast_ifn)); |
1da177e4 LT |
843 | ip_vs_master_syncid = syncid; |
844 | } else { | |
4da62fc7 | 845 | strlcpy(ip_vs_backup_mcast_ifn, mcast_ifn, sizeof(ip_vs_backup_mcast_ifn)); |
1da177e4 LT |
846 | ip_vs_backup_syncid = syncid; |
847 | } | |
848 | ||
849 | repeat: | |
850 | if ((pid = kernel_thread(fork_sync_thread, &startup, 0)) < 0) { | |
851 | IP_VS_ERR("could not create fork_sync_thread due to %d... " | |
852 | "retrying.\n", pid); | |
853 | ssleep(1); | |
854 | goto repeat; | |
855 | } | |
856 | ||
857 | wait_for_completion(&startup); | |
858 | ||
859 | return 0; | |
860 | } | |
861 | ||
862 | ||
863 | int stop_sync_thread(int state) | |
864 | { | |
865 | DECLARE_WAITQUEUE(wait, current); | |
866 | ||
867 | if ((state == IP_VS_STATE_MASTER && !sync_master_pid) || | |
868 | (state == IP_VS_STATE_BACKUP && !sync_backup_pid)) | |
869 | return -ESRCH; | |
870 | ||
871 | IP_VS_DBG(7, "%s: pid %d\n", __FUNCTION__, current->pid); | |
872 | IP_VS_INFO("stopping sync thread %d ...\n", | |
873 | (state == IP_VS_STATE_MASTER) ? sync_master_pid : sync_backup_pid); | |
874 | ||
875 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
876 | add_wait_queue(&stop_sync_wait, &wait); | |
877 | set_stop_sync(state, 1); | |
878 | ip_vs_sync_state -= state; | |
879 | wake_up(&sync_wait); | |
880 | schedule(); | |
881 | __set_current_state(TASK_RUNNING); | |
882 | remove_wait_queue(&stop_sync_wait, &wait); | |
883 | ||
884 | /* Note: no need to reap the sync thread, because its parent | |
885 | process is the init process */ | |
886 | ||
887 | if ((state == IP_VS_STATE_MASTER && stop_master_sync) || | |
888 | (state == IP_VS_STATE_BACKUP && stop_backup_sync)) | |
889 | IP_VS_BUG(); | |
890 | ||
891 | return 0; | |
892 | } |