]>
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 SIMPLE_CONN_SIZE (sizeof(struct ip_vs_sync_conn)) |
76 | #define FULL_CONN_SIZE \ | |
77 | (sizeof(struct ip_vs_sync_conn) + sizeof(struct ip_vs_sync_conn_options)) | |
78 | ||
79 | ||
80 | /* | |
81 | The master mulitcasts messages to the backup load balancers in the | |
82 | following format. | |
83 | ||
84 | 0 1 2 3 | |
85 | 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 | |
86 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
87 | | Count Conns | SyncID | Size | | |
88 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
89 | | | | |
90 | | IPVS Sync Connection (1) | | |
91 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
92 | | . | | |
93 | | . | | |
94 | | . | | |
95 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
96 | | | | |
97 | | IPVS Sync Connection (n) | | |
98 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
99 | */ | |
100 | ||
101 | #define SYNC_MESG_HEADER_LEN 4 | |
102 | ||
103 | struct ip_vs_sync_mesg { | |
104 | __u8 nr_conns; | |
105 | __u8 syncid; | |
106 | __u16 size; | |
107 | ||
108 | /* ip_vs_sync_conn entries start here */ | |
109 | }; | |
110 | ||
111 | /* the maximum length of sync (sending/receiving) message */ | |
112 | static int sync_send_mesg_maxlen; | |
113 | static int sync_recv_mesg_maxlen; | |
114 | ||
115 | struct ip_vs_sync_buff { | |
116 | struct list_head list; | |
117 | unsigned long firstuse; | |
118 | ||
119 | /* pointers for the message data */ | |
120 | struct ip_vs_sync_mesg *mesg; | |
121 | unsigned char *head; | |
122 | unsigned char *end; | |
123 | }; | |
124 | ||
125 | ||
126 | /* the sync_buff list head and the lock */ | |
127 | static LIST_HEAD(ip_vs_sync_queue); | |
128 | static DEFINE_SPINLOCK(ip_vs_sync_lock); | |
129 | ||
130 | /* current sync_buff for accepting new conn entries */ | |
131 | static struct ip_vs_sync_buff *curr_sb = NULL; | |
132 | static DEFINE_SPINLOCK(curr_sb_lock); | |
133 | ||
134 | /* ipvs sync daemon state */ | |
135 | volatile int ip_vs_sync_state = IP_VS_STATE_NONE; | |
136 | volatile int ip_vs_master_syncid = 0; | |
137 | volatile int ip_vs_backup_syncid = 0; | |
138 | ||
139 | /* multicast interface name */ | |
140 | char ip_vs_master_mcast_ifn[IP_VS_IFNAME_MAXLEN]; | |
141 | char ip_vs_backup_mcast_ifn[IP_VS_IFNAME_MAXLEN]; | |
142 | ||
143 | /* multicast addr */ | |
144 | static struct sockaddr_in mcast_addr; | |
145 | ||
146 | ||
147 | static inline void sb_queue_tail(struct ip_vs_sync_buff *sb) | |
148 | { | |
149 | spin_lock(&ip_vs_sync_lock); | |
150 | list_add_tail(&sb->list, &ip_vs_sync_queue); | |
151 | spin_unlock(&ip_vs_sync_lock); | |
152 | } | |
153 | ||
154 | static inline struct ip_vs_sync_buff * sb_dequeue(void) | |
155 | { | |
156 | struct ip_vs_sync_buff *sb; | |
157 | ||
158 | spin_lock_bh(&ip_vs_sync_lock); | |
159 | if (list_empty(&ip_vs_sync_queue)) { | |
160 | sb = NULL; | |
161 | } else { | |
162 | sb = list_entry(ip_vs_sync_queue.next, | |
163 | struct ip_vs_sync_buff, | |
164 | list); | |
165 | list_del(&sb->list); | |
166 | } | |
167 | spin_unlock_bh(&ip_vs_sync_lock); | |
168 | ||
169 | return sb; | |
170 | } | |
171 | ||
172 | static inline struct ip_vs_sync_buff * ip_vs_sync_buff_create(void) | |
173 | { | |
174 | struct ip_vs_sync_buff *sb; | |
175 | ||
176 | if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC))) | |
177 | return NULL; | |
178 | ||
179 | if (!(sb->mesg=kmalloc(sync_send_mesg_maxlen, GFP_ATOMIC))) { | |
180 | kfree(sb); | |
181 | return NULL; | |
182 | } | |
183 | sb->mesg->nr_conns = 0; | |
184 | sb->mesg->syncid = ip_vs_master_syncid; | |
185 | sb->mesg->size = 4; | |
186 | sb->head = (unsigned char *)sb->mesg + 4; | |
187 | sb->end = (unsigned char *)sb->mesg + sync_send_mesg_maxlen; | |
188 | sb->firstuse = jiffies; | |
189 | return sb; | |
190 | } | |
191 | ||
192 | static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb) | |
193 | { | |
194 | kfree(sb->mesg); | |
195 | kfree(sb); | |
196 | } | |
197 | ||
198 | /* | |
199 | * Get the current sync buffer if it has been created for more | |
200 | * than the specified time or the specified time is zero. | |
201 | */ | |
202 | static inline struct ip_vs_sync_buff * | |
203 | get_curr_sync_buff(unsigned long time) | |
204 | { | |
205 | struct ip_vs_sync_buff *sb; | |
206 | ||
207 | spin_lock_bh(&curr_sb_lock); | |
208 | if (curr_sb && (time == 0 || | |
209 | time_before(jiffies - curr_sb->firstuse, time))) { | |
210 | sb = curr_sb; | |
211 | curr_sb = NULL; | |
212 | } else | |
213 | sb = NULL; | |
214 | spin_unlock_bh(&curr_sb_lock); | |
215 | return sb; | |
216 | } | |
217 | ||
218 | ||
219 | /* | |
220 | * Add an ip_vs_conn information into the current sync_buff. | |
221 | * Called by ip_vs_in. | |
222 | */ | |
223 | void ip_vs_sync_conn(struct ip_vs_conn *cp) | |
224 | { | |
225 | struct ip_vs_sync_mesg *m; | |
226 | struct ip_vs_sync_conn *s; | |
227 | int len; | |
228 | ||
229 | spin_lock(&curr_sb_lock); | |
230 | if (!curr_sb) { | |
231 | if (!(curr_sb=ip_vs_sync_buff_create())) { | |
232 | spin_unlock(&curr_sb_lock); | |
233 | IP_VS_ERR("ip_vs_sync_buff_create failed.\n"); | |
234 | return; | |
235 | } | |
236 | } | |
237 | ||
238 | len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE : | |
239 | SIMPLE_CONN_SIZE; | |
240 | m = curr_sb->mesg; | |
241 | s = (struct ip_vs_sync_conn *)curr_sb->head; | |
242 | ||
243 | /* copy members */ | |
244 | s->protocol = cp->protocol; | |
245 | s->cport = cp->cport; | |
246 | s->vport = cp->vport; | |
247 | s->dport = cp->dport; | |
248 | s->caddr = cp->caddr; | |
249 | s->vaddr = cp->vaddr; | |
250 | s->daddr = cp->daddr; | |
251 | s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED); | |
252 | s->state = htons(cp->state); | |
253 | if (cp->flags & IP_VS_CONN_F_SEQ_MASK) { | |
254 | struct ip_vs_sync_conn_options *opt = | |
255 | (struct ip_vs_sync_conn_options *)&s[1]; | |
256 | memcpy(opt, &cp->in_seq, sizeof(*opt)); | |
257 | } | |
258 | ||
259 | m->nr_conns++; | |
260 | m->size += len; | |
261 | curr_sb->head += len; | |
262 | ||
263 | /* check if there is a space for next one */ | |
264 | if (curr_sb->head+FULL_CONN_SIZE > curr_sb->end) { | |
265 | sb_queue_tail(curr_sb); | |
266 | curr_sb = NULL; | |
267 | } | |
268 | spin_unlock(&curr_sb_lock); | |
269 | ||
270 | /* synchronize its controller if it has */ | |
271 | if (cp->control) | |
272 | ip_vs_sync_conn(cp->control); | |
273 | } | |
274 | ||
275 | ||
276 | /* | |
277 | * Process received multicast message and create the corresponding | |
278 | * ip_vs_conn entries. | |
279 | */ | |
280 | static void ip_vs_process_message(const char *buffer, const size_t buflen) | |
281 | { | |
282 | struct ip_vs_sync_mesg *m = (struct ip_vs_sync_mesg *)buffer; | |
283 | struct ip_vs_sync_conn *s; | |
284 | struct ip_vs_sync_conn_options *opt; | |
285 | struct ip_vs_conn *cp; | |
5c81833c | 286 | struct ip_vs_protocol *pp; |
1e356f9c | 287 | struct ip_vs_dest *dest; |
1da177e4 LT |
288 | char *p; |
289 | int i; | |
290 | ||
291 | /* Convert size back to host byte order */ | |
292 | m->size = ntohs(m->size); | |
293 | ||
294 | if (buflen != m->size) { | |
295 | IP_VS_ERR("bogus message\n"); | |
296 | return; | |
297 | } | |
298 | ||
299 | /* SyncID sanity check */ | |
300 | if (ip_vs_backup_syncid != 0 && m->syncid != ip_vs_backup_syncid) { | |
301 | IP_VS_DBG(7, "Ignoring incoming msg with syncid = %d\n", | |
302 | m->syncid); | |
303 | return; | |
304 | } | |
305 | ||
306 | p = (char *)buffer + sizeof(struct ip_vs_sync_mesg); | |
307 | for (i=0; i<m->nr_conns; i++) { | |
87375ab4 JA |
308 | unsigned flags; |
309 | ||
1da177e4 | 310 | s = (struct ip_vs_sync_conn *)p; |
87375ab4 JA |
311 | flags = ntohs(s->flags); |
312 | if (!(flags & IP_VS_CONN_F_TEMPLATE)) | |
313 | cp = ip_vs_conn_in_get(s->protocol, | |
314 | s->caddr, s->cport, | |
315 | s->vaddr, s->vport); | |
316 | else | |
317 | cp = ip_vs_ct_in_get(s->protocol, | |
318 | s->caddr, s->cport, | |
319 | s->vaddr, s->vport); | |
1da177e4 | 320 | if (!cp) { |
1e356f9c RB |
321 | /* |
322 | * Find the appropriate destination for the connection. | |
323 | * If it is not found the connection will remain unbound | |
324 | * but still handled. | |
325 | */ | |
326 | dest = ip_vs_find_dest(s->daddr, s->dport, | |
327 | s->vaddr, s->vport, | |
328 | s->protocol); | |
1da177e4 LT |
329 | cp = ip_vs_conn_new(s->protocol, |
330 | s->caddr, s->cport, | |
331 | s->vaddr, s->vport, | |
332 | s->daddr, s->dport, | |
1e356f9c RB |
333 | flags, dest); |
334 | if (dest) | |
335 | atomic_dec(&dest->refcnt); | |
1da177e4 LT |
336 | if (!cp) { |
337 | IP_VS_ERR("ip_vs_conn_new failed\n"); | |
338 | return; | |
339 | } | |
340 | cp->state = ntohs(s->state); | |
341 | } else if (!cp->dest) { | |
1e356f9c RB |
342 | dest = ip_vs_try_bind_dest(cp); |
343 | if (!dest) { | |
344 | /* it is an unbound entry created by | |
345 | * synchronization */ | |
1e356f9c RB |
346 | cp->flags = flags | IP_VS_CONN_F_HASHED; |
347 | } else | |
348 | atomic_dec(&dest->refcnt); | |
1da177e4 LT |
349 | } /* Note that we don't touch its state and flags |
350 | if it is a normal entry. */ | |
351 | ||
87375ab4 | 352 | if (flags & IP_VS_CONN_F_SEQ_MASK) { |
1da177e4 LT |
353 | opt = (struct ip_vs_sync_conn_options *)&s[1]; |
354 | memcpy(&cp->in_seq, opt, sizeof(*opt)); | |
355 | p += FULL_CONN_SIZE; | |
356 | } else | |
357 | p += SIMPLE_CONN_SIZE; | |
358 | ||
359 | atomic_set(&cp->in_pkts, sysctl_ip_vs_sync_threshold[0]); | |
efac5276 | 360 | cp->state = ntohs(s->state); |
5c81833c AG |
361 | pp = ip_vs_proto_get(s->protocol); |
362 | cp->timeout = pp->timeout_table[cp->state]; | |
1da177e4 LT |
363 | ip_vs_conn_put(cp); |
364 | ||
365 | if (p > buffer+buflen) { | |
366 | IP_VS_ERR("bogus message\n"); | |
367 | return; | |
368 | } | |
369 | } | |
370 | } | |
371 | ||
372 | ||
373 | /* | |
374 | * Setup loopback of outgoing multicasts on a sending socket | |
375 | */ | |
376 | static void set_mcast_loop(struct sock *sk, u_char loop) | |
377 | { | |
378 | struct inet_sock *inet = inet_sk(sk); | |
379 | ||
380 | /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */ | |
381 | lock_sock(sk); | |
382 | inet->mc_loop = loop ? 1 : 0; | |
383 | release_sock(sk); | |
384 | } | |
385 | ||
386 | /* | |
387 | * Specify TTL for outgoing multicasts on a sending socket | |
388 | */ | |
389 | static void set_mcast_ttl(struct sock *sk, u_char ttl) | |
390 | { | |
391 | struct inet_sock *inet = inet_sk(sk); | |
392 | ||
393 | /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */ | |
394 | lock_sock(sk); | |
395 | inet->mc_ttl = ttl; | |
396 | release_sock(sk); | |
397 | } | |
398 | ||
399 | /* | |
400 | * Specifiy default interface for outgoing multicasts | |
401 | */ | |
402 | static int set_mcast_if(struct sock *sk, char *ifname) | |
403 | { | |
404 | struct net_device *dev; | |
405 | struct inet_sock *inet = inet_sk(sk); | |
406 | ||
881d966b | 407 | if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL) |
1da177e4 LT |
408 | return -ENODEV; |
409 | ||
410 | if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) | |
411 | return -EINVAL; | |
412 | ||
413 | lock_sock(sk); | |
414 | inet->mc_index = dev->ifindex; | |
415 | /* inet->mc_addr = 0; */ | |
416 | release_sock(sk); | |
417 | ||
418 | return 0; | |
419 | } | |
420 | ||
421 | ||
422 | /* | |
423 | * Set the maximum length of sync message according to the | |
424 | * specified interface's MTU. | |
425 | */ | |
426 | static int set_sync_mesg_maxlen(int sync_state) | |
427 | { | |
428 | struct net_device *dev; | |
429 | int num; | |
430 | ||
431 | if (sync_state == IP_VS_STATE_MASTER) { | |
881d966b | 432 | if ((dev = __dev_get_by_name(&init_net, ip_vs_master_mcast_ifn)) == NULL) |
1da177e4 LT |
433 | return -ENODEV; |
434 | ||
435 | num = (dev->mtu - sizeof(struct iphdr) - | |
436 | sizeof(struct udphdr) - | |
437 | SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE; | |
438 | sync_send_mesg_maxlen = | |
439 | SYNC_MESG_HEADER_LEN + SIMPLE_CONN_SIZE * num; | |
440 | IP_VS_DBG(7, "setting the maximum length of sync sending " | |
441 | "message %d.\n", sync_send_mesg_maxlen); | |
442 | } else if (sync_state == IP_VS_STATE_BACKUP) { | |
881d966b | 443 | if ((dev = __dev_get_by_name(&init_net, ip_vs_backup_mcast_ifn)) == NULL) |
1da177e4 LT |
444 | return -ENODEV; |
445 | ||
446 | sync_recv_mesg_maxlen = dev->mtu - | |
447 | sizeof(struct iphdr) - sizeof(struct udphdr); | |
448 | IP_VS_DBG(7, "setting the maximum length of sync receiving " | |
449 | "message %d.\n", sync_recv_mesg_maxlen); | |
450 | } | |
451 | ||
452 | return 0; | |
453 | } | |
454 | ||
455 | ||
456 | /* | |
457 | * Join a multicast group. | |
458 | * the group is specified by a class D multicast address 224.0.0.0/8 | |
459 | * in the in_addr structure passed in as a parameter. | |
460 | */ | |
461 | static int | |
462 | join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname) | |
463 | { | |
464 | struct ip_mreqn mreq; | |
465 | struct net_device *dev; | |
466 | int ret; | |
467 | ||
468 | memset(&mreq, 0, sizeof(mreq)); | |
469 | memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr)); | |
470 | ||
881d966b | 471 | if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL) |
1da177e4 LT |
472 | return -ENODEV; |
473 | if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) | |
474 | return -EINVAL; | |
475 | ||
476 | mreq.imr_ifindex = dev->ifindex; | |
477 | ||
478 | lock_sock(sk); | |
479 | ret = ip_mc_join_group(sk, &mreq); | |
480 | release_sock(sk); | |
481 | ||
482 | return ret; | |
483 | } | |
484 | ||
485 | ||
486 | static int bind_mcastif_addr(struct socket *sock, char *ifname) | |
487 | { | |
488 | struct net_device *dev; | |
a61ced5d | 489 | __be32 addr; |
1da177e4 LT |
490 | struct sockaddr_in sin; |
491 | ||
881d966b | 492 | if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL) |
1da177e4 LT |
493 | return -ENODEV; |
494 | ||
495 | addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE); | |
496 | if (!addr) | |
497 | IP_VS_ERR("You probably need to specify IP address on " | |
498 | "multicast interface.\n"); | |
499 | ||
500 | IP_VS_DBG(7, "binding socket with (%s) %u.%u.%u.%u\n", | |
501 | ifname, NIPQUAD(addr)); | |
502 | ||
503 | /* Now bind the socket with the address of multicast interface */ | |
504 | sin.sin_family = AF_INET; | |
505 | sin.sin_addr.s_addr = addr; | |
506 | sin.sin_port = 0; | |
507 | ||
508 | return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin)); | |
509 | } | |
510 | ||
511 | /* | |
512 | * Set up sending multicast socket over UDP | |
513 | */ | |
514 | static struct socket * make_send_sock(void) | |
515 | { | |
516 | struct socket *sock; | |
517 | ||
518 | /* First create a socket */ | |
519 | if (sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock) < 0) { | |
520 | IP_VS_ERR("Error during creation of socket; terminating\n"); | |
521 | return NULL; | |
522 | } | |
523 | ||
524 | if (set_mcast_if(sock->sk, ip_vs_master_mcast_ifn) < 0) { | |
525 | IP_VS_ERR("Error setting outbound mcast interface\n"); | |
526 | goto error; | |
527 | } | |
528 | ||
529 | set_mcast_loop(sock->sk, 0); | |
530 | set_mcast_ttl(sock->sk, 1); | |
531 | ||
532 | if (bind_mcastif_addr(sock, ip_vs_master_mcast_ifn) < 0) { | |
533 | IP_VS_ERR("Error binding address of the mcast interface\n"); | |
534 | goto error; | |
535 | } | |
536 | ||
537 | if (sock->ops->connect(sock, | |
538 | (struct sockaddr*)&mcast_addr, | |
539 | sizeof(struct sockaddr), 0) < 0) { | |
540 | IP_VS_ERR("Error connecting to the multicast addr\n"); | |
541 | goto error; | |
542 | } | |
543 | ||
544 | return sock; | |
545 | ||
546 | error: | |
547 | sock_release(sock); | |
548 | return NULL; | |
549 | } | |
550 | ||
551 | ||
552 | /* | |
553 | * Set up receiving multicast socket over UDP | |
554 | */ | |
555 | static struct socket * make_receive_sock(void) | |
556 | { | |
557 | struct socket *sock; | |
558 | ||
559 | /* First create a socket */ | |
560 | if (sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock) < 0) { | |
561 | IP_VS_ERR("Error during creation of socket; terminating\n"); | |
562 | return NULL; | |
563 | } | |
564 | ||
565 | /* it is equivalent to the REUSEADDR option in user-space */ | |
566 | sock->sk->sk_reuse = 1; | |
567 | ||
568 | if (sock->ops->bind(sock, | |
569 | (struct sockaddr*)&mcast_addr, | |
570 | sizeof(struct sockaddr)) < 0) { | |
571 | IP_VS_ERR("Error binding to the multicast addr\n"); | |
572 | goto error; | |
573 | } | |
574 | ||
575 | /* join the multicast group */ | |
576 | if (join_mcast_group(sock->sk, | |
577 | (struct in_addr*)&mcast_addr.sin_addr, | |
578 | ip_vs_backup_mcast_ifn) < 0) { | |
579 | IP_VS_ERR("Error joining to the multicast group\n"); | |
580 | goto error; | |
581 | } | |
582 | ||
583 | return sock; | |
584 | ||
585 | error: | |
586 | sock_release(sock); | |
587 | return NULL; | |
588 | } | |
589 | ||
590 | ||
591 | static int | |
592 | ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length) | |
593 | { | |
594 | struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL}; | |
595 | struct kvec iov; | |
596 | int len; | |
597 | ||
598 | EnterFunction(7); | |
599 | iov.iov_base = (void *)buffer; | |
600 | iov.iov_len = length; | |
601 | ||
602 | len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length)); | |
603 | ||
604 | LeaveFunction(7); | |
605 | return len; | |
606 | } | |
607 | ||
608 | static void | |
609 | ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg) | |
610 | { | |
611 | int msize; | |
612 | ||
613 | msize = msg->size; | |
614 | ||
615 | /* Put size in network byte order */ | |
616 | msg->size = htons(msg->size); | |
617 | ||
618 | if (ip_vs_send_async(sock, (char *)msg, msize) != msize) | |
619 | IP_VS_ERR("ip_vs_send_async error\n"); | |
620 | } | |
621 | ||
622 | static int | |
623 | ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen) | |
624 | { | |
625 | struct msghdr msg = {NULL,}; | |
626 | struct kvec iov; | |
627 | int len; | |
628 | ||
629 | EnterFunction(7); | |
630 | ||
631 | /* Receive a packet */ | |
632 | iov.iov_base = buffer; | |
633 | iov.iov_len = (size_t)buflen; | |
634 | ||
635 | len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, 0); | |
636 | ||
637 | if (len < 0) | |
638 | return -1; | |
639 | ||
640 | LeaveFunction(7); | |
641 | return len; | |
642 | } | |
643 | ||
644 | ||
645 | static DECLARE_WAIT_QUEUE_HEAD(sync_wait); | |
646 | static pid_t sync_master_pid = 0; | |
647 | static pid_t sync_backup_pid = 0; | |
648 | ||
649 | static DECLARE_WAIT_QUEUE_HEAD(stop_sync_wait); | |
650 | static int stop_master_sync = 0; | |
651 | static int stop_backup_sync = 0; | |
652 | ||
653 | static void sync_master_loop(void) | |
654 | { | |
655 | struct socket *sock; | |
656 | struct ip_vs_sync_buff *sb; | |
657 | ||
658 | /* create the sending multicast socket */ | |
659 | sock = make_send_sock(); | |
660 | if (!sock) | |
661 | return; | |
662 | ||
663 | IP_VS_INFO("sync thread started: state = MASTER, mcast_ifn = %s, " | |
664 | "syncid = %d\n", | |
665 | ip_vs_master_mcast_ifn, ip_vs_master_syncid); | |
666 | ||
667 | for (;;) { | |
668 | while ((sb=sb_dequeue())) { | |
669 | ip_vs_send_sync_msg(sock, sb->mesg); | |
670 | ip_vs_sync_buff_release(sb); | |
671 | } | |
672 | ||
673 | /* check if entries stay in curr_sb for 2 seconds */ | |
674 | if ((sb = get_curr_sync_buff(2*HZ))) { | |
675 | ip_vs_send_sync_msg(sock, sb->mesg); | |
676 | ip_vs_sync_buff_release(sb); | |
677 | } | |
678 | ||
679 | if (stop_master_sync) | |
680 | break; | |
681 | ||
89eaeb09 | 682 | msleep_interruptible(1000); |
1da177e4 LT |
683 | } |
684 | ||
685 | /* clean up the sync_buff queue */ | |
686 | while ((sb=sb_dequeue())) { | |
687 | ip_vs_sync_buff_release(sb); | |
688 | } | |
689 | ||
690 | /* clean up the current sync_buff */ | |
691 | if ((sb = get_curr_sync_buff(0))) { | |
692 | ip_vs_sync_buff_release(sb); | |
693 | } | |
694 | ||
695 | /* release the sending multicast socket */ | |
696 | sock_release(sock); | |
697 | } | |
698 | ||
699 | ||
700 | static void sync_backup_loop(void) | |
701 | { | |
702 | struct socket *sock; | |
703 | char *buf; | |
704 | int len; | |
705 | ||
706 | if (!(buf = kmalloc(sync_recv_mesg_maxlen, GFP_ATOMIC))) { | |
707 | IP_VS_ERR("sync_backup_loop: kmalloc error\n"); | |
708 | return; | |
709 | } | |
710 | ||
711 | /* create the receiving multicast socket */ | |
712 | sock = make_receive_sock(); | |
713 | if (!sock) | |
714 | goto out; | |
715 | ||
716 | IP_VS_INFO("sync thread started: state = BACKUP, mcast_ifn = %s, " | |
717 | "syncid = %d\n", | |
718 | ip_vs_backup_mcast_ifn, ip_vs_backup_syncid); | |
719 | ||
720 | for (;;) { | |
721 | /* do you have data now? */ | |
722 | while (!skb_queue_empty(&(sock->sk->sk_receive_queue))) { | |
723 | if ((len = | |
724 | ip_vs_receive(sock, buf, | |
725 | sync_recv_mesg_maxlen)) <= 0) { | |
726 | IP_VS_ERR("receiving message error\n"); | |
727 | break; | |
728 | } | |
729 | /* disable bottom half, because it accessed the data | |
730 | shared by softirq while getting/creating conns */ | |
731 | local_bh_disable(); | |
732 | ip_vs_process_message(buf, len); | |
733 | local_bh_enable(); | |
734 | } | |
735 | ||
736 | if (stop_backup_sync) | |
737 | break; | |
738 | ||
89eaeb09 | 739 | msleep_interruptible(1000); |
1da177e4 LT |
740 | } |
741 | ||
742 | /* release the sending multicast socket */ | |
743 | sock_release(sock); | |
744 | ||
745 | out: | |
746 | kfree(buf); | |
747 | } | |
748 | ||
749 | ||
750 | static void set_sync_pid(int sync_state, pid_t sync_pid) | |
751 | { | |
752 | if (sync_state == IP_VS_STATE_MASTER) | |
753 | sync_master_pid = sync_pid; | |
754 | else if (sync_state == IP_VS_STATE_BACKUP) | |
755 | sync_backup_pid = sync_pid; | |
756 | } | |
757 | ||
758 | static void set_stop_sync(int sync_state, int set) | |
759 | { | |
760 | if (sync_state == IP_VS_STATE_MASTER) | |
761 | stop_master_sync = set; | |
762 | else if (sync_state == IP_VS_STATE_BACKUP) | |
763 | stop_backup_sync = set; | |
764 | else { | |
765 | stop_master_sync = set; | |
766 | stop_backup_sync = set; | |
767 | } | |
768 | } | |
769 | ||
770 | static int sync_thread(void *startup) | |
771 | { | |
772 | DECLARE_WAITQUEUE(wait, current); | |
773 | mm_segment_t oldmm; | |
774 | int state; | |
775 | const char *name; | |
cc0191ae | 776 | struct ip_vs_sync_thread_data *tinfo = startup; |
1da177e4 LT |
777 | |
778 | /* increase the module use count */ | |
779 | ip_vs_use_count_inc(); | |
780 | ||
781 | if (ip_vs_sync_state & IP_VS_STATE_MASTER && !sync_master_pid) { | |
782 | state = IP_VS_STATE_MASTER; | |
783 | name = "ipvs_syncmaster"; | |
784 | } else if (ip_vs_sync_state & IP_VS_STATE_BACKUP && !sync_backup_pid) { | |
785 | state = IP_VS_STATE_BACKUP; | |
786 | name = "ipvs_syncbackup"; | |
787 | } else { | |
788 | IP_VS_BUG(); | |
789 | ip_vs_use_count_dec(); | |
790 | return -EINVAL; | |
791 | } | |
792 | ||
793 | daemonize(name); | |
794 | ||
795 | oldmm = get_fs(); | |
796 | set_fs(KERNEL_DS); | |
797 | ||
798 | /* Block all signals */ | |
799 | spin_lock_irq(¤t->sighand->siglock); | |
800 | siginitsetinv(¤t->blocked, 0); | |
801 | recalc_sigpending(); | |
802 | spin_unlock_irq(¤t->sighand->siglock); | |
803 | ||
804 | /* set the maximum length of sync message */ | |
805 | set_sync_mesg_maxlen(state); | |
806 | ||
807 | /* set up multicast address */ | |
808 | mcast_addr.sin_family = AF_INET; | |
809 | mcast_addr.sin_port = htons(IP_VS_SYNC_PORT); | |
810 | mcast_addr.sin_addr.s_addr = htonl(IP_VS_SYNC_GROUP); | |
811 | ||
812 | add_wait_queue(&sync_wait, &wait); | |
813 | ||
6651fd56 | 814 | set_sync_pid(state, task_pid_nr(current)); |
cc0191ae NH |
815 | complete(tinfo->startup); |
816 | ||
817 | /* | |
818 | * once we call the completion queue above, we should | |
819 | * null out that reference, since its allocated on the | |
820 | * stack of the creating kernel thread | |
821 | */ | |
822 | tinfo->startup = NULL; | |
1da177e4 LT |
823 | |
824 | /* processing master/backup loop here */ | |
825 | if (state == IP_VS_STATE_MASTER) | |
826 | sync_master_loop(); | |
827 | else if (state == IP_VS_STATE_BACKUP) | |
828 | sync_backup_loop(); | |
829 | else IP_VS_BUG(); | |
830 | ||
831 | remove_wait_queue(&sync_wait, &wait); | |
832 | ||
833 | /* thread exits */ | |
cc0191ae NH |
834 | |
835 | /* | |
836 | * If we weren't explicitly stopped, then we | |
837 | * exited in error, and should undo our state | |
838 | */ | |
839 | if ((!stop_master_sync) && (!stop_backup_sync)) | |
840 | ip_vs_sync_state -= tinfo->state; | |
841 | ||
1da177e4 LT |
842 | set_sync_pid(state, 0); |
843 | IP_VS_INFO("sync thread stopped!\n"); | |
844 | ||
845 | set_fs(oldmm); | |
846 | ||
847 | /* decrease the module use count */ | |
848 | ip_vs_use_count_dec(); | |
849 | ||
850 | set_stop_sync(state, 0); | |
851 | wake_up(&stop_sync_wait); | |
852 | ||
cc0191ae NH |
853 | /* |
854 | * we need to free the structure that was allocated | |
855 | * for us in start_sync_thread | |
856 | */ | |
857 | kfree(tinfo); | |
1da177e4 LT |
858 | return 0; |
859 | } | |
860 | ||
861 | ||
862 | static int fork_sync_thread(void *startup) | |
863 | { | |
864 | pid_t pid; | |
865 | ||
866 | /* fork the sync thread here, then the parent process of the | |
867 | sync thread is the init process after this thread exits. */ | |
868 | repeat: | |
869 | if ((pid = kernel_thread(sync_thread, startup, 0)) < 0) { | |
870 | IP_VS_ERR("could not create sync_thread due to %d... " | |
871 | "retrying.\n", pid); | |
89eaeb09 | 872 | msleep_interruptible(1000); |
1da177e4 LT |
873 | goto repeat; |
874 | } | |
875 | ||
876 | return 0; | |
877 | } | |
878 | ||
879 | ||
880 | int start_sync_thread(int state, char *mcast_ifn, __u8 syncid) | |
881 | { | |
6e9a4738 | 882 | DECLARE_COMPLETION_ONSTACK(startup); |
1da177e4 | 883 | pid_t pid; |
cc0191ae | 884 | struct ip_vs_sync_thread_data *tinfo; |
1da177e4 LT |
885 | |
886 | if ((state == IP_VS_STATE_MASTER && sync_master_pid) || | |
887 | (state == IP_VS_STATE_BACKUP && sync_backup_pid)) | |
888 | return -EEXIST; | |
889 | ||
cc0191ae NH |
890 | /* |
891 | * Note that tinfo will be freed in sync_thread on exit | |
892 | */ | |
893 | tinfo = kmalloc(sizeof(struct ip_vs_sync_thread_data), GFP_KERNEL); | |
894 | if (!tinfo) | |
895 | return -ENOMEM; | |
896 | ||
ba25f9dc | 897 | IP_VS_DBG(7, "%s: pid %d\n", __FUNCTION__, task_pid_nr(current)); |
1da177e4 LT |
898 | IP_VS_DBG(7, "Each ip_vs_sync_conn entry need %Zd bytes\n", |
899 | sizeof(struct ip_vs_sync_conn)); | |
900 | ||
901 | ip_vs_sync_state |= state; | |
902 | if (state == IP_VS_STATE_MASTER) { | |
37004af3 SH |
903 | strlcpy(ip_vs_master_mcast_ifn, mcast_ifn, |
904 | sizeof(ip_vs_master_mcast_ifn)); | |
1da177e4 LT |
905 | ip_vs_master_syncid = syncid; |
906 | } else { | |
37004af3 SH |
907 | strlcpy(ip_vs_backup_mcast_ifn, mcast_ifn, |
908 | sizeof(ip_vs_backup_mcast_ifn)); | |
1da177e4 LT |
909 | ip_vs_backup_syncid = syncid; |
910 | } | |
911 | ||
cc0191ae NH |
912 | tinfo->state = state; |
913 | tinfo->startup = &startup; | |
914 | ||
1da177e4 | 915 | repeat: |
cc0191ae | 916 | if ((pid = kernel_thread(fork_sync_thread, tinfo, 0)) < 0) { |
1da177e4 LT |
917 | IP_VS_ERR("could not create fork_sync_thread due to %d... " |
918 | "retrying.\n", pid); | |
89eaeb09 | 919 | msleep_interruptible(1000); |
1da177e4 LT |
920 | goto repeat; |
921 | } | |
922 | ||
923 | wait_for_completion(&startup); | |
924 | ||
925 | return 0; | |
926 | } | |
927 | ||
928 | ||
929 | int stop_sync_thread(int state) | |
930 | { | |
931 | DECLARE_WAITQUEUE(wait, current); | |
932 | ||
933 | if ((state == IP_VS_STATE_MASTER && !sync_master_pid) || | |
934 | (state == IP_VS_STATE_BACKUP && !sync_backup_pid)) | |
935 | return -ESRCH; | |
936 | ||
ba25f9dc | 937 | IP_VS_DBG(7, "%s: pid %d\n", __FUNCTION__, task_pid_nr(current)); |
1da177e4 | 938 | IP_VS_INFO("stopping sync thread %d ...\n", |
37004af3 SH |
939 | (state == IP_VS_STATE_MASTER) ? |
940 | sync_master_pid : sync_backup_pid); | |
1da177e4 LT |
941 | |
942 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
943 | add_wait_queue(&stop_sync_wait, &wait); | |
944 | set_stop_sync(state, 1); | |
945 | ip_vs_sync_state -= state; | |
946 | wake_up(&sync_wait); | |
947 | schedule(); | |
948 | __set_current_state(TASK_RUNNING); | |
949 | remove_wait_queue(&stop_sync_wait, &wait); | |
950 | ||
951 | /* Note: no need to reap the sync thread, because its parent | |
952 | process is the init process */ | |
953 | ||
954 | if ((state == IP_VS_STATE_MASTER && stop_master_sync) || | |
955 | (state == IP_VS_STATE_BACKUP && stop_backup_sync)) | |
956 | IP_VS_BUG(); | |
957 | ||
958 | return 0; | |
959 | } |