]>
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 | * | |
1da177e4 LT |
8 | * Authors: Wensong Zhang <wensong@linuxvirtualserver.org> |
9 | * | |
10 | * ip_vs_sync: sync connection info from master load balancer to backups | |
11 | * through multicast | |
12 | * | |
13 | * Changes: | |
14 | * Alexandre Cassen : Added master & backup support at a time. | |
15 | * Alexandre Cassen : Added SyncID support for incoming sync | |
16 | * messages filtering. | |
17 | * Justin Ossevoort : Fix endian problem on sync message size. | |
18 | */ | |
19 | ||
20 | #include <linux/module.h> | |
21 | #include <linux/slab.h> | |
14c85021 | 22 | #include <linux/inetdevice.h> |
1da177e4 LT |
23 | #include <linux/net.h> |
24 | #include <linux/completion.h> | |
25 | #include <linux/delay.h> | |
26 | #include <linux/skbuff.h> | |
27 | #include <linux/in.h> | |
28 | #include <linux/igmp.h> /* for ip_mc_join_group */ | |
14c85021 | 29 | #include <linux/udp.h> |
e6dd731c | 30 | #include <linux/err.h> |
998e7a76 | 31 | #include <linux/kthread.h> |
ba6fd850 | 32 | #include <linux/wait.h> |
1da177e4 LT |
33 | |
34 | #include <net/ip.h> | |
35 | #include <net/sock.h> | |
1da177e4 LT |
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 | 70 | struct ip_vs_sync_thread_data { |
998e7a76 SW |
71 | struct socket *sock; |
72 | char *buf; | |
cc0191ae NH |
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 | ||
998e7a76 SW |
143 | /* sync daemon tasks */ |
144 | static struct task_struct *sync_master_thread; | |
145 | static struct task_struct *sync_backup_thread; | |
146 | ||
1da177e4 | 147 | /* multicast addr */ |
d5640050 SW |
148 | static struct sockaddr_in mcast_addr = { |
149 | .sin_family = AF_INET, | |
150 | .sin_port = __constant_htons(IP_VS_SYNC_PORT), | |
151 | .sin_addr.s_addr = __constant_htonl(IP_VS_SYNC_GROUP), | |
152 | }; | |
1da177e4 LT |
153 | |
154 | ||
998e7a76 | 155 | static inline struct ip_vs_sync_buff *sb_dequeue(void) |
1da177e4 LT |
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 | ||
998e7a76 SW |
199 | static inline void sb_queue_tail(struct ip_vs_sync_buff *sb) |
200 | { | |
201 | spin_lock(&ip_vs_sync_lock); | |
202 | if (ip_vs_sync_state & IP_VS_STATE_MASTER) | |
203 | list_add_tail(&sb->list, &ip_vs_sync_queue); | |
204 | else | |
205 | ip_vs_sync_buff_release(sb); | |
206 | spin_unlock(&ip_vs_sync_lock); | |
207 | } | |
208 | ||
1da177e4 LT |
209 | /* |
210 | * Get the current sync buffer if it has been created for more | |
211 | * than the specified time or the specified time is zero. | |
212 | */ | |
213 | static inline struct ip_vs_sync_buff * | |
214 | get_curr_sync_buff(unsigned long time) | |
215 | { | |
216 | struct ip_vs_sync_buff *sb; | |
217 | ||
218 | spin_lock_bh(&curr_sb_lock); | |
219 | if (curr_sb && (time == 0 || | |
220 | time_before(jiffies - curr_sb->firstuse, time))) { | |
221 | sb = curr_sb; | |
222 | curr_sb = NULL; | |
223 | } else | |
224 | sb = NULL; | |
225 | spin_unlock_bh(&curr_sb_lock); | |
226 | return sb; | |
227 | } | |
228 | ||
229 | ||
230 | /* | |
231 | * Add an ip_vs_conn information into the current sync_buff. | |
232 | * Called by ip_vs_in. | |
233 | */ | |
234 | void ip_vs_sync_conn(struct ip_vs_conn *cp) | |
235 | { | |
236 | struct ip_vs_sync_mesg *m; | |
237 | struct ip_vs_sync_conn *s; | |
238 | int len; | |
239 | ||
240 | spin_lock(&curr_sb_lock); | |
241 | if (!curr_sb) { | |
242 | if (!(curr_sb=ip_vs_sync_buff_create())) { | |
243 | spin_unlock(&curr_sb_lock); | |
244 | IP_VS_ERR("ip_vs_sync_buff_create failed.\n"); | |
245 | return; | |
246 | } | |
247 | } | |
248 | ||
249 | len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE : | |
250 | SIMPLE_CONN_SIZE; | |
251 | m = curr_sb->mesg; | |
252 | s = (struct ip_vs_sync_conn *)curr_sb->head; | |
253 | ||
254 | /* copy members */ | |
255 | s->protocol = cp->protocol; | |
256 | s->cport = cp->cport; | |
257 | s->vport = cp->vport; | |
258 | s->dport = cp->dport; | |
e7ade46a JV |
259 | s->caddr = cp->caddr.ip; |
260 | s->vaddr = cp->vaddr.ip; | |
261 | s->daddr = cp->daddr.ip; | |
1da177e4 LT |
262 | s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED); |
263 | s->state = htons(cp->state); | |
264 | if (cp->flags & IP_VS_CONN_F_SEQ_MASK) { | |
265 | struct ip_vs_sync_conn_options *opt = | |
266 | (struct ip_vs_sync_conn_options *)&s[1]; | |
267 | memcpy(opt, &cp->in_seq, sizeof(*opt)); | |
268 | } | |
269 | ||
270 | m->nr_conns++; | |
271 | m->size += len; | |
272 | curr_sb->head += len; | |
273 | ||
274 | /* check if there is a space for next one */ | |
275 | if (curr_sb->head+FULL_CONN_SIZE > curr_sb->end) { | |
276 | sb_queue_tail(curr_sb); | |
277 | curr_sb = NULL; | |
278 | } | |
279 | spin_unlock(&curr_sb_lock); | |
280 | ||
281 | /* synchronize its controller if it has */ | |
282 | if (cp->control) | |
283 | ip_vs_sync_conn(cp->control); | |
284 | } | |
285 | ||
286 | ||
287 | /* | |
288 | * Process received multicast message and create the corresponding | |
289 | * ip_vs_conn entries. | |
290 | */ | |
291 | static void ip_vs_process_message(const char *buffer, const size_t buflen) | |
292 | { | |
293 | struct ip_vs_sync_mesg *m = (struct ip_vs_sync_mesg *)buffer; | |
294 | struct ip_vs_sync_conn *s; | |
295 | struct ip_vs_sync_conn_options *opt; | |
296 | struct ip_vs_conn *cp; | |
5c81833c | 297 | struct ip_vs_protocol *pp; |
1e356f9c | 298 | struct ip_vs_dest *dest; |
1da177e4 LT |
299 | char *p; |
300 | int i; | |
301 | ||
2ad17def JA |
302 | if (buflen < sizeof(struct ip_vs_sync_mesg)) { |
303 | IP_VS_ERR_RL("sync message header too short\n"); | |
304 | return; | |
305 | } | |
306 | ||
1da177e4 LT |
307 | /* Convert size back to host byte order */ |
308 | m->size = ntohs(m->size); | |
309 | ||
310 | if (buflen != m->size) { | |
2ad17def | 311 | IP_VS_ERR_RL("bogus sync message size\n"); |
1da177e4 LT |
312 | return; |
313 | } | |
314 | ||
315 | /* SyncID sanity check */ | |
316 | if (ip_vs_backup_syncid != 0 && m->syncid != ip_vs_backup_syncid) { | |
317 | IP_VS_DBG(7, "Ignoring incoming msg with syncid = %d\n", | |
318 | m->syncid); | |
319 | return; | |
320 | } | |
321 | ||
322 | p = (char *)buffer + sizeof(struct ip_vs_sync_mesg); | |
323 | for (i=0; i<m->nr_conns; i++) { | |
b209639e | 324 | unsigned flags, state; |
87375ab4 | 325 | |
2ad17def JA |
326 | if (p + SIMPLE_CONN_SIZE > buffer+buflen) { |
327 | IP_VS_ERR_RL("bogus conn in sync message\n"); | |
328 | return; | |
329 | } | |
330 | s = (struct ip_vs_sync_conn *) p; | |
7a4fbb1f | 331 | flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC; |
2ad17def JA |
332 | flags &= ~IP_VS_CONN_F_HASHED; |
333 | if (flags & IP_VS_CONN_F_SEQ_MASK) { | |
334 | opt = (struct ip_vs_sync_conn_options *)&s[1]; | |
335 | p += FULL_CONN_SIZE; | |
336 | if (p > buffer+buflen) { | |
337 | IP_VS_ERR_RL("bogus conn options in sync message\n"); | |
338 | return; | |
339 | } | |
340 | } else { | |
341 | opt = NULL; | |
342 | p += SIMPLE_CONN_SIZE; | |
343 | } | |
344 | ||
b209639e | 345 | state = ntohs(s->state); |
2ad17def JA |
346 | if (!(flags & IP_VS_CONN_F_TEMPLATE)) { |
347 | pp = ip_vs_proto_get(s->protocol); | |
348 | if (!pp) { | |
349 | IP_VS_ERR_RL("Unsupported protocol %u in sync msg\n", | |
350 | s->protocol); | |
351 | continue; | |
352 | } | |
353 | if (state >= pp->num_states) { | |
354 | IP_VS_DBG(2, "Invalid %s state %u in sync msg\n", | |
355 | pp->name, state); | |
356 | continue; | |
357 | } | |
358 | } else { | |
359 | /* protocol in templates is not used for state/timeout */ | |
360 | pp = NULL; | |
361 | if (state > 0) { | |
362 | IP_VS_DBG(2, "Invalid template state %u in sync msg\n", | |
363 | state); | |
364 | state = 0; | |
365 | } | |
366 | } | |
367 | ||
87375ab4 | 368 | if (!(flags & IP_VS_CONN_F_TEMPLATE)) |
28364a59 JV |
369 | cp = ip_vs_conn_in_get(AF_INET, s->protocol, |
370 | (union nf_inet_addr *)&s->caddr, | |
371 | s->cport, | |
372 | (union nf_inet_addr *)&s->vaddr, | |
373 | s->vport); | |
87375ab4 | 374 | else |
28364a59 JV |
375 | cp = ip_vs_ct_in_get(AF_INET, s->protocol, |
376 | (union nf_inet_addr *)&s->caddr, | |
377 | s->cport, | |
378 | (union nf_inet_addr *)&s->vaddr, | |
379 | s->vport); | |
1da177e4 | 380 | if (!cp) { |
1e356f9c RB |
381 | /* |
382 | * Find the appropriate destination for the connection. | |
383 | * If it is not found the connection will remain unbound | |
384 | * but still handled. | |
385 | */ | |
7937df15 JV |
386 | dest = ip_vs_find_dest(AF_INET, |
387 | (union nf_inet_addr *)&s->daddr, | |
388 | s->dport, | |
389 | (union nf_inet_addr *)&s->vaddr, | |
390 | s->vport, | |
1e356f9c | 391 | s->protocol); |
b209639e RB |
392 | /* Set the approprite ativity flag */ |
393 | if (s->protocol == IPPROTO_TCP) { | |
394 | if (state != IP_VS_TCP_S_ESTABLISHED) | |
395 | flags |= IP_VS_CONN_F_INACTIVE; | |
396 | else | |
397 | flags &= ~IP_VS_CONN_F_INACTIVE; | |
398 | } | |
28364a59 | 399 | cp = ip_vs_conn_new(AF_INET, s->protocol, |
cd9fe6c4 | 400 | (union nf_inet_addr *)&s->caddr, |
28364a59 | 401 | s->cport, |
cd9fe6c4 | 402 | (union nf_inet_addr *)&s->vaddr, |
28364a59 | 403 | s->vport, |
cd9fe6c4 | 404 | (union nf_inet_addr *)&s->daddr, |
28364a59 | 405 | s->dport, |
1e356f9c RB |
406 | flags, dest); |
407 | if (dest) | |
408 | atomic_dec(&dest->refcnt); | |
1da177e4 LT |
409 | if (!cp) { |
410 | IP_VS_ERR("ip_vs_conn_new failed\n"); | |
411 | return; | |
412 | } | |
1da177e4 | 413 | } else if (!cp->dest) { |
1e356f9c | 414 | dest = ip_vs_try_bind_dest(cp); |
2ad17def | 415 | if (dest) |
1e356f9c | 416 | atomic_dec(&dest->refcnt); |
b209639e RB |
417 | } else if ((cp->dest) && (cp->protocol == IPPROTO_TCP) && |
418 | (cp->state != state)) { | |
419 | /* update active/inactive flag for the connection */ | |
420 | dest = cp->dest; | |
421 | if (!(cp->flags & IP_VS_CONN_F_INACTIVE) && | |
422 | (state != IP_VS_TCP_S_ESTABLISHED)) { | |
423 | atomic_dec(&dest->activeconns); | |
424 | atomic_inc(&dest->inactconns); | |
425 | cp->flags |= IP_VS_CONN_F_INACTIVE; | |
426 | } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) && | |
427 | (state == IP_VS_TCP_S_ESTABLISHED)) { | |
428 | atomic_inc(&dest->activeconns); | |
429 | atomic_dec(&dest->inactconns); | |
430 | cp->flags &= ~IP_VS_CONN_F_INACTIVE; | |
431 | } | |
432 | } | |
1da177e4 | 433 | |
2ad17def | 434 | if (opt) |
1da177e4 | 435 | memcpy(&cp->in_seq, opt, sizeof(*opt)); |
1da177e4 | 436 | atomic_set(&cp->in_pkts, sysctl_ip_vs_sync_threshold[0]); |
b209639e | 437 | cp->state = state; |
2ad17def JA |
438 | cp->old_state = cp->state; |
439 | /* | |
440 | * We can not recover the right timeout for templates | |
441 | * in all cases, we can not find the right fwmark | |
442 | * virtual service. If needed, we can do it for | |
443 | * non-fwmark persistent services. | |
444 | */ | |
445 | if (!(flags & IP_VS_CONN_F_TEMPLATE) && pp->timeout_table) | |
446 | cp->timeout = pp->timeout_table[state]; | |
447 | else | |
448 | cp->timeout = (3*60*HZ); | |
1da177e4 | 449 | ip_vs_conn_put(cp); |
1da177e4 LT |
450 | } |
451 | } | |
452 | ||
453 | ||
454 | /* | |
455 | * Setup loopback of outgoing multicasts on a sending socket | |
456 | */ | |
457 | static void set_mcast_loop(struct sock *sk, u_char loop) | |
458 | { | |
459 | struct inet_sock *inet = inet_sk(sk); | |
460 | ||
461 | /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */ | |
462 | lock_sock(sk); | |
463 | inet->mc_loop = loop ? 1 : 0; | |
464 | release_sock(sk); | |
465 | } | |
466 | ||
467 | /* | |
468 | * Specify TTL for outgoing multicasts on a sending socket | |
469 | */ | |
470 | static void set_mcast_ttl(struct sock *sk, u_char ttl) | |
471 | { | |
472 | struct inet_sock *inet = inet_sk(sk); | |
473 | ||
474 | /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */ | |
475 | lock_sock(sk); | |
476 | inet->mc_ttl = ttl; | |
477 | release_sock(sk); | |
478 | } | |
479 | ||
480 | /* | |
481 | * Specifiy default interface for outgoing multicasts | |
482 | */ | |
483 | static int set_mcast_if(struct sock *sk, char *ifname) | |
484 | { | |
485 | struct net_device *dev; | |
486 | struct inet_sock *inet = inet_sk(sk); | |
487 | ||
881d966b | 488 | if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL) |
1da177e4 LT |
489 | return -ENODEV; |
490 | ||
491 | if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) | |
492 | return -EINVAL; | |
493 | ||
494 | lock_sock(sk); | |
495 | inet->mc_index = dev->ifindex; | |
496 | /* inet->mc_addr = 0; */ | |
497 | release_sock(sk); | |
498 | ||
499 | return 0; | |
500 | } | |
501 | ||
502 | ||
503 | /* | |
504 | * Set the maximum length of sync message according to the | |
505 | * specified interface's MTU. | |
506 | */ | |
507 | static int set_sync_mesg_maxlen(int sync_state) | |
508 | { | |
509 | struct net_device *dev; | |
510 | int num; | |
511 | ||
512 | if (sync_state == IP_VS_STATE_MASTER) { | |
881d966b | 513 | if ((dev = __dev_get_by_name(&init_net, ip_vs_master_mcast_ifn)) == NULL) |
1da177e4 LT |
514 | return -ENODEV; |
515 | ||
516 | num = (dev->mtu - sizeof(struct iphdr) - | |
517 | sizeof(struct udphdr) - | |
518 | SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE; | |
519 | sync_send_mesg_maxlen = | |
520 | SYNC_MESG_HEADER_LEN + SIMPLE_CONN_SIZE * num; | |
521 | IP_VS_DBG(7, "setting the maximum length of sync sending " | |
522 | "message %d.\n", sync_send_mesg_maxlen); | |
523 | } else if (sync_state == IP_VS_STATE_BACKUP) { | |
881d966b | 524 | if ((dev = __dev_get_by_name(&init_net, ip_vs_backup_mcast_ifn)) == NULL) |
1da177e4 LT |
525 | return -ENODEV; |
526 | ||
527 | sync_recv_mesg_maxlen = dev->mtu - | |
528 | sizeof(struct iphdr) - sizeof(struct udphdr); | |
529 | IP_VS_DBG(7, "setting the maximum length of sync receiving " | |
530 | "message %d.\n", sync_recv_mesg_maxlen); | |
531 | } | |
532 | ||
533 | return 0; | |
534 | } | |
535 | ||
536 | ||
537 | /* | |
538 | * Join a multicast group. | |
539 | * the group is specified by a class D multicast address 224.0.0.0/8 | |
540 | * in the in_addr structure passed in as a parameter. | |
541 | */ | |
542 | static int | |
543 | join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname) | |
544 | { | |
545 | struct ip_mreqn mreq; | |
546 | struct net_device *dev; | |
547 | int ret; | |
548 | ||
549 | memset(&mreq, 0, sizeof(mreq)); | |
550 | memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr)); | |
551 | ||
881d966b | 552 | if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL) |
1da177e4 LT |
553 | return -ENODEV; |
554 | if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) | |
555 | return -EINVAL; | |
556 | ||
557 | mreq.imr_ifindex = dev->ifindex; | |
558 | ||
559 | lock_sock(sk); | |
560 | ret = ip_mc_join_group(sk, &mreq); | |
561 | release_sock(sk); | |
562 | ||
563 | return ret; | |
564 | } | |
565 | ||
566 | ||
567 | static int bind_mcastif_addr(struct socket *sock, char *ifname) | |
568 | { | |
569 | struct net_device *dev; | |
a61ced5d | 570 | __be32 addr; |
1da177e4 LT |
571 | struct sockaddr_in sin; |
572 | ||
881d966b | 573 | if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL) |
1da177e4 LT |
574 | return -ENODEV; |
575 | ||
576 | addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE); | |
577 | if (!addr) | |
578 | IP_VS_ERR("You probably need to specify IP address on " | |
579 | "multicast interface.\n"); | |
580 | ||
581 | IP_VS_DBG(7, "binding socket with (%s) %u.%u.%u.%u\n", | |
582 | ifname, NIPQUAD(addr)); | |
583 | ||
584 | /* Now bind the socket with the address of multicast interface */ | |
585 | sin.sin_family = AF_INET; | |
586 | sin.sin_addr.s_addr = addr; | |
587 | sin.sin_port = 0; | |
588 | ||
589 | return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin)); | |
590 | } | |
591 | ||
592 | /* | |
593 | * Set up sending multicast socket over UDP | |
594 | */ | |
595 | static struct socket * make_send_sock(void) | |
596 | { | |
597 | struct socket *sock; | |
e6dd731c | 598 | int result; |
1da177e4 LT |
599 | |
600 | /* First create a socket */ | |
e6dd731c SW |
601 | result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock); |
602 | if (result < 0) { | |
1da177e4 | 603 | IP_VS_ERR("Error during creation of socket; terminating\n"); |
e6dd731c | 604 | return ERR_PTR(result); |
1da177e4 LT |
605 | } |
606 | ||
e6dd731c SW |
607 | result = set_mcast_if(sock->sk, ip_vs_master_mcast_ifn); |
608 | if (result < 0) { | |
1da177e4 LT |
609 | IP_VS_ERR("Error setting outbound mcast interface\n"); |
610 | goto error; | |
611 | } | |
612 | ||
613 | set_mcast_loop(sock->sk, 0); | |
614 | set_mcast_ttl(sock->sk, 1); | |
615 | ||
e6dd731c SW |
616 | result = bind_mcastif_addr(sock, ip_vs_master_mcast_ifn); |
617 | if (result < 0) { | |
1da177e4 LT |
618 | IP_VS_ERR("Error binding address of the mcast interface\n"); |
619 | goto error; | |
620 | } | |
621 | ||
e6dd731c SW |
622 | result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr, |
623 | sizeof(struct sockaddr), 0); | |
624 | if (result < 0) { | |
1da177e4 LT |
625 | IP_VS_ERR("Error connecting to the multicast addr\n"); |
626 | goto error; | |
627 | } | |
628 | ||
629 | return sock; | |
630 | ||
631 | error: | |
632 | sock_release(sock); | |
e6dd731c | 633 | return ERR_PTR(result); |
1da177e4 LT |
634 | } |
635 | ||
636 | ||
637 | /* | |
638 | * Set up receiving multicast socket over UDP | |
639 | */ | |
640 | static struct socket * make_receive_sock(void) | |
641 | { | |
642 | struct socket *sock; | |
e6dd731c | 643 | int result; |
1da177e4 LT |
644 | |
645 | /* First create a socket */ | |
e6dd731c SW |
646 | result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock); |
647 | if (result < 0) { | |
1da177e4 | 648 | IP_VS_ERR("Error during creation of socket; terminating\n"); |
e6dd731c | 649 | return ERR_PTR(result); |
1da177e4 LT |
650 | } |
651 | ||
652 | /* it is equivalent to the REUSEADDR option in user-space */ | |
653 | sock->sk->sk_reuse = 1; | |
654 | ||
e6dd731c SW |
655 | result = sock->ops->bind(sock, (struct sockaddr *) &mcast_addr, |
656 | sizeof(struct sockaddr)); | |
657 | if (result < 0) { | |
1da177e4 LT |
658 | IP_VS_ERR("Error binding to the multicast addr\n"); |
659 | goto error; | |
660 | } | |
661 | ||
662 | /* join the multicast group */ | |
e6dd731c SW |
663 | result = join_mcast_group(sock->sk, |
664 | (struct in_addr *) &mcast_addr.sin_addr, | |
665 | ip_vs_backup_mcast_ifn); | |
666 | if (result < 0) { | |
1da177e4 LT |
667 | IP_VS_ERR("Error joining to the multicast group\n"); |
668 | goto error; | |
669 | } | |
670 | ||
671 | return sock; | |
672 | ||
673 | error: | |
674 | sock_release(sock); | |
e6dd731c | 675 | return ERR_PTR(result); |
1da177e4 LT |
676 | } |
677 | ||
678 | ||
679 | static int | |
680 | ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length) | |
681 | { | |
682 | struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL}; | |
683 | struct kvec iov; | |
684 | int len; | |
685 | ||
686 | EnterFunction(7); | |
687 | iov.iov_base = (void *)buffer; | |
688 | iov.iov_len = length; | |
689 | ||
690 | len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length)); | |
691 | ||
692 | LeaveFunction(7); | |
693 | return len; | |
694 | } | |
695 | ||
696 | static void | |
697 | ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg) | |
698 | { | |
699 | int msize; | |
700 | ||
701 | msize = msg->size; | |
702 | ||
703 | /* Put size in network byte order */ | |
704 | msg->size = htons(msg->size); | |
705 | ||
706 | if (ip_vs_send_async(sock, (char *)msg, msize) != msize) | |
707 | IP_VS_ERR("ip_vs_send_async error\n"); | |
708 | } | |
709 | ||
710 | static int | |
711 | ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen) | |
712 | { | |
713 | struct msghdr msg = {NULL,}; | |
714 | struct kvec iov; | |
715 | int len; | |
716 | ||
717 | EnterFunction(7); | |
718 | ||
719 | /* Receive a packet */ | |
720 | iov.iov_base = buffer; | |
721 | iov.iov_len = (size_t)buflen; | |
722 | ||
723 | len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, 0); | |
724 | ||
725 | if (len < 0) | |
726 | return -1; | |
727 | ||
728 | LeaveFunction(7); | |
729 | return len; | |
730 | } | |
731 | ||
732 | ||
998e7a76 | 733 | static int sync_thread_master(void *data) |
1da177e4 | 734 | { |
998e7a76 | 735 | struct ip_vs_sync_thread_data *tinfo = data; |
1da177e4 LT |
736 | struct ip_vs_sync_buff *sb; |
737 | ||
1da177e4 LT |
738 | IP_VS_INFO("sync thread started: state = MASTER, mcast_ifn = %s, " |
739 | "syncid = %d\n", | |
740 | ip_vs_master_mcast_ifn, ip_vs_master_syncid); | |
741 | ||
998e7a76 SW |
742 | while (!kthread_should_stop()) { |
743 | while ((sb = sb_dequeue())) { | |
744 | ip_vs_send_sync_msg(tinfo->sock, sb->mesg); | |
1da177e4 LT |
745 | ip_vs_sync_buff_release(sb); |
746 | } | |
747 | ||
748 | /* check if entries stay in curr_sb for 2 seconds */ | |
998e7a76 SW |
749 | sb = get_curr_sync_buff(2 * HZ); |
750 | if (sb) { | |
751 | ip_vs_send_sync_msg(tinfo->sock, sb->mesg); | |
1da177e4 LT |
752 | ip_vs_sync_buff_release(sb); |
753 | } | |
754 | ||
375c6bba | 755 | schedule_timeout_interruptible(HZ); |
1da177e4 LT |
756 | } |
757 | ||
758 | /* clean up the sync_buff queue */ | |
759 | while ((sb=sb_dequeue())) { | |
760 | ip_vs_sync_buff_release(sb); | |
761 | } | |
762 | ||
763 | /* clean up the current sync_buff */ | |
764 | if ((sb = get_curr_sync_buff(0))) { | |
765 | ip_vs_sync_buff_release(sb); | |
766 | } | |
767 | ||
768 | /* release the sending multicast socket */ | |
998e7a76 SW |
769 | sock_release(tinfo->sock); |
770 | kfree(tinfo); | |
771 | ||
772 | return 0; | |
1da177e4 LT |
773 | } |
774 | ||
775 | ||
998e7a76 | 776 | static int sync_thread_backup(void *data) |
1da177e4 | 777 | { |
998e7a76 | 778 | struct ip_vs_sync_thread_data *tinfo = data; |
1da177e4 LT |
779 | int len; |
780 | ||
1da177e4 LT |
781 | IP_VS_INFO("sync thread started: state = BACKUP, mcast_ifn = %s, " |
782 | "syncid = %d\n", | |
783 | ip_vs_backup_mcast_ifn, ip_vs_backup_syncid); | |
784 | ||
998e7a76 | 785 | while (!kthread_should_stop()) { |
ba6fd850 SW |
786 | wait_event_interruptible(*tinfo->sock->sk->sk_sleep, |
787 | !skb_queue_empty(&tinfo->sock->sk->sk_receive_queue) | |
788 | || kthread_should_stop()); | |
789 | ||
998e7a76 SW |
790 | /* do we have data now? */ |
791 | while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) { | |
792 | len = ip_vs_receive(tinfo->sock, tinfo->buf, | |
793 | sync_recv_mesg_maxlen); | |
794 | if (len <= 0) { | |
1da177e4 LT |
795 | IP_VS_ERR("receiving message error\n"); |
796 | break; | |
797 | } | |
998e7a76 SW |
798 | |
799 | /* disable bottom half, because it accesses the data | |
1da177e4 LT |
800 | shared by softirq while getting/creating conns */ |
801 | local_bh_disable(); | |
998e7a76 | 802 | ip_vs_process_message(tinfo->buf, len); |
1da177e4 LT |
803 | local_bh_enable(); |
804 | } | |
1da177e4 LT |
805 | } |
806 | ||
807 | /* release the sending multicast socket */ | |
998e7a76 SW |
808 | sock_release(tinfo->sock); |
809 | kfree(tinfo->buf); | |
810 | kfree(tinfo); | |
1da177e4 | 811 | |
998e7a76 | 812 | return 0; |
1da177e4 LT |
813 | } |
814 | ||
815 | ||
998e7a76 | 816 | int start_sync_thread(int state, char *mcast_ifn, __u8 syncid) |
1da177e4 | 817 | { |
998e7a76 SW |
818 | struct ip_vs_sync_thread_data *tinfo; |
819 | struct task_struct **realtask, *task; | |
820 | struct socket *sock; | |
821 | char *name, *buf = NULL; | |
822 | int (*threadfn)(void *data); | |
823 | int result = -ENOMEM; | |
1da177e4 | 824 | |
998e7a76 SW |
825 | IP_VS_DBG(7, "%s: pid %d\n", __func__, task_pid_nr(current)); |
826 | IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n", | |
827 | sizeof(struct ip_vs_sync_conn)); | |
1da177e4 | 828 | |
998e7a76 SW |
829 | if (state == IP_VS_STATE_MASTER) { |
830 | if (sync_master_thread) | |
831 | return -EEXIST; | |
1da177e4 | 832 | |
998e7a76 SW |
833 | strlcpy(ip_vs_master_mcast_ifn, mcast_ifn, |
834 | sizeof(ip_vs_master_mcast_ifn)); | |
835 | ip_vs_master_syncid = syncid; | |
836 | realtask = &sync_master_thread; | |
1da177e4 | 837 | name = "ipvs_syncmaster"; |
998e7a76 SW |
838 | threadfn = sync_thread_master; |
839 | sock = make_send_sock(); | |
840 | } else if (state == IP_VS_STATE_BACKUP) { | |
841 | if (sync_backup_thread) | |
842 | return -EEXIST; | |
843 | ||
844 | strlcpy(ip_vs_backup_mcast_ifn, mcast_ifn, | |
845 | sizeof(ip_vs_backup_mcast_ifn)); | |
846 | ip_vs_backup_syncid = syncid; | |
847 | realtask = &sync_backup_thread; | |
1da177e4 | 848 | name = "ipvs_syncbackup"; |
998e7a76 SW |
849 | threadfn = sync_thread_backup; |
850 | sock = make_receive_sock(); | |
1da177e4 | 851 | } else { |
1da177e4 LT |
852 | return -EINVAL; |
853 | } | |
854 | ||
998e7a76 SW |
855 | if (IS_ERR(sock)) { |
856 | result = PTR_ERR(sock); | |
857 | goto out; | |
858 | } | |
1da177e4 | 859 | |
1da177e4 | 860 | set_sync_mesg_maxlen(state); |
998e7a76 SW |
861 | if (state == IP_VS_STATE_BACKUP) { |
862 | buf = kmalloc(sync_recv_mesg_maxlen, GFP_KERNEL); | |
863 | if (!buf) | |
864 | goto outsocket; | |
865 | } | |
1da177e4 | 866 | |
998e7a76 SW |
867 | tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL); |
868 | if (!tinfo) | |
869 | goto outbuf; | |
cc0191ae | 870 | |
998e7a76 SW |
871 | tinfo->sock = sock; |
872 | tinfo->buf = buf; | |
1da177e4 | 873 | |
998e7a76 SW |
874 | task = kthread_run(threadfn, tinfo, name); |
875 | if (IS_ERR(task)) { | |
876 | result = PTR_ERR(task); | |
877 | goto outtinfo; | |
878 | } | |
1da177e4 | 879 | |
998e7a76 SW |
880 | /* mark as active */ |
881 | *realtask = task; | |
882 | ip_vs_sync_state |= state; | |
1da177e4 | 883 | |
998e7a76 SW |
884 | /* increase the module use count */ |
885 | ip_vs_use_count_inc(); | |
1da177e4 LT |
886 | |
887 | return 0; | |
1da177e4 | 888 | |
998e7a76 SW |
889 | outtinfo: |
890 | kfree(tinfo); | |
891 | outbuf: | |
892 | kfree(buf); | |
893 | outsocket: | |
894 | sock_release(sock); | |
895 | out: | |
896 | return result; | |
1da177e4 LT |
897 | } |
898 | ||
899 | ||
998e7a76 | 900 | int stop_sync_thread(int state) |
1da177e4 | 901 | { |
0dc47877 | 902 | IP_VS_DBG(7, "%s: pid %d\n", __func__, task_pid_nr(current)); |
1da177e4 | 903 | |
1da177e4 | 904 | if (state == IP_VS_STATE_MASTER) { |
998e7a76 SW |
905 | if (!sync_master_thread) |
906 | return -ESRCH; | |
cc0191ae | 907 | |
998e7a76 SW |
908 | IP_VS_INFO("stopping master sync thread %d ...\n", |
909 | task_pid_nr(sync_master_thread)); | |
1da177e4 | 910 | |
998e7a76 SW |
911 | /* |
912 | * The lock synchronizes with sb_queue_tail(), so that we don't | |
913 | * add sync buffers to the queue, when we are already in | |
914 | * progress of stopping the master sync daemon. | |
915 | */ | |
1da177e4 | 916 | |
bc0fde2f | 917 | spin_lock_bh(&ip_vs_sync_lock); |
998e7a76 | 918 | ip_vs_sync_state &= ~IP_VS_STATE_MASTER; |
bc0fde2f | 919 | spin_unlock_bh(&ip_vs_sync_lock); |
998e7a76 SW |
920 | kthread_stop(sync_master_thread); |
921 | sync_master_thread = NULL; | |
922 | } else if (state == IP_VS_STATE_BACKUP) { | |
923 | if (!sync_backup_thread) | |
924 | return -ESRCH; | |
925 | ||
926 | IP_VS_INFO("stopping backup sync thread %d ...\n", | |
927 | task_pid_nr(sync_backup_thread)); | |
928 | ||
929 | ip_vs_sync_state &= ~IP_VS_STATE_BACKUP; | |
930 | kthread_stop(sync_backup_thread); | |
931 | sync_backup_thread = NULL; | |
932 | } else { | |
933 | return -EINVAL; | |
934 | } | |
1da177e4 | 935 | |
998e7a76 SW |
936 | /* decrease the module use count */ |
937 | ip_vs_use_count_dec(); | |
1da177e4 LT |
938 | |
939 | return 0; | |
940 | } |