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> |
1da177e4 LT |
32 | |
33 | #include <net/ip.h> | |
34 | #include <net/sock.h> | |
1da177e4 LT |
35 | |
36 | #include <net/ip_vs.h> | |
37 | ||
38 | #define IP_VS_SYNC_GROUP 0xe0000051 /* multicast addr - 224.0.0.81 */ | |
39 | #define IP_VS_SYNC_PORT 8848 /* multicast port */ | |
40 | ||
41 | ||
42 | /* | |
43 | * IPVS sync connection entry | |
44 | */ | |
45 | struct ip_vs_sync_conn { | |
46 | __u8 reserved; | |
47 | ||
48 | /* Protocol, addresses and port numbers */ | |
49 | __u8 protocol; /* Which protocol (TCP/UDP) */ | |
014d730d AV |
50 | __be16 cport; |
51 | __be16 vport; | |
52 | __be16 dport; | |
53 | __be32 caddr; /* client address */ | |
54 | __be32 vaddr; /* virtual address */ | |
55 | __be32 daddr; /* destination address */ | |
1da177e4 LT |
56 | |
57 | /* Flags and state transition */ | |
014d730d AV |
58 | __be16 flags; /* status flags */ |
59 | __be16 state; /* state info */ | |
1da177e4 LT |
60 | |
61 | /* The sequence options start here */ | |
62 | }; | |
63 | ||
64 | struct ip_vs_sync_conn_options { | |
65 | struct ip_vs_seq in_seq; /* incoming seq. struct */ | |
66 | struct ip_vs_seq out_seq; /* outgoing seq. struct */ | |
67 | }; | |
68 | ||
cc0191ae | 69 | struct ip_vs_sync_thread_data { |
998e7a76 SW |
70 | struct socket *sock; |
71 | char *buf; | |
cc0191ae NH |
72 | }; |
73 | ||
1da177e4 LT |
74 | #define SIMPLE_CONN_SIZE (sizeof(struct ip_vs_sync_conn)) |
75 | #define FULL_CONN_SIZE \ | |
76 | (sizeof(struct ip_vs_sync_conn) + sizeof(struct ip_vs_sync_conn_options)) | |
77 | ||
78 | ||
79 | /* | |
80 | The master mulitcasts messages to the backup load balancers in the | |
81 | following format. | |
82 | ||
83 | 0 1 2 3 | |
84 | 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 | |
85 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
86 | | Count Conns | SyncID | Size | | |
87 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
88 | | | | |
89 | | IPVS Sync Connection (1) | | |
90 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
91 | | . | | |
92 | | . | | |
93 | | . | | |
94 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
95 | | | | |
96 | | IPVS Sync Connection (n) | | |
97 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
98 | */ | |
99 | ||
100 | #define SYNC_MESG_HEADER_LEN 4 | |
101 | ||
102 | struct ip_vs_sync_mesg { | |
103 | __u8 nr_conns; | |
104 | __u8 syncid; | |
105 | __u16 size; | |
106 | ||
107 | /* ip_vs_sync_conn entries start here */ | |
108 | }; | |
109 | ||
110 | /* the maximum length of sync (sending/receiving) message */ | |
111 | static int sync_send_mesg_maxlen; | |
112 | static int sync_recv_mesg_maxlen; | |
113 | ||
114 | struct ip_vs_sync_buff { | |
115 | struct list_head list; | |
116 | unsigned long firstuse; | |
117 | ||
118 | /* pointers for the message data */ | |
119 | struct ip_vs_sync_mesg *mesg; | |
120 | unsigned char *head; | |
121 | unsigned char *end; | |
122 | }; | |
123 | ||
124 | ||
125 | /* the sync_buff list head and the lock */ | |
126 | static LIST_HEAD(ip_vs_sync_queue); | |
127 | static DEFINE_SPINLOCK(ip_vs_sync_lock); | |
128 | ||
129 | /* current sync_buff for accepting new conn entries */ | |
130 | static struct ip_vs_sync_buff *curr_sb = NULL; | |
131 | static DEFINE_SPINLOCK(curr_sb_lock); | |
132 | ||
133 | /* ipvs sync daemon state */ | |
134 | volatile int ip_vs_sync_state = IP_VS_STATE_NONE; | |
135 | volatile int ip_vs_master_syncid = 0; | |
136 | volatile int ip_vs_backup_syncid = 0; | |
137 | ||
138 | /* multicast interface name */ | |
139 | char ip_vs_master_mcast_ifn[IP_VS_IFNAME_MAXLEN]; | |
140 | char ip_vs_backup_mcast_ifn[IP_VS_IFNAME_MAXLEN]; | |
141 | ||
998e7a76 SW |
142 | /* sync daemon tasks */ |
143 | static struct task_struct *sync_master_thread; | |
144 | static struct task_struct *sync_backup_thread; | |
145 | ||
1da177e4 | 146 | /* multicast addr */ |
d5640050 SW |
147 | static struct sockaddr_in mcast_addr = { |
148 | .sin_family = AF_INET, | |
149 | .sin_port = __constant_htons(IP_VS_SYNC_PORT), | |
150 | .sin_addr.s_addr = __constant_htonl(IP_VS_SYNC_GROUP), | |
151 | }; | |
1da177e4 LT |
152 | |
153 | ||
998e7a76 | 154 | static inline struct ip_vs_sync_buff *sb_dequeue(void) |
1da177e4 LT |
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 | ||
998e7a76 SW |
198 | static inline void sb_queue_tail(struct ip_vs_sync_buff *sb) |
199 | { | |
200 | spin_lock(&ip_vs_sync_lock); | |
201 | if (ip_vs_sync_state & IP_VS_STATE_MASTER) | |
202 | list_add_tail(&sb->list, &ip_vs_sync_queue); | |
203 | else | |
204 | ip_vs_sync_buff_release(sb); | |
205 | spin_unlock(&ip_vs_sync_lock); | |
206 | } | |
207 | ||
1da177e4 LT |
208 | /* |
209 | * Get the current sync buffer if it has been created for more | |
210 | * than the specified time or the specified time is zero. | |
211 | */ | |
212 | static inline struct ip_vs_sync_buff * | |
213 | get_curr_sync_buff(unsigned long time) | |
214 | { | |
215 | struct ip_vs_sync_buff *sb; | |
216 | ||
217 | spin_lock_bh(&curr_sb_lock); | |
218 | if (curr_sb && (time == 0 || | |
219 | time_before(jiffies - curr_sb->firstuse, time))) { | |
220 | sb = curr_sb; | |
221 | curr_sb = NULL; | |
222 | } else | |
223 | sb = NULL; | |
224 | spin_unlock_bh(&curr_sb_lock); | |
225 | return sb; | |
226 | } | |
227 | ||
228 | ||
229 | /* | |
230 | * Add an ip_vs_conn information into the current sync_buff. | |
231 | * Called by ip_vs_in. | |
232 | */ | |
233 | void ip_vs_sync_conn(struct ip_vs_conn *cp) | |
234 | { | |
235 | struct ip_vs_sync_mesg *m; | |
236 | struct ip_vs_sync_conn *s; | |
237 | int len; | |
238 | ||
239 | spin_lock(&curr_sb_lock); | |
240 | if (!curr_sb) { | |
241 | if (!(curr_sb=ip_vs_sync_buff_create())) { | |
242 | spin_unlock(&curr_sb_lock); | |
243 | IP_VS_ERR("ip_vs_sync_buff_create failed.\n"); | |
244 | return; | |
245 | } | |
246 | } | |
247 | ||
248 | len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE : | |
249 | SIMPLE_CONN_SIZE; | |
250 | m = curr_sb->mesg; | |
251 | s = (struct ip_vs_sync_conn *)curr_sb->head; | |
252 | ||
253 | /* copy members */ | |
254 | s->protocol = cp->protocol; | |
255 | s->cport = cp->cport; | |
256 | s->vport = cp->vport; | |
257 | s->dport = cp->dport; | |
258 | s->caddr = cp->caddr; | |
259 | s->vaddr = cp->vaddr; | |
260 | s->daddr = cp->daddr; | |
261 | s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED); | |
262 | s->state = htons(cp->state); | |
263 | if (cp->flags & IP_VS_CONN_F_SEQ_MASK) { | |
264 | struct ip_vs_sync_conn_options *opt = | |
265 | (struct ip_vs_sync_conn_options *)&s[1]; | |
266 | memcpy(opt, &cp->in_seq, sizeof(*opt)); | |
267 | } | |
268 | ||
269 | m->nr_conns++; | |
270 | m->size += len; | |
271 | curr_sb->head += len; | |
272 | ||
273 | /* check if there is a space for next one */ | |
274 | if (curr_sb->head+FULL_CONN_SIZE > curr_sb->end) { | |
275 | sb_queue_tail(curr_sb); | |
276 | curr_sb = NULL; | |
277 | } | |
278 | spin_unlock(&curr_sb_lock); | |
279 | ||
280 | /* synchronize its controller if it has */ | |
281 | if (cp->control) | |
282 | ip_vs_sync_conn(cp->control); | |
283 | } | |
284 | ||
285 | ||
286 | /* | |
287 | * Process received multicast message and create the corresponding | |
288 | * ip_vs_conn entries. | |
289 | */ | |
290 | static void ip_vs_process_message(const char *buffer, const size_t buflen) | |
291 | { | |
292 | struct ip_vs_sync_mesg *m = (struct ip_vs_sync_mesg *)buffer; | |
293 | struct ip_vs_sync_conn *s; | |
294 | struct ip_vs_sync_conn_options *opt; | |
295 | struct ip_vs_conn *cp; | |
5c81833c | 296 | struct ip_vs_protocol *pp; |
1e356f9c | 297 | struct ip_vs_dest *dest; |
1da177e4 LT |
298 | char *p; |
299 | int i; | |
300 | ||
2ad17def JA |
301 | if (buflen < sizeof(struct ip_vs_sync_mesg)) { |
302 | IP_VS_ERR_RL("sync message header too short\n"); | |
303 | return; | |
304 | } | |
305 | ||
1da177e4 LT |
306 | /* Convert size back to host byte order */ |
307 | m->size = ntohs(m->size); | |
308 | ||
309 | if (buflen != m->size) { | |
2ad17def | 310 | IP_VS_ERR_RL("bogus sync message size\n"); |
1da177e4 LT |
311 | return; |
312 | } | |
313 | ||
314 | /* SyncID sanity check */ | |
315 | if (ip_vs_backup_syncid != 0 && m->syncid != ip_vs_backup_syncid) { | |
316 | IP_VS_DBG(7, "Ignoring incoming msg with syncid = %d\n", | |
317 | m->syncid); | |
318 | return; | |
319 | } | |
320 | ||
321 | p = (char *)buffer + sizeof(struct ip_vs_sync_mesg); | |
322 | for (i=0; i<m->nr_conns; i++) { | |
b209639e | 323 | unsigned flags, state; |
87375ab4 | 324 | |
2ad17def JA |
325 | if (p + SIMPLE_CONN_SIZE > buffer+buflen) { |
326 | IP_VS_ERR_RL("bogus conn in sync message\n"); | |
327 | return; | |
328 | } | |
329 | s = (struct ip_vs_sync_conn *) p; | |
7a4fbb1f | 330 | flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC; |
2ad17def JA |
331 | flags &= ~IP_VS_CONN_F_HASHED; |
332 | if (flags & IP_VS_CONN_F_SEQ_MASK) { | |
333 | opt = (struct ip_vs_sync_conn_options *)&s[1]; | |
334 | p += FULL_CONN_SIZE; | |
335 | if (p > buffer+buflen) { | |
336 | IP_VS_ERR_RL("bogus conn options in sync message\n"); | |
337 | return; | |
338 | } | |
339 | } else { | |
340 | opt = NULL; | |
341 | p += SIMPLE_CONN_SIZE; | |
342 | } | |
343 | ||
b209639e | 344 | state = ntohs(s->state); |
2ad17def JA |
345 | if (!(flags & IP_VS_CONN_F_TEMPLATE)) { |
346 | pp = ip_vs_proto_get(s->protocol); | |
347 | if (!pp) { | |
348 | IP_VS_ERR_RL("Unsupported protocol %u in sync msg\n", | |
349 | s->protocol); | |
350 | continue; | |
351 | } | |
352 | if (state >= pp->num_states) { | |
353 | IP_VS_DBG(2, "Invalid %s state %u in sync msg\n", | |
354 | pp->name, state); | |
355 | continue; | |
356 | } | |
357 | } else { | |
358 | /* protocol in templates is not used for state/timeout */ | |
359 | pp = NULL; | |
360 | if (state > 0) { | |
361 | IP_VS_DBG(2, "Invalid template state %u in sync msg\n", | |
362 | state); | |
363 | state = 0; | |
364 | } | |
365 | } | |
366 | ||
87375ab4 JA |
367 | if (!(flags & IP_VS_CONN_F_TEMPLATE)) |
368 | cp = ip_vs_conn_in_get(s->protocol, | |
369 | s->caddr, s->cport, | |
370 | s->vaddr, s->vport); | |
371 | else | |
372 | cp = ip_vs_ct_in_get(s->protocol, | |
373 | s->caddr, s->cport, | |
374 | s->vaddr, s->vport); | |
1da177e4 | 375 | if (!cp) { |
1e356f9c RB |
376 | /* |
377 | * Find the appropriate destination for the connection. | |
378 | * If it is not found the connection will remain unbound | |
379 | * but still handled. | |
380 | */ | |
381 | dest = ip_vs_find_dest(s->daddr, s->dport, | |
382 | s->vaddr, s->vport, | |
383 | s->protocol); | |
b209639e RB |
384 | /* Set the approprite ativity flag */ |
385 | if (s->protocol == IPPROTO_TCP) { | |
386 | if (state != IP_VS_TCP_S_ESTABLISHED) | |
387 | flags |= IP_VS_CONN_F_INACTIVE; | |
388 | else | |
389 | flags &= ~IP_VS_CONN_F_INACTIVE; | |
390 | } | |
1da177e4 LT |
391 | cp = ip_vs_conn_new(s->protocol, |
392 | s->caddr, s->cport, | |
393 | s->vaddr, s->vport, | |
394 | s->daddr, s->dport, | |
1e356f9c RB |
395 | flags, dest); |
396 | if (dest) | |
397 | atomic_dec(&dest->refcnt); | |
1da177e4 LT |
398 | if (!cp) { |
399 | IP_VS_ERR("ip_vs_conn_new failed\n"); | |
400 | return; | |
401 | } | |
1da177e4 | 402 | } else if (!cp->dest) { |
1e356f9c | 403 | dest = ip_vs_try_bind_dest(cp); |
2ad17def | 404 | if (dest) |
1e356f9c | 405 | atomic_dec(&dest->refcnt); |
b209639e RB |
406 | } else if ((cp->dest) && (cp->protocol == IPPROTO_TCP) && |
407 | (cp->state != state)) { | |
408 | /* update active/inactive flag for the connection */ | |
409 | dest = cp->dest; | |
410 | if (!(cp->flags & IP_VS_CONN_F_INACTIVE) && | |
411 | (state != IP_VS_TCP_S_ESTABLISHED)) { | |
412 | atomic_dec(&dest->activeconns); | |
413 | atomic_inc(&dest->inactconns); | |
414 | cp->flags |= IP_VS_CONN_F_INACTIVE; | |
415 | } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) && | |
416 | (state == IP_VS_TCP_S_ESTABLISHED)) { | |
417 | atomic_inc(&dest->activeconns); | |
418 | atomic_dec(&dest->inactconns); | |
419 | cp->flags &= ~IP_VS_CONN_F_INACTIVE; | |
420 | } | |
421 | } | |
1da177e4 | 422 | |
2ad17def | 423 | if (opt) |
1da177e4 | 424 | memcpy(&cp->in_seq, opt, sizeof(*opt)); |
1da177e4 | 425 | atomic_set(&cp->in_pkts, sysctl_ip_vs_sync_threshold[0]); |
b209639e | 426 | cp->state = state; |
2ad17def JA |
427 | cp->old_state = cp->state; |
428 | /* | |
429 | * We can not recover the right timeout for templates | |
430 | * in all cases, we can not find the right fwmark | |
431 | * virtual service. If needed, we can do it for | |
432 | * non-fwmark persistent services. | |
433 | */ | |
434 | if (!(flags & IP_VS_CONN_F_TEMPLATE) && pp->timeout_table) | |
435 | cp->timeout = pp->timeout_table[state]; | |
436 | else | |
437 | cp->timeout = (3*60*HZ); | |
1da177e4 | 438 | ip_vs_conn_put(cp); |
1da177e4 LT |
439 | } |
440 | } | |
441 | ||
442 | ||
443 | /* | |
444 | * Setup loopback of outgoing multicasts on a sending socket | |
445 | */ | |
446 | static void set_mcast_loop(struct sock *sk, u_char loop) | |
447 | { | |
448 | struct inet_sock *inet = inet_sk(sk); | |
449 | ||
450 | /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */ | |
451 | lock_sock(sk); | |
452 | inet->mc_loop = loop ? 1 : 0; | |
453 | release_sock(sk); | |
454 | } | |
455 | ||
456 | /* | |
457 | * Specify TTL for outgoing multicasts on a sending socket | |
458 | */ | |
459 | static void set_mcast_ttl(struct sock *sk, u_char ttl) | |
460 | { | |
461 | struct inet_sock *inet = inet_sk(sk); | |
462 | ||
463 | /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */ | |
464 | lock_sock(sk); | |
465 | inet->mc_ttl = ttl; | |
466 | release_sock(sk); | |
467 | } | |
468 | ||
469 | /* | |
470 | * Specifiy default interface for outgoing multicasts | |
471 | */ | |
472 | static int set_mcast_if(struct sock *sk, char *ifname) | |
473 | { | |
474 | struct net_device *dev; | |
475 | struct inet_sock *inet = inet_sk(sk); | |
476 | ||
881d966b | 477 | if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL) |
1da177e4 LT |
478 | return -ENODEV; |
479 | ||
480 | if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) | |
481 | return -EINVAL; | |
482 | ||
483 | lock_sock(sk); | |
484 | inet->mc_index = dev->ifindex; | |
485 | /* inet->mc_addr = 0; */ | |
486 | release_sock(sk); | |
487 | ||
488 | return 0; | |
489 | } | |
490 | ||
491 | ||
492 | /* | |
493 | * Set the maximum length of sync message according to the | |
494 | * specified interface's MTU. | |
495 | */ | |
496 | static int set_sync_mesg_maxlen(int sync_state) | |
497 | { | |
498 | struct net_device *dev; | |
499 | int num; | |
500 | ||
501 | if (sync_state == IP_VS_STATE_MASTER) { | |
881d966b | 502 | if ((dev = __dev_get_by_name(&init_net, ip_vs_master_mcast_ifn)) == NULL) |
1da177e4 LT |
503 | return -ENODEV; |
504 | ||
505 | num = (dev->mtu - sizeof(struct iphdr) - | |
506 | sizeof(struct udphdr) - | |
507 | SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE; | |
508 | sync_send_mesg_maxlen = | |
509 | SYNC_MESG_HEADER_LEN + SIMPLE_CONN_SIZE * num; | |
510 | IP_VS_DBG(7, "setting the maximum length of sync sending " | |
511 | "message %d.\n", sync_send_mesg_maxlen); | |
512 | } else if (sync_state == IP_VS_STATE_BACKUP) { | |
881d966b | 513 | if ((dev = __dev_get_by_name(&init_net, ip_vs_backup_mcast_ifn)) == NULL) |
1da177e4 LT |
514 | return -ENODEV; |
515 | ||
516 | sync_recv_mesg_maxlen = dev->mtu - | |
517 | sizeof(struct iphdr) - sizeof(struct udphdr); | |
518 | IP_VS_DBG(7, "setting the maximum length of sync receiving " | |
519 | "message %d.\n", sync_recv_mesg_maxlen); | |
520 | } | |
521 | ||
522 | return 0; | |
523 | } | |
524 | ||
525 | ||
526 | /* | |
527 | * Join a multicast group. | |
528 | * the group is specified by a class D multicast address 224.0.0.0/8 | |
529 | * in the in_addr structure passed in as a parameter. | |
530 | */ | |
531 | static int | |
532 | join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname) | |
533 | { | |
534 | struct ip_mreqn mreq; | |
535 | struct net_device *dev; | |
536 | int ret; | |
537 | ||
538 | memset(&mreq, 0, sizeof(mreq)); | |
539 | memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr)); | |
540 | ||
881d966b | 541 | if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL) |
1da177e4 LT |
542 | return -ENODEV; |
543 | if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) | |
544 | return -EINVAL; | |
545 | ||
546 | mreq.imr_ifindex = dev->ifindex; | |
547 | ||
548 | lock_sock(sk); | |
549 | ret = ip_mc_join_group(sk, &mreq); | |
550 | release_sock(sk); | |
551 | ||
552 | return ret; | |
553 | } | |
554 | ||
555 | ||
556 | static int bind_mcastif_addr(struct socket *sock, char *ifname) | |
557 | { | |
558 | struct net_device *dev; | |
a61ced5d | 559 | __be32 addr; |
1da177e4 LT |
560 | struct sockaddr_in sin; |
561 | ||
881d966b | 562 | if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL) |
1da177e4 LT |
563 | return -ENODEV; |
564 | ||
565 | addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE); | |
566 | if (!addr) | |
567 | IP_VS_ERR("You probably need to specify IP address on " | |
568 | "multicast interface.\n"); | |
569 | ||
570 | IP_VS_DBG(7, "binding socket with (%s) %u.%u.%u.%u\n", | |
571 | ifname, NIPQUAD(addr)); | |
572 | ||
573 | /* Now bind the socket with the address of multicast interface */ | |
574 | sin.sin_family = AF_INET; | |
575 | sin.sin_addr.s_addr = addr; | |
576 | sin.sin_port = 0; | |
577 | ||
578 | return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin)); | |
579 | } | |
580 | ||
581 | /* | |
582 | * Set up sending multicast socket over UDP | |
583 | */ | |
584 | static struct socket * make_send_sock(void) | |
585 | { | |
586 | struct socket *sock; | |
e6dd731c | 587 | int result; |
1da177e4 LT |
588 | |
589 | /* First create a socket */ | |
e6dd731c SW |
590 | result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock); |
591 | if (result < 0) { | |
1da177e4 | 592 | IP_VS_ERR("Error during creation of socket; terminating\n"); |
e6dd731c | 593 | return ERR_PTR(result); |
1da177e4 LT |
594 | } |
595 | ||
e6dd731c SW |
596 | result = set_mcast_if(sock->sk, ip_vs_master_mcast_ifn); |
597 | if (result < 0) { | |
1da177e4 LT |
598 | IP_VS_ERR("Error setting outbound mcast interface\n"); |
599 | goto error; | |
600 | } | |
601 | ||
602 | set_mcast_loop(sock->sk, 0); | |
603 | set_mcast_ttl(sock->sk, 1); | |
604 | ||
e6dd731c SW |
605 | result = bind_mcastif_addr(sock, ip_vs_master_mcast_ifn); |
606 | if (result < 0) { | |
1da177e4 LT |
607 | IP_VS_ERR("Error binding address of the mcast interface\n"); |
608 | goto error; | |
609 | } | |
610 | ||
e6dd731c SW |
611 | result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr, |
612 | sizeof(struct sockaddr), 0); | |
613 | if (result < 0) { | |
1da177e4 LT |
614 | IP_VS_ERR("Error connecting to the multicast addr\n"); |
615 | goto error; | |
616 | } | |
617 | ||
618 | return sock; | |
619 | ||
620 | error: | |
621 | sock_release(sock); | |
e6dd731c | 622 | return ERR_PTR(result); |
1da177e4 LT |
623 | } |
624 | ||
625 | ||
626 | /* | |
627 | * Set up receiving multicast socket over UDP | |
628 | */ | |
629 | static struct socket * make_receive_sock(void) | |
630 | { | |
631 | struct socket *sock; | |
e6dd731c | 632 | int result; |
1da177e4 LT |
633 | |
634 | /* First create a socket */ | |
e6dd731c SW |
635 | result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock); |
636 | if (result < 0) { | |
1da177e4 | 637 | IP_VS_ERR("Error during creation of socket; terminating\n"); |
e6dd731c | 638 | return ERR_PTR(result); |
1da177e4 LT |
639 | } |
640 | ||
641 | /* it is equivalent to the REUSEADDR option in user-space */ | |
642 | sock->sk->sk_reuse = 1; | |
643 | ||
e6dd731c SW |
644 | result = sock->ops->bind(sock, (struct sockaddr *) &mcast_addr, |
645 | sizeof(struct sockaddr)); | |
646 | if (result < 0) { | |
1da177e4 LT |
647 | IP_VS_ERR("Error binding to the multicast addr\n"); |
648 | goto error; | |
649 | } | |
650 | ||
651 | /* join the multicast group */ | |
e6dd731c SW |
652 | result = join_mcast_group(sock->sk, |
653 | (struct in_addr *) &mcast_addr.sin_addr, | |
654 | ip_vs_backup_mcast_ifn); | |
655 | if (result < 0) { | |
1da177e4 LT |
656 | IP_VS_ERR("Error joining to the multicast group\n"); |
657 | goto error; | |
658 | } | |
659 | ||
660 | return sock; | |
661 | ||
662 | error: | |
663 | sock_release(sock); | |
e6dd731c | 664 | return ERR_PTR(result); |
1da177e4 LT |
665 | } |
666 | ||
667 | ||
668 | static int | |
669 | ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length) | |
670 | { | |
671 | struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL}; | |
672 | struct kvec iov; | |
673 | int len; | |
674 | ||
675 | EnterFunction(7); | |
676 | iov.iov_base = (void *)buffer; | |
677 | iov.iov_len = length; | |
678 | ||
679 | len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length)); | |
680 | ||
681 | LeaveFunction(7); | |
682 | return len; | |
683 | } | |
684 | ||
685 | static void | |
686 | ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg) | |
687 | { | |
688 | int msize; | |
689 | ||
690 | msize = msg->size; | |
691 | ||
692 | /* Put size in network byte order */ | |
693 | msg->size = htons(msg->size); | |
694 | ||
695 | if (ip_vs_send_async(sock, (char *)msg, msize) != msize) | |
696 | IP_VS_ERR("ip_vs_send_async error\n"); | |
697 | } | |
698 | ||
699 | static int | |
700 | ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen) | |
701 | { | |
702 | struct msghdr msg = {NULL,}; | |
703 | struct kvec iov; | |
704 | int len; | |
705 | ||
706 | EnterFunction(7); | |
707 | ||
708 | /* Receive a packet */ | |
709 | iov.iov_base = buffer; | |
710 | iov.iov_len = (size_t)buflen; | |
711 | ||
712 | len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, 0); | |
713 | ||
714 | if (len < 0) | |
715 | return -1; | |
716 | ||
717 | LeaveFunction(7); | |
718 | return len; | |
719 | } | |
720 | ||
721 | ||
998e7a76 | 722 | static int sync_thread_master(void *data) |
1da177e4 | 723 | { |
998e7a76 | 724 | struct ip_vs_sync_thread_data *tinfo = data; |
1da177e4 LT |
725 | struct ip_vs_sync_buff *sb; |
726 | ||
1da177e4 LT |
727 | IP_VS_INFO("sync thread started: state = MASTER, mcast_ifn = %s, " |
728 | "syncid = %d\n", | |
729 | ip_vs_master_mcast_ifn, ip_vs_master_syncid); | |
730 | ||
998e7a76 SW |
731 | while (!kthread_should_stop()) { |
732 | while ((sb = sb_dequeue())) { | |
733 | ip_vs_send_sync_msg(tinfo->sock, sb->mesg); | |
1da177e4 LT |
734 | ip_vs_sync_buff_release(sb); |
735 | } | |
736 | ||
737 | /* check if entries stay in curr_sb for 2 seconds */ | |
998e7a76 SW |
738 | sb = get_curr_sync_buff(2 * HZ); |
739 | if (sb) { | |
740 | ip_vs_send_sync_msg(tinfo->sock, sb->mesg); | |
1da177e4 LT |
741 | ip_vs_sync_buff_release(sb); |
742 | } | |
743 | ||
89eaeb09 | 744 | msleep_interruptible(1000); |
1da177e4 LT |
745 | } |
746 | ||
747 | /* clean up the sync_buff queue */ | |
748 | while ((sb=sb_dequeue())) { | |
749 | ip_vs_sync_buff_release(sb); | |
750 | } | |
751 | ||
752 | /* clean up the current sync_buff */ | |
753 | if ((sb = get_curr_sync_buff(0))) { | |
754 | ip_vs_sync_buff_release(sb); | |
755 | } | |
756 | ||
757 | /* release the sending multicast socket */ | |
998e7a76 SW |
758 | sock_release(tinfo->sock); |
759 | kfree(tinfo); | |
760 | ||
761 | return 0; | |
1da177e4 LT |
762 | } |
763 | ||
764 | ||
998e7a76 | 765 | static int sync_thread_backup(void *data) |
1da177e4 | 766 | { |
998e7a76 | 767 | struct ip_vs_sync_thread_data *tinfo = data; |
1da177e4 LT |
768 | int len; |
769 | ||
1da177e4 LT |
770 | IP_VS_INFO("sync thread started: state = BACKUP, mcast_ifn = %s, " |
771 | "syncid = %d\n", | |
772 | ip_vs_backup_mcast_ifn, ip_vs_backup_syncid); | |
773 | ||
998e7a76 SW |
774 | while (!kthread_should_stop()) { |
775 | /* do we have data now? */ | |
776 | while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) { | |
777 | len = ip_vs_receive(tinfo->sock, tinfo->buf, | |
778 | sync_recv_mesg_maxlen); | |
779 | if (len <= 0) { | |
1da177e4 LT |
780 | IP_VS_ERR("receiving message error\n"); |
781 | break; | |
782 | } | |
998e7a76 SW |
783 | |
784 | /* disable bottom half, because it accesses the data | |
1da177e4 LT |
785 | shared by softirq while getting/creating conns */ |
786 | local_bh_disable(); | |
998e7a76 | 787 | ip_vs_process_message(tinfo->buf, len); |
1da177e4 LT |
788 | local_bh_enable(); |
789 | } | |
790 | ||
89eaeb09 | 791 | msleep_interruptible(1000); |
1da177e4 LT |
792 | } |
793 | ||
794 | /* release the sending multicast socket */ | |
998e7a76 SW |
795 | sock_release(tinfo->sock); |
796 | kfree(tinfo->buf); | |
797 | kfree(tinfo); | |
1da177e4 | 798 | |
998e7a76 | 799 | return 0; |
1da177e4 LT |
800 | } |
801 | ||
802 | ||
998e7a76 | 803 | int start_sync_thread(int state, char *mcast_ifn, __u8 syncid) |
1da177e4 | 804 | { |
998e7a76 SW |
805 | struct ip_vs_sync_thread_data *tinfo; |
806 | struct task_struct **realtask, *task; | |
807 | struct socket *sock; | |
808 | char *name, *buf = NULL; | |
809 | int (*threadfn)(void *data); | |
810 | int result = -ENOMEM; | |
1da177e4 | 811 | |
998e7a76 SW |
812 | IP_VS_DBG(7, "%s: pid %d\n", __func__, task_pid_nr(current)); |
813 | IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n", | |
814 | sizeof(struct ip_vs_sync_conn)); | |
1da177e4 | 815 | |
998e7a76 SW |
816 | if (state == IP_VS_STATE_MASTER) { |
817 | if (sync_master_thread) | |
818 | return -EEXIST; | |
1da177e4 | 819 | |
998e7a76 SW |
820 | strlcpy(ip_vs_master_mcast_ifn, mcast_ifn, |
821 | sizeof(ip_vs_master_mcast_ifn)); | |
822 | ip_vs_master_syncid = syncid; | |
823 | realtask = &sync_master_thread; | |
1da177e4 | 824 | name = "ipvs_syncmaster"; |
998e7a76 SW |
825 | threadfn = sync_thread_master; |
826 | sock = make_send_sock(); | |
827 | } else if (state == IP_VS_STATE_BACKUP) { | |
828 | if (sync_backup_thread) | |
829 | return -EEXIST; | |
830 | ||
831 | strlcpy(ip_vs_backup_mcast_ifn, mcast_ifn, | |
832 | sizeof(ip_vs_backup_mcast_ifn)); | |
833 | ip_vs_backup_syncid = syncid; | |
834 | realtask = &sync_backup_thread; | |
1da177e4 | 835 | name = "ipvs_syncbackup"; |
998e7a76 SW |
836 | threadfn = sync_thread_backup; |
837 | sock = make_receive_sock(); | |
1da177e4 | 838 | } else { |
1da177e4 LT |
839 | return -EINVAL; |
840 | } | |
841 | ||
998e7a76 SW |
842 | if (IS_ERR(sock)) { |
843 | result = PTR_ERR(sock); | |
844 | goto out; | |
845 | } | |
1da177e4 | 846 | |
1da177e4 | 847 | set_sync_mesg_maxlen(state); |
998e7a76 SW |
848 | if (state == IP_VS_STATE_BACKUP) { |
849 | buf = kmalloc(sync_recv_mesg_maxlen, GFP_KERNEL); | |
850 | if (!buf) | |
851 | goto outsocket; | |
852 | } | |
1da177e4 | 853 | |
998e7a76 SW |
854 | tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL); |
855 | if (!tinfo) | |
856 | goto outbuf; | |
cc0191ae | 857 | |
998e7a76 SW |
858 | tinfo->sock = sock; |
859 | tinfo->buf = buf; | |
1da177e4 | 860 | |
998e7a76 SW |
861 | task = kthread_run(threadfn, tinfo, name); |
862 | if (IS_ERR(task)) { | |
863 | result = PTR_ERR(task); | |
864 | goto outtinfo; | |
865 | } | |
1da177e4 | 866 | |
998e7a76 SW |
867 | /* mark as active */ |
868 | *realtask = task; | |
869 | ip_vs_sync_state |= state; | |
1da177e4 | 870 | |
998e7a76 SW |
871 | /* increase the module use count */ |
872 | ip_vs_use_count_inc(); | |
1da177e4 LT |
873 | |
874 | return 0; | |
1da177e4 | 875 | |
998e7a76 SW |
876 | outtinfo: |
877 | kfree(tinfo); | |
878 | outbuf: | |
879 | kfree(buf); | |
880 | outsocket: | |
881 | sock_release(sock); | |
882 | out: | |
883 | return result; | |
1da177e4 LT |
884 | } |
885 | ||
886 | ||
998e7a76 | 887 | int stop_sync_thread(int state) |
1da177e4 | 888 | { |
0dc47877 | 889 | IP_VS_DBG(7, "%s: pid %d\n", __func__, task_pid_nr(current)); |
1da177e4 | 890 | |
1da177e4 | 891 | if (state == IP_VS_STATE_MASTER) { |
998e7a76 SW |
892 | if (!sync_master_thread) |
893 | return -ESRCH; | |
cc0191ae | 894 | |
998e7a76 SW |
895 | IP_VS_INFO("stopping master sync thread %d ...\n", |
896 | task_pid_nr(sync_master_thread)); | |
1da177e4 | 897 | |
998e7a76 SW |
898 | /* |
899 | * The lock synchronizes with sb_queue_tail(), so that we don't | |
900 | * add sync buffers to the queue, when we are already in | |
901 | * progress of stopping the master sync daemon. | |
902 | */ | |
1da177e4 | 903 | |
998e7a76 SW |
904 | spin_lock(&ip_vs_sync_lock); |
905 | ip_vs_sync_state &= ~IP_VS_STATE_MASTER; | |
906 | spin_unlock(&ip_vs_sync_lock); | |
907 | kthread_stop(sync_master_thread); | |
908 | sync_master_thread = NULL; | |
909 | } else if (state == IP_VS_STATE_BACKUP) { | |
910 | if (!sync_backup_thread) | |
911 | return -ESRCH; | |
912 | ||
913 | IP_VS_INFO("stopping backup sync thread %d ...\n", | |
914 | task_pid_nr(sync_backup_thread)); | |
915 | ||
916 | ip_vs_sync_state &= ~IP_VS_STATE_BACKUP; | |
917 | kthread_stop(sync_backup_thread); | |
918 | sync_backup_thread = NULL; | |
919 | } else { | |
920 | return -EINVAL; | |
921 | } | |
1da177e4 | 922 | |
998e7a76 SW |
923 | /* decrease the module use count */ |
924 | ip_vs_use_count_dec(); | |
1da177e4 LT |
925 | |
926 | return 0; | |
927 | } |