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 | ||
9aada7ac HE |
20 | #define KMSG_COMPONENT "IPVS" |
21 | #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt | |
22 | ||
1da177e4 LT |
23 | #include <linux/module.h> |
24 | #include <linux/slab.h> | |
14c85021 | 25 | #include <linux/inetdevice.h> |
1da177e4 LT |
26 | #include <linux/net.h> |
27 | #include <linux/completion.h> | |
28 | #include <linux/delay.h> | |
29 | #include <linux/skbuff.h> | |
30 | #include <linux/in.h> | |
31 | #include <linux/igmp.h> /* for ip_mc_join_group */ | |
14c85021 | 32 | #include <linux/udp.h> |
e6dd731c | 33 | #include <linux/err.h> |
998e7a76 | 34 | #include <linux/kthread.h> |
ba6fd850 | 35 | #include <linux/wait.h> |
e6f225eb | 36 | #include <linux/kernel.h> |
1da177e4 LT |
37 | |
38 | #include <net/ip.h> | |
39 | #include <net/sock.h> | |
1da177e4 LT |
40 | |
41 | #include <net/ip_vs.h> | |
42 | ||
43 | #define IP_VS_SYNC_GROUP 0xe0000051 /* multicast addr - 224.0.0.81 */ | |
44 | #define IP_VS_SYNC_PORT 8848 /* multicast port */ | |
45 | ||
46 | ||
47 | /* | |
48 | * IPVS sync connection entry | |
49 | */ | |
50 | struct ip_vs_sync_conn { | |
51 | __u8 reserved; | |
52 | ||
53 | /* Protocol, addresses and port numbers */ | |
54 | __u8 protocol; /* Which protocol (TCP/UDP) */ | |
014d730d AV |
55 | __be16 cport; |
56 | __be16 vport; | |
57 | __be16 dport; | |
58 | __be32 caddr; /* client address */ | |
59 | __be32 vaddr; /* virtual address */ | |
60 | __be32 daddr; /* destination address */ | |
1da177e4 LT |
61 | |
62 | /* Flags and state transition */ | |
014d730d AV |
63 | __be16 flags; /* status flags */ |
64 | __be16 state; /* state info */ | |
1da177e4 LT |
65 | |
66 | /* The sequence options start here */ | |
67 | }; | |
68 | ||
69 | struct ip_vs_sync_conn_options { | |
70 | struct ip_vs_seq in_seq; /* incoming seq. struct */ | |
71 | struct ip_vs_seq out_seq; /* outgoing seq. struct */ | |
72 | }; | |
73 | ||
cc0191ae | 74 | struct ip_vs_sync_thread_data { |
998e7a76 SW |
75 | struct socket *sock; |
76 | char *buf; | |
cc0191ae NH |
77 | }; |
78 | ||
1da177e4 LT |
79 | #define SIMPLE_CONN_SIZE (sizeof(struct ip_vs_sync_conn)) |
80 | #define FULL_CONN_SIZE \ | |
81 | (sizeof(struct ip_vs_sync_conn) + sizeof(struct ip_vs_sync_conn_options)) | |
82 | ||
83 | ||
84 | /* | |
85 | The master mulitcasts messages to the backup load balancers in the | |
86 | following format. | |
87 | ||
88 | 0 1 2 3 | |
89 | 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 | |
90 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
91 | | Count Conns | SyncID | Size | | |
92 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
93 | | | | |
94 | | IPVS Sync Connection (1) | | |
95 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
96 | | . | | |
97 | | . | | |
98 | | . | | |
99 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
100 | | | | |
101 | | IPVS Sync Connection (n) | | |
102 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
103 | */ | |
104 | ||
105 | #define SYNC_MESG_HEADER_LEN 4 | |
e6f225eb | 106 | #define MAX_CONNS_PER_SYNCBUFF 255 /* nr_conns in ip_vs_sync_mesg is 8 bit */ |
1da177e4 LT |
107 | |
108 | struct ip_vs_sync_mesg { | |
109 | __u8 nr_conns; | |
110 | __u8 syncid; | |
111 | __u16 size; | |
112 | ||
113 | /* ip_vs_sync_conn entries start here */ | |
114 | }; | |
115 | ||
116 | /* the maximum length of sync (sending/receiving) message */ | |
117 | static int sync_send_mesg_maxlen; | |
118 | static int sync_recv_mesg_maxlen; | |
119 | ||
120 | struct ip_vs_sync_buff { | |
121 | struct list_head list; | |
122 | unsigned long firstuse; | |
123 | ||
124 | /* pointers for the message data */ | |
125 | struct ip_vs_sync_mesg *mesg; | |
126 | unsigned char *head; | |
127 | unsigned char *end; | |
128 | }; | |
129 | ||
130 | ||
131 | /* the sync_buff list head and the lock */ | |
132 | static LIST_HEAD(ip_vs_sync_queue); | |
133 | static DEFINE_SPINLOCK(ip_vs_sync_lock); | |
134 | ||
135 | /* current sync_buff for accepting new conn entries */ | |
136 | static struct ip_vs_sync_buff *curr_sb = NULL; | |
137 | static DEFINE_SPINLOCK(curr_sb_lock); | |
138 | ||
139 | /* ipvs sync daemon state */ | |
140 | volatile int ip_vs_sync_state = IP_VS_STATE_NONE; | |
141 | volatile int ip_vs_master_syncid = 0; | |
142 | volatile int ip_vs_backup_syncid = 0; | |
143 | ||
144 | /* multicast interface name */ | |
145 | char ip_vs_master_mcast_ifn[IP_VS_IFNAME_MAXLEN]; | |
146 | char ip_vs_backup_mcast_ifn[IP_VS_IFNAME_MAXLEN]; | |
147 | ||
998e7a76 SW |
148 | /* sync daemon tasks */ |
149 | static struct task_struct *sync_master_thread; | |
150 | static struct task_struct *sync_backup_thread; | |
151 | ||
1da177e4 | 152 | /* multicast addr */ |
d5640050 SW |
153 | static struct sockaddr_in mcast_addr = { |
154 | .sin_family = AF_INET, | |
09640e63 HH |
155 | .sin_port = cpu_to_be16(IP_VS_SYNC_PORT), |
156 | .sin_addr.s_addr = cpu_to_be32(IP_VS_SYNC_GROUP), | |
d5640050 | 157 | }; |
1da177e4 LT |
158 | |
159 | ||
998e7a76 | 160 | static inline struct ip_vs_sync_buff *sb_dequeue(void) |
1da177e4 LT |
161 | { |
162 | struct ip_vs_sync_buff *sb; | |
163 | ||
164 | spin_lock_bh(&ip_vs_sync_lock); | |
165 | if (list_empty(&ip_vs_sync_queue)) { | |
166 | sb = NULL; | |
167 | } else { | |
168 | sb = list_entry(ip_vs_sync_queue.next, | |
169 | struct ip_vs_sync_buff, | |
170 | list); | |
171 | list_del(&sb->list); | |
172 | } | |
173 | spin_unlock_bh(&ip_vs_sync_lock); | |
174 | ||
175 | return sb; | |
176 | } | |
177 | ||
178 | static inline struct ip_vs_sync_buff * ip_vs_sync_buff_create(void) | |
179 | { | |
180 | struct ip_vs_sync_buff *sb; | |
181 | ||
182 | if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC))) | |
183 | return NULL; | |
184 | ||
185 | if (!(sb->mesg=kmalloc(sync_send_mesg_maxlen, GFP_ATOMIC))) { | |
186 | kfree(sb); | |
187 | return NULL; | |
188 | } | |
189 | sb->mesg->nr_conns = 0; | |
190 | sb->mesg->syncid = ip_vs_master_syncid; | |
191 | sb->mesg->size = 4; | |
192 | sb->head = (unsigned char *)sb->mesg + 4; | |
193 | sb->end = (unsigned char *)sb->mesg + sync_send_mesg_maxlen; | |
194 | sb->firstuse = jiffies; | |
195 | return sb; | |
196 | } | |
197 | ||
198 | static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb) | |
199 | { | |
200 | kfree(sb->mesg); | |
201 | kfree(sb); | |
202 | } | |
203 | ||
998e7a76 SW |
204 | static inline void sb_queue_tail(struct ip_vs_sync_buff *sb) |
205 | { | |
206 | spin_lock(&ip_vs_sync_lock); | |
207 | if (ip_vs_sync_state & IP_VS_STATE_MASTER) | |
208 | list_add_tail(&sb->list, &ip_vs_sync_queue); | |
209 | else | |
210 | ip_vs_sync_buff_release(sb); | |
211 | spin_unlock(&ip_vs_sync_lock); | |
212 | } | |
213 | ||
1da177e4 LT |
214 | /* |
215 | * Get the current sync buffer if it has been created for more | |
216 | * than the specified time or the specified time is zero. | |
217 | */ | |
218 | static inline struct ip_vs_sync_buff * | |
219 | get_curr_sync_buff(unsigned long time) | |
220 | { | |
221 | struct ip_vs_sync_buff *sb; | |
222 | ||
223 | spin_lock_bh(&curr_sb_lock); | |
224 | if (curr_sb && (time == 0 || | |
225 | time_before(jiffies - curr_sb->firstuse, time))) { | |
226 | sb = curr_sb; | |
227 | curr_sb = NULL; | |
228 | } else | |
229 | sb = NULL; | |
230 | spin_unlock_bh(&curr_sb_lock); | |
231 | return sb; | |
232 | } | |
233 | ||
234 | ||
235 | /* | |
236 | * Add an ip_vs_conn information into the current sync_buff. | |
237 | * Called by ip_vs_in. | |
238 | */ | |
239 | void ip_vs_sync_conn(struct ip_vs_conn *cp) | |
240 | { | |
241 | struct ip_vs_sync_mesg *m; | |
242 | struct ip_vs_sync_conn *s; | |
243 | int len; | |
244 | ||
245 | spin_lock(&curr_sb_lock); | |
246 | if (!curr_sb) { | |
247 | if (!(curr_sb=ip_vs_sync_buff_create())) { | |
248 | spin_unlock(&curr_sb_lock); | |
1e3e238e | 249 | pr_err("ip_vs_sync_buff_create failed.\n"); |
1da177e4 LT |
250 | return; |
251 | } | |
252 | } | |
253 | ||
254 | len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE : | |
255 | SIMPLE_CONN_SIZE; | |
256 | m = curr_sb->mesg; | |
257 | s = (struct ip_vs_sync_conn *)curr_sb->head; | |
258 | ||
259 | /* copy members */ | |
260 | s->protocol = cp->protocol; | |
261 | s->cport = cp->cport; | |
262 | s->vport = cp->vport; | |
263 | s->dport = cp->dport; | |
e7ade46a JV |
264 | s->caddr = cp->caddr.ip; |
265 | s->vaddr = cp->vaddr.ip; | |
266 | s->daddr = cp->daddr.ip; | |
1da177e4 LT |
267 | s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED); |
268 | s->state = htons(cp->state); | |
269 | if (cp->flags & IP_VS_CONN_F_SEQ_MASK) { | |
270 | struct ip_vs_sync_conn_options *opt = | |
271 | (struct ip_vs_sync_conn_options *)&s[1]; | |
272 | memcpy(opt, &cp->in_seq, sizeof(*opt)); | |
273 | } | |
274 | ||
275 | m->nr_conns++; | |
276 | m->size += len; | |
277 | curr_sb->head += len; | |
278 | ||
279 | /* check if there is a space for next one */ | |
280 | if (curr_sb->head+FULL_CONN_SIZE > curr_sb->end) { | |
281 | sb_queue_tail(curr_sb); | |
282 | curr_sb = NULL; | |
283 | } | |
284 | spin_unlock(&curr_sb_lock); | |
285 | ||
286 | /* synchronize its controller if it has */ | |
287 | if (cp->control) | |
288 | ip_vs_sync_conn(cp->control); | |
289 | } | |
290 | ||
291 | ||
292 | /* | |
293 | * Process received multicast message and create the corresponding | |
294 | * ip_vs_conn entries. | |
295 | */ | |
296 | static void ip_vs_process_message(const char *buffer, const size_t buflen) | |
297 | { | |
298 | struct ip_vs_sync_mesg *m = (struct ip_vs_sync_mesg *)buffer; | |
299 | struct ip_vs_sync_conn *s; | |
300 | struct ip_vs_sync_conn_options *opt; | |
301 | struct ip_vs_conn *cp; | |
5c81833c | 302 | struct ip_vs_protocol *pp; |
1e356f9c | 303 | struct ip_vs_dest *dest; |
f11017ec | 304 | struct ip_vs_conn_param param; |
1da177e4 LT |
305 | char *p; |
306 | int i; | |
307 | ||
2ad17def JA |
308 | if (buflen < sizeof(struct ip_vs_sync_mesg)) { |
309 | IP_VS_ERR_RL("sync message header too short\n"); | |
310 | return; | |
311 | } | |
312 | ||
1da177e4 LT |
313 | /* Convert size back to host byte order */ |
314 | m->size = ntohs(m->size); | |
315 | ||
316 | if (buflen != m->size) { | |
2ad17def | 317 | IP_VS_ERR_RL("bogus sync message size\n"); |
1da177e4 LT |
318 | return; |
319 | } | |
320 | ||
321 | /* SyncID sanity check */ | |
322 | if (ip_vs_backup_syncid != 0 && m->syncid != ip_vs_backup_syncid) { | |
323 | IP_VS_DBG(7, "Ignoring incoming msg with syncid = %d\n", | |
324 | m->syncid); | |
325 | return; | |
326 | } | |
327 | ||
328 | p = (char *)buffer + sizeof(struct ip_vs_sync_mesg); | |
329 | for (i=0; i<m->nr_conns; i++) { | |
b209639e | 330 | unsigned flags, state; |
87375ab4 | 331 | |
2ad17def JA |
332 | if (p + SIMPLE_CONN_SIZE > buffer+buflen) { |
333 | IP_VS_ERR_RL("bogus conn in sync message\n"); | |
334 | return; | |
335 | } | |
336 | s = (struct ip_vs_sync_conn *) p; | |
7a4fbb1f | 337 | flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC; |
2ad17def JA |
338 | flags &= ~IP_VS_CONN_F_HASHED; |
339 | if (flags & IP_VS_CONN_F_SEQ_MASK) { | |
340 | opt = (struct ip_vs_sync_conn_options *)&s[1]; | |
341 | p += FULL_CONN_SIZE; | |
342 | if (p > buffer+buflen) { | |
343 | IP_VS_ERR_RL("bogus conn options in sync message\n"); | |
344 | return; | |
345 | } | |
346 | } else { | |
347 | opt = NULL; | |
348 | p += SIMPLE_CONN_SIZE; | |
349 | } | |
350 | ||
b209639e | 351 | state = ntohs(s->state); |
2ad17def JA |
352 | if (!(flags & IP_VS_CONN_F_TEMPLATE)) { |
353 | pp = ip_vs_proto_get(s->protocol); | |
354 | if (!pp) { | |
355 | IP_VS_ERR_RL("Unsupported protocol %u in sync msg\n", | |
356 | s->protocol); | |
357 | continue; | |
358 | } | |
359 | if (state >= pp->num_states) { | |
360 | IP_VS_DBG(2, "Invalid %s state %u in sync msg\n", | |
361 | pp->name, state); | |
362 | continue; | |
363 | } | |
364 | } else { | |
365 | /* protocol in templates is not used for state/timeout */ | |
366 | pp = NULL; | |
367 | if (state > 0) { | |
368 | IP_VS_DBG(2, "Invalid template state %u in sync msg\n", | |
369 | state); | |
370 | state = 0; | |
371 | } | |
372 | } | |
373 | ||
f11017ec SH |
374 | { |
375 | ip_vs_conn_fill_param(AF_INET, s->protocol, | |
376 | (union nf_inet_addr *)&s->caddr, | |
377 | s->cport, | |
378 | (union nf_inet_addr *)&s->vaddr, | |
379 | s->vport, ¶m); | |
380 | if (!(flags & IP_VS_CONN_F_TEMPLATE)) | |
381 | cp = ip_vs_conn_in_get(¶m); | |
382 | else | |
383 | cp = ip_vs_ct_in_get(¶m); | |
384 | } | |
1da177e4 | 385 | if (!cp) { |
1e356f9c RB |
386 | /* |
387 | * Find the appropriate destination for the connection. | |
388 | * If it is not found the connection will remain unbound | |
389 | * but still handled. | |
390 | */ | |
7937df15 JV |
391 | dest = ip_vs_find_dest(AF_INET, |
392 | (union nf_inet_addr *)&s->daddr, | |
393 | s->dport, | |
394 | (union nf_inet_addr *)&s->vaddr, | |
395 | s->vport, | |
1e356f9c | 396 | s->protocol); |
b209639e RB |
397 | /* Set the approprite ativity flag */ |
398 | if (s->protocol == IPPROTO_TCP) { | |
399 | if (state != IP_VS_TCP_S_ESTABLISHED) | |
400 | flags |= IP_VS_CONN_F_INACTIVE; | |
401 | else | |
402 | flags &= ~IP_VS_CONN_F_INACTIVE; | |
2906f66a VMR |
403 | } else if (s->protocol == IPPROTO_SCTP) { |
404 | if (state != IP_VS_SCTP_S_ESTABLISHED) | |
405 | flags |= IP_VS_CONN_F_INACTIVE; | |
406 | else | |
407 | flags &= ~IP_VS_CONN_F_INACTIVE; | |
b209639e | 408 | } |
f11017ec | 409 | cp = ip_vs_conn_new(¶m, |
cd9fe6c4 | 410 | (union nf_inet_addr *)&s->daddr, |
f11017ec | 411 | s->dport, flags, dest); |
1e356f9c RB |
412 | if (dest) |
413 | atomic_dec(&dest->refcnt); | |
1da177e4 | 414 | if (!cp) { |
1e3e238e | 415 | pr_err("ip_vs_conn_new failed\n"); |
1da177e4 LT |
416 | return; |
417 | } | |
1da177e4 | 418 | } else if (!cp->dest) { |
1e356f9c | 419 | dest = ip_vs_try_bind_dest(cp); |
2ad17def | 420 | if (dest) |
1e356f9c | 421 | atomic_dec(&dest->refcnt); |
b209639e RB |
422 | } else if ((cp->dest) && (cp->protocol == IPPROTO_TCP) && |
423 | (cp->state != state)) { | |
424 | /* update active/inactive flag for the connection */ | |
425 | dest = cp->dest; | |
426 | if (!(cp->flags & IP_VS_CONN_F_INACTIVE) && | |
427 | (state != IP_VS_TCP_S_ESTABLISHED)) { | |
428 | atomic_dec(&dest->activeconns); | |
429 | atomic_inc(&dest->inactconns); | |
430 | cp->flags |= IP_VS_CONN_F_INACTIVE; | |
431 | } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) && | |
432 | (state == IP_VS_TCP_S_ESTABLISHED)) { | |
433 | atomic_inc(&dest->activeconns); | |
434 | atomic_dec(&dest->inactconns); | |
435 | cp->flags &= ~IP_VS_CONN_F_INACTIVE; | |
436 | } | |
2906f66a VMR |
437 | } else if ((cp->dest) && (cp->protocol == IPPROTO_SCTP) && |
438 | (cp->state != state)) { | |
439 | dest = cp->dest; | |
440 | if (!(cp->flags & IP_VS_CONN_F_INACTIVE) && | |
441 | (state != IP_VS_SCTP_S_ESTABLISHED)) { | |
442 | atomic_dec(&dest->activeconns); | |
443 | atomic_inc(&dest->inactconns); | |
444 | cp->flags &= ~IP_VS_CONN_F_INACTIVE; | |
445 | } | |
b209639e | 446 | } |
1da177e4 | 447 | |
2ad17def | 448 | if (opt) |
1da177e4 | 449 | memcpy(&cp->in_seq, opt, sizeof(*opt)); |
1da177e4 | 450 | atomic_set(&cp->in_pkts, sysctl_ip_vs_sync_threshold[0]); |
b209639e | 451 | cp->state = state; |
2ad17def JA |
452 | cp->old_state = cp->state; |
453 | /* | |
454 | * We can not recover the right timeout for templates | |
455 | * in all cases, we can not find the right fwmark | |
456 | * virtual service. If needed, we can do it for | |
457 | * non-fwmark persistent services. | |
458 | */ | |
459 | if (!(flags & IP_VS_CONN_F_TEMPLATE) && pp->timeout_table) | |
460 | cp->timeout = pp->timeout_table[state]; | |
461 | else | |
462 | cp->timeout = (3*60*HZ); | |
1da177e4 | 463 | ip_vs_conn_put(cp); |
1da177e4 LT |
464 | } |
465 | } | |
466 | ||
467 | ||
468 | /* | |
469 | * Setup loopback of outgoing multicasts on a sending socket | |
470 | */ | |
471 | static void set_mcast_loop(struct sock *sk, u_char loop) | |
472 | { | |
473 | struct inet_sock *inet = inet_sk(sk); | |
474 | ||
475 | /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */ | |
476 | lock_sock(sk); | |
477 | inet->mc_loop = loop ? 1 : 0; | |
478 | release_sock(sk); | |
479 | } | |
480 | ||
481 | /* | |
482 | * Specify TTL for outgoing multicasts on a sending socket | |
483 | */ | |
484 | static void set_mcast_ttl(struct sock *sk, u_char ttl) | |
485 | { | |
486 | struct inet_sock *inet = inet_sk(sk); | |
487 | ||
488 | /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */ | |
489 | lock_sock(sk); | |
490 | inet->mc_ttl = ttl; | |
491 | release_sock(sk); | |
492 | } | |
493 | ||
494 | /* | |
495 | * Specifiy default interface for outgoing multicasts | |
496 | */ | |
497 | static int set_mcast_if(struct sock *sk, char *ifname) | |
498 | { | |
499 | struct net_device *dev; | |
500 | struct inet_sock *inet = inet_sk(sk); | |
501 | ||
881d966b | 502 | if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL) |
1da177e4 LT |
503 | return -ENODEV; |
504 | ||
505 | if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) | |
506 | return -EINVAL; | |
507 | ||
508 | lock_sock(sk); | |
509 | inet->mc_index = dev->ifindex; | |
510 | /* inet->mc_addr = 0; */ | |
511 | release_sock(sk); | |
512 | ||
513 | return 0; | |
514 | } | |
515 | ||
516 | ||
517 | /* | |
518 | * Set the maximum length of sync message according to the | |
519 | * specified interface's MTU. | |
520 | */ | |
521 | static int set_sync_mesg_maxlen(int sync_state) | |
522 | { | |
523 | struct net_device *dev; | |
524 | int num; | |
525 | ||
526 | if (sync_state == IP_VS_STATE_MASTER) { | |
881d966b | 527 | if ((dev = __dev_get_by_name(&init_net, ip_vs_master_mcast_ifn)) == NULL) |
1da177e4 LT |
528 | return -ENODEV; |
529 | ||
530 | num = (dev->mtu - sizeof(struct iphdr) - | |
531 | sizeof(struct udphdr) - | |
532 | SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE; | |
e6f225eb SW |
533 | sync_send_mesg_maxlen = SYNC_MESG_HEADER_LEN + |
534 | SIMPLE_CONN_SIZE * min(num, MAX_CONNS_PER_SYNCBUFF); | |
1da177e4 LT |
535 | IP_VS_DBG(7, "setting the maximum length of sync sending " |
536 | "message %d.\n", sync_send_mesg_maxlen); | |
537 | } else if (sync_state == IP_VS_STATE_BACKUP) { | |
881d966b | 538 | if ((dev = __dev_get_by_name(&init_net, ip_vs_backup_mcast_ifn)) == NULL) |
1da177e4 LT |
539 | return -ENODEV; |
540 | ||
541 | sync_recv_mesg_maxlen = dev->mtu - | |
542 | sizeof(struct iphdr) - sizeof(struct udphdr); | |
543 | IP_VS_DBG(7, "setting the maximum length of sync receiving " | |
544 | "message %d.\n", sync_recv_mesg_maxlen); | |
545 | } | |
546 | ||
547 | return 0; | |
548 | } | |
549 | ||
550 | ||
551 | /* | |
552 | * Join a multicast group. | |
553 | * the group is specified by a class D multicast address 224.0.0.0/8 | |
554 | * in the in_addr structure passed in as a parameter. | |
555 | */ | |
556 | static int | |
557 | join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname) | |
558 | { | |
559 | struct ip_mreqn mreq; | |
560 | struct net_device *dev; | |
561 | int ret; | |
562 | ||
563 | memset(&mreq, 0, sizeof(mreq)); | |
564 | memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr)); | |
565 | ||
881d966b | 566 | if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL) |
1da177e4 LT |
567 | return -ENODEV; |
568 | if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) | |
569 | return -EINVAL; | |
570 | ||
571 | mreq.imr_ifindex = dev->ifindex; | |
572 | ||
573 | lock_sock(sk); | |
574 | ret = ip_mc_join_group(sk, &mreq); | |
575 | release_sock(sk); | |
576 | ||
577 | return ret; | |
578 | } | |
579 | ||
580 | ||
581 | static int bind_mcastif_addr(struct socket *sock, char *ifname) | |
582 | { | |
583 | struct net_device *dev; | |
a61ced5d | 584 | __be32 addr; |
1da177e4 LT |
585 | struct sockaddr_in sin; |
586 | ||
881d966b | 587 | if ((dev = __dev_get_by_name(&init_net, ifname)) == NULL) |
1da177e4 LT |
588 | return -ENODEV; |
589 | ||
590 | addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE); | |
591 | if (!addr) | |
1e3e238e HE |
592 | pr_err("You probably need to specify IP address on " |
593 | "multicast interface.\n"); | |
1da177e4 | 594 | |
14d5e834 HH |
595 | IP_VS_DBG(7, "binding socket with (%s) %pI4\n", |
596 | ifname, &addr); | |
1da177e4 LT |
597 | |
598 | /* Now bind the socket with the address of multicast interface */ | |
599 | sin.sin_family = AF_INET; | |
600 | sin.sin_addr.s_addr = addr; | |
601 | sin.sin_port = 0; | |
602 | ||
603 | return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin)); | |
604 | } | |
605 | ||
606 | /* | |
607 | * Set up sending multicast socket over UDP | |
608 | */ | |
609 | static struct socket * make_send_sock(void) | |
610 | { | |
611 | struct socket *sock; | |
e6dd731c | 612 | int result; |
1da177e4 LT |
613 | |
614 | /* First create a socket */ | |
e6dd731c SW |
615 | result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock); |
616 | if (result < 0) { | |
1e3e238e | 617 | pr_err("Error during creation of socket; terminating\n"); |
e6dd731c | 618 | return ERR_PTR(result); |
1da177e4 LT |
619 | } |
620 | ||
e6dd731c SW |
621 | result = set_mcast_if(sock->sk, ip_vs_master_mcast_ifn); |
622 | if (result < 0) { | |
1e3e238e | 623 | pr_err("Error setting outbound mcast interface\n"); |
1da177e4 LT |
624 | goto error; |
625 | } | |
626 | ||
627 | set_mcast_loop(sock->sk, 0); | |
628 | set_mcast_ttl(sock->sk, 1); | |
629 | ||
e6dd731c SW |
630 | result = bind_mcastif_addr(sock, ip_vs_master_mcast_ifn); |
631 | if (result < 0) { | |
1e3e238e | 632 | pr_err("Error binding address of the mcast interface\n"); |
1da177e4 LT |
633 | goto error; |
634 | } | |
635 | ||
e6dd731c SW |
636 | result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr, |
637 | sizeof(struct sockaddr), 0); | |
638 | if (result < 0) { | |
1e3e238e | 639 | pr_err("Error connecting to the multicast addr\n"); |
1da177e4 LT |
640 | goto error; |
641 | } | |
642 | ||
643 | return sock; | |
644 | ||
645 | error: | |
646 | sock_release(sock); | |
e6dd731c | 647 | return ERR_PTR(result); |
1da177e4 LT |
648 | } |
649 | ||
650 | ||
651 | /* | |
652 | * Set up receiving multicast socket over UDP | |
653 | */ | |
654 | static struct socket * make_receive_sock(void) | |
655 | { | |
656 | struct socket *sock; | |
e6dd731c | 657 | int result; |
1da177e4 LT |
658 | |
659 | /* First create a socket */ | |
e6dd731c SW |
660 | result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock); |
661 | if (result < 0) { | |
1e3e238e | 662 | pr_err("Error during creation of socket; terminating\n"); |
e6dd731c | 663 | return ERR_PTR(result); |
1da177e4 LT |
664 | } |
665 | ||
666 | /* it is equivalent to the REUSEADDR option in user-space */ | |
667 | sock->sk->sk_reuse = 1; | |
668 | ||
e6dd731c SW |
669 | result = sock->ops->bind(sock, (struct sockaddr *) &mcast_addr, |
670 | sizeof(struct sockaddr)); | |
671 | if (result < 0) { | |
1e3e238e | 672 | pr_err("Error binding to the multicast addr\n"); |
1da177e4 LT |
673 | goto error; |
674 | } | |
675 | ||
676 | /* join the multicast group */ | |
e6dd731c SW |
677 | result = join_mcast_group(sock->sk, |
678 | (struct in_addr *) &mcast_addr.sin_addr, | |
679 | ip_vs_backup_mcast_ifn); | |
680 | if (result < 0) { | |
1e3e238e | 681 | pr_err("Error joining to the multicast group\n"); |
1da177e4 LT |
682 | goto error; |
683 | } | |
684 | ||
685 | return sock; | |
686 | ||
687 | error: | |
688 | sock_release(sock); | |
e6dd731c | 689 | return ERR_PTR(result); |
1da177e4 LT |
690 | } |
691 | ||
692 | ||
693 | static int | |
694 | ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length) | |
695 | { | |
696 | struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL}; | |
697 | struct kvec iov; | |
698 | int len; | |
699 | ||
700 | EnterFunction(7); | |
701 | iov.iov_base = (void *)buffer; | |
702 | iov.iov_len = length; | |
703 | ||
704 | len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length)); | |
705 | ||
706 | LeaveFunction(7); | |
707 | return len; | |
708 | } | |
709 | ||
710 | static void | |
711 | ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg) | |
712 | { | |
713 | int msize; | |
714 | ||
715 | msize = msg->size; | |
716 | ||
717 | /* Put size in network byte order */ | |
718 | msg->size = htons(msg->size); | |
719 | ||
720 | if (ip_vs_send_async(sock, (char *)msg, msize) != msize) | |
1e3e238e | 721 | pr_err("ip_vs_send_async error\n"); |
1da177e4 LT |
722 | } |
723 | ||
724 | static int | |
725 | ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen) | |
726 | { | |
727 | struct msghdr msg = {NULL,}; | |
728 | struct kvec iov; | |
729 | int len; | |
730 | ||
731 | EnterFunction(7); | |
732 | ||
733 | /* Receive a packet */ | |
734 | iov.iov_base = buffer; | |
735 | iov.iov_len = (size_t)buflen; | |
736 | ||
737 | len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, 0); | |
738 | ||
739 | if (len < 0) | |
740 | return -1; | |
741 | ||
742 | LeaveFunction(7); | |
743 | return len; | |
744 | } | |
745 | ||
746 | ||
998e7a76 | 747 | static int sync_thread_master(void *data) |
1da177e4 | 748 | { |
998e7a76 | 749 | struct ip_vs_sync_thread_data *tinfo = data; |
1da177e4 LT |
750 | struct ip_vs_sync_buff *sb; |
751 | ||
1e3e238e HE |
752 | pr_info("sync thread started: state = MASTER, mcast_ifn = %s, " |
753 | "syncid = %d\n", | |
754 | ip_vs_master_mcast_ifn, ip_vs_master_syncid); | |
1da177e4 | 755 | |
998e7a76 SW |
756 | while (!kthread_should_stop()) { |
757 | while ((sb = sb_dequeue())) { | |
758 | ip_vs_send_sync_msg(tinfo->sock, sb->mesg); | |
1da177e4 LT |
759 | ip_vs_sync_buff_release(sb); |
760 | } | |
761 | ||
762 | /* check if entries stay in curr_sb for 2 seconds */ | |
998e7a76 SW |
763 | sb = get_curr_sync_buff(2 * HZ); |
764 | if (sb) { | |
765 | ip_vs_send_sync_msg(tinfo->sock, sb->mesg); | |
1da177e4 LT |
766 | ip_vs_sync_buff_release(sb); |
767 | } | |
768 | ||
375c6bba | 769 | schedule_timeout_interruptible(HZ); |
1da177e4 LT |
770 | } |
771 | ||
772 | /* clean up the sync_buff queue */ | |
773 | while ((sb=sb_dequeue())) { | |
774 | ip_vs_sync_buff_release(sb); | |
775 | } | |
776 | ||
777 | /* clean up the current sync_buff */ | |
778 | if ((sb = get_curr_sync_buff(0))) { | |
779 | ip_vs_sync_buff_release(sb); | |
780 | } | |
781 | ||
782 | /* release the sending multicast socket */ | |
998e7a76 SW |
783 | sock_release(tinfo->sock); |
784 | kfree(tinfo); | |
785 | ||
786 | return 0; | |
1da177e4 LT |
787 | } |
788 | ||
789 | ||
998e7a76 | 790 | static int sync_thread_backup(void *data) |
1da177e4 | 791 | { |
998e7a76 | 792 | struct ip_vs_sync_thread_data *tinfo = data; |
1da177e4 LT |
793 | int len; |
794 | ||
1e3e238e HE |
795 | pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, " |
796 | "syncid = %d\n", | |
797 | ip_vs_backup_mcast_ifn, ip_vs_backup_syncid); | |
1da177e4 | 798 | |
998e7a76 | 799 | while (!kthread_should_stop()) { |
aa395145 | 800 | wait_event_interruptible(*sk_sleep(tinfo->sock->sk), |
ba6fd850 SW |
801 | !skb_queue_empty(&tinfo->sock->sk->sk_receive_queue) |
802 | || kthread_should_stop()); | |
803 | ||
998e7a76 SW |
804 | /* do we have data now? */ |
805 | while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) { | |
806 | len = ip_vs_receive(tinfo->sock, tinfo->buf, | |
807 | sync_recv_mesg_maxlen); | |
808 | if (len <= 0) { | |
1e3e238e | 809 | pr_err("receiving message error\n"); |
1da177e4 LT |
810 | break; |
811 | } | |
998e7a76 SW |
812 | |
813 | /* disable bottom half, because it accesses the data | |
1da177e4 LT |
814 | shared by softirq while getting/creating conns */ |
815 | local_bh_disable(); | |
998e7a76 | 816 | ip_vs_process_message(tinfo->buf, len); |
1da177e4 LT |
817 | local_bh_enable(); |
818 | } | |
1da177e4 LT |
819 | } |
820 | ||
821 | /* release the sending multicast socket */ | |
998e7a76 SW |
822 | sock_release(tinfo->sock); |
823 | kfree(tinfo->buf); | |
824 | kfree(tinfo); | |
1da177e4 | 825 | |
998e7a76 | 826 | return 0; |
1da177e4 LT |
827 | } |
828 | ||
829 | ||
998e7a76 | 830 | int start_sync_thread(int state, char *mcast_ifn, __u8 syncid) |
1da177e4 | 831 | { |
998e7a76 SW |
832 | struct ip_vs_sync_thread_data *tinfo; |
833 | struct task_struct **realtask, *task; | |
834 | struct socket *sock; | |
835 | char *name, *buf = NULL; | |
836 | int (*threadfn)(void *data); | |
837 | int result = -ENOMEM; | |
1da177e4 | 838 | |
1e3e238e | 839 | IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current)); |
998e7a76 SW |
840 | IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n", |
841 | sizeof(struct ip_vs_sync_conn)); | |
1da177e4 | 842 | |
998e7a76 SW |
843 | if (state == IP_VS_STATE_MASTER) { |
844 | if (sync_master_thread) | |
845 | return -EEXIST; | |
1da177e4 | 846 | |
998e7a76 SW |
847 | strlcpy(ip_vs_master_mcast_ifn, mcast_ifn, |
848 | sizeof(ip_vs_master_mcast_ifn)); | |
849 | ip_vs_master_syncid = syncid; | |
850 | realtask = &sync_master_thread; | |
1da177e4 | 851 | name = "ipvs_syncmaster"; |
998e7a76 SW |
852 | threadfn = sync_thread_master; |
853 | sock = make_send_sock(); | |
854 | } else if (state == IP_VS_STATE_BACKUP) { | |
855 | if (sync_backup_thread) | |
856 | return -EEXIST; | |
857 | ||
858 | strlcpy(ip_vs_backup_mcast_ifn, mcast_ifn, | |
859 | sizeof(ip_vs_backup_mcast_ifn)); | |
860 | ip_vs_backup_syncid = syncid; | |
861 | realtask = &sync_backup_thread; | |
1da177e4 | 862 | name = "ipvs_syncbackup"; |
998e7a76 SW |
863 | threadfn = sync_thread_backup; |
864 | sock = make_receive_sock(); | |
1da177e4 | 865 | } else { |
1da177e4 LT |
866 | return -EINVAL; |
867 | } | |
868 | ||
998e7a76 SW |
869 | if (IS_ERR(sock)) { |
870 | result = PTR_ERR(sock); | |
871 | goto out; | |
872 | } | |
1da177e4 | 873 | |
1da177e4 | 874 | set_sync_mesg_maxlen(state); |
998e7a76 SW |
875 | if (state == IP_VS_STATE_BACKUP) { |
876 | buf = kmalloc(sync_recv_mesg_maxlen, GFP_KERNEL); | |
877 | if (!buf) | |
878 | goto outsocket; | |
879 | } | |
1da177e4 | 880 | |
998e7a76 SW |
881 | tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL); |
882 | if (!tinfo) | |
883 | goto outbuf; | |
cc0191ae | 884 | |
998e7a76 SW |
885 | tinfo->sock = sock; |
886 | tinfo->buf = buf; | |
1da177e4 | 887 | |
998e7a76 SW |
888 | task = kthread_run(threadfn, tinfo, name); |
889 | if (IS_ERR(task)) { | |
890 | result = PTR_ERR(task); | |
891 | goto outtinfo; | |
892 | } | |
1da177e4 | 893 | |
998e7a76 SW |
894 | /* mark as active */ |
895 | *realtask = task; | |
896 | ip_vs_sync_state |= state; | |
1da177e4 | 897 | |
998e7a76 SW |
898 | /* increase the module use count */ |
899 | ip_vs_use_count_inc(); | |
1da177e4 LT |
900 | |
901 | return 0; | |
1da177e4 | 902 | |
998e7a76 SW |
903 | outtinfo: |
904 | kfree(tinfo); | |
905 | outbuf: | |
906 | kfree(buf); | |
907 | outsocket: | |
908 | sock_release(sock); | |
909 | out: | |
910 | return result; | |
1da177e4 LT |
911 | } |
912 | ||
913 | ||
998e7a76 | 914 | int stop_sync_thread(int state) |
1da177e4 | 915 | { |
1e3e238e | 916 | IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current)); |
1da177e4 | 917 | |
1da177e4 | 918 | if (state == IP_VS_STATE_MASTER) { |
998e7a76 SW |
919 | if (!sync_master_thread) |
920 | return -ESRCH; | |
cc0191ae | 921 | |
1e3e238e HE |
922 | pr_info("stopping master sync thread %d ...\n", |
923 | task_pid_nr(sync_master_thread)); | |
1da177e4 | 924 | |
998e7a76 SW |
925 | /* |
926 | * The lock synchronizes with sb_queue_tail(), so that we don't | |
927 | * add sync buffers to the queue, when we are already in | |
928 | * progress of stopping the master sync daemon. | |
929 | */ | |
1da177e4 | 930 | |
bc0fde2f | 931 | spin_lock_bh(&ip_vs_sync_lock); |
998e7a76 | 932 | ip_vs_sync_state &= ~IP_VS_STATE_MASTER; |
bc0fde2f | 933 | spin_unlock_bh(&ip_vs_sync_lock); |
998e7a76 SW |
934 | kthread_stop(sync_master_thread); |
935 | sync_master_thread = NULL; | |
936 | } else if (state == IP_VS_STATE_BACKUP) { | |
937 | if (!sync_backup_thread) | |
938 | return -ESRCH; | |
939 | ||
1e3e238e HE |
940 | pr_info("stopping backup sync thread %d ...\n", |
941 | task_pid_nr(sync_backup_thread)); | |
998e7a76 SW |
942 | |
943 | ip_vs_sync_state &= ~IP_VS_STATE_BACKUP; | |
944 | kthread_stop(sync_backup_thread); | |
945 | sync_backup_thread = NULL; | |
946 | } else { | |
947 | return -EINVAL; | |
948 | } | |
1da177e4 | 949 | |
998e7a76 SW |
950 | /* decrease the module use count */ |
951 | ip_vs_use_count_dec(); | |
1da177e4 LT |
952 | |
953 | return 0; | |
954 | } |