1 /*
2 * IPVS An implementation of the IP virtual server support for the
3 * LINUX operating system. IPVS is now implemented as a module
4 * over the NetFilter framework. IPVS can be used to build a
5 * high-performance and highly available server based on a
6 * cluster of servers.
7 *
8 * Version 1, is capable of handling both version 0 and 1 messages.
9 * Version 0 is the plain old format.
10 * Note Version 0 receivers will just drop Ver 1 messages.
11 * Version 1 is capable of handle IPv6, Persistence data,
12 * time-outs, and firewall marks.
13 * In ver.1 "ip_vs_sync_conn_options" will be sent in netw. order.
14 * Ver. 0 can be turned on by sysctl -w net.ipv4.vs.sync_version=0
15 *
16 * Definitions Message: is a complete datagram
17 * Sync_conn: is a part of a Message
18 * Param Data is an option to a Sync_conn.
19 *
20 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
21 *
22 * ip_vs_sync: sync connection info from master load balancer to backups
23 * through multicast
24 *
25 * Changes:
26 * Alexandre Cassen : Added master & backup support at a time.
27 * Alexandre Cassen : Added SyncID support for incoming sync
28 * messages filtering.
29 * Justin Ossevoort : Fix endian problem on sync message size.
30 * Hans Schillstrom : Added Version 1: i.e. IPv6,
31 * Persistence support, fwmark and time-out.
32 */
33
34 #define KMSG_COMPONENT "IPVS"
35 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
36
37 #include <linux/module.h>
38 #include <linux/slab.h>
39 #include <linux/inetdevice.h>
40 #include <linux/net.h>
41 #include <linux/completion.h>
42 #include <linux/delay.h>
43 #include <linux/skbuff.h>
44 #include <linux/in.h>
45 #include <linux/igmp.h> /* for ip_mc_join_group */
46 #include <linux/udp.h>
47 #include <linux/err.h>
48 #include <linux/kthread.h>
49 #include <linux/wait.h>
50 #include <linux/kernel.h>
51
52 #include <asm/unaligned.h> /* Used for ntoh_seq and hton_seq */
53
54 #include <net/ip.h>
55 #include <net/sock.h>
56
57 #include <net/ip_vs.h>
58
59 #define IP_VS_SYNC_GROUP 0xe0000051 /* multicast addr - 224.0.0.81 */
60 #define IP_VS_SYNC_PORT 8848 /* multicast port */
61
62 #define SYNC_PROTO_VER 1 /* Protocol version in header */
63
64 static struct lock_class_key __ipvs_sync_key;
65 /*
66 * IPVS sync connection entry
67 * Version 0, i.e. original version.
68 */
69 struct ip_vs_sync_conn_v0 {
70 __u8 reserved;
71
72 /* Protocol, addresses and port numbers */
73 __u8 protocol; /* Which protocol (TCP/UDP) */
74 __be16 cport;
75 __be16 vport;
76 __be16 dport;
77 __be32 caddr; /* client address */
78 __be32 vaddr; /* virtual address */
79 __be32 daddr; /* destination address */
80
81 /* Flags and state transition */
82 __be16 flags; /* status flags */
83 __be16 state; /* state info */
84
85 /* The sequence options start here */
86 };
87
88 struct ip_vs_sync_conn_options {
89 struct ip_vs_seq in_seq; /* incoming seq. struct */
90 struct ip_vs_seq out_seq; /* outgoing seq. struct */
91 };
92
93 /*
94 Sync Connection format (sync_conn)
95
96 0 1 2 3
97 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
98 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
99 | Type | Protocol | Ver. | Size |
100 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
101 | Flags |
102 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
103 | State | cport |
104 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
105 | vport | dport |
106 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
107 | fwmark |
108 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
109 | timeout (in sec.) |
110 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
111 | ... |
112 | IP-Addresses (v4 or v6) |
113 | ... |
114 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
115 Optional Parameters.
116 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
117 | Param. Type | Param. Length | Param. data |
118 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
119 | ... |
120 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
121 | | Param Type | Param. Length |
122 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
123 | Param data |
124 | Last Param data should be padded for 32 bit alignment |
125 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
126 */
127
128 /*
129 * Type 0, IPv4 sync connection format
130 */
131 struct ip_vs_sync_v4 {
132 __u8 type;
133 __u8 protocol; /* Which protocol (TCP/UDP) */
134 __be16 ver_size; /* Version msb 4 bits */
135 /* Flags and state transition */
136 __be32 flags; /* status flags */
137 __be16 state; /* state info */
138 /* Protocol, addresses and port numbers */
139 __be16 cport;
140 __be16 vport;
141 __be16 dport;
142 __be32 fwmark; /* Firewall mark from skb */
143 __be32 timeout; /* cp timeout */
144 __be32 caddr; /* client address */
145 __be32 vaddr; /* virtual address */
146 __be32 daddr; /* destination address */
147 /* The sequence options start here */
148 /* PE data padded to 32bit alignment after seq. options */
149 };
150 /*
151 * Type 2 messages IPv6
152 */
153 struct ip_vs_sync_v6 {
154 __u8 type;
155 __u8 protocol; /* Which protocol (TCP/UDP) */
156 __be16 ver_size; /* Version msb 4 bits */
157 /* Flags and state transition */
158 __be32 flags; /* status flags */
159 __be16 state; /* state info */
160 /* Protocol, addresses and port numbers */
161 __be16 cport;
162 __be16 vport;
163 __be16 dport;
164 __be32 fwmark; /* Firewall mark from skb */
165 __be32 timeout; /* cp timeout */
166 struct in6_addr caddr; /* client address */
167 struct in6_addr vaddr; /* virtual address */
168 struct in6_addr daddr; /* destination address */
169 /* The sequence options start here */
170 /* PE data padded to 32bit alignment after seq. options */
171 };
172
173 union ip_vs_sync_conn {
174 struct ip_vs_sync_v4 v4;
175 struct ip_vs_sync_v6 v6;
176 };
177
178 /* Bits in Type field in above */
179 #define STYPE_INET6 0
180 #define STYPE_F_INET6 (1 << STYPE_INET6)
181
182 #define SVER_SHIFT 12 /* Shift to get version */
183 #define SVER_MASK 0x0fff /* Mask to strip version */
184
185 #define IPVS_OPT_SEQ_DATA 1
186 #define IPVS_OPT_PE_DATA 2
187 #define IPVS_OPT_PE_NAME 3
188 #define IPVS_OPT_PARAM 7
189
190 #define IPVS_OPT_F_SEQ_DATA (1 << (IPVS_OPT_SEQ_DATA-1))
191 #define IPVS_OPT_F_PE_DATA (1 << (IPVS_OPT_PE_DATA-1))
192 #define IPVS_OPT_F_PE_NAME (1 << (IPVS_OPT_PE_NAME-1))
193 #define IPVS_OPT_F_PARAM (1 << (IPVS_OPT_PARAM-1))
194
195 struct ip_vs_sync_thread_data {
196 struct net *net;
197 struct socket *sock;
198 char *buf;
199 };
200
201 /* Version 0 definition of packet sizes */
202 #define SIMPLE_CONN_SIZE (sizeof(struct ip_vs_sync_conn_v0))
203 #define FULL_CONN_SIZE \
204 (sizeof(struct ip_vs_sync_conn_v0) + sizeof(struct ip_vs_sync_conn_options))
205
206
207 /*
208 The master mulitcasts messages (Datagrams) to the backup load balancers
209 in the following format.
210
211 Version 1:
212 Note, first byte should be Zero, so ver 0 receivers will drop the packet.
213
214 0 1 2 3
215 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
216 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
217 | 0 | SyncID | Size |
218 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
219 | Count Conns | Version | Reserved, set to Zero |
220 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
221 | |
222 | IPVS Sync Connection (1) |
223 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
224 | . |
225 ~ . ~
226 | . |
227 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
228 | |
229 | IPVS Sync Connection (n) |
230 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
231
232 Version 0 Header
233 0 1 2 3
234 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
235 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
236 | Count Conns | SyncID | Size |
237 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
238 | IPVS Sync Connection (1) |
239 */
240
241 #define SYNC_MESG_HEADER_LEN 4
242 #define MAX_CONNS_PER_SYNCBUFF 255 /* nr_conns in ip_vs_sync_mesg is 8 bit */
243
244 /* Version 0 header */
245 struct ip_vs_sync_mesg_v0 {
246 __u8 nr_conns;
247 __u8 syncid;
248 __u16 size;
249
250 /* ip_vs_sync_conn entries start here */
251 };
252
253 /* Version 1 header */
254 struct ip_vs_sync_mesg {
255 __u8 reserved; /* must be zero */
256 __u8 syncid;
257 __u16 size;
258 __u8 nr_conns;
259 __s8 version; /* SYNC_PROTO_VER */
260 __u16 spare;
261 /* ip_vs_sync_conn entries start here */
262 };
263
264 struct ip_vs_sync_buff {
265 struct list_head list;
266 unsigned long firstuse;
267
268 /* pointers for the message data */
269 struct ip_vs_sync_mesg *mesg;
270 unsigned char *head;
271 unsigned char *end;
272 };
273
274 /* multicast addr */
275 static struct sockaddr_in mcast_addr = {
276 .sin_family = AF_INET,
277 .sin_port = cpu_to_be16(IP_VS_SYNC_PORT),
278 .sin_addr.s_addr = cpu_to_be32(IP_VS_SYNC_GROUP),
279 };
280
281 /*
282 * Copy of struct ip_vs_seq
283 * From unaligned network order to aligned host order
284 */
ntoh_seq(struct ip_vs_seq * no,struct ip_vs_seq * ho)285 static void ntoh_seq(struct ip_vs_seq *no, struct ip_vs_seq *ho)
286 {
287 ho->init_seq = get_unaligned_be32(&no->init_seq);
288 ho->delta = get_unaligned_be32(&no->delta);
289 ho->previous_delta = get_unaligned_be32(&no->previous_delta);
290 }
291
292 /*
293 * Copy of struct ip_vs_seq
294 * From Aligned host order to unaligned network order
295 */
hton_seq(struct ip_vs_seq * ho,struct ip_vs_seq * no)296 static void hton_seq(struct ip_vs_seq *ho, struct ip_vs_seq *no)
297 {
298 put_unaligned_be32(ho->init_seq, &no->init_seq);
299 put_unaligned_be32(ho->delta, &no->delta);
300 put_unaligned_be32(ho->previous_delta, &no->previous_delta);
301 }
302
sb_dequeue(struct netns_ipvs * ipvs)303 static inline struct ip_vs_sync_buff *sb_dequeue(struct netns_ipvs *ipvs)
304 {
305 struct ip_vs_sync_buff *sb;
306
307 spin_lock_bh(&ipvs->sync_lock);
308 if (list_empty(&ipvs->sync_queue)) {
309 sb = NULL;
310 } else {
311 sb = list_entry(ipvs->sync_queue.next,
312 struct ip_vs_sync_buff,
313 list);
314 list_del(&sb->list);
315 }
316 spin_unlock_bh(&ipvs->sync_lock);
317
318 return sb;
319 }
320
321 /*
322 * Create a new sync buffer for Version 1 proto.
323 */
324 static inline struct ip_vs_sync_buff *
ip_vs_sync_buff_create(struct netns_ipvs * ipvs)325 ip_vs_sync_buff_create(struct netns_ipvs *ipvs)
326 {
327 struct ip_vs_sync_buff *sb;
328
329 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
330 return NULL;
331
332 sb->mesg = kmalloc(ipvs->send_mesg_maxlen, GFP_ATOMIC);
333 if (!sb->mesg) {
334 kfree(sb);
335 return NULL;
336 }
337 sb->mesg->reserved = 0; /* old nr_conns i.e. must be zeo now */
338 sb->mesg->version = SYNC_PROTO_VER;
339 sb->mesg->syncid = ipvs->master_syncid;
340 sb->mesg->size = sizeof(struct ip_vs_sync_mesg);
341 sb->mesg->nr_conns = 0;
342 sb->mesg->spare = 0;
343 sb->head = (unsigned char *)sb->mesg + sizeof(struct ip_vs_sync_mesg);
344 sb->end = (unsigned char *)sb->mesg + ipvs->send_mesg_maxlen;
345
346 sb->firstuse = jiffies;
347 return sb;
348 }
349
ip_vs_sync_buff_release(struct ip_vs_sync_buff * sb)350 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
351 {
352 kfree(sb->mesg);
353 kfree(sb);
354 }
355
sb_queue_tail(struct netns_ipvs * ipvs)356 static inline void sb_queue_tail(struct netns_ipvs *ipvs)
357 {
358 struct ip_vs_sync_buff *sb = ipvs->sync_buff;
359
360 spin_lock(&ipvs->sync_lock);
361 if (ipvs->sync_state & IP_VS_STATE_MASTER)
362 list_add_tail(&sb->list, &ipvs->sync_queue);
363 else
364 ip_vs_sync_buff_release(sb);
365 spin_unlock(&ipvs->sync_lock);
366 }
367
368 /*
369 * Get the current sync buffer if it has been created for more
370 * than the specified time or the specified time is zero.
371 */
372 static inline struct ip_vs_sync_buff *
get_curr_sync_buff(struct netns_ipvs * ipvs,unsigned long time)373 get_curr_sync_buff(struct netns_ipvs *ipvs, unsigned long time)
374 {
375 struct ip_vs_sync_buff *sb;
376
377 spin_lock_bh(&ipvs->sync_buff_lock);
378 if (ipvs->sync_buff &&
379 time_after_eq(jiffies - ipvs->sync_buff->firstuse, time)) {
380 sb = ipvs->sync_buff;
381 ipvs->sync_buff = NULL;
382 } else
383 sb = NULL;
384 spin_unlock_bh(&ipvs->sync_buff_lock);
385 return sb;
386 }
387
388 /*
389 * Switch mode from sending version 0 or 1
390 * - must handle sync_buf
391 */
ip_vs_sync_switch_mode(struct net * net,int mode)392 void ip_vs_sync_switch_mode(struct net *net, int mode)
393 {
394 struct netns_ipvs *ipvs = net_ipvs(net);
395
396 if (!(ipvs->sync_state & IP_VS_STATE_MASTER))
397 return;
398 if (mode == sysctl_sync_ver(ipvs) || !ipvs->sync_buff)
399 return;
400
401 spin_lock_bh(&ipvs->sync_buff_lock);
402 /* Buffer empty ? then let buf_create do the job */
403 if (ipvs->sync_buff->mesg->size <= sizeof(struct ip_vs_sync_mesg)) {
404 kfree(ipvs->sync_buff);
405 ipvs->sync_buff = NULL;
406 } else {
407 spin_lock_bh(&ipvs->sync_lock);
408 if (ipvs->sync_state & IP_VS_STATE_MASTER)
409 list_add_tail(&ipvs->sync_buff->list,
410 &ipvs->sync_queue);
411 else
412 ip_vs_sync_buff_release(ipvs->sync_buff);
413 spin_unlock_bh(&ipvs->sync_lock);
414 }
415 spin_unlock_bh(&ipvs->sync_buff_lock);
416 }
417
418 /*
419 * Create a new sync buffer for Version 0 proto.
420 */
421 static inline struct ip_vs_sync_buff *
ip_vs_sync_buff_create_v0(struct netns_ipvs * ipvs)422 ip_vs_sync_buff_create_v0(struct netns_ipvs *ipvs)
423 {
424 struct ip_vs_sync_buff *sb;
425 struct ip_vs_sync_mesg_v0 *mesg;
426
427 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
428 return NULL;
429
430 sb->mesg = kmalloc(ipvs->send_mesg_maxlen, GFP_ATOMIC);
431 if (!sb->mesg) {
432 kfree(sb);
433 return NULL;
434 }
435 mesg = (struct ip_vs_sync_mesg_v0 *)sb->mesg;
436 mesg->nr_conns = 0;
437 mesg->syncid = ipvs->master_syncid;
438 mesg->size = sizeof(struct ip_vs_sync_mesg_v0);
439 sb->head = (unsigned char *)mesg + sizeof(struct ip_vs_sync_mesg_v0);
440 sb->end = (unsigned char *)mesg + ipvs->send_mesg_maxlen;
441 sb->firstuse = jiffies;
442 return sb;
443 }
444
445 /*
446 * Version 0 , could be switched in by sys_ctl.
447 * Add an ip_vs_conn information into the current sync_buff.
448 */
ip_vs_sync_conn_v0(struct net * net,struct ip_vs_conn * cp)449 void ip_vs_sync_conn_v0(struct net *net, struct ip_vs_conn *cp)
450 {
451 struct netns_ipvs *ipvs = net_ipvs(net);
452 struct ip_vs_sync_mesg_v0 *m;
453 struct ip_vs_sync_conn_v0 *s;
454 int len;
455
456 if (unlikely(cp->af != AF_INET))
457 return;
458 /* Do not sync ONE PACKET */
459 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
460 return;
461
462 spin_lock(&ipvs->sync_buff_lock);
463 if (!ipvs->sync_buff) {
464 ipvs->sync_buff =
465 ip_vs_sync_buff_create_v0(ipvs);
466 if (!ipvs->sync_buff) {
467 spin_unlock(&ipvs->sync_buff_lock);
468 pr_err("ip_vs_sync_buff_create failed.\n");
469 return;
470 }
471 }
472
473 len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
474 SIMPLE_CONN_SIZE;
475 m = (struct ip_vs_sync_mesg_v0 *)ipvs->sync_buff->mesg;
476 s = (struct ip_vs_sync_conn_v0 *)ipvs->sync_buff->head;
477
478 /* copy members */
479 s->reserved = 0;
480 s->protocol = cp->protocol;
481 s->cport = cp->cport;
482 s->vport = cp->vport;
483 s->dport = cp->dport;
484 s->caddr = cp->caddr.ip;
485 s->vaddr = cp->vaddr.ip;
486 s->daddr = cp->daddr.ip;
487 s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
488 s->state = htons(cp->state);
489 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
490 struct ip_vs_sync_conn_options *opt =
491 (struct ip_vs_sync_conn_options *)&s[1];
492 memcpy(opt, &cp->in_seq, sizeof(*opt));
493 }
494
495 m->nr_conns++;
496 m->size += len;
497 ipvs->sync_buff->head += len;
498
499 /* check if there is a space for next one */
500 if (ipvs->sync_buff->head + FULL_CONN_SIZE > ipvs->sync_buff->end) {
501 sb_queue_tail(ipvs);
502 ipvs->sync_buff = NULL;
503 }
504 spin_unlock(&ipvs->sync_buff_lock);
505
506 /* synchronize its controller if it has */
507 if (cp->control)
508 ip_vs_sync_conn(net, cp->control);
509 }
510
511 /*
512 * Add an ip_vs_conn information into the current sync_buff.
513 * Called by ip_vs_in.
514 * Sending Version 1 messages
515 */
ip_vs_sync_conn(struct net * net,struct ip_vs_conn * cp)516 void ip_vs_sync_conn(struct net *net, struct ip_vs_conn *cp)
517 {
518 struct netns_ipvs *ipvs = net_ipvs(net);
519 struct ip_vs_sync_mesg *m;
520 union ip_vs_sync_conn *s;
521 __u8 *p;
522 unsigned int len, pe_name_len, pad;
523
524 /* Handle old version of the protocol */
525 if (sysctl_sync_ver(ipvs) == 0) {
526 ip_vs_sync_conn_v0(net, cp);
527 return;
528 }
529 /* Do not sync ONE PACKET */
530 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
531 goto control;
532 sloop:
533 /* Sanity checks */
534 pe_name_len = 0;
535 if (cp->pe_data_len) {
536 if (!cp->pe_data || !cp->dest) {
537 IP_VS_ERR_RL("SYNC, connection pe_data invalid\n");
538 return;
539 }
540 pe_name_len = strnlen(cp->pe->name, IP_VS_PENAME_MAXLEN);
541 }
542
543 spin_lock(&ipvs->sync_buff_lock);
544
545 #ifdef CONFIG_IP_VS_IPV6
546 if (cp->af == AF_INET6)
547 len = sizeof(struct ip_vs_sync_v6);
548 else
549 #endif
550 len = sizeof(struct ip_vs_sync_v4);
551
552 if (cp->flags & IP_VS_CONN_F_SEQ_MASK)
553 len += sizeof(struct ip_vs_sync_conn_options) + 2;
554
555 if (cp->pe_data_len)
556 len += cp->pe_data_len + 2; /* + Param hdr field */
557 if (pe_name_len)
558 len += pe_name_len + 2;
559
560 /* check if there is a space for this one */
561 pad = 0;
562 if (ipvs->sync_buff) {
563 pad = (4 - (size_t)ipvs->sync_buff->head) & 3;
564 if (ipvs->sync_buff->head + len + pad > ipvs->sync_buff->end) {
565 sb_queue_tail(ipvs);
566 ipvs->sync_buff = NULL;
567 pad = 0;
568 }
569 }
570
571 if (!ipvs->sync_buff) {
572 ipvs->sync_buff = ip_vs_sync_buff_create(ipvs);
573 if (!ipvs->sync_buff) {
574 spin_unlock(&ipvs->sync_buff_lock);
575 pr_err("ip_vs_sync_buff_create failed.\n");
576 return;
577 }
578 }
579
580 m = ipvs->sync_buff->mesg;
581 p = ipvs->sync_buff->head;
582 ipvs->sync_buff->head += pad + len;
583 m->size += pad + len;
584 /* Add ev. padding from prev. sync_conn */
585 while (pad--)
586 *(p++) = 0;
587
588 s = (union ip_vs_sync_conn *)p;
589
590 /* Set message type & copy members */
591 s->v4.type = (cp->af == AF_INET6 ? STYPE_F_INET6 : 0);
592 s->v4.ver_size = htons(len & SVER_MASK); /* Version 0 */
593 s->v4.flags = htonl(cp->flags & ~IP_VS_CONN_F_HASHED);
594 s->v4.state = htons(cp->state);
595 s->v4.protocol = cp->protocol;
596 s->v4.cport = cp->cport;
597 s->v4.vport = cp->vport;
598 s->v4.dport = cp->dport;
599 s->v4.fwmark = htonl(cp->fwmark);
600 s->v4.timeout = htonl(cp->timeout / HZ);
601 m->nr_conns++;
602
603 #ifdef CONFIG_IP_VS_IPV6
604 if (cp->af == AF_INET6) {
605 p += sizeof(struct ip_vs_sync_v6);
606 s->v6.caddr = cp->caddr.in6;
607 s->v6.vaddr = cp->vaddr.in6;
608 s->v6.daddr = cp->daddr.in6;
609 } else
610 #endif
611 {
612 p += sizeof(struct ip_vs_sync_v4); /* options ptr */
613 s->v4.caddr = cp->caddr.ip;
614 s->v4.vaddr = cp->vaddr.ip;
615 s->v4.daddr = cp->daddr.ip;
616 }
617 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
618 *(p++) = IPVS_OPT_SEQ_DATA;
619 *(p++) = sizeof(struct ip_vs_sync_conn_options);
620 hton_seq((struct ip_vs_seq *)p, &cp->in_seq);
621 p += sizeof(struct ip_vs_seq);
622 hton_seq((struct ip_vs_seq *)p, &cp->out_seq);
623 p += sizeof(struct ip_vs_seq);
624 }
625 /* Handle pe data */
626 if (cp->pe_data_len && cp->pe_data) {
627 *(p++) = IPVS_OPT_PE_DATA;
628 *(p++) = cp->pe_data_len;
629 memcpy(p, cp->pe_data, cp->pe_data_len);
630 p += cp->pe_data_len;
631 if (pe_name_len) {
632 /* Add PE_NAME */
633 *(p++) = IPVS_OPT_PE_NAME;
634 *(p++) = pe_name_len;
635 memcpy(p, cp->pe->name, pe_name_len);
636 p += pe_name_len;
637 }
638 }
639
640 spin_unlock(&ipvs->sync_buff_lock);
641
642 control:
643 /* synchronize its controller if it has */
644 cp = cp->control;
645 if (!cp)
646 return;
647 /*
648 * Reduce sync rate for templates
649 * i.e only increment in_pkts for Templates.
650 */
651 if (cp->flags & IP_VS_CONN_F_TEMPLATE) {
652 int pkts = atomic_add_return(1, &cp->in_pkts);
653
654 if (pkts % sysctl_sync_period(ipvs) != 1)
655 return;
656 }
657 goto sloop;
658 }
659
660 /*
661 * fill_param used by version 1
662 */
663 static inline int
ip_vs_conn_fill_param_sync(struct net * net,int af,union ip_vs_sync_conn * sc,struct ip_vs_conn_param * p,__u8 * pe_data,unsigned int pe_data_len,__u8 * pe_name,unsigned int pe_name_len)664 ip_vs_conn_fill_param_sync(struct net *net, int af, union ip_vs_sync_conn *sc,
665 struct ip_vs_conn_param *p,
666 __u8 *pe_data, unsigned int pe_data_len,
667 __u8 *pe_name, unsigned int pe_name_len)
668 {
669 #ifdef CONFIG_IP_VS_IPV6
670 if (af == AF_INET6)
671 ip_vs_conn_fill_param(net, af, sc->v6.protocol,
672 (const union nf_inet_addr *)&sc->v6.caddr,
673 sc->v6.cport,
674 (const union nf_inet_addr *)&sc->v6.vaddr,
675 sc->v6.vport, p);
676 else
677 #endif
678 ip_vs_conn_fill_param(net, af, sc->v4.protocol,
679 (const union nf_inet_addr *)&sc->v4.caddr,
680 sc->v4.cport,
681 (const union nf_inet_addr *)&sc->v4.vaddr,
682 sc->v4.vport, p);
683 /* Handle pe data */
684 if (pe_data_len) {
685 if (pe_name_len) {
686 char buff[IP_VS_PENAME_MAXLEN+1];
687
688 memcpy(buff, pe_name, pe_name_len);
689 buff[pe_name_len]=0;
690 p->pe = __ip_vs_pe_getbyname(buff);
691 if (!p->pe) {
692 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n",
693 buff);
694 return 1;
695 }
696 } else {
697 IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n");
698 return 1;
699 }
700
701 p->pe_data = kmemdup(pe_data, pe_data_len, GFP_ATOMIC);
702 if (!p->pe_data) {
703 if (p->pe->module)
704 module_put(p->pe->module);
705 return -ENOMEM;
706 }
707 p->pe_data_len = pe_data_len;
708 }
709 return 0;
710 }
711
712 /*
713 * Connection Add / Update.
714 * Common for version 0 and 1 reception of backup sync_conns.
715 * Param: ...
716 * timeout is in sec.
717 */
ip_vs_proc_conn(struct net * net,struct ip_vs_conn_param * param,unsigned int flags,unsigned int state,unsigned int protocol,unsigned int type,const union nf_inet_addr * daddr,__be16 dport,unsigned long timeout,__u32 fwmark,struct ip_vs_sync_conn_options * opt)718 static void ip_vs_proc_conn(struct net *net, struct ip_vs_conn_param *param,
719 unsigned int flags, unsigned int state,
720 unsigned int protocol, unsigned int type,
721 const union nf_inet_addr *daddr, __be16 dport,
722 unsigned long timeout, __u32 fwmark,
723 struct ip_vs_sync_conn_options *opt)
724 {
725 struct ip_vs_dest *dest;
726 struct ip_vs_conn *cp;
727 struct netns_ipvs *ipvs = net_ipvs(net);
728
729 if (!(flags & IP_VS_CONN_F_TEMPLATE))
730 cp = ip_vs_conn_in_get(param);
731 else
732 cp = ip_vs_ct_in_get(param);
733
734 if (cp && param->pe_data) /* Free pe_data */
735 kfree(param->pe_data);
736 if (!cp) {
737 /*
738 * Find the appropriate destination for the connection.
739 * If it is not found the connection will remain unbound
740 * but still handled.
741 */
742 dest = ip_vs_find_dest(net, type, daddr, dport, param->vaddr,
743 param->vport, protocol, fwmark, flags);
744
745 /* Set the approprite ativity flag */
746 if (protocol == IPPROTO_TCP) {
747 if (state != IP_VS_TCP_S_ESTABLISHED)
748 flags |= IP_VS_CONN_F_INACTIVE;
749 else
750 flags &= ~IP_VS_CONN_F_INACTIVE;
751 } else if (protocol == IPPROTO_SCTP) {
752 if (state != IP_VS_SCTP_S_ESTABLISHED)
753 flags |= IP_VS_CONN_F_INACTIVE;
754 else
755 flags &= ~IP_VS_CONN_F_INACTIVE;
756 }
757 cp = ip_vs_conn_new(param, daddr, dport, flags, dest, fwmark);
758 if (dest)
759 atomic_dec(&dest->refcnt);
760 if (!cp) {
761 if (param->pe_data)
762 kfree(param->pe_data);
763 IP_VS_DBG(2, "BACKUP, add new conn. failed\n");
764 return;
765 }
766 } else if (!cp->dest) {
767 dest = ip_vs_try_bind_dest(cp);
768 if (dest)
769 atomic_dec(&dest->refcnt);
770 } else if ((cp->dest) && (cp->protocol == IPPROTO_TCP) &&
771 (cp->state != state)) {
772 /* update active/inactive flag for the connection */
773 dest = cp->dest;
774 if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
775 (state != IP_VS_TCP_S_ESTABLISHED)) {
776 atomic_dec(&dest->activeconns);
777 atomic_inc(&dest->inactconns);
778 cp->flags |= IP_VS_CONN_F_INACTIVE;
779 } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) &&
780 (state == IP_VS_TCP_S_ESTABLISHED)) {
781 atomic_inc(&dest->activeconns);
782 atomic_dec(&dest->inactconns);
783 cp->flags &= ~IP_VS_CONN_F_INACTIVE;
784 }
785 } else if ((cp->dest) && (cp->protocol == IPPROTO_SCTP) &&
786 (cp->state != state)) {
787 dest = cp->dest;
788 if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
789 (state != IP_VS_SCTP_S_ESTABLISHED)) {
790 atomic_dec(&dest->activeconns);
791 atomic_inc(&dest->inactconns);
792 cp->flags &= ~IP_VS_CONN_F_INACTIVE;
793 }
794 }
795
796 if (opt)
797 memcpy(&cp->in_seq, opt, sizeof(*opt));
798 atomic_set(&cp->in_pkts, sysctl_sync_threshold(ipvs));
799 cp->state = state;
800 cp->old_state = cp->state;
801 /*
802 * For Ver 0 messages style
803 * - Not possible to recover the right timeout for templates
804 * - can not find the right fwmark
805 * virtual service. If needed, we can do it for
806 * non-fwmark persistent services.
807 * Ver 1 messages style.
808 * - No problem.
809 */
810 if (timeout) {
811 if (timeout > MAX_SCHEDULE_TIMEOUT / HZ)
812 timeout = MAX_SCHEDULE_TIMEOUT / HZ;
813 cp->timeout = timeout*HZ;
814 } else {
815 struct ip_vs_proto_data *pd;
816
817 pd = ip_vs_proto_data_get(net, protocol);
818 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pd && pd->timeout_table)
819 cp->timeout = pd->timeout_table[state];
820 else
821 cp->timeout = (3*60*HZ);
822 }
823 ip_vs_conn_put(cp);
824 }
825
826 /*
827 * Process received multicast message for Version 0
828 */
ip_vs_process_message_v0(struct net * net,const char * buffer,const size_t buflen)829 static void ip_vs_process_message_v0(struct net *net, const char *buffer,
830 const size_t buflen)
831 {
832 struct ip_vs_sync_mesg_v0 *m = (struct ip_vs_sync_mesg_v0 *)buffer;
833 struct ip_vs_sync_conn_v0 *s;
834 struct ip_vs_sync_conn_options *opt;
835 struct ip_vs_protocol *pp;
836 struct ip_vs_conn_param param;
837 char *p;
838 int i;
839
840 p = (char *)buffer + sizeof(struct ip_vs_sync_mesg_v0);
841 for (i=0; i<m->nr_conns; i++) {
842 unsigned flags, state;
843
844 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
845 IP_VS_ERR_RL("BACKUP v0, bogus conn\n");
846 return;
847 }
848 s = (struct ip_vs_sync_conn_v0 *) p;
849 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
850 flags &= ~IP_VS_CONN_F_HASHED;
851 if (flags & IP_VS_CONN_F_SEQ_MASK) {
852 opt = (struct ip_vs_sync_conn_options *)&s[1];
853 p += FULL_CONN_SIZE;
854 if (p > buffer+buflen) {
855 IP_VS_ERR_RL("BACKUP v0, Dropping buffer bogus conn options\n");
856 return;
857 }
858 } else {
859 opt = NULL;
860 p += SIMPLE_CONN_SIZE;
861 }
862
863 state = ntohs(s->state);
864 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
865 pp = ip_vs_proto_get(s->protocol);
866 if (!pp) {
867 IP_VS_DBG(2, "BACKUP v0, Unsupported protocol %u\n",
868 s->protocol);
869 continue;
870 }
871 if (state >= pp->num_states) {
872 IP_VS_DBG(2, "BACKUP v0, Invalid %s state %u\n",
873 pp->name, state);
874 continue;
875 }
876 } else {
877 /* protocol in templates is not used for state/timeout */
878 if (state > 0) {
879 IP_VS_DBG(2, "BACKUP v0, Invalid template state %u\n",
880 state);
881 state = 0;
882 }
883 }
884
885 ip_vs_conn_fill_param(net, AF_INET, s->protocol,
886 (const union nf_inet_addr *)&s->caddr,
887 s->cport,
888 (const union nf_inet_addr *)&s->vaddr,
889 s->vport, ¶m);
890
891 /* Send timeout as Zero */
892 ip_vs_proc_conn(net, ¶m, flags, state, s->protocol, AF_INET,
893 (union nf_inet_addr *)&s->daddr, s->dport,
894 0, 0, opt);
895 }
896 }
897
898 /*
899 * Handle options
900 */
ip_vs_proc_seqopt(__u8 * p,unsigned int plen,__u32 * opt_flags,struct ip_vs_sync_conn_options * opt)901 static inline int ip_vs_proc_seqopt(__u8 *p, unsigned int plen,
902 __u32 *opt_flags,
903 struct ip_vs_sync_conn_options *opt)
904 {
905 struct ip_vs_sync_conn_options *topt;
906
907 topt = (struct ip_vs_sync_conn_options *)p;
908
909 if (plen != sizeof(struct ip_vs_sync_conn_options)) {
910 IP_VS_DBG(2, "BACKUP, bogus conn options length\n");
911 return -EINVAL;
912 }
913 if (*opt_flags & IPVS_OPT_F_SEQ_DATA) {
914 IP_VS_DBG(2, "BACKUP, conn options found twice\n");
915 return -EINVAL;
916 }
917 ntoh_seq(&topt->in_seq, &opt->in_seq);
918 ntoh_seq(&topt->out_seq, &opt->out_seq);
919 *opt_flags |= IPVS_OPT_F_SEQ_DATA;
920 return 0;
921 }
922
ip_vs_proc_str(__u8 * p,unsigned int plen,unsigned int * data_len,__u8 ** data,unsigned int maxlen,__u32 * opt_flags,__u32 flag)923 static int ip_vs_proc_str(__u8 *p, unsigned int plen, unsigned int *data_len,
924 __u8 **data, unsigned int maxlen,
925 __u32 *opt_flags, __u32 flag)
926 {
927 if (plen > maxlen) {
928 IP_VS_DBG(2, "BACKUP, bogus par.data len > %d\n", maxlen);
929 return -EINVAL;
930 }
931 if (*opt_flags & flag) {
932 IP_VS_DBG(2, "BACKUP, Par.data found twice 0x%x\n", flag);
933 return -EINVAL;
934 }
935 *data_len = plen;
936 *data = p;
937 *opt_flags |= flag;
938 return 0;
939 }
940 /*
941 * Process a Version 1 sync. connection
942 */
ip_vs_proc_sync_conn(struct net * net,__u8 * p,__u8 * msg_end)943 static inline int ip_vs_proc_sync_conn(struct net *net, __u8 *p, __u8 *msg_end)
944 {
945 struct ip_vs_sync_conn_options opt;
946 union ip_vs_sync_conn *s;
947 struct ip_vs_protocol *pp;
948 struct ip_vs_conn_param param;
949 __u32 flags;
950 unsigned int af, state, pe_data_len=0, pe_name_len=0;
951 __u8 *pe_data=NULL, *pe_name=NULL;
952 __u32 opt_flags=0;
953 int retc=0;
954
955 s = (union ip_vs_sync_conn *) p;
956
957 if (s->v6.type & STYPE_F_INET6) {
958 #ifdef CONFIG_IP_VS_IPV6
959 af = AF_INET6;
960 p += sizeof(struct ip_vs_sync_v6);
961 #else
962 IP_VS_DBG(3,"BACKUP, IPv6 msg received, and IPVS is not compiled for IPv6\n");
963 retc = 10;
964 goto out;
965 #endif
966 } else if (!s->v4.type) {
967 af = AF_INET;
968 p += sizeof(struct ip_vs_sync_v4);
969 } else {
970 return -10;
971 }
972 if (p > msg_end)
973 return -20;
974
975 /* Process optional params check Type & Len. */
976 while (p < msg_end) {
977 int ptype;
978 int plen;
979
980 if (p+2 > msg_end)
981 return -30;
982 ptype = *(p++);
983 plen = *(p++);
984
985 if (!plen || ((p + plen) > msg_end))
986 return -40;
987 /* Handle seq option p = param data */
988 switch (ptype & ~IPVS_OPT_F_PARAM) {
989 case IPVS_OPT_SEQ_DATA:
990 if (ip_vs_proc_seqopt(p, plen, &opt_flags, &opt))
991 return -50;
992 break;
993
994 case IPVS_OPT_PE_DATA:
995 if (ip_vs_proc_str(p, plen, &pe_data_len, &pe_data,
996 IP_VS_PEDATA_MAXLEN, &opt_flags,
997 IPVS_OPT_F_PE_DATA))
998 return -60;
999 break;
1000
1001 case IPVS_OPT_PE_NAME:
1002 if (ip_vs_proc_str(p, plen,&pe_name_len, &pe_name,
1003 IP_VS_PENAME_MAXLEN, &opt_flags,
1004 IPVS_OPT_F_PE_NAME))
1005 return -70;
1006 break;
1007
1008 default:
1009 /* Param data mandatory ? */
1010 if (!(ptype & IPVS_OPT_F_PARAM)) {
1011 IP_VS_DBG(3, "BACKUP, Unknown mandatory param %d found\n",
1012 ptype & ~IPVS_OPT_F_PARAM);
1013 retc = 20;
1014 goto out;
1015 }
1016 }
1017 p += plen; /* Next option */
1018 }
1019
1020 /* Get flags and Mask off unsupported */
1021 flags = ntohl(s->v4.flags) & IP_VS_CONN_F_BACKUP_MASK;
1022 flags |= IP_VS_CONN_F_SYNC;
1023 state = ntohs(s->v4.state);
1024
1025 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
1026 pp = ip_vs_proto_get(s->v4.protocol);
1027 if (!pp) {
1028 IP_VS_DBG(3,"BACKUP, Unsupported protocol %u\n",
1029 s->v4.protocol);
1030 retc = 30;
1031 goto out;
1032 }
1033 if (state >= pp->num_states) {
1034 IP_VS_DBG(3, "BACKUP, Invalid %s state %u\n",
1035 pp->name, state);
1036 retc = 40;
1037 goto out;
1038 }
1039 } else {
1040 /* protocol in templates is not used for state/timeout */
1041 if (state > 0) {
1042 IP_VS_DBG(3, "BACKUP, Invalid template state %u\n",
1043 state);
1044 state = 0;
1045 }
1046 }
1047 if (ip_vs_conn_fill_param_sync(net, af, s, ¶m, pe_data,
1048 pe_data_len, pe_name, pe_name_len)) {
1049 retc = 50;
1050 goto out;
1051 }
1052 /* If only IPv4, just silent skip IPv6 */
1053 if (af == AF_INET)
1054 ip_vs_proc_conn(net, ¶m, flags, state, s->v4.protocol, af,
1055 (union nf_inet_addr *)&s->v4.daddr, s->v4.dport,
1056 ntohl(s->v4.timeout), ntohl(s->v4.fwmark),
1057 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1058 );
1059 #ifdef CONFIG_IP_VS_IPV6
1060 else
1061 ip_vs_proc_conn(net, ¶m, flags, state, s->v6.protocol, af,
1062 (union nf_inet_addr *)&s->v6.daddr, s->v6.dport,
1063 ntohl(s->v6.timeout), ntohl(s->v6.fwmark),
1064 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1065 );
1066 #endif
1067 return 0;
1068 /* Error exit */
1069 out:
1070 IP_VS_DBG(2, "BACKUP, Single msg dropped err:%d\n", retc);
1071 return retc;
1072
1073 }
1074 /*
1075 * Process received multicast message and create the corresponding
1076 * ip_vs_conn entries.
1077 * Handles Version 0 & 1
1078 */
ip_vs_process_message(struct net * net,__u8 * buffer,const size_t buflen)1079 static void ip_vs_process_message(struct net *net, __u8 *buffer,
1080 const size_t buflen)
1081 {
1082 struct netns_ipvs *ipvs = net_ipvs(net);
1083 struct ip_vs_sync_mesg *m2 = (struct ip_vs_sync_mesg *)buffer;
1084 __u8 *p, *msg_end;
1085 int i, nr_conns;
1086
1087 if (buflen < sizeof(struct ip_vs_sync_mesg_v0)) {
1088 IP_VS_DBG(2, "BACKUP, message header too short\n");
1089 return;
1090 }
1091 /* Convert size back to host byte order */
1092 m2->size = ntohs(m2->size);
1093
1094 if (buflen != m2->size) {
1095 IP_VS_DBG(2, "BACKUP, bogus message size\n");
1096 return;
1097 }
1098 /* SyncID sanity check */
1099 if (ipvs->backup_syncid != 0 && m2->syncid != ipvs->backup_syncid) {
1100 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2->syncid);
1101 return;
1102 }
1103 /* Handle version 1 message */
1104 if ((m2->version == SYNC_PROTO_VER) && (m2->reserved == 0)
1105 && (m2->spare == 0)) {
1106
1107 msg_end = buffer + sizeof(struct ip_vs_sync_mesg);
1108 nr_conns = m2->nr_conns;
1109
1110 for (i=0; i<nr_conns; i++) {
1111 union ip_vs_sync_conn *s;
1112 unsigned size;
1113 int retc;
1114
1115 p = msg_end;
1116 if (p + sizeof(s->v4) > buffer+buflen) {
1117 IP_VS_ERR_RL("BACKUP, Dropping buffer, to small\n");
1118 return;
1119 }
1120 s = (union ip_vs_sync_conn *)p;
1121 size = ntohs(s->v4.ver_size) & SVER_MASK;
1122 msg_end = p + size;
1123 /* Basic sanity checks */
1124 if (msg_end > buffer+buflen) {
1125 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n");
1126 return;
1127 }
1128 if (ntohs(s->v4.ver_size) >> SVER_SHIFT) {
1129 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n",
1130 ntohs(s->v4.ver_size) >> SVER_SHIFT);
1131 return;
1132 }
1133 /* Process a single sync_conn */
1134 retc = ip_vs_proc_sync_conn(net, p, msg_end);
1135 if (retc < 0) {
1136 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n",
1137 retc);
1138 return;
1139 }
1140 /* Make sure we have 32 bit alignment */
1141 msg_end = p + ((size + 3) & ~3);
1142 }
1143 } else {
1144 /* Old type of message */
1145 ip_vs_process_message_v0(net, buffer, buflen);
1146 return;
1147 }
1148 }
1149
1150
1151 /*
1152 * Setup loopback of outgoing multicasts on a sending socket
1153 */
set_mcast_loop(struct sock * sk,u_char loop)1154 static void set_mcast_loop(struct sock *sk, u_char loop)
1155 {
1156 struct inet_sock *inet = inet_sk(sk);
1157
1158 /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
1159 lock_sock(sk);
1160 inet->mc_loop = loop ? 1 : 0;
1161 release_sock(sk);
1162 }
1163
1164 /*
1165 * Specify TTL for outgoing multicasts on a sending socket
1166 */
set_mcast_ttl(struct sock * sk,u_char ttl)1167 static void set_mcast_ttl(struct sock *sk, u_char ttl)
1168 {
1169 struct inet_sock *inet = inet_sk(sk);
1170
1171 /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
1172 lock_sock(sk);
1173 inet->mc_ttl = ttl;
1174 release_sock(sk);
1175 }
1176
1177 /*
1178 * Specifiy default interface for outgoing multicasts
1179 */
set_mcast_if(struct sock * sk,char * ifname)1180 static int set_mcast_if(struct sock *sk, char *ifname)
1181 {
1182 struct net_device *dev;
1183 struct inet_sock *inet = inet_sk(sk);
1184 struct net *net = sock_net(sk);
1185
1186 dev = __dev_get_by_name(net, ifname);
1187 if (!dev)
1188 return -ENODEV;
1189
1190 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1191 return -EINVAL;
1192
1193 lock_sock(sk);
1194 inet->mc_index = dev->ifindex;
1195 /* inet->mc_addr = 0; */
1196 release_sock(sk);
1197
1198 return 0;
1199 }
1200
1201
1202 /*
1203 * Set the maximum length of sync message according to the
1204 * specified interface's MTU.
1205 */
set_sync_mesg_maxlen(struct net * net,int sync_state)1206 static int set_sync_mesg_maxlen(struct net *net, int sync_state)
1207 {
1208 struct netns_ipvs *ipvs = net_ipvs(net);
1209 struct net_device *dev;
1210 int num;
1211
1212 if (sync_state == IP_VS_STATE_MASTER) {
1213 dev = __dev_get_by_name(net, ipvs->master_mcast_ifn);
1214 if (!dev)
1215 return -ENODEV;
1216
1217 num = (dev->mtu - sizeof(struct iphdr) -
1218 sizeof(struct udphdr) -
1219 SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE;
1220 ipvs->send_mesg_maxlen = SYNC_MESG_HEADER_LEN +
1221 SIMPLE_CONN_SIZE * min(num, MAX_CONNS_PER_SYNCBUFF);
1222 IP_VS_DBG(7, "setting the maximum length of sync sending "
1223 "message %d.\n", ipvs->send_mesg_maxlen);
1224 } else if (sync_state == IP_VS_STATE_BACKUP) {
1225 dev = __dev_get_by_name(net, ipvs->backup_mcast_ifn);
1226 if (!dev)
1227 return -ENODEV;
1228
1229 ipvs->recv_mesg_maxlen = dev->mtu -
1230 sizeof(struct iphdr) - sizeof(struct udphdr);
1231 IP_VS_DBG(7, "setting the maximum length of sync receiving "
1232 "message %d.\n", ipvs->recv_mesg_maxlen);
1233 }
1234
1235 return 0;
1236 }
1237
1238
1239 /*
1240 * Join a multicast group.
1241 * the group is specified by a class D multicast address 224.0.0.0/8
1242 * in the in_addr structure passed in as a parameter.
1243 */
1244 static int
join_mcast_group(struct sock * sk,struct in_addr * addr,char * ifname)1245 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
1246 {
1247 struct net *net = sock_net(sk);
1248 struct ip_mreqn mreq;
1249 struct net_device *dev;
1250 int ret;
1251
1252 memset(&mreq, 0, sizeof(mreq));
1253 memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
1254
1255 dev = __dev_get_by_name(net, ifname);
1256 if (!dev)
1257 return -ENODEV;
1258 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1259 return -EINVAL;
1260
1261 mreq.imr_ifindex = dev->ifindex;
1262
1263 lock_sock(sk);
1264 ret = ip_mc_join_group(sk, &mreq);
1265 release_sock(sk);
1266
1267 return ret;
1268 }
1269
1270
bind_mcastif_addr(struct socket * sock,char * ifname)1271 static int bind_mcastif_addr(struct socket *sock, char *ifname)
1272 {
1273 struct net *net = sock_net(sock->sk);
1274 struct net_device *dev;
1275 __be32 addr;
1276 struct sockaddr_in sin;
1277
1278 dev = __dev_get_by_name(net, ifname);
1279 if (!dev)
1280 return -ENODEV;
1281
1282 addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
1283 if (!addr)
1284 pr_err("You probably need to specify IP address on "
1285 "multicast interface.\n");
1286
1287 IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
1288 ifname, &addr);
1289
1290 /* Now bind the socket with the address of multicast interface */
1291 sin.sin_family = AF_INET;
1292 sin.sin_addr.s_addr = addr;
1293 sin.sin_port = 0;
1294
1295 return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
1296 }
1297
1298 /*
1299 * Set up sending multicast socket over UDP
1300 */
make_send_sock(struct net * net)1301 static struct socket *make_send_sock(struct net *net)
1302 {
1303 struct netns_ipvs *ipvs = net_ipvs(net);
1304 struct socket *sock;
1305 int result;
1306
1307 /* First create a socket move it to right name space later */
1308 result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1309 if (result < 0) {
1310 pr_err("Error during creation of socket; terminating\n");
1311 return ERR_PTR(result);
1312 }
1313 /*
1314 * Kernel sockets that are a part of a namespace, should not
1315 * hold a reference to a namespace in order to allow to stop it.
1316 * After sk_change_net should be released using sk_release_kernel.
1317 */
1318 sk_change_net(sock->sk, net);
1319 result = set_mcast_if(sock->sk, ipvs->master_mcast_ifn);
1320 if (result < 0) {
1321 pr_err("Error setting outbound mcast interface\n");
1322 goto error;
1323 }
1324
1325 set_mcast_loop(sock->sk, 0);
1326 set_mcast_ttl(sock->sk, 1);
1327
1328 result = bind_mcastif_addr(sock, ipvs->master_mcast_ifn);
1329 if (result < 0) {
1330 pr_err("Error binding address of the mcast interface\n");
1331 goto error;
1332 }
1333
1334 result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr,
1335 sizeof(struct sockaddr), 0);
1336 if (result < 0) {
1337 pr_err("Error connecting to the multicast addr\n");
1338 goto error;
1339 }
1340
1341 return sock;
1342
1343 error:
1344 sk_release_kernel(sock->sk);
1345 return ERR_PTR(result);
1346 }
1347
1348
1349 /*
1350 * Set up receiving multicast socket over UDP
1351 */
make_receive_sock(struct net * net)1352 static struct socket *make_receive_sock(struct net *net)
1353 {
1354 struct netns_ipvs *ipvs = net_ipvs(net);
1355 struct socket *sock;
1356 int result;
1357
1358 /* First create a socket */
1359 result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1360 if (result < 0) {
1361 pr_err("Error during creation of socket; terminating\n");
1362 return ERR_PTR(result);
1363 }
1364 /*
1365 * Kernel sockets that are a part of a namespace, should not
1366 * hold a reference to a namespace in order to allow to stop it.
1367 * After sk_change_net should be released using sk_release_kernel.
1368 */
1369 sk_change_net(sock->sk, net);
1370 /* it is equivalent to the REUSEADDR option in user-space */
1371 sock->sk->sk_reuse = 1;
1372
1373 result = sock->ops->bind(sock, (struct sockaddr *) &mcast_addr,
1374 sizeof(struct sockaddr));
1375 if (result < 0) {
1376 pr_err("Error binding to the multicast addr\n");
1377 goto error;
1378 }
1379
1380 /* join the multicast group */
1381 result = join_mcast_group(sock->sk,
1382 (struct in_addr *) &mcast_addr.sin_addr,
1383 ipvs->backup_mcast_ifn);
1384 if (result < 0) {
1385 pr_err("Error joining to the multicast group\n");
1386 goto error;
1387 }
1388
1389 return sock;
1390
1391 error:
1392 sk_release_kernel(sock->sk);
1393 return ERR_PTR(result);
1394 }
1395
1396
1397 static int
ip_vs_send_async(struct socket * sock,const char * buffer,const size_t length)1398 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
1399 {
1400 struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
1401 struct kvec iov;
1402 int len;
1403
1404 EnterFunction(7);
1405 iov.iov_base = (void *)buffer;
1406 iov.iov_len = length;
1407
1408 len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
1409
1410 LeaveFunction(7);
1411 return len;
1412 }
1413
1414 static void
ip_vs_send_sync_msg(struct socket * sock,struct ip_vs_sync_mesg * msg)1415 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
1416 {
1417 int msize;
1418
1419 msize = msg->size;
1420
1421 /* Put size in network byte order */
1422 msg->size = htons(msg->size);
1423
1424 if (ip_vs_send_async(sock, (char *)msg, msize) != msize)
1425 pr_err("ip_vs_send_async error\n");
1426 }
1427
1428 static int
ip_vs_receive(struct socket * sock,char * buffer,const size_t buflen)1429 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
1430 {
1431 struct msghdr msg = {NULL,};
1432 struct kvec iov;
1433 int len;
1434
1435 EnterFunction(7);
1436
1437 /* Receive a packet */
1438 iov.iov_base = buffer;
1439 iov.iov_len = (size_t)buflen;
1440
1441 len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, 0);
1442
1443 if (len < 0)
1444 return -1;
1445
1446 LeaveFunction(7);
1447 return len;
1448 }
1449
1450
sync_thread_master(void * data)1451 static int sync_thread_master(void *data)
1452 {
1453 struct ip_vs_sync_thread_data *tinfo = data;
1454 struct netns_ipvs *ipvs = net_ipvs(tinfo->net);
1455 struct ip_vs_sync_buff *sb;
1456
1457 pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
1458 "syncid = %d\n",
1459 ipvs->master_mcast_ifn, ipvs->master_syncid);
1460
1461 while (!kthread_should_stop()) {
1462 while ((sb = sb_dequeue(ipvs))) {
1463 ip_vs_send_sync_msg(tinfo->sock, sb->mesg);
1464 ip_vs_sync_buff_release(sb);
1465 }
1466
1467 /* check if entries stay in ipvs->sync_buff for 2 seconds */
1468 sb = get_curr_sync_buff(ipvs, 2 * HZ);
1469 if (sb) {
1470 ip_vs_send_sync_msg(tinfo->sock, sb->mesg);
1471 ip_vs_sync_buff_release(sb);
1472 }
1473
1474 schedule_timeout_interruptible(HZ);
1475 }
1476
1477 /* clean up the sync_buff queue */
1478 while ((sb = sb_dequeue(ipvs)))
1479 ip_vs_sync_buff_release(sb);
1480
1481 /* clean up the current sync_buff */
1482 sb = get_curr_sync_buff(ipvs, 0);
1483 if (sb)
1484 ip_vs_sync_buff_release(sb);
1485
1486 /* release the sending multicast socket */
1487 sk_release_kernel(tinfo->sock->sk);
1488 kfree(tinfo);
1489
1490 return 0;
1491 }
1492
1493
sync_thread_backup(void * data)1494 static int sync_thread_backup(void *data)
1495 {
1496 struct ip_vs_sync_thread_data *tinfo = data;
1497 struct netns_ipvs *ipvs = net_ipvs(tinfo->net);
1498 int len;
1499
1500 pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
1501 "syncid = %d\n",
1502 ipvs->backup_mcast_ifn, ipvs->backup_syncid);
1503
1504 while (!kthread_should_stop()) {
1505 wait_event_interruptible(*sk_sleep(tinfo->sock->sk),
1506 !skb_queue_empty(&tinfo->sock->sk->sk_receive_queue)
1507 || kthread_should_stop());
1508
1509 /* do we have data now? */
1510 while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) {
1511 len = ip_vs_receive(tinfo->sock, tinfo->buf,
1512 ipvs->recv_mesg_maxlen);
1513 if (len <= 0) {
1514 pr_err("receiving message error\n");
1515 break;
1516 }
1517
1518 /* disable bottom half, because it accesses the data
1519 shared by softirq while getting/creating conns */
1520 local_bh_disable();
1521 ip_vs_process_message(tinfo->net, tinfo->buf, len);
1522 local_bh_enable();
1523 }
1524 }
1525
1526 /* release the sending multicast socket */
1527 sk_release_kernel(tinfo->sock->sk);
1528 kfree(tinfo->buf);
1529 kfree(tinfo);
1530
1531 return 0;
1532 }
1533
1534
start_sync_thread(struct net * net,int state,char * mcast_ifn,__u8 syncid)1535 int start_sync_thread(struct net *net, int state, char *mcast_ifn, __u8 syncid)
1536 {
1537 struct ip_vs_sync_thread_data *tinfo;
1538 struct task_struct **realtask, *task;
1539 struct socket *sock;
1540 struct netns_ipvs *ipvs = net_ipvs(net);
1541 char *name, *buf = NULL;
1542 int (*threadfn)(void *data);
1543 int result = -ENOMEM;
1544
1545 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1546 IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n",
1547 sizeof(struct ip_vs_sync_conn_v0));
1548
1549
1550 if (state == IP_VS_STATE_MASTER) {
1551 if (ipvs->master_thread)
1552 return -EEXIST;
1553
1554 strlcpy(ipvs->master_mcast_ifn, mcast_ifn,
1555 sizeof(ipvs->master_mcast_ifn));
1556 ipvs->master_syncid = syncid;
1557 realtask = &ipvs->master_thread;
1558 name = "ipvs_master:%d";
1559 threadfn = sync_thread_master;
1560 sock = make_send_sock(net);
1561 } else if (state == IP_VS_STATE_BACKUP) {
1562 if (ipvs->backup_thread)
1563 return -EEXIST;
1564
1565 strlcpy(ipvs->backup_mcast_ifn, mcast_ifn,
1566 sizeof(ipvs->backup_mcast_ifn));
1567 ipvs->backup_syncid = syncid;
1568 realtask = &ipvs->backup_thread;
1569 name = "ipvs_backup:%d";
1570 threadfn = sync_thread_backup;
1571 sock = make_receive_sock(net);
1572 } else {
1573 return -EINVAL;
1574 }
1575
1576 if (IS_ERR(sock)) {
1577 result = PTR_ERR(sock);
1578 goto out;
1579 }
1580
1581 set_sync_mesg_maxlen(net, state);
1582 if (state == IP_VS_STATE_BACKUP) {
1583 buf = kmalloc(ipvs->recv_mesg_maxlen, GFP_KERNEL);
1584 if (!buf)
1585 goto outsocket;
1586 }
1587
1588 tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL);
1589 if (!tinfo)
1590 goto outbuf;
1591
1592 tinfo->net = net;
1593 tinfo->sock = sock;
1594 tinfo->buf = buf;
1595
1596 task = kthread_run(threadfn, tinfo, name, ipvs->gen);
1597 if (IS_ERR(task)) {
1598 result = PTR_ERR(task);
1599 goto outtinfo;
1600 }
1601
1602 /* mark as active */
1603 *realtask = task;
1604 ipvs->sync_state |= state;
1605
1606 /* increase the module use count */
1607 ip_vs_use_count_inc();
1608
1609 return 0;
1610
1611 outtinfo:
1612 kfree(tinfo);
1613 outbuf:
1614 kfree(buf);
1615 outsocket:
1616 sk_release_kernel(sock->sk);
1617 out:
1618 return result;
1619 }
1620
1621
stop_sync_thread(struct net * net,int state)1622 int stop_sync_thread(struct net *net, int state)
1623 {
1624 struct netns_ipvs *ipvs = net_ipvs(net);
1625 int retc = -EINVAL;
1626
1627 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1628
1629 if (state == IP_VS_STATE_MASTER) {
1630 if (!ipvs->master_thread)
1631 return -ESRCH;
1632
1633 pr_info("stopping master sync thread %d ...\n",
1634 task_pid_nr(ipvs->master_thread));
1635
1636 /*
1637 * The lock synchronizes with sb_queue_tail(), so that we don't
1638 * add sync buffers to the queue, when we are already in
1639 * progress of stopping the master sync daemon.
1640 */
1641
1642 spin_lock_bh(&ipvs->sync_lock);
1643 ipvs->sync_state &= ~IP_VS_STATE_MASTER;
1644 spin_unlock_bh(&ipvs->sync_lock);
1645 retc = kthread_stop(ipvs->master_thread);
1646 ipvs->master_thread = NULL;
1647 } else if (state == IP_VS_STATE_BACKUP) {
1648 if (!ipvs->backup_thread)
1649 return -ESRCH;
1650
1651 pr_info("stopping backup sync thread %d ...\n",
1652 task_pid_nr(ipvs->backup_thread));
1653
1654 ipvs->sync_state &= ~IP_VS_STATE_BACKUP;
1655 retc = kthread_stop(ipvs->backup_thread);
1656 ipvs->backup_thread = NULL;
1657 }
1658
1659 /* decrease the module use count */
1660 ip_vs_use_count_dec();
1661
1662 return retc;
1663 }
1664
1665 /*
1666 * Initialize data struct for each netns
1667 */
ip_vs_sync_net_init(struct net * net)1668 int __net_init ip_vs_sync_net_init(struct net *net)
1669 {
1670 struct netns_ipvs *ipvs = net_ipvs(net);
1671
1672 __mutex_init(&ipvs->sync_mutex, "ipvs->sync_mutex", &__ipvs_sync_key);
1673 INIT_LIST_HEAD(&ipvs->sync_queue);
1674 spin_lock_init(&ipvs->sync_lock);
1675 spin_lock_init(&ipvs->sync_buff_lock);
1676
1677 ipvs->sync_mcast_addr.sin_family = AF_INET;
1678 ipvs->sync_mcast_addr.sin_port = cpu_to_be16(IP_VS_SYNC_PORT);
1679 ipvs->sync_mcast_addr.sin_addr.s_addr = cpu_to_be32(IP_VS_SYNC_GROUP);
1680 return 0;
1681 }
1682
ip_vs_sync_net_cleanup(struct net * net)1683 void ip_vs_sync_net_cleanup(struct net *net)
1684 {
1685 int retc;
1686 struct netns_ipvs *ipvs = net_ipvs(net);
1687
1688 mutex_lock(&ipvs->sync_mutex);
1689 retc = stop_sync_thread(net, IP_VS_STATE_MASTER);
1690 if (retc && retc != -ESRCH)
1691 pr_err("Failed to stop Master Daemon\n");
1692
1693 retc = stop_sync_thread(net, IP_VS_STATE_BACKUP);
1694 if (retc && retc != -ESRCH)
1695 pr_err("Failed to stop Backup Daemon\n");
1696 mutex_unlock(&ipvs->sync_mutex);
1697 }
1698