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 int id;
200 };
201
202 /* Version 0 definition of packet sizes */
203 #define SIMPLE_CONN_SIZE (sizeof(struct ip_vs_sync_conn_v0))
204 #define FULL_CONN_SIZE \
205 (sizeof(struct ip_vs_sync_conn_v0) + sizeof(struct ip_vs_sync_conn_options))
206
207
208 /*
209 The master mulitcasts messages (Datagrams) to the backup load balancers
210 in the following format.
211
212 Version 1:
213 Note, first byte should be Zero, so ver 0 receivers will drop the packet.
214
215 0 1 2 3
216 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
217 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
218 | 0 | SyncID | Size |
219 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
220 | Count Conns | Version | Reserved, set to Zero |
221 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
222 | |
223 | IPVS Sync Connection (1) |
224 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
225 | . |
226 ~ . ~
227 | . |
228 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
229 | |
230 | IPVS Sync Connection (n) |
231 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
232
233 Version 0 Header
234 0 1 2 3
235 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
236 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
237 | Count Conns | SyncID | Size |
238 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
239 | IPVS Sync Connection (1) |
240 */
241
242 #define SYNC_MESG_HEADER_LEN 4
243 #define MAX_CONNS_PER_SYNCBUFF 255 /* nr_conns in ip_vs_sync_mesg is 8 bit */
244
245 /* Version 0 header */
246 struct ip_vs_sync_mesg_v0 {
247 __u8 nr_conns;
248 __u8 syncid;
249 __be16 size;
250
251 /* ip_vs_sync_conn entries start here */
252 };
253
254 /* Version 1 header */
255 struct ip_vs_sync_mesg {
256 __u8 reserved; /* must be zero */
257 __u8 syncid;
258 __be16 size;
259 __u8 nr_conns;
260 __s8 version; /* SYNC_PROTO_VER */
261 __u16 spare;
262 /* ip_vs_sync_conn entries start here */
263 };
264
265 struct ip_vs_sync_buff {
266 struct list_head list;
267 unsigned long firstuse;
268
269 /* pointers for the message data */
270 struct ip_vs_sync_mesg *mesg;
271 unsigned char *head;
272 unsigned char *end;
273 };
274
275 /*
276 * Copy of struct ip_vs_seq
277 * From unaligned network order to aligned host order
278 */
ntoh_seq(struct ip_vs_seq * no,struct ip_vs_seq * ho)279 static void ntoh_seq(struct ip_vs_seq *no, struct ip_vs_seq *ho)
280 {
281 ho->init_seq = get_unaligned_be32(&no->init_seq);
282 ho->delta = get_unaligned_be32(&no->delta);
283 ho->previous_delta = get_unaligned_be32(&no->previous_delta);
284 }
285
286 /*
287 * Copy of struct ip_vs_seq
288 * From Aligned host order to unaligned network order
289 */
hton_seq(struct ip_vs_seq * ho,struct ip_vs_seq * no)290 static void hton_seq(struct ip_vs_seq *ho, struct ip_vs_seq *no)
291 {
292 put_unaligned_be32(ho->init_seq, &no->init_seq);
293 put_unaligned_be32(ho->delta, &no->delta);
294 put_unaligned_be32(ho->previous_delta, &no->previous_delta);
295 }
296
297 static inline struct ip_vs_sync_buff *
sb_dequeue(struct netns_ipvs * ipvs,struct ipvs_master_sync_state * ms)298 sb_dequeue(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
299 {
300 struct ip_vs_sync_buff *sb;
301
302 spin_lock_bh(&ipvs->sync_lock);
303 if (list_empty(&ms->sync_queue)) {
304 sb = NULL;
305 __set_current_state(TASK_INTERRUPTIBLE);
306 } else {
307 sb = list_entry(ms->sync_queue.next, struct ip_vs_sync_buff,
308 list);
309 list_del(&sb->list);
310 ms->sync_queue_len--;
311 if (!ms->sync_queue_len)
312 ms->sync_queue_delay = 0;
313 }
314 spin_unlock_bh(&ipvs->sync_lock);
315
316 return sb;
317 }
318
319 /*
320 * Create a new sync buffer for Version 1 proto.
321 */
322 static inline struct ip_vs_sync_buff *
ip_vs_sync_buff_create(struct netns_ipvs * ipvs)323 ip_vs_sync_buff_create(struct netns_ipvs *ipvs)
324 {
325 struct ip_vs_sync_buff *sb;
326
327 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
328 return NULL;
329
330 sb->mesg = kmalloc(ipvs->send_mesg_maxlen, GFP_ATOMIC);
331 if (!sb->mesg) {
332 kfree(sb);
333 return NULL;
334 }
335 sb->mesg->reserved = 0; /* old nr_conns i.e. must be zero now */
336 sb->mesg->version = SYNC_PROTO_VER;
337 sb->mesg->syncid = ipvs->master_syncid;
338 sb->mesg->size = htons(sizeof(struct ip_vs_sync_mesg));
339 sb->mesg->nr_conns = 0;
340 sb->mesg->spare = 0;
341 sb->head = (unsigned char *)sb->mesg + sizeof(struct ip_vs_sync_mesg);
342 sb->end = (unsigned char *)sb->mesg + ipvs->send_mesg_maxlen;
343
344 sb->firstuse = jiffies;
345 return sb;
346 }
347
ip_vs_sync_buff_release(struct ip_vs_sync_buff * sb)348 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
349 {
350 kfree(sb->mesg);
351 kfree(sb);
352 }
353
sb_queue_tail(struct netns_ipvs * ipvs,struct ipvs_master_sync_state * ms)354 static inline void sb_queue_tail(struct netns_ipvs *ipvs,
355 struct ipvs_master_sync_state *ms)
356 {
357 struct ip_vs_sync_buff *sb = ms->sync_buff;
358
359 spin_lock(&ipvs->sync_lock);
360 if (ipvs->sync_state & IP_VS_STATE_MASTER &&
361 ms->sync_queue_len < sysctl_sync_qlen_max(ipvs)) {
362 if (!ms->sync_queue_len)
363 schedule_delayed_work(&ms->master_wakeup_work,
364 max(IPVS_SYNC_SEND_DELAY, 1));
365 ms->sync_queue_len++;
366 list_add_tail(&sb->list, &ms->sync_queue);
367 if ((++ms->sync_queue_delay) == IPVS_SYNC_WAKEUP_RATE)
368 wake_up_process(ms->master_thread);
369 } else
370 ip_vs_sync_buff_release(sb);
371 spin_unlock(&ipvs->sync_lock);
372 }
373
374 /*
375 * Get the current sync buffer if it has been created for more
376 * than the specified time or the specified time is zero.
377 */
378 static inline struct ip_vs_sync_buff *
get_curr_sync_buff(struct netns_ipvs * ipvs,struct ipvs_master_sync_state * ms,unsigned long time)379 get_curr_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms,
380 unsigned long time)
381 {
382 struct ip_vs_sync_buff *sb;
383
384 spin_lock_bh(&ipvs->sync_buff_lock);
385 sb = ms->sync_buff;
386 if (sb && time_after_eq(jiffies - sb->firstuse, time)) {
387 ms->sync_buff = NULL;
388 __set_current_state(TASK_RUNNING);
389 } else
390 sb = NULL;
391 spin_unlock_bh(&ipvs->sync_buff_lock);
392 return sb;
393 }
394
395 static inline int
select_master_thread_id(struct netns_ipvs * ipvs,struct ip_vs_conn * cp)396 select_master_thread_id(struct netns_ipvs *ipvs, struct ip_vs_conn *cp)
397 {
398 return ((long) cp >> (1 + ilog2(sizeof(*cp)))) & ipvs->threads_mask;
399 }
400
401 /*
402 * Create a new sync buffer for Version 0 proto.
403 */
404 static inline struct ip_vs_sync_buff *
ip_vs_sync_buff_create_v0(struct netns_ipvs * ipvs)405 ip_vs_sync_buff_create_v0(struct netns_ipvs *ipvs)
406 {
407 struct ip_vs_sync_buff *sb;
408 struct ip_vs_sync_mesg_v0 *mesg;
409
410 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
411 return NULL;
412
413 sb->mesg = kmalloc(ipvs->send_mesg_maxlen, GFP_ATOMIC);
414 if (!sb->mesg) {
415 kfree(sb);
416 return NULL;
417 }
418 mesg = (struct ip_vs_sync_mesg_v0 *)sb->mesg;
419 mesg->nr_conns = 0;
420 mesg->syncid = ipvs->master_syncid;
421 mesg->size = htons(sizeof(struct ip_vs_sync_mesg_v0));
422 sb->head = (unsigned char *)mesg + sizeof(struct ip_vs_sync_mesg_v0);
423 sb->end = (unsigned char *)mesg + ipvs->send_mesg_maxlen;
424 sb->firstuse = jiffies;
425 return sb;
426 }
427
428 /* Check if conn should be synced.
429 * pkts: conn packets, use sysctl_sync_threshold to avoid packet check
430 * - (1) sync_refresh_period: reduce sync rate. Additionally, retry
431 * sync_retries times with period of sync_refresh_period/8
432 * - (2) if both sync_refresh_period and sync_period are 0 send sync only
433 * for state changes or only once when pkts matches sync_threshold
434 * - (3) templates: rate can be reduced only with sync_refresh_period or
435 * with (2)
436 */
ip_vs_sync_conn_needed(struct netns_ipvs * ipvs,struct ip_vs_conn * cp,int pkts)437 static int ip_vs_sync_conn_needed(struct netns_ipvs *ipvs,
438 struct ip_vs_conn *cp, int pkts)
439 {
440 unsigned long orig = ACCESS_ONCE(cp->sync_endtime);
441 unsigned long now = jiffies;
442 unsigned long n = (now + cp->timeout) & ~3UL;
443 unsigned int sync_refresh_period;
444 int sync_period;
445 int force;
446
447 /* Check if we sync in current state */
448 if (unlikely(cp->flags & IP_VS_CONN_F_TEMPLATE))
449 force = 0;
450 else if (likely(cp->protocol == IPPROTO_TCP)) {
451 if (!((1 << cp->state) &
452 ((1 << IP_VS_TCP_S_ESTABLISHED) |
453 (1 << IP_VS_TCP_S_FIN_WAIT) |
454 (1 << IP_VS_TCP_S_CLOSE) |
455 (1 << IP_VS_TCP_S_CLOSE_WAIT) |
456 (1 << IP_VS_TCP_S_TIME_WAIT))))
457 return 0;
458 force = cp->state != cp->old_state;
459 if (force && cp->state != IP_VS_TCP_S_ESTABLISHED)
460 goto set;
461 } else if (unlikely(cp->protocol == IPPROTO_SCTP)) {
462 if (!((1 << cp->state) &
463 ((1 << IP_VS_SCTP_S_ESTABLISHED) |
464 (1 << IP_VS_SCTP_S_CLOSED) |
465 (1 << IP_VS_SCTP_S_SHUT_ACK_CLI) |
466 (1 << IP_VS_SCTP_S_SHUT_ACK_SER))))
467 return 0;
468 force = cp->state != cp->old_state;
469 if (force && cp->state != IP_VS_SCTP_S_ESTABLISHED)
470 goto set;
471 } else {
472 /* UDP or another protocol with single state */
473 force = 0;
474 }
475
476 sync_refresh_period = sysctl_sync_refresh_period(ipvs);
477 if (sync_refresh_period > 0) {
478 long diff = n - orig;
479 long min_diff = max(cp->timeout >> 1, 10UL * HZ);
480
481 /* Avoid sync if difference is below sync_refresh_period
482 * and below the half timeout.
483 */
484 if (abs(diff) < min_t(long, sync_refresh_period, min_diff)) {
485 int retries = orig & 3;
486
487 if (retries >= sysctl_sync_retries(ipvs))
488 return 0;
489 if (time_before(now, orig - cp->timeout +
490 (sync_refresh_period >> 3)))
491 return 0;
492 n |= retries + 1;
493 }
494 }
495 sync_period = sysctl_sync_period(ipvs);
496 if (sync_period > 0) {
497 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE) &&
498 pkts % sync_period != sysctl_sync_threshold(ipvs))
499 return 0;
500 } else if (sync_refresh_period <= 0 &&
501 pkts != sysctl_sync_threshold(ipvs))
502 return 0;
503
504 set:
505 cp->old_state = cp->state;
506 n = cmpxchg(&cp->sync_endtime, orig, n);
507 return n == orig || force;
508 }
509
510 /*
511 * Version 0 , could be switched in by sys_ctl.
512 * Add an ip_vs_conn information into the current sync_buff.
513 */
ip_vs_sync_conn_v0(struct net * net,struct ip_vs_conn * cp,int pkts)514 static void ip_vs_sync_conn_v0(struct net *net, struct ip_vs_conn *cp,
515 int pkts)
516 {
517 struct netns_ipvs *ipvs = net_ipvs(net);
518 struct ip_vs_sync_mesg_v0 *m;
519 struct ip_vs_sync_conn_v0 *s;
520 struct ip_vs_sync_buff *buff;
521 struct ipvs_master_sync_state *ms;
522 int id;
523 int len;
524
525 if (unlikely(cp->af != AF_INET))
526 return;
527 /* Do not sync ONE PACKET */
528 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
529 return;
530
531 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
532 return;
533
534 spin_lock_bh(&ipvs->sync_buff_lock);
535 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
536 spin_unlock_bh(&ipvs->sync_buff_lock);
537 return;
538 }
539
540 id = select_master_thread_id(ipvs, cp);
541 ms = &ipvs->ms[id];
542 buff = ms->sync_buff;
543 if (buff) {
544 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
545 /* Send buffer if it is for v1 */
546 if (!m->nr_conns) {
547 sb_queue_tail(ipvs, ms);
548 ms->sync_buff = NULL;
549 buff = NULL;
550 }
551 }
552 if (!buff) {
553 buff = ip_vs_sync_buff_create_v0(ipvs);
554 if (!buff) {
555 spin_unlock_bh(&ipvs->sync_buff_lock);
556 pr_err("ip_vs_sync_buff_create failed.\n");
557 return;
558 }
559 ms->sync_buff = buff;
560 }
561
562 len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
563 SIMPLE_CONN_SIZE;
564 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
565 s = (struct ip_vs_sync_conn_v0 *) buff->head;
566
567 /* copy members */
568 s->reserved = 0;
569 s->protocol = cp->protocol;
570 s->cport = cp->cport;
571 s->vport = cp->vport;
572 s->dport = cp->dport;
573 s->caddr = cp->caddr.ip;
574 s->vaddr = cp->vaddr.ip;
575 s->daddr = cp->daddr.ip;
576 s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
577 s->state = htons(cp->state);
578 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
579 struct ip_vs_sync_conn_options *opt =
580 (struct ip_vs_sync_conn_options *)&s[1];
581 memcpy(opt, &cp->in_seq, sizeof(*opt));
582 }
583
584 m->nr_conns++;
585 m->size = htons(ntohs(m->size) + len);
586 buff->head += len;
587
588 /* check if there is a space for next one */
589 if (buff->head + FULL_CONN_SIZE > buff->end) {
590 sb_queue_tail(ipvs, ms);
591 ms->sync_buff = NULL;
592 }
593 spin_unlock_bh(&ipvs->sync_buff_lock);
594
595 /* synchronize its controller if it has */
596 cp = cp->control;
597 if (cp) {
598 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
599 pkts = atomic_add_return(1, &cp->in_pkts);
600 else
601 pkts = sysctl_sync_threshold(ipvs);
602 ip_vs_sync_conn(net, cp->control, pkts);
603 }
604 }
605
606 /*
607 * Add an ip_vs_conn information into the current sync_buff.
608 * Called by ip_vs_in.
609 * Sending Version 1 messages
610 */
ip_vs_sync_conn(struct net * net,struct ip_vs_conn * cp,int pkts)611 void ip_vs_sync_conn(struct net *net, struct ip_vs_conn *cp, int pkts)
612 {
613 struct netns_ipvs *ipvs = net_ipvs(net);
614 struct ip_vs_sync_mesg *m;
615 union ip_vs_sync_conn *s;
616 struct ip_vs_sync_buff *buff;
617 struct ipvs_master_sync_state *ms;
618 int id;
619 __u8 *p;
620 unsigned int len, pe_name_len, pad;
621
622 /* Handle old version of the protocol */
623 if (sysctl_sync_ver(ipvs) == 0) {
624 ip_vs_sync_conn_v0(net, cp, pkts);
625 return;
626 }
627 /* Do not sync ONE PACKET */
628 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
629 goto control;
630 sloop:
631 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
632 goto control;
633
634 /* Sanity checks */
635 pe_name_len = 0;
636 if (cp->pe_data_len) {
637 if (!cp->pe_data || !cp->dest) {
638 IP_VS_ERR_RL("SYNC, connection pe_data invalid\n");
639 return;
640 }
641 pe_name_len = strnlen(cp->pe->name, IP_VS_PENAME_MAXLEN);
642 }
643
644 spin_lock_bh(&ipvs->sync_buff_lock);
645 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
646 spin_unlock_bh(&ipvs->sync_buff_lock);
647 return;
648 }
649
650 id = select_master_thread_id(ipvs, cp);
651 ms = &ipvs->ms[id];
652
653 #ifdef CONFIG_IP_VS_IPV6
654 if (cp->af == AF_INET6)
655 len = sizeof(struct ip_vs_sync_v6);
656 else
657 #endif
658 len = sizeof(struct ip_vs_sync_v4);
659
660 if (cp->flags & IP_VS_CONN_F_SEQ_MASK)
661 len += sizeof(struct ip_vs_sync_conn_options) + 2;
662
663 if (cp->pe_data_len)
664 len += cp->pe_data_len + 2; /* + Param hdr field */
665 if (pe_name_len)
666 len += pe_name_len + 2;
667
668 /* check if there is a space for this one */
669 pad = 0;
670 buff = ms->sync_buff;
671 if (buff) {
672 m = buff->mesg;
673 pad = (4 - (size_t) buff->head) & 3;
674 /* Send buffer if it is for v0 */
675 if (buff->head + len + pad > buff->end || m->reserved) {
676 sb_queue_tail(ipvs, ms);
677 ms->sync_buff = NULL;
678 buff = NULL;
679 pad = 0;
680 }
681 }
682
683 if (!buff) {
684 buff = ip_vs_sync_buff_create(ipvs);
685 if (!buff) {
686 spin_unlock_bh(&ipvs->sync_buff_lock);
687 pr_err("ip_vs_sync_buff_create failed.\n");
688 return;
689 }
690 ms->sync_buff = buff;
691 m = buff->mesg;
692 }
693
694 p = buff->head;
695 buff->head += pad + len;
696 m->size = htons(ntohs(m->size) + pad + len);
697 /* Add ev. padding from prev. sync_conn */
698 while (pad--)
699 *(p++) = 0;
700
701 s = (union ip_vs_sync_conn *)p;
702
703 /* Set message type & copy members */
704 s->v4.type = (cp->af == AF_INET6 ? STYPE_F_INET6 : 0);
705 s->v4.ver_size = htons(len & SVER_MASK); /* Version 0 */
706 s->v4.flags = htonl(cp->flags & ~IP_VS_CONN_F_HASHED);
707 s->v4.state = htons(cp->state);
708 s->v4.protocol = cp->protocol;
709 s->v4.cport = cp->cport;
710 s->v4.vport = cp->vport;
711 s->v4.dport = cp->dport;
712 s->v4.fwmark = htonl(cp->fwmark);
713 s->v4.timeout = htonl(cp->timeout / HZ);
714 m->nr_conns++;
715
716 #ifdef CONFIG_IP_VS_IPV6
717 if (cp->af == AF_INET6) {
718 p += sizeof(struct ip_vs_sync_v6);
719 s->v6.caddr = cp->caddr.in6;
720 s->v6.vaddr = cp->vaddr.in6;
721 s->v6.daddr = cp->daddr.in6;
722 } else
723 #endif
724 {
725 p += sizeof(struct ip_vs_sync_v4); /* options ptr */
726 s->v4.caddr = cp->caddr.ip;
727 s->v4.vaddr = cp->vaddr.ip;
728 s->v4.daddr = cp->daddr.ip;
729 }
730 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
731 *(p++) = IPVS_OPT_SEQ_DATA;
732 *(p++) = sizeof(struct ip_vs_sync_conn_options);
733 hton_seq((struct ip_vs_seq *)p, &cp->in_seq);
734 p += sizeof(struct ip_vs_seq);
735 hton_seq((struct ip_vs_seq *)p, &cp->out_seq);
736 p += sizeof(struct ip_vs_seq);
737 }
738 /* Handle pe data */
739 if (cp->pe_data_len && cp->pe_data) {
740 *(p++) = IPVS_OPT_PE_DATA;
741 *(p++) = cp->pe_data_len;
742 memcpy(p, cp->pe_data, cp->pe_data_len);
743 p += cp->pe_data_len;
744 if (pe_name_len) {
745 /* Add PE_NAME */
746 *(p++) = IPVS_OPT_PE_NAME;
747 *(p++) = pe_name_len;
748 memcpy(p, cp->pe->name, pe_name_len);
749 p += pe_name_len;
750 }
751 }
752
753 spin_unlock_bh(&ipvs->sync_buff_lock);
754
755 control:
756 /* synchronize its controller if it has */
757 cp = cp->control;
758 if (!cp)
759 return;
760 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
761 pkts = atomic_add_return(1, &cp->in_pkts);
762 else
763 pkts = sysctl_sync_threshold(ipvs);
764 goto sloop;
765 }
766
767 /*
768 * fill_param used by version 1
769 */
770 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)771 ip_vs_conn_fill_param_sync(struct net *net, int af, union ip_vs_sync_conn *sc,
772 struct ip_vs_conn_param *p,
773 __u8 *pe_data, unsigned int pe_data_len,
774 __u8 *pe_name, unsigned int pe_name_len)
775 {
776 #ifdef CONFIG_IP_VS_IPV6
777 if (af == AF_INET6)
778 ip_vs_conn_fill_param(net, af, sc->v6.protocol,
779 (const union nf_inet_addr *)&sc->v6.caddr,
780 sc->v6.cport,
781 (const union nf_inet_addr *)&sc->v6.vaddr,
782 sc->v6.vport, p);
783 else
784 #endif
785 ip_vs_conn_fill_param(net, af, sc->v4.protocol,
786 (const union nf_inet_addr *)&sc->v4.caddr,
787 sc->v4.cport,
788 (const union nf_inet_addr *)&sc->v4.vaddr,
789 sc->v4.vport, p);
790 /* Handle pe data */
791 if (pe_data_len) {
792 if (pe_name_len) {
793 char buff[IP_VS_PENAME_MAXLEN+1];
794
795 memcpy(buff, pe_name, pe_name_len);
796 buff[pe_name_len]=0;
797 p->pe = __ip_vs_pe_getbyname(buff);
798 if (!p->pe) {
799 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n",
800 buff);
801 return 1;
802 }
803 } else {
804 IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n");
805 return 1;
806 }
807
808 p->pe_data = kmemdup(pe_data, pe_data_len, GFP_ATOMIC);
809 if (!p->pe_data) {
810 if (p->pe->module)
811 module_put(p->pe->module);
812 return -ENOMEM;
813 }
814 p->pe_data_len = pe_data_len;
815 }
816 return 0;
817 }
818
819 /*
820 * Connection Add / Update.
821 * Common for version 0 and 1 reception of backup sync_conns.
822 * Param: ...
823 * timeout is in sec.
824 */
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)825 static void ip_vs_proc_conn(struct net *net, struct ip_vs_conn_param *param,
826 unsigned int flags, unsigned int state,
827 unsigned int protocol, unsigned int type,
828 const union nf_inet_addr *daddr, __be16 dport,
829 unsigned long timeout, __u32 fwmark,
830 struct ip_vs_sync_conn_options *opt)
831 {
832 struct ip_vs_dest *dest;
833 struct ip_vs_conn *cp;
834 struct netns_ipvs *ipvs = net_ipvs(net);
835
836 if (!(flags & IP_VS_CONN_F_TEMPLATE))
837 cp = ip_vs_conn_in_get(param);
838 else
839 cp = ip_vs_ct_in_get(param);
840
841 if (cp) {
842 /* Free pe_data */
843 kfree(param->pe_data);
844
845 dest = cp->dest;
846 spin_lock_bh(&cp->lock);
847 if ((cp->flags ^ flags) & IP_VS_CONN_F_INACTIVE &&
848 !(flags & IP_VS_CONN_F_TEMPLATE) && dest) {
849 if (flags & IP_VS_CONN_F_INACTIVE) {
850 atomic_dec(&dest->activeconns);
851 atomic_inc(&dest->inactconns);
852 } else {
853 atomic_inc(&dest->activeconns);
854 atomic_dec(&dest->inactconns);
855 }
856 }
857 flags &= IP_VS_CONN_F_BACKUP_UPD_MASK;
858 flags |= cp->flags & ~IP_VS_CONN_F_BACKUP_UPD_MASK;
859 cp->flags = flags;
860 spin_unlock_bh(&cp->lock);
861 if (!dest)
862 ip_vs_try_bind_dest(cp);
863 } else {
864 /*
865 * Find the appropriate destination for the connection.
866 * If it is not found the connection will remain unbound
867 * but still handled.
868 */
869 rcu_read_lock();
870 dest = ip_vs_find_dest(net, type, daddr, dport, param->vaddr,
871 param->vport, protocol, fwmark, flags);
872
873 cp = ip_vs_conn_new(param, daddr, dport, flags, dest, fwmark);
874 rcu_read_unlock();
875 if (!cp) {
876 if (param->pe_data)
877 kfree(param->pe_data);
878 IP_VS_DBG(2, "BACKUP, add new conn. failed\n");
879 return;
880 }
881 }
882
883 if (opt)
884 memcpy(&cp->in_seq, opt, sizeof(*opt));
885 atomic_set(&cp->in_pkts, sysctl_sync_threshold(ipvs));
886 cp->state = state;
887 cp->old_state = cp->state;
888 /*
889 * For Ver 0 messages style
890 * - Not possible to recover the right timeout for templates
891 * - can not find the right fwmark
892 * virtual service. If needed, we can do it for
893 * non-fwmark persistent services.
894 * Ver 1 messages style.
895 * - No problem.
896 */
897 if (timeout) {
898 if (timeout > MAX_SCHEDULE_TIMEOUT / HZ)
899 timeout = MAX_SCHEDULE_TIMEOUT / HZ;
900 cp->timeout = timeout*HZ;
901 } else {
902 struct ip_vs_proto_data *pd;
903
904 pd = ip_vs_proto_data_get(net, protocol);
905 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pd && pd->timeout_table)
906 cp->timeout = pd->timeout_table[state];
907 else
908 cp->timeout = (3*60*HZ);
909 }
910 ip_vs_conn_put(cp);
911 }
912
913 /*
914 * Process received multicast message for Version 0
915 */
ip_vs_process_message_v0(struct net * net,const char * buffer,const size_t buflen)916 static void ip_vs_process_message_v0(struct net *net, const char *buffer,
917 const size_t buflen)
918 {
919 struct ip_vs_sync_mesg_v0 *m = (struct ip_vs_sync_mesg_v0 *)buffer;
920 struct ip_vs_sync_conn_v0 *s;
921 struct ip_vs_sync_conn_options *opt;
922 struct ip_vs_protocol *pp;
923 struct ip_vs_conn_param param;
924 char *p;
925 int i;
926
927 p = (char *)buffer + sizeof(struct ip_vs_sync_mesg_v0);
928 for (i=0; i<m->nr_conns; i++) {
929 unsigned int flags, state;
930
931 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
932 IP_VS_ERR_RL("BACKUP v0, bogus conn\n");
933 return;
934 }
935 s = (struct ip_vs_sync_conn_v0 *) p;
936 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
937 flags &= ~IP_VS_CONN_F_HASHED;
938 if (flags & IP_VS_CONN_F_SEQ_MASK) {
939 opt = (struct ip_vs_sync_conn_options *)&s[1];
940 p += FULL_CONN_SIZE;
941 if (p > buffer+buflen) {
942 IP_VS_ERR_RL("BACKUP v0, Dropping buffer bogus conn options\n");
943 return;
944 }
945 } else {
946 opt = NULL;
947 p += SIMPLE_CONN_SIZE;
948 }
949
950 state = ntohs(s->state);
951 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
952 pp = ip_vs_proto_get(s->protocol);
953 if (!pp) {
954 IP_VS_DBG(2, "BACKUP v0, Unsupported protocol %u\n",
955 s->protocol);
956 continue;
957 }
958 if (state >= pp->num_states) {
959 IP_VS_DBG(2, "BACKUP v0, Invalid %s state %u\n",
960 pp->name, state);
961 continue;
962 }
963 } else {
964 /* protocol in templates is not used for state/timeout */
965 if (state > 0) {
966 IP_VS_DBG(2, "BACKUP v0, Invalid template state %u\n",
967 state);
968 state = 0;
969 }
970 }
971
972 ip_vs_conn_fill_param(net, AF_INET, s->protocol,
973 (const union nf_inet_addr *)&s->caddr,
974 s->cport,
975 (const union nf_inet_addr *)&s->vaddr,
976 s->vport, ¶m);
977
978 /* Send timeout as Zero */
979 ip_vs_proc_conn(net, ¶m, flags, state, s->protocol, AF_INET,
980 (union nf_inet_addr *)&s->daddr, s->dport,
981 0, 0, opt);
982 }
983 }
984
985 /*
986 * Handle options
987 */
ip_vs_proc_seqopt(__u8 * p,unsigned int plen,__u32 * opt_flags,struct ip_vs_sync_conn_options * opt)988 static inline int ip_vs_proc_seqopt(__u8 *p, unsigned int plen,
989 __u32 *opt_flags,
990 struct ip_vs_sync_conn_options *opt)
991 {
992 struct ip_vs_sync_conn_options *topt;
993
994 topt = (struct ip_vs_sync_conn_options *)p;
995
996 if (plen != sizeof(struct ip_vs_sync_conn_options)) {
997 IP_VS_DBG(2, "BACKUP, bogus conn options length\n");
998 return -EINVAL;
999 }
1000 if (*opt_flags & IPVS_OPT_F_SEQ_DATA) {
1001 IP_VS_DBG(2, "BACKUP, conn options found twice\n");
1002 return -EINVAL;
1003 }
1004 ntoh_seq(&topt->in_seq, &opt->in_seq);
1005 ntoh_seq(&topt->out_seq, &opt->out_seq);
1006 *opt_flags |= IPVS_OPT_F_SEQ_DATA;
1007 return 0;
1008 }
1009
ip_vs_proc_str(__u8 * p,unsigned int plen,unsigned int * data_len,__u8 ** data,unsigned int maxlen,__u32 * opt_flags,__u32 flag)1010 static int ip_vs_proc_str(__u8 *p, unsigned int plen, unsigned int *data_len,
1011 __u8 **data, unsigned int maxlen,
1012 __u32 *opt_flags, __u32 flag)
1013 {
1014 if (plen > maxlen) {
1015 IP_VS_DBG(2, "BACKUP, bogus par.data len > %d\n", maxlen);
1016 return -EINVAL;
1017 }
1018 if (*opt_flags & flag) {
1019 IP_VS_DBG(2, "BACKUP, Par.data found twice 0x%x\n", flag);
1020 return -EINVAL;
1021 }
1022 *data_len = plen;
1023 *data = p;
1024 *opt_flags |= flag;
1025 return 0;
1026 }
1027 /*
1028 * Process a Version 1 sync. connection
1029 */
ip_vs_proc_sync_conn(struct net * net,__u8 * p,__u8 * msg_end)1030 static inline int ip_vs_proc_sync_conn(struct net *net, __u8 *p, __u8 *msg_end)
1031 {
1032 struct ip_vs_sync_conn_options opt;
1033 union ip_vs_sync_conn *s;
1034 struct ip_vs_protocol *pp;
1035 struct ip_vs_conn_param param;
1036 __u32 flags;
1037 unsigned int af, state, pe_data_len=0, pe_name_len=0;
1038 __u8 *pe_data=NULL, *pe_name=NULL;
1039 __u32 opt_flags=0;
1040 int retc=0;
1041
1042 s = (union ip_vs_sync_conn *) p;
1043
1044 if (s->v6.type & STYPE_F_INET6) {
1045 #ifdef CONFIG_IP_VS_IPV6
1046 af = AF_INET6;
1047 p += sizeof(struct ip_vs_sync_v6);
1048 #else
1049 IP_VS_DBG(3,"BACKUP, IPv6 msg received, and IPVS is not compiled for IPv6\n");
1050 retc = 10;
1051 goto out;
1052 #endif
1053 } else if (!s->v4.type) {
1054 af = AF_INET;
1055 p += sizeof(struct ip_vs_sync_v4);
1056 } else {
1057 return -10;
1058 }
1059 if (p > msg_end)
1060 return -20;
1061
1062 /* Process optional params check Type & Len. */
1063 while (p < msg_end) {
1064 int ptype;
1065 int plen;
1066
1067 if (p+2 > msg_end)
1068 return -30;
1069 ptype = *(p++);
1070 plen = *(p++);
1071
1072 if (!plen || ((p + plen) > msg_end))
1073 return -40;
1074 /* Handle seq option p = param data */
1075 switch (ptype & ~IPVS_OPT_F_PARAM) {
1076 case IPVS_OPT_SEQ_DATA:
1077 if (ip_vs_proc_seqopt(p, plen, &opt_flags, &opt))
1078 return -50;
1079 break;
1080
1081 case IPVS_OPT_PE_DATA:
1082 if (ip_vs_proc_str(p, plen, &pe_data_len, &pe_data,
1083 IP_VS_PEDATA_MAXLEN, &opt_flags,
1084 IPVS_OPT_F_PE_DATA))
1085 return -60;
1086 break;
1087
1088 case IPVS_OPT_PE_NAME:
1089 if (ip_vs_proc_str(p, plen,&pe_name_len, &pe_name,
1090 IP_VS_PENAME_MAXLEN, &opt_flags,
1091 IPVS_OPT_F_PE_NAME))
1092 return -70;
1093 break;
1094
1095 default:
1096 /* Param data mandatory ? */
1097 if (!(ptype & IPVS_OPT_F_PARAM)) {
1098 IP_VS_DBG(3, "BACKUP, Unknown mandatory param %d found\n",
1099 ptype & ~IPVS_OPT_F_PARAM);
1100 retc = 20;
1101 goto out;
1102 }
1103 }
1104 p += plen; /* Next option */
1105 }
1106
1107 /* Get flags and Mask off unsupported */
1108 flags = ntohl(s->v4.flags) & IP_VS_CONN_F_BACKUP_MASK;
1109 flags |= IP_VS_CONN_F_SYNC;
1110 state = ntohs(s->v4.state);
1111
1112 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
1113 pp = ip_vs_proto_get(s->v4.protocol);
1114 if (!pp) {
1115 IP_VS_DBG(3,"BACKUP, Unsupported protocol %u\n",
1116 s->v4.protocol);
1117 retc = 30;
1118 goto out;
1119 }
1120 if (state >= pp->num_states) {
1121 IP_VS_DBG(3, "BACKUP, Invalid %s state %u\n",
1122 pp->name, state);
1123 retc = 40;
1124 goto out;
1125 }
1126 } else {
1127 /* protocol in templates is not used for state/timeout */
1128 if (state > 0) {
1129 IP_VS_DBG(3, "BACKUP, Invalid template state %u\n",
1130 state);
1131 state = 0;
1132 }
1133 }
1134 if (ip_vs_conn_fill_param_sync(net, af, s, ¶m, pe_data,
1135 pe_data_len, pe_name, pe_name_len)) {
1136 retc = 50;
1137 goto out;
1138 }
1139 /* If only IPv4, just silent skip IPv6 */
1140 if (af == AF_INET)
1141 ip_vs_proc_conn(net, ¶m, flags, state, s->v4.protocol, af,
1142 (union nf_inet_addr *)&s->v4.daddr, s->v4.dport,
1143 ntohl(s->v4.timeout), ntohl(s->v4.fwmark),
1144 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1145 );
1146 #ifdef CONFIG_IP_VS_IPV6
1147 else
1148 ip_vs_proc_conn(net, ¶m, flags, state, s->v6.protocol, af,
1149 (union nf_inet_addr *)&s->v6.daddr, s->v6.dport,
1150 ntohl(s->v6.timeout), ntohl(s->v6.fwmark),
1151 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1152 );
1153 #endif
1154 return 0;
1155 /* Error exit */
1156 out:
1157 IP_VS_DBG(2, "BACKUP, Single msg dropped err:%d\n", retc);
1158 return retc;
1159
1160 }
1161 /*
1162 * Process received multicast message and create the corresponding
1163 * ip_vs_conn entries.
1164 * Handles Version 0 & 1
1165 */
ip_vs_process_message(struct net * net,__u8 * buffer,const size_t buflen)1166 static void ip_vs_process_message(struct net *net, __u8 *buffer,
1167 const size_t buflen)
1168 {
1169 struct netns_ipvs *ipvs = net_ipvs(net);
1170 struct ip_vs_sync_mesg *m2 = (struct ip_vs_sync_mesg *)buffer;
1171 __u8 *p, *msg_end;
1172 int i, nr_conns;
1173
1174 if (buflen < sizeof(struct ip_vs_sync_mesg_v0)) {
1175 IP_VS_DBG(2, "BACKUP, message header too short\n");
1176 return;
1177 }
1178
1179 if (buflen != ntohs(m2->size)) {
1180 IP_VS_DBG(2, "BACKUP, bogus message size\n");
1181 return;
1182 }
1183 /* SyncID sanity check */
1184 if (ipvs->backup_syncid != 0 && m2->syncid != ipvs->backup_syncid) {
1185 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2->syncid);
1186 return;
1187 }
1188 /* Handle version 1 message */
1189 if ((m2->version == SYNC_PROTO_VER) && (m2->reserved == 0)
1190 && (m2->spare == 0)) {
1191
1192 msg_end = buffer + sizeof(struct ip_vs_sync_mesg);
1193 nr_conns = m2->nr_conns;
1194
1195 for (i=0; i<nr_conns; i++) {
1196 union ip_vs_sync_conn *s;
1197 unsigned int size;
1198 int retc;
1199
1200 p = msg_end;
1201 if (p + sizeof(s->v4) > buffer+buflen) {
1202 IP_VS_ERR_RL("BACKUP, Dropping buffer, to small\n");
1203 return;
1204 }
1205 s = (union ip_vs_sync_conn *)p;
1206 size = ntohs(s->v4.ver_size) & SVER_MASK;
1207 msg_end = p + size;
1208 /* Basic sanity checks */
1209 if (msg_end > buffer+buflen) {
1210 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n");
1211 return;
1212 }
1213 if (ntohs(s->v4.ver_size) >> SVER_SHIFT) {
1214 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n",
1215 ntohs(s->v4.ver_size) >> SVER_SHIFT);
1216 return;
1217 }
1218 /* Process a single sync_conn */
1219 retc = ip_vs_proc_sync_conn(net, p, msg_end);
1220 if (retc < 0) {
1221 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n",
1222 retc);
1223 return;
1224 }
1225 /* Make sure we have 32 bit alignment */
1226 msg_end = p + ((size + 3) & ~3);
1227 }
1228 } else {
1229 /* Old type of message */
1230 ip_vs_process_message_v0(net, buffer, buflen);
1231 return;
1232 }
1233 }
1234
1235
1236 /*
1237 * Setup sndbuf (mode=1) or rcvbuf (mode=0)
1238 */
set_sock_size(struct sock * sk,int mode,int val)1239 static void set_sock_size(struct sock *sk, int mode, int val)
1240 {
1241 /* setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &val, sizeof(val)); */
1242 /* setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)); */
1243 lock_sock(sk);
1244 if (mode) {
1245 val = clamp_t(int, val, (SOCK_MIN_SNDBUF + 1) / 2,
1246 sysctl_wmem_max);
1247 sk->sk_sndbuf = val * 2;
1248 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1249 } else {
1250 val = clamp_t(int, val, (SOCK_MIN_RCVBUF + 1) / 2,
1251 sysctl_rmem_max);
1252 sk->sk_rcvbuf = val * 2;
1253 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1254 }
1255 release_sock(sk);
1256 }
1257
1258 /*
1259 * Setup loopback of outgoing multicasts on a sending socket
1260 */
set_mcast_loop(struct sock * sk,u_char loop)1261 static void set_mcast_loop(struct sock *sk, u_char loop)
1262 {
1263 struct inet_sock *inet = inet_sk(sk);
1264
1265 /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
1266 lock_sock(sk);
1267 inet->mc_loop = loop ? 1 : 0;
1268 release_sock(sk);
1269 }
1270
1271 /*
1272 * Specify TTL for outgoing multicasts on a sending socket
1273 */
set_mcast_ttl(struct sock * sk,u_char ttl)1274 static void set_mcast_ttl(struct sock *sk, u_char ttl)
1275 {
1276 struct inet_sock *inet = inet_sk(sk);
1277
1278 /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
1279 lock_sock(sk);
1280 inet->mc_ttl = ttl;
1281 release_sock(sk);
1282 }
1283
1284 /*
1285 * Specifiy default interface for outgoing multicasts
1286 */
set_mcast_if(struct sock * sk,char * ifname)1287 static int set_mcast_if(struct sock *sk, char *ifname)
1288 {
1289 struct net_device *dev;
1290 struct inet_sock *inet = inet_sk(sk);
1291 struct net *net = sock_net(sk);
1292
1293 dev = __dev_get_by_name(net, ifname);
1294 if (!dev)
1295 return -ENODEV;
1296
1297 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1298 return -EINVAL;
1299
1300 lock_sock(sk);
1301 inet->mc_index = dev->ifindex;
1302 /* inet->mc_addr = 0; */
1303 release_sock(sk);
1304
1305 return 0;
1306 }
1307
1308
1309 /*
1310 * Set the maximum length of sync message according to the
1311 * specified interface's MTU.
1312 */
set_sync_mesg_maxlen(struct net * net,int sync_state)1313 static int set_sync_mesg_maxlen(struct net *net, int sync_state)
1314 {
1315 struct netns_ipvs *ipvs = net_ipvs(net);
1316 struct net_device *dev;
1317 int num;
1318
1319 if (sync_state == IP_VS_STATE_MASTER) {
1320 dev = __dev_get_by_name(net, ipvs->master_mcast_ifn);
1321 if (!dev)
1322 return -ENODEV;
1323
1324 num = (dev->mtu - sizeof(struct iphdr) -
1325 sizeof(struct udphdr) -
1326 SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE;
1327 ipvs->send_mesg_maxlen = SYNC_MESG_HEADER_LEN +
1328 SIMPLE_CONN_SIZE * min(num, MAX_CONNS_PER_SYNCBUFF);
1329 IP_VS_DBG(7, "setting the maximum length of sync sending "
1330 "message %d.\n", ipvs->send_mesg_maxlen);
1331 } else if (sync_state == IP_VS_STATE_BACKUP) {
1332 dev = __dev_get_by_name(net, ipvs->backup_mcast_ifn);
1333 if (!dev)
1334 return -ENODEV;
1335
1336 ipvs->recv_mesg_maxlen = dev->mtu -
1337 sizeof(struct iphdr) - sizeof(struct udphdr);
1338 IP_VS_DBG(7, "setting the maximum length of sync receiving "
1339 "message %d.\n", ipvs->recv_mesg_maxlen);
1340 }
1341
1342 return 0;
1343 }
1344
1345
1346 /*
1347 * Join a multicast group.
1348 * the group is specified by a class D multicast address 224.0.0.0/8
1349 * in the in_addr structure passed in as a parameter.
1350 */
1351 static int
join_mcast_group(struct sock * sk,struct in_addr * addr,char * ifname)1352 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
1353 {
1354 struct net *net = sock_net(sk);
1355 struct ip_mreqn mreq;
1356 struct net_device *dev;
1357 int ret;
1358
1359 memset(&mreq, 0, sizeof(mreq));
1360 memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
1361
1362 dev = __dev_get_by_name(net, ifname);
1363 if (!dev)
1364 return -ENODEV;
1365 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1366 return -EINVAL;
1367
1368 mreq.imr_ifindex = dev->ifindex;
1369
1370 lock_sock(sk);
1371 ret = ip_mc_join_group(sk, &mreq);
1372 release_sock(sk);
1373
1374 return ret;
1375 }
1376
1377
bind_mcastif_addr(struct socket * sock,char * ifname)1378 static int bind_mcastif_addr(struct socket *sock, char *ifname)
1379 {
1380 struct net *net = sock_net(sock->sk);
1381 struct net_device *dev;
1382 __be32 addr;
1383 struct sockaddr_in sin;
1384
1385 dev = __dev_get_by_name(net, ifname);
1386 if (!dev)
1387 return -ENODEV;
1388
1389 addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
1390 if (!addr)
1391 pr_err("You probably need to specify IP address on "
1392 "multicast interface.\n");
1393
1394 IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
1395 ifname, &addr);
1396
1397 /* Now bind the socket with the address of multicast interface */
1398 sin.sin_family = AF_INET;
1399 sin.sin_addr.s_addr = addr;
1400 sin.sin_port = 0;
1401
1402 return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
1403 }
1404
1405 /*
1406 * Set up sending multicast socket over UDP
1407 */
make_send_sock(struct net * net,int id)1408 static struct socket *make_send_sock(struct net *net, int id)
1409 {
1410 struct netns_ipvs *ipvs = net_ipvs(net);
1411 /* multicast addr */
1412 struct sockaddr_in mcast_addr = {
1413 .sin_family = AF_INET,
1414 .sin_port = cpu_to_be16(IP_VS_SYNC_PORT + id),
1415 .sin_addr.s_addr = cpu_to_be32(IP_VS_SYNC_GROUP),
1416 };
1417 struct socket *sock;
1418 int result;
1419
1420 /* First create a socket move it to right name space later */
1421 result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1422 if (result < 0) {
1423 pr_err("Error during creation of socket; terminating\n");
1424 return ERR_PTR(result);
1425 }
1426 /*
1427 * Kernel sockets that are a part of a namespace, should not
1428 * hold a reference to a namespace in order to allow to stop it.
1429 * After sk_change_net should be released using sk_release_kernel.
1430 */
1431 sk_change_net(sock->sk, net);
1432 result = set_mcast_if(sock->sk, ipvs->master_mcast_ifn);
1433 if (result < 0) {
1434 pr_err("Error setting outbound mcast interface\n");
1435 goto error;
1436 }
1437
1438 set_mcast_loop(sock->sk, 0);
1439 set_mcast_ttl(sock->sk, 1);
1440 result = sysctl_sync_sock_size(ipvs);
1441 if (result > 0)
1442 set_sock_size(sock->sk, 1, result);
1443
1444 result = bind_mcastif_addr(sock, ipvs->master_mcast_ifn);
1445 if (result < 0) {
1446 pr_err("Error binding address of the mcast interface\n");
1447 goto error;
1448 }
1449
1450 result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr,
1451 sizeof(struct sockaddr), 0);
1452 if (result < 0) {
1453 pr_err("Error connecting to the multicast addr\n");
1454 goto error;
1455 }
1456
1457 return sock;
1458
1459 error:
1460 sk_release_kernel(sock->sk);
1461 return ERR_PTR(result);
1462 }
1463
1464
1465 /*
1466 * Set up receiving multicast socket over UDP
1467 */
make_receive_sock(struct net * net,int id)1468 static struct socket *make_receive_sock(struct net *net, int id)
1469 {
1470 struct netns_ipvs *ipvs = net_ipvs(net);
1471 /* multicast addr */
1472 struct sockaddr_in mcast_addr = {
1473 .sin_family = AF_INET,
1474 .sin_port = cpu_to_be16(IP_VS_SYNC_PORT + id),
1475 .sin_addr.s_addr = cpu_to_be32(IP_VS_SYNC_GROUP),
1476 };
1477 struct socket *sock;
1478 int result;
1479
1480 /* First create a socket */
1481 result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1482 if (result < 0) {
1483 pr_err("Error during creation of socket; terminating\n");
1484 return ERR_PTR(result);
1485 }
1486 /*
1487 * Kernel sockets that are a part of a namespace, should not
1488 * hold a reference to a namespace in order to allow to stop it.
1489 * After sk_change_net should be released using sk_release_kernel.
1490 */
1491 sk_change_net(sock->sk, net);
1492 /* it is equivalent to the REUSEADDR option in user-space */
1493 sock->sk->sk_reuse = SK_CAN_REUSE;
1494 result = sysctl_sync_sock_size(ipvs);
1495 if (result > 0)
1496 set_sock_size(sock->sk, 0, result);
1497
1498 result = sock->ops->bind(sock, (struct sockaddr *) &mcast_addr,
1499 sizeof(struct sockaddr));
1500 if (result < 0) {
1501 pr_err("Error binding to the multicast addr\n");
1502 goto error;
1503 }
1504
1505 /* join the multicast group */
1506 result = join_mcast_group(sock->sk,
1507 (struct in_addr *) &mcast_addr.sin_addr,
1508 ipvs->backup_mcast_ifn);
1509 if (result < 0) {
1510 pr_err("Error joining to the multicast group\n");
1511 goto error;
1512 }
1513
1514 return sock;
1515
1516 error:
1517 sk_release_kernel(sock->sk);
1518 return ERR_PTR(result);
1519 }
1520
1521
1522 static int
ip_vs_send_async(struct socket * sock,const char * buffer,const size_t length)1523 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
1524 {
1525 struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
1526 struct kvec iov;
1527 int len;
1528
1529 EnterFunction(7);
1530 iov.iov_base = (void *)buffer;
1531 iov.iov_len = length;
1532
1533 len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
1534
1535 LeaveFunction(7);
1536 return len;
1537 }
1538
1539 static int
ip_vs_send_sync_msg(struct socket * sock,struct ip_vs_sync_mesg * msg)1540 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
1541 {
1542 int msize;
1543 int ret;
1544
1545 msize = ntohs(msg->size);
1546
1547 ret = ip_vs_send_async(sock, (char *)msg, msize);
1548 if (ret >= 0 || ret == -EAGAIN)
1549 return ret;
1550 pr_err("ip_vs_send_async error %d\n", ret);
1551 return 0;
1552 }
1553
1554 static int
ip_vs_receive(struct socket * sock,char * buffer,const size_t buflen)1555 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
1556 {
1557 struct msghdr msg = {NULL,};
1558 struct kvec iov;
1559 int len;
1560
1561 EnterFunction(7);
1562
1563 /* Receive a packet */
1564 iov.iov_base = buffer;
1565 iov.iov_len = (size_t)buflen;
1566
1567 len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, MSG_DONTWAIT);
1568
1569 if (len < 0)
1570 return len;
1571
1572 LeaveFunction(7);
1573 return len;
1574 }
1575
1576 /* Wakeup the master thread for sending */
master_wakeup_work_handler(struct work_struct * work)1577 static void master_wakeup_work_handler(struct work_struct *work)
1578 {
1579 struct ipvs_master_sync_state *ms =
1580 container_of(work, struct ipvs_master_sync_state,
1581 master_wakeup_work.work);
1582 struct netns_ipvs *ipvs = ms->ipvs;
1583
1584 spin_lock_bh(&ipvs->sync_lock);
1585 if (ms->sync_queue_len &&
1586 ms->sync_queue_delay < IPVS_SYNC_WAKEUP_RATE) {
1587 ms->sync_queue_delay = IPVS_SYNC_WAKEUP_RATE;
1588 wake_up_process(ms->master_thread);
1589 }
1590 spin_unlock_bh(&ipvs->sync_lock);
1591 }
1592
1593 /* Get next buffer to send */
1594 static inline struct ip_vs_sync_buff *
next_sync_buff(struct netns_ipvs * ipvs,struct ipvs_master_sync_state * ms)1595 next_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
1596 {
1597 struct ip_vs_sync_buff *sb;
1598
1599 sb = sb_dequeue(ipvs, ms);
1600 if (sb)
1601 return sb;
1602 /* Do not delay entries in buffer for more than 2 seconds */
1603 return get_curr_sync_buff(ipvs, ms, IPVS_SYNC_FLUSH_TIME);
1604 }
1605
sync_thread_master(void * data)1606 static int sync_thread_master(void *data)
1607 {
1608 struct ip_vs_sync_thread_data *tinfo = data;
1609 struct netns_ipvs *ipvs = net_ipvs(tinfo->net);
1610 struct ipvs_master_sync_state *ms = &ipvs->ms[tinfo->id];
1611 struct sock *sk = tinfo->sock->sk;
1612 struct ip_vs_sync_buff *sb;
1613
1614 pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
1615 "syncid = %d, id = %d\n",
1616 ipvs->master_mcast_ifn, ipvs->master_syncid, tinfo->id);
1617
1618 for (;;) {
1619 sb = next_sync_buff(ipvs, ms);
1620 if (unlikely(kthread_should_stop()))
1621 break;
1622 if (!sb) {
1623 schedule_timeout(IPVS_SYNC_CHECK_PERIOD);
1624 continue;
1625 }
1626 while (ip_vs_send_sync_msg(tinfo->sock, sb->mesg) < 0) {
1627 int ret = 0;
1628
1629 __wait_event_interruptible(*sk_sleep(sk),
1630 sock_writeable(sk) ||
1631 kthread_should_stop(),
1632 ret);
1633 if (unlikely(kthread_should_stop()))
1634 goto done;
1635 }
1636 ip_vs_sync_buff_release(sb);
1637 }
1638
1639 done:
1640 __set_current_state(TASK_RUNNING);
1641 if (sb)
1642 ip_vs_sync_buff_release(sb);
1643
1644 /* clean up the sync_buff queue */
1645 while ((sb = sb_dequeue(ipvs, ms)))
1646 ip_vs_sync_buff_release(sb);
1647 __set_current_state(TASK_RUNNING);
1648
1649 /* clean up the current sync_buff */
1650 sb = get_curr_sync_buff(ipvs, ms, 0);
1651 if (sb)
1652 ip_vs_sync_buff_release(sb);
1653
1654 /* release the sending multicast socket */
1655 sk_release_kernel(tinfo->sock->sk);
1656 kfree(tinfo);
1657
1658 return 0;
1659 }
1660
1661
sync_thread_backup(void * data)1662 static int sync_thread_backup(void *data)
1663 {
1664 struct ip_vs_sync_thread_data *tinfo = data;
1665 struct netns_ipvs *ipvs = net_ipvs(tinfo->net);
1666 int len;
1667
1668 pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
1669 "syncid = %d, id = %d\n",
1670 ipvs->backup_mcast_ifn, ipvs->backup_syncid, tinfo->id);
1671
1672 while (!kthread_should_stop()) {
1673 wait_event_interruptible(*sk_sleep(tinfo->sock->sk),
1674 !skb_queue_empty(&tinfo->sock->sk->sk_receive_queue)
1675 || kthread_should_stop());
1676
1677 /* do we have data now? */
1678 while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) {
1679 len = ip_vs_receive(tinfo->sock, tinfo->buf,
1680 ipvs->recv_mesg_maxlen);
1681 if (len <= 0) {
1682 if (len != -EAGAIN)
1683 pr_err("receiving message error\n");
1684 break;
1685 }
1686
1687 ip_vs_process_message(tinfo->net, tinfo->buf, len);
1688 }
1689 }
1690
1691 /* release the sending multicast socket */
1692 sk_release_kernel(tinfo->sock->sk);
1693 kfree(tinfo->buf);
1694 kfree(tinfo);
1695
1696 return 0;
1697 }
1698
1699
start_sync_thread(struct net * net,int state,char * mcast_ifn,__u8 syncid)1700 int start_sync_thread(struct net *net, int state, char *mcast_ifn, __u8 syncid)
1701 {
1702 struct ip_vs_sync_thread_data *tinfo;
1703 struct task_struct **array = NULL, *task;
1704 struct socket *sock;
1705 struct netns_ipvs *ipvs = net_ipvs(net);
1706 char *name;
1707 int (*threadfn)(void *data);
1708 int id, count;
1709 int result = -ENOMEM;
1710
1711 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1712 IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n",
1713 sizeof(struct ip_vs_sync_conn_v0));
1714
1715 if (!ipvs->sync_state) {
1716 count = clamp(sysctl_sync_ports(ipvs), 1, IPVS_SYNC_PORTS_MAX);
1717 ipvs->threads_mask = count - 1;
1718 } else
1719 count = ipvs->threads_mask + 1;
1720
1721 if (state == IP_VS_STATE_MASTER) {
1722 if (ipvs->ms)
1723 return -EEXIST;
1724
1725 strlcpy(ipvs->master_mcast_ifn, mcast_ifn,
1726 sizeof(ipvs->master_mcast_ifn));
1727 ipvs->master_syncid = syncid;
1728 name = "ipvs-m:%d:%d";
1729 threadfn = sync_thread_master;
1730 } else if (state == IP_VS_STATE_BACKUP) {
1731 if (ipvs->backup_threads)
1732 return -EEXIST;
1733
1734 strlcpy(ipvs->backup_mcast_ifn, mcast_ifn,
1735 sizeof(ipvs->backup_mcast_ifn));
1736 ipvs->backup_syncid = syncid;
1737 name = "ipvs-b:%d:%d";
1738 threadfn = sync_thread_backup;
1739 } else {
1740 return -EINVAL;
1741 }
1742
1743 if (state == IP_VS_STATE_MASTER) {
1744 struct ipvs_master_sync_state *ms;
1745
1746 ipvs->ms = kzalloc(count * sizeof(ipvs->ms[0]), GFP_KERNEL);
1747 if (!ipvs->ms)
1748 goto out;
1749 ms = ipvs->ms;
1750 for (id = 0; id < count; id++, ms++) {
1751 INIT_LIST_HEAD(&ms->sync_queue);
1752 ms->sync_queue_len = 0;
1753 ms->sync_queue_delay = 0;
1754 INIT_DELAYED_WORK(&ms->master_wakeup_work,
1755 master_wakeup_work_handler);
1756 ms->ipvs = ipvs;
1757 }
1758 } else {
1759 array = kzalloc(count * sizeof(struct task_struct *),
1760 GFP_KERNEL);
1761 if (!array)
1762 goto out;
1763 }
1764 set_sync_mesg_maxlen(net, state);
1765
1766 tinfo = NULL;
1767 for (id = 0; id < count; id++) {
1768 if (state == IP_VS_STATE_MASTER)
1769 sock = make_send_sock(net, id);
1770 else
1771 sock = make_receive_sock(net, id);
1772 if (IS_ERR(sock)) {
1773 result = PTR_ERR(sock);
1774 goto outtinfo;
1775 }
1776 tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL);
1777 if (!tinfo)
1778 goto outsocket;
1779 tinfo->net = net;
1780 tinfo->sock = sock;
1781 if (state == IP_VS_STATE_BACKUP) {
1782 tinfo->buf = kmalloc(ipvs->recv_mesg_maxlen,
1783 GFP_KERNEL);
1784 if (!tinfo->buf)
1785 goto outtinfo;
1786 } else {
1787 tinfo->buf = NULL;
1788 }
1789 tinfo->id = id;
1790
1791 task = kthread_run(threadfn, tinfo, name, ipvs->gen, id);
1792 if (IS_ERR(task)) {
1793 result = PTR_ERR(task);
1794 goto outtinfo;
1795 }
1796 tinfo = NULL;
1797 if (state == IP_VS_STATE_MASTER)
1798 ipvs->ms[id].master_thread = task;
1799 else
1800 array[id] = task;
1801 }
1802
1803 /* mark as active */
1804
1805 if (state == IP_VS_STATE_BACKUP)
1806 ipvs->backup_threads = array;
1807 spin_lock_bh(&ipvs->sync_buff_lock);
1808 ipvs->sync_state |= state;
1809 spin_unlock_bh(&ipvs->sync_buff_lock);
1810
1811 /* increase the module use count */
1812 ip_vs_use_count_inc();
1813
1814 return 0;
1815
1816 outsocket:
1817 sk_release_kernel(sock->sk);
1818
1819 outtinfo:
1820 if (tinfo) {
1821 sk_release_kernel(tinfo->sock->sk);
1822 kfree(tinfo->buf);
1823 kfree(tinfo);
1824 }
1825 count = id;
1826 while (count-- > 0) {
1827 if (state == IP_VS_STATE_MASTER)
1828 kthread_stop(ipvs->ms[count].master_thread);
1829 else
1830 kthread_stop(array[count]);
1831 }
1832 kfree(array);
1833
1834 out:
1835 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
1836 kfree(ipvs->ms);
1837 ipvs->ms = NULL;
1838 }
1839 return result;
1840 }
1841
1842
stop_sync_thread(struct net * net,int state)1843 int stop_sync_thread(struct net *net, int state)
1844 {
1845 struct netns_ipvs *ipvs = net_ipvs(net);
1846 struct task_struct **array;
1847 int id;
1848 int retc = -EINVAL;
1849
1850 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1851
1852 if (state == IP_VS_STATE_MASTER) {
1853 if (!ipvs->ms)
1854 return -ESRCH;
1855
1856 /*
1857 * The lock synchronizes with sb_queue_tail(), so that we don't
1858 * add sync buffers to the queue, when we are already in
1859 * progress of stopping the master sync daemon.
1860 */
1861
1862 spin_lock_bh(&ipvs->sync_buff_lock);
1863 spin_lock(&ipvs->sync_lock);
1864 ipvs->sync_state &= ~IP_VS_STATE_MASTER;
1865 spin_unlock(&ipvs->sync_lock);
1866 spin_unlock_bh(&ipvs->sync_buff_lock);
1867
1868 retc = 0;
1869 for (id = ipvs->threads_mask; id >= 0; id--) {
1870 struct ipvs_master_sync_state *ms = &ipvs->ms[id];
1871 int ret;
1872
1873 pr_info("stopping master sync thread %d ...\n",
1874 task_pid_nr(ms->master_thread));
1875 cancel_delayed_work_sync(&ms->master_wakeup_work);
1876 ret = kthread_stop(ms->master_thread);
1877 if (retc >= 0)
1878 retc = ret;
1879 }
1880 kfree(ipvs->ms);
1881 ipvs->ms = NULL;
1882 } else if (state == IP_VS_STATE_BACKUP) {
1883 if (!ipvs->backup_threads)
1884 return -ESRCH;
1885
1886 ipvs->sync_state &= ~IP_VS_STATE_BACKUP;
1887 array = ipvs->backup_threads;
1888 retc = 0;
1889 for (id = ipvs->threads_mask; id >= 0; id--) {
1890 int ret;
1891
1892 pr_info("stopping backup sync thread %d ...\n",
1893 task_pid_nr(array[id]));
1894 ret = kthread_stop(array[id]);
1895 if (retc >= 0)
1896 retc = ret;
1897 }
1898 kfree(array);
1899 ipvs->backup_threads = NULL;
1900 }
1901
1902 /* decrease the module use count */
1903 ip_vs_use_count_dec();
1904
1905 return retc;
1906 }
1907
1908 /*
1909 * Initialize data struct for each netns
1910 */
ip_vs_sync_net_init(struct net * net)1911 int __net_init ip_vs_sync_net_init(struct net *net)
1912 {
1913 struct netns_ipvs *ipvs = net_ipvs(net);
1914
1915 __mutex_init(&ipvs->sync_mutex, "ipvs->sync_mutex", &__ipvs_sync_key);
1916 spin_lock_init(&ipvs->sync_lock);
1917 spin_lock_init(&ipvs->sync_buff_lock);
1918 return 0;
1919 }
1920
ip_vs_sync_net_cleanup(struct net * net)1921 void ip_vs_sync_net_cleanup(struct net *net)
1922 {
1923 int retc;
1924 struct netns_ipvs *ipvs = net_ipvs(net);
1925
1926 mutex_lock(&ipvs->sync_mutex);
1927 retc = stop_sync_thread(net, IP_VS_STATE_MASTER);
1928 if (retc && retc != -ESRCH)
1929 pr_err("Failed to stop Master Daemon\n");
1930
1931 retc = stop_sync_thread(net, IP_VS_STATE_BACKUP);
1932 if (retc && retc != -ESRCH)
1933 pr_err("Failed to stop Backup Daemon\n");
1934 mutex_unlock(&ipvs->sync_mutex);
1935 }
1936