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 connection is controlled by persistence */
in_persistence(struct ip_vs_conn * cp)429 static inline bool in_persistence(struct ip_vs_conn *cp)
430 {
431 for (cp = cp->control; cp; cp = cp->control) {
432 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
433 return true;
434 }
435 return false;
436 }
437
438 /* Check if conn should be synced.
439 * pkts: conn packets, use sysctl_sync_threshold to avoid packet check
440 * - (1) sync_refresh_period: reduce sync rate. Additionally, retry
441 * sync_retries times with period of sync_refresh_period/8
442 * - (2) if both sync_refresh_period and sync_period are 0 send sync only
443 * for state changes or only once when pkts matches sync_threshold
444 * - (3) templates: rate can be reduced only with sync_refresh_period or
445 * with (2)
446 */
ip_vs_sync_conn_needed(struct netns_ipvs * ipvs,struct ip_vs_conn * cp,int pkts)447 static int ip_vs_sync_conn_needed(struct netns_ipvs *ipvs,
448 struct ip_vs_conn *cp, int pkts)
449 {
450 unsigned long orig = ACCESS_ONCE(cp->sync_endtime);
451 unsigned long now = jiffies;
452 unsigned long n = (now + cp->timeout) & ~3UL;
453 unsigned int sync_refresh_period;
454 int sync_period;
455 int force;
456
457 /* Check if we sync in current state */
458 if (unlikely(cp->flags & IP_VS_CONN_F_TEMPLATE))
459 force = 0;
460 else if (unlikely(sysctl_sync_persist_mode(ipvs) && in_persistence(cp)))
461 return 0;
462 else if (likely(cp->protocol == IPPROTO_TCP)) {
463 if (!((1 << cp->state) &
464 ((1 << IP_VS_TCP_S_ESTABLISHED) |
465 (1 << IP_VS_TCP_S_FIN_WAIT) |
466 (1 << IP_VS_TCP_S_CLOSE) |
467 (1 << IP_VS_TCP_S_CLOSE_WAIT) |
468 (1 << IP_VS_TCP_S_TIME_WAIT))))
469 return 0;
470 force = cp->state != cp->old_state;
471 if (force && cp->state != IP_VS_TCP_S_ESTABLISHED)
472 goto set;
473 } else if (unlikely(cp->protocol == IPPROTO_SCTP)) {
474 if (!((1 << cp->state) &
475 ((1 << IP_VS_SCTP_S_ESTABLISHED) |
476 (1 << IP_VS_SCTP_S_SHUTDOWN_SENT) |
477 (1 << IP_VS_SCTP_S_SHUTDOWN_RECEIVED) |
478 (1 << IP_VS_SCTP_S_SHUTDOWN_ACK_SENT) |
479 (1 << IP_VS_SCTP_S_CLOSED))))
480 return 0;
481 force = cp->state != cp->old_state;
482 if (force && cp->state != IP_VS_SCTP_S_ESTABLISHED)
483 goto set;
484 } else {
485 /* UDP or another protocol with single state */
486 force = 0;
487 }
488
489 sync_refresh_period = sysctl_sync_refresh_period(ipvs);
490 if (sync_refresh_period > 0) {
491 long diff = n - orig;
492 long min_diff = max(cp->timeout >> 1, 10UL * HZ);
493
494 /* Avoid sync if difference is below sync_refresh_period
495 * and below the half timeout.
496 */
497 if (abs(diff) < min_t(long, sync_refresh_period, min_diff)) {
498 int retries = orig & 3;
499
500 if (retries >= sysctl_sync_retries(ipvs))
501 return 0;
502 if (time_before(now, orig - cp->timeout +
503 (sync_refresh_period >> 3)))
504 return 0;
505 n |= retries + 1;
506 }
507 }
508 sync_period = sysctl_sync_period(ipvs);
509 if (sync_period > 0) {
510 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE) &&
511 pkts % sync_period != sysctl_sync_threshold(ipvs))
512 return 0;
513 } else if (sync_refresh_period <= 0 &&
514 pkts != sysctl_sync_threshold(ipvs))
515 return 0;
516
517 set:
518 cp->old_state = cp->state;
519 n = cmpxchg(&cp->sync_endtime, orig, n);
520 return n == orig || force;
521 }
522
523 /*
524 * Version 0 , could be switched in by sys_ctl.
525 * Add an ip_vs_conn information into the current sync_buff.
526 */
ip_vs_sync_conn_v0(struct net * net,struct ip_vs_conn * cp,int pkts)527 static void ip_vs_sync_conn_v0(struct net *net, struct ip_vs_conn *cp,
528 int pkts)
529 {
530 struct netns_ipvs *ipvs = net_ipvs(net);
531 struct ip_vs_sync_mesg_v0 *m;
532 struct ip_vs_sync_conn_v0 *s;
533 struct ip_vs_sync_buff *buff;
534 struct ipvs_master_sync_state *ms;
535 int id;
536 int len;
537
538 if (unlikely(cp->af != AF_INET))
539 return;
540 /* Do not sync ONE PACKET */
541 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
542 return;
543
544 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
545 return;
546
547 spin_lock_bh(&ipvs->sync_buff_lock);
548 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
549 spin_unlock_bh(&ipvs->sync_buff_lock);
550 return;
551 }
552
553 id = select_master_thread_id(ipvs, cp);
554 ms = &ipvs->ms[id];
555 buff = ms->sync_buff;
556 if (buff) {
557 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
558 /* Send buffer if it is for v1 */
559 if (!m->nr_conns) {
560 sb_queue_tail(ipvs, ms);
561 ms->sync_buff = NULL;
562 buff = NULL;
563 }
564 }
565 if (!buff) {
566 buff = ip_vs_sync_buff_create_v0(ipvs);
567 if (!buff) {
568 spin_unlock_bh(&ipvs->sync_buff_lock);
569 pr_err("ip_vs_sync_buff_create failed.\n");
570 return;
571 }
572 ms->sync_buff = buff;
573 }
574
575 len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
576 SIMPLE_CONN_SIZE;
577 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
578 s = (struct ip_vs_sync_conn_v0 *) buff->head;
579
580 /* copy members */
581 s->reserved = 0;
582 s->protocol = cp->protocol;
583 s->cport = cp->cport;
584 s->vport = cp->vport;
585 s->dport = cp->dport;
586 s->caddr = cp->caddr.ip;
587 s->vaddr = cp->vaddr.ip;
588 s->daddr = cp->daddr.ip;
589 s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
590 s->state = htons(cp->state);
591 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
592 struct ip_vs_sync_conn_options *opt =
593 (struct ip_vs_sync_conn_options *)&s[1];
594 memcpy(opt, &cp->in_seq, sizeof(*opt));
595 }
596
597 m->nr_conns++;
598 m->size = htons(ntohs(m->size) + len);
599 buff->head += len;
600
601 /* check if there is a space for next one */
602 if (buff->head + FULL_CONN_SIZE > buff->end) {
603 sb_queue_tail(ipvs, ms);
604 ms->sync_buff = NULL;
605 }
606 spin_unlock_bh(&ipvs->sync_buff_lock);
607
608 /* synchronize its controller if it has */
609 cp = cp->control;
610 if (cp) {
611 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
612 pkts = atomic_add_return(1, &cp->in_pkts);
613 else
614 pkts = sysctl_sync_threshold(ipvs);
615 ip_vs_sync_conn(net, cp, pkts);
616 }
617 }
618
619 /*
620 * Add an ip_vs_conn information into the current sync_buff.
621 * Called by ip_vs_in.
622 * Sending Version 1 messages
623 */
ip_vs_sync_conn(struct net * net,struct ip_vs_conn * cp,int pkts)624 void ip_vs_sync_conn(struct net *net, struct ip_vs_conn *cp, int pkts)
625 {
626 struct netns_ipvs *ipvs = net_ipvs(net);
627 struct ip_vs_sync_mesg *m;
628 union ip_vs_sync_conn *s;
629 struct ip_vs_sync_buff *buff;
630 struct ipvs_master_sync_state *ms;
631 int id;
632 __u8 *p;
633 unsigned int len, pe_name_len, pad;
634
635 /* Handle old version of the protocol */
636 if (sysctl_sync_ver(ipvs) == 0) {
637 ip_vs_sync_conn_v0(net, cp, pkts);
638 return;
639 }
640 /* Do not sync ONE PACKET */
641 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
642 goto control;
643 sloop:
644 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
645 goto control;
646
647 /* Sanity checks */
648 pe_name_len = 0;
649 if (cp->pe_data_len) {
650 if (!cp->pe_data || !cp->dest) {
651 IP_VS_ERR_RL("SYNC, connection pe_data invalid\n");
652 return;
653 }
654 pe_name_len = strnlen(cp->pe->name, IP_VS_PENAME_MAXLEN);
655 }
656
657 spin_lock_bh(&ipvs->sync_buff_lock);
658 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
659 spin_unlock_bh(&ipvs->sync_buff_lock);
660 return;
661 }
662
663 id = select_master_thread_id(ipvs, cp);
664 ms = &ipvs->ms[id];
665
666 #ifdef CONFIG_IP_VS_IPV6
667 if (cp->af == AF_INET6)
668 len = sizeof(struct ip_vs_sync_v6);
669 else
670 #endif
671 len = sizeof(struct ip_vs_sync_v4);
672
673 if (cp->flags & IP_VS_CONN_F_SEQ_MASK)
674 len += sizeof(struct ip_vs_sync_conn_options) + 2;
675
676 if (cp->pe_data_len)
677 len += cp->pe_data_len + 2; /* + Param hdr field */
678 if (pe_name_len)
679 len += pe_name_len + 2;
680
681 /* check if there is a space for this one */
682 pad = 0;
683 buff = ms->sync_buff;
684 if (buff) {
685 m = buff->mesg;
686 pad = (4 - (size_t) buff->head) & 3;
687 /* Send buffer if it is for v0 */
688 if (buff->head + len + pad > buff->end || m->reserved) {
689 sb_queue_tail(ipvs, ms);
690 ms->sync_buff = NULL;
691 buff = NULL;
692 pad = 0;
693 }
694 }
695
696 if (!buff) {
697 buff = ip_vs_sync_buff_create(ipvs);
698 if (!buff) {
699 spin_unlock_bh(&ipvs->sync_buff_lock);
700 pr_err("ip_vs_sync_buff_create failed.\n");
701 return;
702 }
703 ms->sync_buff = buff;
704 m = buff->mesg;
705 }
706
707 p = buff->head;
708 buff->head += pad + len;
709 m->size = htons(ntohs(m->size) + pad + len);
710 /* Add ev. padding from prev. sync_conn */
711 while (pad--)
712 *(p++) = 0;
713
714 s = (union ip_vs_sync_conn *)p;
715
716 /* Set message type & copy members */
717 s->v4.type = (cp->af == AF_INET6 ? STYPE_F_INET6 : 0);
718 s->v4.ver_size = htons(len & SVER_MASK); /* Version 0 */
719 s->v4.flags = htonl(cp->flags & ~IP_VS_CONN_F_HASHED);
720 s->v4.state = htons(cp->state);
721 s->v4.protocol = cp->protocol;
722 s->v4.cport = cp->cport;
723 s->v4.vport = cp->vport;
724 s->v4.dport = cp->dport;
725 s->v4.fwmark = htonl(cp->fwmark);
726 s->v4.timeout = htonl(cp->timeout / HZ);
727 m->nr_conns++;
728
729 #ifdef CONFIG_IP_VS_IPV6
730 if (cp->af == AF_INET6) {
731 p += sizeof(struct ip_vs_sync_v6);
732 s->v6.caddr = cp->caddr.in6;
733 s->v6.vaddr = cp->vaddr.in6;
734 s->v6.daddr = cp->daddr.in6;
735 } else
736 #endif
737 {
738 p += sizeof(struct ip_vs_sync_v4); /* options ptr */
739 s->v4.caddr = cp->caddr.ip;
740 s->v4.vaddr = cp->vaddr.ip;
741 s->v4.daddr = cp->daddr.ip;
742 }
743 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
744 *(p++) = IPVS_OPT_SEQ_DATA;
745 *(p++) = sizeof(struct ip_vs_sync_conn_options);
746 hton_seq((struct ip_vs_seq *)p, &cp->in_seq);
747 p += sizeof(struct ip_vs_seq);
748 hton_seq((struct ip_vs_seq *)p, &cp->out_seq);
749 p += sizeof(struct ip_vs_seq);
750 }
751 /* Handle pe data */
752 if (cp->pe_data_len && cp->pe_data) {
753 *(p++) = IPVS_OPT_PE_DATA;
754 *(p++) = cp->pe_data_len;
755 memcpy(p, cp->pe_data, cp->pe_data_len);
756 p += cp->pe_data_len;
757 if (pe_name_len) {
758 /* Add PE_NAME */
759 *(p++) = IPVS_OPT_PE_NAME;
760 *(p++) = pe_name_len;
761 memcpy(p, cp->pe->name, pe_name_len);
762 p += pe_name_len;
763 }
764 }
765
766 spin_unlock_bh(&ipvs->sync_buff_lock);
767
768 control:
769 /* synchronize its controller if it has */
770 cp = cp->control;
771 if (!cp)
772 return;
773 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
774 pkts = atomic_add_return(1, &cp->in_pkts);
775 else
776 pkts = sysctl_sync_threshold(ipvs);
777 goto sloop;
778 }
779
780 /*
781 * fill_param used by version 1
782 */
783 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)784 ip_vs_conn_fill_param_sync(struct net *net, int af, union ip_vs_sync_conn *sc,
785 struct ip_vs_conn_param *p,
786 __u8 *pe_data, unsigned int pe_data_len,
787 __u8 *pe_name, unsigned int pe_name_len)
788 {
789 #ifdef CONFIG_IP_VS_IPV6
790 if (af == AF_INET6)
791 ip_vs_conn_fill_param(net, af, sc->v6.protocol,
792 (const union nf_inet_addr *)&sc->v6.caddr,
793 sc->v6.cport,
794 (const union nf_inet_addr *)&sc->v6.vaddr,
795 sc->v6.vport, p);
796 else
797 #endif
798 ip_vs_conn_fill_param(net, af, sc->v4.protocol,
799 (const union nf_inet_addr *)&sc->v4.caddr,
800 sc->v4.cport,
801 (const union nf_inet_addr *)&sc->v4.vaddr,
802 sc->v4.vport, p);
803 /* Handle pe data */
804 if (pe_data_len) {
805 if (pe_name_len) {
806 char buff[IP_VS_PENAME_MAXLEN+1];
807
808 memcpy(buff, pe_name, pe_name_len);
809 buff[pe_name_len]=0;
810 p->pe = __ip_vs_pe_getbyname(buff);
811 if (!p->pe) {
812 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n",
813 buff);
814 return 1;
815 }
816 } else {
817 IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n");
818 return 1;
819 }
820
821 p->pe_data = kmemdup(pe_data, pe_data_len, GFP_ATOMIC);
822 if (!p->pe_data) {
823 if (p->pe->module)
824 module_put(p->pe->module);
825 return -ENOMEM;
826 }
827 p->pe_data_len = pe_data_len;
828 }
829 return 0;
830 }
831
832 /*
833 * Connection Add / Update.
834 * Common for version 0 and 1 reception of backup sync_conns.
835 * Param: ...
836 * timeout is in sec.
837 */
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)838 static void ip_vs_proc_conn(struct net *net, struct ip_vs_conn_param *param,
839 unsigned int flags, unsigned int state,
840 unsigned int protocol, unsigned int type,
841 const union nf_inet_addr *daddr, __be16 dport,
842 unsigned long timeout, __u32 fwmark,
843 struct ip_vs_sync_conn_options *opt)
844 {
845 struct ip_vs_dest *dest;
846 struct ip_vs_conn *cp;
847 struct netns_ipvs *ipvs = net_ipvs(net);
848
849 if (!(flags & IP_VS_CONN_F_TEMPLATE))
850 cp = ip_vs_conn_in_get(param);
851 else
852 cp = ip_vs_ct_in_get(param);
853
854 if (cp) {
855 /* Free pe_data */
856 kfree(param->pe_data);
857
858 dest = cp->dest;
859 spin_lock_bh(&cp->lock);
860 if ((cp->flags ^ flags) & IP_VS_CONN_F_INACTIVE &&
861 !(flags & IP_VS_CONN_F_TEMPLATE) && dest) {
862 if (flags & IP_VS_CONN_F_INACTIVE) {
863 atomic_dec(&dest->activeconns);
864 atomic_inc(&dest->inactconns);
865 } else {
866 atomic_inc(&dest->activeconns);
867 atomic_dec(&dest->inactconns);
868 }
869 }
870 flags &= IP_VS_CONN_F_BACKUP_UPD_MASK;
871 flags |= cp->flags & ~IP_VS_CONN_F_BACKUP_UPD_MASK;
872 cp->flags = flags;
873 spin_unlock_bh(&cp->lock);
874 if (!dest)
875 ip_vs_try_bind_dest(cp);
876 } else {
877 /*
878 * Find the appropriate destination for the connection.
879 * If it is not found the connection will remain unbound
880 * but still handled.
881 */
882 rcu_read_lock();
883 /* This function is only invoked by the synchronization
884 * code. We do not currently support heterogeneous pools
885 * with synchronization, so we can make the assumption that
886 * the svc_af is the same as the dest_af
887 */
888 dest = ip_vs_find_dest(net, type, type, daddr, dport,
889 param->vaddr, param->vport, protocol,
890 fwmark, flags);
891
892 cp = ip_vs_conn_new(param, type, daddr, dport, flags, dest,
893 fwmark);
894 rcu_read_unlock();
895 if (!cp) {
896 kfree(param->pe_data);
897 IP_VS_DBG(2, "BACKUP, add new conn. failed\n");
898 return;
899 }
900 }
901
902 if (opt)
903 memcpy(&cp->in_seq, opt, sizeof(*opt));
904 atomic_set(&cp->in_pkts, sysctl_sync_threshold(ipvs));
905 cp->state = state;
906 cp->old_state = cp->state;
907 /*
908 * For Ver 0 messages style
909 * - Not possible to recover the right timeout for templates
910 * - can not find the right fwmark
911 * virtual service. If needed, we can do it for
912 * non-fwmark persistent services.
913 * Ver 1 messages style.
914 * - No problem.
915 */
916 if (timeout) {
917 if (timeout > MAX_SCHEDULE_TIMEOUT / HZ)
918 timeout = MAX_SCHEDULE_TIMEOUT / HZ;
919 cp->timeout = timeout*HZ;
920 } else {
921 struct ip_vs_proto_data *pd;
922
923 pd = ip_vs_proto_data_get(net, protocol);
924 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pd && pd->timeout_table)
925 cp->timeout = pd->timeout_table[state];
926 else
927 cp->timeout = (3*60*HZ);
928 }
929 ip_vs_conn_put(cp);
930 }
931
932 /*
933 * Process received multicast message for Version 0
934 */
ip_vs_process_message_v0(struct net * net,const char * buffer,const size_t buflen)935 static void ip_vs_process_message_v0(struct net *net, const char *buffer,
936 const size_t buflen)
937 {
938 struct ip_vs_sync_mesg_v0 *m = (struct ip_vs_sync_mesg_v0 *)buffer;
939 struct ip_vs_sync_conn_v0 *s;
940 struct ip_vs_sync_conn_options *opt;
941 struct ip_vs_protocol *pp;
942 struct ip_vs_conn_param param;
943 char *p;
944 int i;
945
946 p = (char *)buffer + sizeof(struct ip_vs_sync_mesg_v0);
947 for (i=0; i<m->nr_conns; i++) {
948 unsigned int flags, state;
949
950 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
951 IP_VS_ERR_RL("BACKUP v0, bogus conn\n");
952 return;
953 }
954 s = (struct ip_vs_sync_conn_v0 *) p;
955 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
956 flags &= ~IP_VS_CONN_F_HASHED;
957 if (flags & IP_VS_CONN_F_SEQ_MASK) {
958 opt = (struct ip_vs_sync_conn_options *)&s[1];
959 p += FULL_CONN_SIZE;
960 if (p > buffer+buflen) {
961 IP_VS_ERR_RL("BACKUP v0, Dropping buffer bogus conn options\n");
962 return;
963 }
964 } else {
965 opt = NULL;
966 p += SIMPLE_CONN_SIZE;
967 }
968
969 state = ntohs(s->state);
970 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
971 pp = ip_vs_proto_get(s->protocol);
972 if (!pp) {
973 IP_VS_DBG(2, "BACKUP v0, Unsupported protocol %u\n",
974 s->protocol);
975 continue;
976 }
977 if (state >= pp->num_states) {
978 IP_VS_DBG(2, "BACKUP v0, Invalid %s state %u\n",
979 pp->name, state);
980 continue;
981 }
982 } else {
983 /* protocol in templates is not used for state/timeout */
984 if (state > 0) {
985 IP_VS_DBG(2, "BACKUP v0, Invalid template state %u\n",
986 state);
987 state = 0;
988 }
989 }
990
991 ip_vs_conn_fill_param(net, AF_INET, s->protocol,
992 (const union nf_inet_addr *)&s->caddr,
993 s->cport,
994 (const union nf_inet_addr *)&s->vaddr,
995 s->vport, ¶m);
996
997 /* Send timeout as Zero */
998 ip_vs_proc_conn(net, ¶m, flags, state, s->protocol, AF_INET,
999 (union nf_inet_addr *)&s->daddr, s->dport,
1000 0, 0, opt);
1001 }
1002 }
1003
1004 /*
1005 * Handle options
1006 */
ip_vs_proc_seqopt(__u8 * p,unsigned int plen,__u32 * opt_flags,struct ip_vs_sync_conn_options * opt)1007 static inline int ip_vs_proc_seqopt(__u8 *p, unsigned int plen,
1008 __u32 *opt_flags,
1009 struct ip_vs_sync_conn_options *opt)
1010 {
1011 struct ip_vs_sync_conn_options *topt;
1012
1013 topt = (struct ip_vs_sync_conn_options *)p;
1014
1015 if (plen != sizeof(struct ip_vs_sync_conn_options)) {
1016 IP_VS_DBG(2, "BACKUP, bogus conn options length\n");
1017 return -EINVAL;
1018 }
1019 if (*opt_flags & IPVS_OPT_F_SEQ_DATA) {
1020 IP_VS_DBG(2, "BACKUP, conn options found twice\n");
1021 return -EINVAL;
1022 }
1023 ntoh_seq(&topt->in_seq, &opt->in_seq);
1024 ntoh_seq(&topt->out_seq, &opt->out_seq);
1025 *opt_flags |= IPVS_OPT_F_SEQ_DATA;
1026 return 0;
1027 }
1028
ip_vs_proc_str(__u8 * p,unsigned int plen,unsigned int * data_len,__u8 ** data,unsigned int maxlen,__u32 * opt_flags,__u32 flag)1029 static int ip_vs_proc_str(__u8 *p, unsigned int plen, unsigned int *data_len,
1030 __u8 **data, unsigned int maxlen,
1031 __u32 *opt_flags, __u32 flag)
1032 {
1033 if (plen > maxlen) {
1034 IP_VS_DBG(2, "BACKUP, bogus par.data len > %d\n", maxlen);
1035 return -EINVAL;
1036 }
1037 if (*opt_flags & flag) {
1038 IP_VS_DBG(2, "BACKUP, Par.data found twice 0x%x\n", flag);
1039 return -EINVAL;
1040 }
1041 *data_len = plen;
1042 *data = p;
1043 *opt_flags |= flag;
1044 return 0;
1045 }
1046 /*
1047 * Process a Version 1 sync. connection
1048 */
ip_vs_proc_sync_conn(struct net * net,__u8 * p,__u8 * msg_end)1049 static inline int ip_vs_proc_sync_conn(struct net *net, __u8 *p, __u8 *msg_end)
1050 {
1051 struct ip_vs_sync_conn_options opt;
1052 union ip_vs_sync_conn *s;
1053 struct ip_vs_protocol *pp;
1054 struct ip_vs_conn_param param;
1055 __u32 flags;
1056 unsigned int af, state, pe_data_len=0, pe_name_len=0;
1057 __u8 *pe_data=NULL, *pe_name=NULL;
1058 __u32 opt_flags=0;
1059 int retc=0;
1060
1061 s = (union ip_vs_sync_conn *) p;
1062
1063 if (s->v6.type & STYPE_F_INET6) {
1064 #ifdef CONFIG_IP_VS_IPV6
1065 af = AF_INET6;
1066 p += sizeof(struct ip_vs_sync_v6);
1067 #else
1068 IP_VS_DBG(3,"BACKUP, IPv6 msg received, and IPVS is not compiled for IPv6\n");
1069 retc = 10;
1070 goto out;
1071 #endif
1072 } else if (!s->v4.type) {
1073 af = AF_INET;
1074 p += sizeof(struct ip_vs_sync_v4);
1075 } else {
1076 return -10;
1077 }
1078 if (p > msg_end)
1079 return -20;
1080
1081 /* Process optional params check Type & Len. */
1082 while (p < msg_end) {
1083 int ptype;
1084 int plen;
1085
1086 if (p+2 > msg_end)
1087 return -30;
1088 ptype = *(p++);
1089 plen = *(p++);
1090
1091 if (!plen || ((p + plen) > msg_end))
1092 return -40;
1093 /* Handle seq option p = param data */
1094 switch (ptype & ~IPVS_OPT_F_PARAM) {
1095 case IPVS_OPT_SEQ_DATA:
1096 if (ip_vs_proc_seqopt(p, plen, &opt_flags, &opt))
1097 return -50;
1098 break;
1099
1100 case IPVS_OPT_PE_DATA:
1101 if (ip_vs_proc_str(p, plen, &pe_data_len, &pe_data,
1102 IP_VS_PEDATA_MAXLEN, &opt_flags,
1103 IPVS_OPT_F_PE_DATA))
1104 return -60;
1105 break;
1106
1107 case IPVS_OPT_PE_NAME:
1108 if (ip_vs_proc_str(p, plen,&pe_name_len, &pe_name,
1109 IP_VS_PENAME_MAXLEN, &opt_flags,
1110 IPVS_OPT_F_PE_NAME))
1111 return -70;
1112 break;
1113
1114 default:
1115 /* Param data mandatory ? */
1116 if (!(ptype & IPVS_OPT_F_PARAM)) {
1117 IP_VS_DBG(3, "BACKUP, Unknown mandatory param %d found\n",
1118 ptype & ~IPVS_OPT_F_PARAM);
1119 retc = 20;
1120 goto out;
1121 }
1122 }
1123 p += plen; /* Next option */
1124 }
1125
1126 /* Get flags and Mask off unsupported */
1127 flags = ntohl(s->v4.flags) & IP_VS_CONN_F_BACKUP_MASK;
1128 flags |= IP_VS_CONN_F_SYNC;
1129 state = ntohs(s->v4.state);
1130
1131 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
1132 pp = ip_vs_proto_get(s->v4.protocol);
1133 if (!pp) {
1134 IP_VS_DBG(3,"BACKUP, Unsupported protocol %u\n",
1135 s->v4.protocol);
1136 retc = 30;
1137 goto out;
1138 }
1139 if (state >= pp->num_states) {
1140 IP_VS_DBG(3, "BACKUP, Invalid %s state %u\n",
1141 pp->name, state);
1142 retc = 40;
1143 goto out;
1144 }
1145 } else {
1146 /* protocol in templates is not used for state/timeout */
1147 if (state > 0) {
1148 IP_VS_DBG(3, "BACKUP, Invalid template state %u\n",
1149 state);
1150 state = 0;
1151 }
1152 }
1153 if (ip_vs_conn_fill_param_sync(net, af, s, ¶m, pe_data,
1154 pe_data_len, pe_name, pe_name_len)) {
1155 retc = 50;
1156 goto out;
1157 }
1158 /* If only IPv4, just silent skip IPv6 */
1159 if (af == AF_INET)
1160 ip_vs_proc_conn(net, ¶m, flags, state, s->v4.protocol, af,
1161 (union nf_inet_addr *)&s->v4.daddr, s->v4.dport,
1162 ntohl(s->v4.timeout), ntohl(s->v4.fwmark),
1163 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1164 );
1165 #ifdef CONFIG_IP_VS_IPV6
1166 else
1167 ip_vs_proc_conn(net, ¶m, flags, state, s->v6.protocol, af,
1168 (union nf_inet_addr *)&s->v6.daddr, s->v6.dport,
1169 ntohl(s->v6.timeout), ntohl(s->v6.fwmark),
1170 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1171 );
1172 #endif
1173 return 0;
1174 /* Error exit */
1175 out:
1176 IP_VS_DBG(2, "BACKUP, Single msg dropped err:%d\n", retc);
1177 return retc;
1178
1179 }
1180 /*
1181 * Process received multicast message and create the corresponding
1182 * ip_vs_conn entries.
1183 * Handles Version 0 & 1
1184 */
ip_vs_process_message(struct net * net,__u8 * buffer,const size_t buflen)1185 static void ip_vs_process_message(struct net *net, __u8 *buffer,
1186 const size_t buflen)
1187 {
1188 struct netns_ipvs *ipvs = net_ipvs(net);
1189 struct ip_vs_sync_mesg *m2 = (struct ip_vs_sync_mesg *)buffer;
1190 __u8 *p, *msg_end;
1191 int i, nr_conns;
1192
1193 if (buflen < sizeof(struct ip_vs_sync_mesg_v0)) {
1194 IP_VS_DBG(2, "BACKUP, message header too short\n");
1195 return;
1196 }
1197
1198 if (buflen != ntohs(m2->size)) {
1199 IP_VS_DBG(2, "BACKUP, bogus message size\n");
1200 return;
1201 }
1202 /* SyncID sanity check */
1203 if (ipvs->backup_syncid != 0 && m2->syncid != ipvs->backup_syncid) {
1204 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2->syncid);
1205 return;
1206 }
1207 /* Handle version 1 message */
1208 if ((m2->version == SYNC_PROTO_VER) && (m2->reserved == 0)
1209 && (m2->spare == 0)) {
1210
1211 msg_end = buffer + sizeof(struct ip_vs_sync_mesg);
1212 nr_conns = m2->nr_conns;
1213
1214 for (i=0; i<nr_conns; i++) {
1215 union ip_vs_sync_conn *s;
1216 unsigned int size;
1217 int retc;
1218
1219 p = msg_end;
1220 if (p + sizeof(s->v4) > buffer+buflen) {
1221 IP_VS_ERR_RL("BACKUP, Dropping buffer, to small\n");
1222 return;
1223 }
1224 s = (union ip_vs_sync_conn *)p;
1225 size = ntohs(s->v4.ver_size) & SVER_MASK;
1226 msg_end = p + size;
1227 /* Basic sanity checks */
1228 if (msg_end > buffer+buflen) {
1229 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n");
1230 return;
1231 }
1232 if (ntohs(s->v4.ver_size) >> SVER_SHIFT) {
1233 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n",
1234 ntohs(s->v4.ver_size) >> SVER_SHIFT);
1235 return;
1236 }
1237 /* Process a single sync_conn */
1238 retc = ip_vs_proc_sync_conn(net, p, msg_end);
1239 if (retc < 0) {
1240 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n",
1241 retc);
1242 return;
1243 }
1244 /* Make sure we have 32 bit alignment */
1245 msg_end = p + ((size + 3) & ~3);
1246 }
1247 } else {
1248 /* Old type of message */
1249 ip_vs_process_message_v0(net, buffer, buflen);
1250 return;
1251 }
1252 }
1253
1254
1255 /*
1256 * Setup sndbuf (mode=1) or rcvbuf (mode=0)
1257 */
set_sock_size(struct sock * sk,int mode,int val)1258 static void set_sock_size(struct sock *sk, int mode, int val)
1259 {
1260 /* setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &val, sizeof(val)); */
1261 /* setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)); */
1262 lock_sock(sk);
1263 if (mode) {
1264 val = clamp_t(int, val, (SOCK_MIN_SNDBUF + 1) / 2,
1265 sysctl_wmem_max);
1266 sk->sk_sndbuf = val * 2;
1267 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1268 } else {
1269 val = clamp_t(int, val, (SOCK_MIN_RCVBUF + 1) / 2,
1270 sysctl_rmem_max);
1271 sk->sk_rcvbuf = val * 2;
1272 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1273 }
1274 release_sock(sk);
1275 }
1276
1277 /*
1278 * Setup loopback of outgoing multicasts on a sending socket
1279 */
set_mcast_loop(struct sock * sk,u_char loop)1280 static void set_mcast_loop(struct sock *sk, u_char loop)
1281 {
1282 struct inet_sock *inet = inet_sk(sk);
1283
1284 /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
1285 lock_sock(sk);
1286 inet->mc_loop = loop ? 1 : 0;
1287 release_sock(sk);
1288 }
1289
1290 /*
1291 * Specify TTL for outgoing multicasts on a sending socket
1292 */
set_mcast_ttl(struct sock * sk,u_char ttl)1293 static void set_mcast_ttl(struct sock *sk, u_char ttl)
1294 {
1295 struct inet_sock *inet = inet_sk(sk);
1296
1297 /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
1298 lock_sock(sk);
1299 inet->mc_ttl = ttl;
1300 release_sock(sk);
1301 }
1302
1303 /*
1304 * Specifiy default interface for outgoing multicasts
1305 */
set_mcast_if(struct sock * sk,char * ifname)1306 static int set_mcast_if(struct sock *sk, char *ifname)
1307 {
1308 struct net_device *dev;
1309 struct inet_sock *inet = inet_sk(sk);
1310 struct net *net = sock_net(sk);
1311
1312 dev = __dev_get_by_name(net, ifname);
1313 if (!dev)
1314 return -ENODEV;
1315
1316 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1317 return -EINVAL;
1318
1319 lock_sock(sk);
1320 inet->mc_index = dev->ifindex;
1321 /* inet->mc_addr = 0; */
1322 release_sock(sk);
1323
1324 return 0;
1325 }
1326
1327
1328 /*
1329 * Set the maximum length of sync message according to the
1330 * specified interface's MTU.
1331 */
set_sync_mesg_maxlen(struct net * net,int sync_state)1332 static int set_sync_mesg_maxlen(struct net *net, int sync_state)
1333 {
1334 struct netns_ipvs *ipvs = net_ipvs(net);
1335 struct net_device *dev;
1336 int num;
1337
1338 if (sync_state == IP_VS_STATE_MASTER) {
1339 dev = __dev_get_by_name(net, ipvs->master_mcast_ifn);
1340 if (!dev)
1341 return -ENODEV;
1342
1343 num = (dev->mtu - sizeof(struct iphdr) -
1344 sizeof(struct udphdr) -
1345 SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE;
1346 ipvs->send_mesg_maxlen = SYNC_MESG_HEADER_LEN +
1347 SIMPLE_CONN_SIZE * min(num, MAX_CONNS_PER_SYNCBUFF);
1348 IP_VS_DBG(7, "setting the maximum length of sync sending "
1349 "message %d.\n", ipvs->send_mesg_maxlen);
1350 } else if (sync_state == IP_VS_STATE_BACKUP) {
1351 dev = __dev_get_by_name(net, ipvs->backup_mcast_ifn);
1352 if (!dev)
1353 return -ENODEV;
1354
1355 ipvs->recv_mesg_maxlen = dev->mtu -
1356 sizeof(struct iphdr) - sizeof(struct udphdr);
1357 IP_VS_DBG(7, "setting the maximum length of sync receiving "
1358 "message %d.\n", ipvs->recv_mesg_maxlen);
1359 }
1360
1361 return 0;
1362 }
1363
1364
1365 /*
1366 * Join a multicast group.
1367 * the group is specified by a class D multicast address 224.0.0.0/8
1368 * in the in_addr structure passed in as a parameter.
1369 */
1370 static int
join_mcast_group(struct sock * sk,struct in_addr * addr,char * ifname)1371 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
1372 {
1373 struct net *net = sock_net(sk);
1374 struct ip_mreqn mreq;
1375 struct net_device *dev;
1376 int ret;
1377
1378 memset(&mreq, 0, sizeof(mreq));
1379 memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
1380
1381 dev = __dev_get_by_name(net, ifname);
1382 if (!dev)
1383 return -ENODEV;
1384 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1385 return -EINVAL;
1386
1387 mreq.imr_ifindex = dev->ifindex;
1388
1389 lock_sock(sk);
1390 ret = ip_mc_join_group(sk, &mreq);
1391 release_sock(sk);
1392
1393 return ret;
1394 }
1395
1396
bind_mcastif_addr(struct socket * sock,char * ifname)1397 static int bind_mcastif_addr(struct socket *sock, char *ifname)
1398 {
1399 struct net *net = sock_net(sock->sk);
1400 struct net_device *dev;
1401 __be32 addr;
1402 struct sockaddr_in sin;
1403
1404 dev = __dev_get_by_name(net, ifname);
1405 if (!dev)
1406 return -ENODEV;
1407
1408 addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
1409 if (!addr)
1410 pr_err("You probably need to specify IP address on "
1411 "multicast interface.\n");
1412
1413 IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
1414 ifname, &addr);
1415
1416 /* Now bind the socket with the address of multicast interface */
1417 sin.sin_family = AF_INET;
1418 sin.sin_addr.s_addr = addr;
1419 sin.sin_port = 0;
1420
1421 return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
1422 }
1423
1424 /*
1425 * Set up sending multicast socket over UDP
1426 */
make_send_sock(struct net * net,int id)1427 static struct socket *make_send_sock(struct net *net, int id)
1428 {
1429 struct netns_ipvs *ipvs = net_ipvs(net);
1430 /* multicast addr */
1431 struct sockaddr_in mcast_addr = {
1432 .sin_family = AF_INET,
1433 .sin_port = cpu_to_be16(IP_VS_SYNC_PORT + id),
1434 .sin_addr.s_addr = cpu_to_be32(IP_VS_SYNC_GROUP),
1435 };
1436 struct socket *sock;
1437 int result;
1438
1439 /* First create a socket move it to right name space later */
1440 result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1441 if (result < 0) {
1442 pr_err("Error during creation of socket; terminating\n");
1443 return ERR_PTR(result);
1444 }
1445 /*
1446 * Kernel sockets that are a part of a namespace, should not
1447 * hold a reference to a namespace in order to allow to stop it.
1448 * After sk_change_net should be released using sk_release_kernel.
1449 */
1450 sk_change_net(sock->sk, net);
1451 result = set_mcast_if(sock->sk, ipvs->master_mcast_ifn);
1452 if (result < 0) {
1453 pr_err("Error setting outbound mcast interface\n");
1454 goto error;
1455 }
1456
1457 set_mcast_loop(sock->sk, 0);
1458 set_mcast_ttl(sock->sk, 1);
1459 result = sysctl_sync_sock_size(ipvs);
1460 if (result > 0)
1461 set_sock_size(sock->sk, 1, result);
1462
1463 result = bind_mcastif_addr(sock, ipvs->master_mcast_ifn);
1464 if (result < 0) {
1465 pr_err("Error binding address of the mcast interface\n");
1466 goto error;
1467 }
1468
1469 result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr,
1470 sizeof(struct sockaddr), 0);
1471 if (result < 0) {
1472 pr_err("Error connecting to the multicast addr\n");
1473 goto error;
1474 }
1475
1476 return sock;
1477
1478 error:
1479 sk_release_kernel(sock->sk);
1480 return ERR_PTR(result);
1481 }
1482
1483
1484 /*
1485 * Set up receiving multicast socket over UDP
1486 */
make_receive_sock(struct net * net,int id)1487 static struct socket *make_receive_sock(struct net *net, int id)
1488 {
1489 struct netns_ipvs *ipvs = net_ipvs(net);
1490 /* multicast addr */
1491 struct sockaddr_in mcast_addr = {
1492 .sin_family = AF_INET,
1493 .sin_port = cpu_to_be16(IP_VS_SYNC_PORT + id),
1494 .sin_addr.s_addr = cpu_to_be32(IP_VS_SYNC_GROUP),
1495 };
1496 struct socket *sock;
1497 int result;
1498
1499 /* First create a socket */
1500 result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1501 if (result < 0) {
1502 pr_err("Error during creation of socket; terminating\n");
1503 return ERR_PTR(result);
1504 }
1505 /*
1506 * Kernel sockets that are a part of a namespace, should not
1507 * hold a reference to a namespace in order to allow to stop it.
1508 * After sk_change_net should be released using sk_release_kernel.
1509 */
1510 sk_change_net(sock->sk, net);
1511 /* it is equivalent to the REUSEADDR option in user-space */
1512 sock->sk->sk_reuse = SK_CAN_REUSE;
1513 result = sysctl_sync_sock_size(ipvs);
1514 if (result > 0)
1515 set_sock_size(sock->sk, 0, result);
1516
1517 result = sock->ops->bind(sock, (struct sockaddr *) &mcast_addr,
1518 sizeof(struct sockaddr));
1519 if (result < 0) {
1520 pr_err("Error binding to the multicast addr\n");
1521 goto error;
1522 }
1523
1524 /* join the multicast group */
1525 result = join_mcast_group(sock->sk,
1526 (struct in_addr *) &mcast_addr.sin_addr,
1527 ipvs->backup_mcast_ifn);
1528 if (result < 0) {
1529 pr_err("Error joining to the multicast group\n");
1530 goto error;
1531 }
1532
1533 return sock;
1534
1535 error:
1536 sk_release_kernel(sock->sk);
1537 return ERR_PTR(result);
1538 }
1539
1540
1541 static int
ip_vs_send_async(struct socket * sock,const char * buffer,const size_t length)1542 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
1543 {
1544 struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
1545 struct kvec iov;
1546 int len;
1547
1548 EnterFunction(7);
1549 iov.iov_base = (void *)buffer;
1550 iov.iov_len = length;
1551
1552 len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
1553
1554 LeaveFunction(7);
1555 return len;
1556 }
1557
1558 static int
ip_vs_send_sync_msg(struct socket * sock,struct ip_vs_sync_mesg * msg)1559 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
1560 {
1561 int msize;
1562 int ret;
1563
1564 msize = ntohs(msg->size);
1565
1566 ret = ip_vs_send_async(sock, (char *)msg, msize);
1567 if (ret >= 0 || ret == -EAGAIN)
1568 return ret;
1569 pr_err("ip_vs_send_async error %d\n", ret);
1570 return 0;
1571 }
1572
1573 static int
ip_vs_receive(struct socket * sock,char * buffer,const size_t buflen)1574 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
1575 {
1576 struct msghdr msg = {NULL,};
1577 struct kvec iov;
1578 int len;
1579
1580 EnterFunction(7);
1581
1582 /* Receive a packet */
1583 iov.iov_base = buffer;
1584 iov.iov_len = (size_t)buflen;
1585
1586 len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, MSG_DONTWAIT);
1587
1588 if (len < 0)
1589 return len;
1590
1591 LeaveFunction(7);
1592 return len;
1593 }
1594
1595 /* Wakeup the master thread for sending */
master_wakeup_work_handler(struct work_struct * work)1596 static void master_wakeup_work_handler(struct work_struct *work)
1597 {
1598 struct ipvs_master_sync_state *ms =
1599 container_of(work, struct ipvs_master_sync_state,
1600 master_wakeup_work.work);
1601 struct netns_ipvs *ipvs = ms->ipvs;
1602
1603 spin_lock_bh(&ipvs->sync_lock);
1604 if (ms->sync_queue_len &&
1605 ms->sync_queue_delay < IPVS_SYNC_WAKEUP_RATE) {
1606 ms->sync_queue_delay = IPVS_SYNC_WAKEUP_RATE;
1607 wake_up_process(ms->master_thread);
1608 }
1609 spin_unlock_bh(&ipvs->sync_lock);
1610 }
1611
1612 /* Get next buffer to send */
1613 static inline struct ip_vs_sync_buff *
next_sync_buff(struct netns_ipvs * ipvs,struct ipvs_master_sync_state * ms)1614 next_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
1615 {
1616 struct ip_vs_sync_buff *sb;
1617
1618 sb = sb_dequeue(ipvs, ms);
1619 if (sb)
1620 return sb;
1621 /* Do not delay entries in buffer for more than 2 seconds */
1622 return get_curr_sync_buff(ipvs, ms, IPVS_SYNC_FLUSH_TIME);
1623 }
1624
sync_thread_master(void * data)1625 static int sync_thread_master(void *data)
1626 {
1627 struct ip_vs_sync_thread_data *tinfo = data;
1628 struct netns_ipvs *ipvs = net_ipvs(tinfo->net);
1629 struct ipvs_master_sync_state *ms = &ipvs->ms[tinfo->id];
1630 struct sock *sk = tinfo->sock->sk;
1631 struct ip_vs_sync_buff *sb;
1632
1633 pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
1634 "syncid = %d, id = %d\n",
1635 ipvs->master_mcast_ifn, ipvs->master_syncid, tinfo->id);
1636
1637 for (;;) {
1638 sb = next_sync_buff(ipvs, ms);
1639 if (unlikely(kthread_should_stop()))
1640 break;
1641 if (!sb) {
1642 schedule_timeout(IPVS_SYNC_CHECK_PERIOD);
1643 continue;
1644 }
1645 while (ip_vs_send_sync_msg(tinfo->sock, sb->mesg) < 0) {
1646 /* (Ab)use interruptible sleep to avoid increasing
1647 * the load avg.
1648 */
1649 __wait_event_interruptible(*sk_sleep(sk),
1650 sock_writeable(sk) ||
1651 kthread_should_stop());
1652 if (unlikely(kthread_should_stop()))
1653 goto done;
1654 }
1655 ip_vs_sync_buff_release(sb);
1656 }
1657
1658 done:
1659 __set_current_state(TASK_RUNNING);
1660 if (sb)
1661 ip_vs_sync_buff_release(sb);
1662
1663 /* clean up the sync_buff queue */
1664 while ((sb = sb_dequeue(ipvs, ms)))
1665 ip_vs_sync_buff_release(sb);
1666 __set_current_state(TASK_RUNNING);
1667
1668 /* clean up the current sync_buff */
1669 sb = get_curr_sync_buff(ipvs, ms, 0);
1670 if (sb)
1671 ip_vs_sync_buff_release(sb);
1672
1673 /* release the sending multicast socket */
1674 sk_release_kernel(tinfo->sock->sk);
1675 kfree(tinfo);
1676
1677 return 0;
1678 }
1679
1680
sync_thread_backup(void * data)1681 static int sync_thread_backup(void *data)
1682 {
1683 struct ip_vs_sync_thread_data *tinfo = data;
1684 struct netns_ipvs *ipvs = net_ipvs(tinfo->net);
1685 int len;
1686
1687 pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
1688 "syncid = %d, id = %d\n",
1689 ipvs->backup_mcast_ifn, ipvs->backup_syncid, tinfo->id);
1690
1691 while (!kthread_should_stop()) {
1692 wait_event_interruptible(*sk_sleep(tinfo->sock->sk),
1693 !skb_queue_empty(&tinfo->sock->sk->sk_receive_queue)
1694 || kthread_should_stop());
1695
1696 /* do we have data now? */
1697 while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) {
1698 len = ip_vs_receive(tinfo->sock, tinfo->buf,
1699 ipvs->recv_mesg_maxlen);
1700 if (len <= 0) {
1701 if (len != -EAGAIN)
1702 pr_err("receiving message error\n");
1703 break;
1704 }
1705
1706 ip_vs_process_message(tinfo->net, tinfo->buf, len);
1707 }
1708 }
1709
1710 /* release the sending multicast socket */
1711 sk_release_kernel(tinfo->sock->sk);
1712 kfree(tinfo->buf);
1713 kfree(tinfo);
1714
1715 return 0;
1716 }
1717
1718
start_sync_thread(struct net * net,int state,char * mcast_ifn,__u8 syncid)1719 int start_sync_thread(struct net *net, int state, char *mcast_ifn, __u8 syncid)
1720 {
1721 struct ip_vs_sync_thread_data *tinfo;
1722 struct task_struct **array = NULL, *task;
1723 struct socket *sock;
1724 struct netns_ipvs *ipvs = net_ipvs(net);
1725 char *name;
1726 int (*threadfn)(void *data);
1727 int id, count;
1728 int result = -ENOMEM;
1729
1730 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1731 IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n",
1732 sizeof(struct ip_vs_sync_conn_v0));
1733
1734 if (!ipvs->sync_state) {
1735 count = clamp(sysctl_sync_ports(ipvs), 1, IPVS_SYNC_PORTS_MAX);
1736 ipvs->threads_mask = count - 1;
1737 } else
1738 count = ipvs->threads_mask + 1;
1739
1740 if (state == IP_VS_STATE_MASTER) {
1741 if (ipvs->ms)
1742 return -EEXIST;
1743
1744 strlcpy(ipvs->master_mcast_ifn, mcast_ifn,
1745 sizeof(ipvs->master_mcast_ifn));
1746 ipvs->master_syncid = syncid;
1747 name = "ipvs-m:%d:%d";
1748 threadfn = sync_thread_master;
1749 } else if (state == IP_VS_STATE_BACKUP) {
1750 if (ipvs->backup_threads)
1751 return -EEXIST;
1752
1753 strlcpy(ipvs->backup_mcast_ifn, mcast_ifn,
1754 sizeof(ipvs->backup_mcast_ifn));
1755 ipvs->backup_syncid = syncid;
1756 name = "ipvs-b:%d:%d";
1757 threadfn = sync_thread_backup;
1758 } else {
1759 return -EINVAL;
1760 }
1761
1762 if (state == IP_VS_STATE_MASTER) {
1763 struct ipvs_master_sync_state *ms;
1764
1765 ipvs->ms = kzalloc(count * sizeof(ipvs->ms[0]), GFP_KERNEL);
1766 if (!ipvs->ms)
1767 goto out;
1768 ms = ipvs->ms;
1769 for (id = 0; id < count; id++, ms++) {
1770 INIT_LIST_HEAD(&ms->sync_queue);
1771 ms->sync_queue_len = 0;
1772 ms->sync_queue_delay = 0;
1773 INIT_DELAYED_WORK(&ms->master_wakeup_work,
1774 master_wakeup_work_handler);
1775 ms->ipvs = ipvs;
1776 }
1777 } else {
1778 array = kzalloc(count * sizeof(struct task_struct *),
1779 GFP_KERNEL);
1780 if (!array)
1781 goto out;
1782 }
1783 set_sync_mesg_maxlen(net, state);
1784
1785 tinfo = NULL;
1786 for (id = 0; id < count; id++) {
1787 if (state == IP_VS_STATE_MASTER)
1788 sock = make_send_sock(net, id);
1789 else
1790 sock = make_receive_sock(net, id);
1791 if (IS_ERR(sock)) {
1792 result = PTR_ERR(sock);
1793 goto outtinfo;
1794 }
1795 tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL);
1796 if (!tinfo)
1797 goto outsocket;
1798 tinfo->net = net;
1799 tinfo->sock = sock;
1800 if (state == IP_VS_STATE_BACKUP) {
1801 tinfo->buf = kmalloc(ipvs->recv_mesg_maxlen,
1802 GFP_KERNEL);
1803 if (!tinfo->buf)
1804 goto outtinfo;
1805 } else {
1806 tinfo->buf = NULL;
1807 }
1808 tinfo->id = id;
1809
1810 task = kthread_run(threadfn, tinfo, name, ipvs->gen, id);
1811 if (IS_ERR(task)) {
1812 result = PTR_ERR(task);
1813 goto outtinfo;
1814 }
1815 tinfo = NULL;
1816 if (state == IP_VS_STATE_MASTER)
1817 ipvs->ms[id].master_thread = task;
1818 else
1819 array[id] = task;
1820 }
1821
1822 /* mark as active */
1823
1824 if (state == IP_VS_STATE_BACKUP)
1825 ipvs->backup_threads = array;
1826 spin_lock_bh(&ipvs->sync_buff_lock);
1827 ipvs->sync_state |= state;
1828 spin_unlock_bh(&ipvs->sync_buff_lock);
1829
1830 /* increase the module use count */
1831 ip_vs_use_count_inc();
1832
1833 return 0;
1834
1835 outsocket:
1836 sk_release_kernel(sock->sk);
1837
1838 outtinfo:
1839 if (tinfo) {
1840 sk_release_kernel(tinfo->sock->sk);
1841 kfree(tinfo->buf);
1842 kfree(tinfo);
1843 }
1844 count = id;
1845 while (count-- > 0) {
1846 if (state == IP_VS_STATE_MASTER)
1847 kthread_stop(ipvs->ms[count].master_thread);
1848 else
1849 kthread_stop(array[count]);
1850 }
1851 kfree(array);
1852
1853 out:
1854 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
1855 kfree(ipvs->ms);
1856 ipvs->ms = NULL;
1857 }
1858 return result;
1859 }
1860
1861
stop_sync_thread(struct net * net,int state)1862 int stop_sync_thread(struct net *net, int state)
1863 {
1864 struct netns_ipvs *ipvs = net_ipvs(net);
1865 struct task_struct **array;
1866 int id;
1867 int retc = -EINVAL;
1868
1869 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1870
1871 if (state == IP_VS_STATE_MASTER) {
1872 if (!ipvs->ms)
1873 return -ESRCH;
1874
1875 /*
1876 * The lock synchronizes with sb_queue_tail(), so that we don't
1877 * add sync buffers to the queue, when we are already in
1878 * progress of stopping the master sync daemon.
1879 */
1880
1881 spin_lock_bh(&ipvs->sync_buff_lock);
1882 spin_lock(&ipvs->sync_lock);
1883 ipvs->sync_state &= ~IP_VS_STATE_MASTER;
1884 spin_unlock(&ipvs->sync_lock);
1885 spin_unlock_bh(&ipvs->sync_buff_lock);
1886
1887 retc = 0;
1888 for (id = ipvs->threads_mask; id >= 0; id--) {
1889 struct ipvs_master_sync_state *ms = &ipvs->ms[id];
1890 int ret;
1891
1892 pr_info("stopping master sync thread %d ...\n",
1893 task_pid_nr(ms->master_thread));
1894 cancel_delayed_work_sync(&ms->master_wakeup_work);
1895 ret = kthread_stop(ms->master_thread);
1896 if (retc >= 0)
1897 retc = ret;
1898 }
1899 kfree(ipvs->ms);
1900 ipvs->ms = NULL;
1901 } else if (state == IP_VS_STATE_BACKUP) {
1902 if (!ipvs->backup_threads)
1903 return -ESRCH;
1904
1905 ipvs->sync_state &= ~IP_VS_STATE_BACKUP;
1906 array = ipvs->backup_threads;
1907 retc = 0;
1908 for (id = ipvs->threads_mask; id >= 0; id--) {
1909 int ret;
1910
1911 pr_info("stopping backup sync thread %d ...\n",
1912 task_pid_nr(array[id]));
1913 ret = kthread_stop(array[id]);
1914 if (retc >= 0)
1915 retc = ret;
1916 }
1917 kfree(array);
1918 ipvs->backup_threads = NULL;
1919 }
1920
1921 /* decrease the module use count */
1922 ip_vs_use_count_dec();
1923
1924 return retc;
1925 }
1926
1927 /*
1928 * Initialize data struct for each netns
1929 */
ip_vs_sync_net_init(struct net * net)1930 int __net_init ip_vs_sync_net_init(struct net *net)
1931 {
1932 struct netns_ipvs *ipvs = net_ipvs(net);
1933
1934 __mutex_init(&ipvs->sync_mutex, "ipvs->sync_mutex", &__ipvs_sync_key);
1935 spin_lock_init(&ipvs->sync_lock);
1936 spin_lock_init(&ipvs->sync_buff_lock);
1937 return 0;
1938 }
1939
ip_vs_sync_net_cleanup(struct net * net)1940 void ip_vs_sync_net_cleanup(struct net *net)
1941 {
1942 int retc;
1943 struct netns_ipvs *ipvs = net_ipvs(net);
1944
1945 mutex_lock(&ipvs->sync_mutex);
1946 retc = stop_sync_thread(net, IP_VS_STATE_MASTER);
1947 if (retc && retc != -ESRCH)
1948 pr_err("Failed to stop Master Daemon\n");
1949
1950 retc = stop_sync_thread(net, IP_VS_STATE_BACKUP);
1951 if (retc && retc != -ESRCH)
1952 pr_err("Failed to stop Backup Daemon\n");
1953 mutex_unlock(&ipvs->sync_mutex);
1954 }
1955