1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * IPVS An implementation of the IP virtual server support for the
4 * LINUX operating system. IPVS is now implemented as a module
5 * over the NetFilter framework. IPVS can be used to build a
6 * high-performance and highly available server based on a
7 * cluster of servers.
8 *
9 * Version 1, is capable of handling both version 0 and 1 messages.
10 * Version 0 is the plain old format.
11 * Note Version 0 receivers will just drop Ver 1 messages.
12 * Version 1 is capable of handle IPv6, Persistence data,
13 * time-outs, and firewall marks.
14 * In ver.1 "ip_vs_sync_conn_options" will be sent in netw. order.
15 * Ver. 0 can be turned on by sysctl -w net.ipv4.vs.sync_version=0
16 *
17 * Definitions Message: is a complete datagram
18 * Sync_conn: is a part of a Message
19 * Param Data is an option to a Sync_conn.
20 *
21 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
22 *
23 * ip_vs_sync: sync connection info from master load balancer to backups
24 * through multicast
25 *
26 * Changes:
27 * Alexandre Cassen : Added master & backup support at a time.
28 * Alexandre Cassen : Added SyncID support for incoming sync
29 * messages filtering.
30 * Justin Ossevoort : Fix endian problem on sync message size.
31 * Hans Schillstrom : Added Version 1: i.e. IPv6,
32 * Persistence support, fwmark and time-out.
33 */
34
35 #define pr_fmt(fmt) "IPVS: " 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 #include <linux/sched/signal.h>
52
53 #include <linux/unaligned.h> /* Used for ntoh_seq and hton_seq */
54
55 #include <net/ip.h>
56 #include <net/sock.h>
57
58 #include <net/ip_vs.h>
59
60 #define IP_VS_SYNC_GROUP 0xe0000051 /* multicast addr - 224.0.0.81 */
61 #define IP_VS_SYNC_PORT 8848 /* multicast port */
62
63 #define SYNC_PROTO_VER 1 /* Protocol version in header */
64
65 static struct lock_class_key __ipvs_sync_key;
66 /*
67 * IPVS sync connection entry
68 * Version 0, i.e. original version.
69 */
70 struct ip_vs_sync_conn_v0 {
71 __u8 reserved;
72
73 /* Protocol, addresses and port numbers */
74 __u8 protocol; /* Which protocol (TCP/UDP) */
75 __be16 cport;
76 __be16 vport;
77 __be16 dport;
78 __be32 caddr; /* client address */
79 __be32 vaddr; /* virtual address */
80 __be32 daddr; /* destination address */
81
82 /* Flags and state transition */
83 __be16 flags; /* status flags */
84 __be16 state; /* state info */
85
86 /* The sequence options start here */
87 };
88
89 struct ip_vs_sync_conn_options {
90 struct ip_vs_seq in_seq; /* incoming seq. struct */
91 struct ip_vs_seq out_seq; /* outgoing seq. struct */
92 };
93
94 /*
95 Sync Connection format (sync_conn)
96
97 0 1 2 3
98 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
99 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
100 | Type | Protocol | Ver. | Size |
101 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
102 | Flags |
103 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
104 | State | cport |
105 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
106 | vport | dport |
107 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
108 | fwmark |
109 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
110 | timeout (in sec.) |
111 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
112 | ... |
113 | IP-Addresses (v4 or v6) |
114 | ... |
115 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
116 Optional Parameters.
117 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
118 | Param. Type | Param. Length | Param. data |
119 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
120 | ... |
121 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
122 | | Param Type | Param. Length |
123 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
124 | Param data |
125 | Last Param data should be padded for 32 bit alignment |
126 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
127 */
128
129 /*
130 * Type 0, IPv4 sync connection format
131 */
132 struct ip_vs_sync_v4 {
133 __u8 type;
134 __u8 protocol; /* Which protocol (TCP/UDP) */
135 __be16 ver_size; /* Version msb 4 bits */
136 /* Flags and state transition */
137 __be32 flags; /* status flags */
138 __be16 state; /* state info */
139 /* Protocol, addresses and port numbers */
140 __be16 cport;
141 __be16 vport;
142 __be16 dport;
143 __be32 fwmark; /* Firewall mark from skb */
144 __be32 timeout; /* cp timeout */
145 __be32 caddr; /* client address */
146 __be32 vaddr; /* virtual address */
147 __be32 daddr; /* destination address */
148 /* The sequence options start here */
149 /* PE data padded to 32bit alignment after seq. options */
150 };
151 /*
152 * Type 2 messages IPv6
153 */
154 struct ip_vs_sync_v6 {
155 __u8 type;
156 __u8 protocol; /* Which protocol (TCP/UDP) */
157 __be16 ver_size; /* Version msb 4 bits */
158 /* Flags and state transition */
159 __be32 flags; /* status flags */
160 __be16 state; /* state info */
161 /* Protocol, addresses and port numbers */
162 __be16 cport;
163 __be16 vport;
164 __be16 dport;
165 __be32 fwmark; /* Firewall mark from skb */
166 __be32 timeout; /* cp timeout */
167 struct in6_addr caddr; /* client address */
168 struct in6_addr vaddr; /* virtual address */
169 struct in6_addr daddr; /* destination address */
170 /* The sequence options start here */
171 /* PE data padded to 32bit alignment after seq. options */
172 };
173
174 union ip_vs_sync_conn {
175 struct ip_vs_sync_v4 v4;
176 struct ip_vs_sync_v6 v6;
177 };
178
179 /* Bits in Type field in above */
180 #define STYPE_INET6 0
181 #define STYPE_F_INET6 (1 << STYPE_INET6)
182
183 #define SVER_SHIFT 12 /* Shift to get version */
184 #define SVER_MASK 0x0fff /* Mask to strip version */
185
186 #define IPVS_OPT_SEQ_DATA 1
187 #define IPVS_OPT_PE_DATA 2
188 #define IPVS_OPT_PE_NAME 3
189 #define IPVS_OPT_PARAM 7
190
191 #define IPVS_OPT_F_SEQ_DATA (1 << (IPVS_OPT_SEQ_DATA-1))
192 #define IPVS_OPT_F_PE_DATA (1 << (IPVS_OPT_PE_DATA-1))
193 #define IPVS_OPT_F_PE_NAME (1 << (IPVS_OPT_PE_NAME-1))
194 #define IPVS_OPT_F_PARAM (1 << (IPVS_OPT_PARAM-1))
195
196 struct ip_vs_sync_thread_data {
197 struct task_struct *task;
198 struct netns_ipvs *ipvs;
199 struct socket *sock;
200 char *buf;
201 int id;
202 };
203
204 /* Version 0 definition of packet sizes */
205 #define SIMPLE_CONN_SIZE (sizeof(struct ip_vs_sync_conn_v0))
206 #define FULL_CONN_SIZE \
207 (sizeof(struct ip_vs_sync_conn_v0) + sizeof(struct ip_vs_sync_conn_options))
208
209
210 /*
211 The master mulitcasts messages (Datagrams) to the backup load balancers
212 in the following format.
213
214 Version 1:
215 Note, first byte should be Zero, so ver 0 receivers will drop the packet.
216
217 0 1 2 3
218 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
219 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
220 | 0 | SyncID | Size |
221 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
222 | Count Conns | Version | Reserved, set to Zero |
223 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
224 | |
225 | IPVS Sync Connection (1) |
226 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
227 | . |
228 ~ . ~
229 | . |
230 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
231 | |
232 | IPVS Sync Connection (n) |
233 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
234
235 Version 0 Header
236 0 1 2 3
237 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
238 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
239 | Count Conns | SyncID | Size |
240 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
241 | IPVS Sync Connection (1) |
242 */
243
244 /* Version 0 header */
245 struct ip_vs_sync_mesg_v0 {
246 __u8 nr_conns;
247 __u8 syncid;
248 __be16 size;
249
250 /* ip_vs_sync_conn entries start here */
251 };
252
253 /* Version 1 header */
254 struct ip_vs_sync_mesg {
255 __u8 reserved; /* must be zero */
256 __u8 syncid;
257 __be16 size;
258 __u8 nr_conns;
259 __s8 version; /* SYNC_PROTO_VER */
260 __u16 spare;
261 /* ip_vs_sync_conn entries start here */
262 };
263
264 union ipvs_sockaddr {
265 struct sockaddr_in in;
266 struct sockaddr_in6 in6;
267 };
268
269 struct ip_vs_sync_buff {
270 struct list_head list;
271 unsigned long firstuse;
272
273 /* pointers for the message data */
274 struct ip_vs_sync_mesg *mesg;
275 unsigned char *head;
276 unsigned char *end;
277 };
278
279 /*
280 * Copy of struct ip_vs_seq
281 * From unaligned network order to aligned host order
282 */
ntoh_seq(struct ip_vs_seq * no,struct ip_vs_seq * ho)283 static void ntoh_seq(struct ip_vs_seq *no, struct ip_vs_seq *ho)
284 {
285 memset(ho, 0, sizeof(*ho));
286 ho->init_seq = get_unaligned_be32(&no->init_seq);
287 ho->delta = get_unaligned_be32(&no->delta);
288 ho->previous_delta = get_unaligned_be32(&no->previous_delta);
289 }
290
291 /*
292 * Copy of struct ip_vs_seq
293 * From Aligned host order to unaligned network order
294 */
hton_seq(struct ip_vs_seq * ho,struct ip_vs_seq * no)295 static void hton_seq(struct ip_vs_seq *ho, struct ip_vs_seq *no)
296 {
297 put_unaligned_be32(ho->init_seq, &no->init_seq);
298 put_unaligned_be32(ho->delta, &no->delta);
299 put_unaligned_be32(ho->previous_delta, &no->previous_delta);
300 }
301
302 static inline struct ip_vs_sync_buff *
sb_dequeue(struct netns_ipvs * ipvs,struct ipvs_master_sync_state * ms)303 sb_dequeue(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
304 {
305 struct ip_vs_sync_buff *sb;
306
307 spin_lock_bh(&ipvs->sync_lock);
308 if (list_empty(&ms->sync_queue)) {
309 sb = NULL;
310 __set_current_state(TASK_INTERRUPTIBLE);
311 } else {
312 sb = list_entry(ms->sync_queue.next, struct ip_vs_sync_buff,
313 list);
314 list_del(&sb->list);
315 ms->sync_queue_len--;
316 if (!ms->sync_queue_len)
317 ms->sync_queue_delay = 0;
318 }
319 spin_unlock_bh(&ipvs->sync_lock);
320
321 return sb;
322 }
323
324 /*
325 * Create a new sync buffer for Version 1 proto.
326 */
327 static inline struct ip_vs_sync_buff *
ip_vs_sync_buff_create(struct netns_ipvs * ipvs,unsigned int len)328 ip_vs_sync_buff_create(struct netns_ipvs *ipvs, unsigned int len)
329 {
330 struct ip_vs_sync_buff *sb;
331
332 if (!(sb=kmalloc_obj(struct ip_vs_sync_buff, GFP_ATOMIC)))
333 return NULL;
334
335 len = max_t(unsigned int, len + sizeof(struct ip_vs_sync_mesg),
336 ipvs->mcfg.sync_maxlen);
337 sb->mesg = kmalloc(len, GFP_ATOMIC);
338 if (!sb->mesg) {
339 kfree(sb);
340 return NULL;
341 }
342 sb->mesg->reserved = 0; /* old nr_conns i.e. must be zero now */
343 sb->mesg->version = SYNC_PROTO_VER;
344 sb->mesg->syncid = ipvs->mcfg.syncid;
345 sb->mesg->size = htons(sizeof(struct ip_vs_sync_mesg));
346 sb->mesg->nr_conns = 0;
347 sb->mesg->spare = 0;
348 sb->head = (unsigned char *)sb->mesg + sizeof(struct ip_vs_sync_mesg);
349 sb->end = (unsigned char *)sb->mesg + len;
350
351 sb->firstuse = jiffies;
352 return sb;
353 }
354
ip_vs_sync_buff_release(struct ip_vs_sync_buff * sb)355 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
356 {
357 kfree(sb->mesg);
358 kfree(sb);
359 }
360
sb_queue_tail(struct netns_ipvs * ipvs,struct ipvs_master_sync_state * ms)361 static inline void sb_queue_tail(struct netns_ipvs *ipvs,
362 struct ipvs_master_sync_state *ms)
363 {
364 struct ip_vs_sync_buff *sb = ms->sync_buff;
365
366 spin_lock(&ipvs->sync_lock);
367 if (ipvs->sync_state & IP_VS_STATE_MASTER &&
368 ms->sync_queue_len < sysctl_sync_qlen_max(ipvs)) {
369 if (!ms->sync_queue_len)
370 schedule_delayed_work(&ms->master_wakeup_work,
371 max(IPVS_SYNC_SEND_DELAY, 1));
372 ms->sync_queue_len++;
373 list_add_tail(&sb->list, &ms->sync_queue);
374 if ((++ms->sync_queue_delay) == IPVS_SYNC_WAKEUP_RATE) {
375 int id = (int)(ms - ipvs->ms);
376
377 wake_up_process(ipvs->master_tinfo[id].task);
378 }
379 } else
380 ip_vs_sync_buff_release(sb);
381 spin_unlock(&ipvs->sync_lock);
382 }
383
384 /*
385 * Get the current sync buffer if it has been created for more
386 * than the specified time or the specified time is zero.
387 */
388 static inline struct ip_vs_sync_buff *
get_curr_sync_buff(struct netns_ipvs * ipvs,struct ipvs_master_sync_state * ms,unsigned long time)389 get_curr_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms,
390 unsigned long time)
391 {
392 struct ip_vs_sync_buff *sb;
393
394 spin_lock_bh(&ipvs->sync_buff_lock);
395 sb = ms->sync_buff;
396 if (sb && time_after_eq(jiffies - sb->firstuse, time)) {
397 ms->sync_buff = NULL;
398 __set_current_state(TASK_RUNNING);
399 } else
400 sb = NULL;
401 spin_unlock_bh(&ipvs->sync_buff_lock);
402 return sb;
403 }
404
405 static inline int
select_master_thread_id(struct netns_ipvs * ipvs,struct ip_vs_conn * cp)406 select_master_thread_id(struct netns_ipvs *ipvs, struct ip_vs_conn *cp)
407 {
408 return ((long) cp >> (1 + ilog2(sizeof(*cp)))) & ipvs->threads_mask;
409 }
410
411 /*
412 * Create a new sync buffer for Version 0 proto.
413 */
414 static inline struct ip_vs_sync_buff *
ip_vs_sync_buff_create_v0(struct netns_ipvs * ipvs,unsigned int len)415 ip_vs_sync_buff_create_v0(struct netns_ipvs *ipvs, unsigned int len)
416 {
417 struct ip_vs_sync_buff *sb;
418 struct ip_vs_sync_mesg_v0 *mesg;
419
420 if (!(sb=kmalloc_obj(struct ip_vs_sync_buff, GFP_ATOMIC)))
421 return NULL;
422
423 len = max_t(unsigned int, len + sizeof(struct ip_vs_sync_mesg_v0),
424 ipvs->mcfg.sync_maxlen);
425 sb->mesg = kmalloc(len, GFP_ATOMIC);
426 if (!sb->mesg) {
427 kfree(sb);
428 return NULL;
429 }
430 mesg = (struct ip_vs_sync_mesg_v0 *)sb->mesg;
431 mesg->nr_conns = 0;
432 mesg->syncid = ipvs->mcfg.syncid;
433 mesg->size = htons(sizeof(struct ip_vs_sync_mesg_v0));
434 sb->head = (unsigned char *)mesg + sizeof(struct ip_vs_sync_mesg_v0);
435 sb->end = (unsigned char *)mesg + len;
436 sb->firstuse = jiffies;
437 return sb;
438 }
439
440 /* Check if connection is controlled by persistence */
in_persistence(struct ip_vs_conn * cp)441 static inline bool in_persistence(struct ip_vs_conn *cp)
442 {
443 for (cp = cp->control; cp; cp = cp->control) {
444 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
445 return true;
446 }
447 return false;
448 }
449
450 /* Check if conn should be synced.
451 * pkts: conn packets, use sysctl_sync_threshold to avoid packet check
452 * - (1) sync_refresh_period: reduce sync rate. Additionally, retry
453 * sync_retries times with period of sync_refresh_period/8
454 * - (2) if both sync_refresh_period and sync_period are 0 send sync only
455 * for state changes or only once when pkts matches sync_threshold
456 * - (3) templates: rate can be reduced only with sync_refresh_period or
457 * with (2)
458 */
ip_vs_sync_conn_needed(struct netns_ipvs * ipvs,struct ip_vs_conn * cp,int pkts)459 static int ip_vs_sync_conn_needed(struct netns_ipvs *ipvs,
460 struct ip_vs_conn *cp, int pkts)
461 {
462 unsigned long orig = READ_ONCE(cp->sync_endtime);
463 unsigned long now = jiffies;
464 unsigned long n = (now + cp->timeout) & ~3UL;
465 unsigned int sync_refresh_period;
466 int sync_period;
467 int force;
468
469 /* Check if we sync in current state */
470 if (unlikely(cp->flags & IP_VS_CONN_F_TEMPLATE))
471 force = 0;
472 else if (unlikely(sysctl_sync_persist_mode(ipvs) && in_persistence(cp)))
473 return 0;
474 else if (likely(cp->protocol == IPPROTO_TCP)) {
475 if (!((1 << cp->state) &
476 ((1 << IP_VS_TCP_S_ESTABLISHED) |
477 (1 << IP_VS_TCP_S_FIN_WAIT) |
478 (1 << IP_VS_TCP_S_CLOSE) |
479 (1 << IP_VS_TCP_S_CLOSE_WAIT) |
480 (1 << IP_VS_TCP_S_TIME_WAIT))))
481 return 0;
482 force = cp->state != cp->old_state;
483 if (force && cp->state != IP_VS_TCP_S_ESTABLISHED)
484 goto set;
485 } else if (unlikely(cp->protocol == IPPROTO_SCTP)) {
486 if (!((1 << cp->state) &
487 ((1 << IP_VS_SCTP_S_ESTABLISHED) |
488 (1 << IP_VS_SCTP_S_SHUTDOWN_SENT) |
489 (1 << IP_VS_SCTP_S_SHUTDOWN_RECEIVED) |
490 (1 << IP_VS_SCTP_S_SHUTDOWN_ACK_SENT) |
491 (1 << IP_VS_SCTP_S_CLOSED))))
492 return 0;
493 force = cp->state != cp->old_state;
494 if (force && cp->state != IP_VS_SCTP_S_ESTABLISHED)
495 goto set;
496 } else {
497 /* UDP or another protocol with single state */
498 force = 0;
499 }
500
501 sync_refresh_period = sysctl_sync_refresh_period(ipvs);
502 if (sync_refresh_period > 0) {
503 long diff = n - orig;
504 long min_diff = max(cp->timeout >> 1, 10UL * HZ);
505
506 /* Avoid sync if difference is below sync_refresh_period
507 * and below the half timeout.
508 */
509 if (abs(diff) < min_t(long, sync_refresh_period, min_diff)) {
510 int retries = orig & 3;
511
512 if (retries >= sysctl_sync_retries(ipvs))
513 return 0;
514 if (time_before(now, orig - cp->timeout +
515 (sync_refresh_period >> 3)))
516 return 0;
517 n |= retries + 1;
518 }
519 }
520 sync_period = sysctl_sync_period(ipvs);
521 if (sync_period > 0) {
522 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE) &&
523 pkts % sync_period != sysctl_sync_threshold(ipvs))
524 return 0;
525 } else if (!sync_refresh_period &&
526 pkts != sysctl_sync_threshold(ipvs))
527 return 0;
528
529 set:
530 cp->old_state = cp->state;
531 n = cmpxchg(&cp->sync_endtime, orig, n);
532 return n == orig || force;
533 }
534
535 /*
536 * Version 0 , could be switched in by sys_ctl.
537 * Add an ip_vs_conn information into the current sync_buff.
538 */
ip_vs_sync_conn_v0(struct netns_ipvs * ipvs,struct ip_vs_conn * cp,int pkts)539 static void ip_vs_sync_conn_v0(struct netns_ipvs *ipvs, struct ip_vs_conn *cp,
540 int pkts)
541 {
542 struct ip_vs_sync_mesg_v0 *m;
543 struct ip_vs_sync_conn_v0 *s;
544 struct ip_vs_sync_buff *buff;
545 struct ipvs_master_sync_state *ms;
546 int id;
547 unsigned int len;
548
549 if (unlikely(cp->af != AF_INET))
550 return;
551 /* Do not sync ONE PACKET */
552 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
553 return;
554
555 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
556 return;
557
558 spin_lock_bh(&ipvs->sync_buff_lock);
559 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
560 spin_unlock_bh(&ipvs->sync_buff_lock);
561 return;
562 }
563
564 id = select_master_thread_id(ipvs, cp);
565 ms = &ipvs->ms[id];
566 buff = ms->sync_buff;
567 len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
568 SIMPLE_CONN_SIZE;
569 if (buff) {
570 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
571 /* Send buffer if it is for v1 */
572 if (buff->head + len > buff->end || !m->nr_conns) {
573 sb_queue_tail(ipvs, ms);
574 ms->sync_buff = NULL;
575 buff = NULL;
576 }
577 }
578 if (!buff) {
579 buff = ip_vs_sync_buff_create_v0(ipvs, len);
580 if (!buff) {
581 spin_unlock_bh(&ipvs->sync_buff_lock);
582 pr_err("ip_vs_sync_buff_create failed.\n");
583 return;
584 }
585 ms->sync_buff = buff;
586 }
587
588 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
589 s = (struct ip_vs_sync_conn_v0 *) buff->head;
590
591 /* copy members */
592 s->reserved = 0;
593 s->protocol = cp->protocol;
594 s->cport = cp->cport;
595 s->vport = cp->vport;
596 s->dport = cp->dport;
597 s->caddr = cp->caddr.ip;
598 s->vaddr = cp->vaddr.ip;
599 s->daddr = cp->daddr.ip;
600 s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
601 s->state = htons(cp->state);
602 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
603 struct ip_vs_sync_conn_options *opt =
604 (struct ip_vs_sync_conn_options *)&s[1];
605 memcpy(opt, &cp->sync_conn_opt, sizeof(*opt));
606 }
607
608 m->nr_conns++;
609 m->size = htons(ntohs(m->size) + len);
610 buff->head += len;
611 spin_unlock_bh(&ipvs->sync_buff_lock);
612
613 /* synchronize its controller if it has */
614 cp = cp->control;
615 if (cp) {
616 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
617 pkts = atomic_inc_return(&cp->in_pkts);
618 else
619 pkts = sysctl_sync_threshold(ipvs);
620 ip_vs_sync_conn(ipvs, cp, pkts);
621 }
622 }
623
624 /*
625 * Add an ip_vs_conn information into the current sync_buff.
626 * Called by ip_vs_in.
627 * Sending Version 1 messages
628 */
ip_vs_sync_conn(struct netns_ipvs * ipvs,struct ip_vs_conn * cp,int pkts)629 void ip_vs_sync_conn(struct netns_ipvs *ipvs, struct ip_vs_conn *cp, int pkts)
630 {
631 struct ip_vs_sync_mesg *m;
632 union ip_vs_sync_conn *s;
633 struct ip_vs_sync_buff *buff;
634 struct ipvs_master_sync_state *ms;
635 int id;
636 __u8 *p;
637 unsigned int len, pe_name_len, pad;
638
639 /* Handle old version of the protocol */
640 if (sysctl_sync_ver(ipvs) == 0) {
641 ip_vs_sync_conn_v0(ipvs, cp, pkts);
642 return;
643 }
644 /* Do not sync ONE PACKET */
645 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
646 goto control;
647 sloop:
648 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
649 goto control;
650
651 /* Sanity checks */
652 pe_name_len = 0;
653 if (cp->pe_data_len) {
654 if (!cp->pe_data || !cp->dest) {
655 IP_VS_ERR_RL("SYNC, connection pe_data invalid\n");
656 return;
657 }
658 pe_name_len = strnlen(cp->pe->name, IP_VS_PENAME_MAXLEN);
659 }
660
661 spin_lock_bh(&ipvs->sync_buff_lock);
662 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
663 spin_unlock_bh(&ipvs->sync_buff_lock);
664 return;
665 }
666
667 id = select_master_thread_id(ipvs, cp);
668 ms = &ipvs->ms[id];
669
670 #ifdef CONFIG_IP_VS_IPV6
671 if (cp->af == AF_INET6)
672 len = sizeof(struct ip_vs_sync_v6);
673 else
674 #endif
675 len = sizeof(struct ip_vs_sync_v4);
676
677 if (cp->flags & IP_VS_CONN_F_SEQ_MASK)
678 len += sizeof(struct ip_vs_sync_conn_options) + 2;
679
680 if (cp->pe_data_len)
681 len += cp->pe_data_len + 2; /* + Param hdr field */
682 if (pe_name_len)
683 len += pe_name_len + 2;
684
685 /* check if there is a space for this one */
686 pad = 0;
687 buff = ms->sync_buff;
688 if (buff) {
689 m = buff->mesg;
690 pad = (4 - (size_t) buff->head) & 3;
691 /* Send buffer if it is for v0 */
692 if (buff->head + len + pad > buff->end || m->reserved) {
693 sb_queue_tail(ipvs, ms);
694 ms->sync_buff = NULL;
695 buff = NULL;
696 pad = 0;
697 }
698 }
699
700 if (!buff) {
701 buff = ip_vs_sync_buff_create(ipvs, len);
702 if (!buff) {
703 spin_unlock_bh(&ipvs->sync_buff_lock);
704 pr_err("ip_vs_sync_buff_create failed.\n");
705 return;
706 }
707 ms->sync_buff = buff;
708 m = buff->mesg;
709 }
710
711 p = buff->head;
712 buff->head += pad + len;
713 m->size = htons(ntohs(m->size) + pad + len);
714 /* Add ev. padding from prev. sync_conn */
715 while (pad--)
716 *(p++) = 0;
717
718 s = (union ip_vs_sync_conn *)p;
719
720 /* Set message type & copy members */
721 s->v4.type = (cp->af == AF_INET6 ? STYPE_F_INET6 : 0);
722 s->v4.ver_size = htons(len & SVER_MASK); /* Version 0 */
723 s->v4.flags = htonl(cp->flags & ~IP_VS_CONN_F_HASHED);
724 s->v4.state = htons(cp->state);
725 s->v4.protocol = cp->protocol;
726 s->v4.cport = cp->cport;
727 s->v4.vport = cp->vport;
728 s->v4.dport = cp->dport;
729 s->v4.fwmark = htonl(cp->fwmark);
730 s->v4.timeout = htonl(cp->timeout / HZ);
731 m->nr_conns++;
732
733 #ifdef CONFIG_IP_VS_IPV6
734 if (cp->af == AF_INET6) {
735 p += sizeof(struct ip_vs_sync_v6);
736 s->v6.caddr = cp->caddr.in6;
737 s->v6.vaddr = cp->vaddr.in6;
738 s->v6.daddr = cp->daddr.in6;
739 } else
740 #endif
741 {
742 p += sizeof(struct ip_vs_sync_v4); /* options ptr */
743 s->v4.caddr = cp->caddr.ip;
744 s->v4.vaddr = cp->vaddr.ip;
745 s->v4.daddr = cp->daddr.ip;
746 }
747 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
748 *(p++) = IPVS_OPT_SEQ_DATA;
749 *(p++) = sizeof(struct ip_vs_sync_conn_options);
750 hton_seq((struct ip_vs_seq *)p, &cp->in_seq);
751 p += sizeof(struct ip_vs_seq);
752 hton_seq((struct ip_vs_seq *)p, &cp->out_seq);
753 p += sizeof(struct ip_vs_seq);
754 }
755 /* Handle pe data */
756 if (cp->pe_data_len && cp->pe_data) {
757 *(p++) = IPVS_OPT_PE_DATA;
758 *(p++) = cp->pe_data_len;
759 memcpy(p, cp->pe_data, cp->pe_data_len);
760 p += cp->pe_data_len;
761 if (pe_name_len) {
762 /* Add PE_NAME */
763 *(p++) = IPVS_OPT_PE_NAME;
764 *(p++) = pe_name_len;
765 memcpy(p, cp->pe->name, pe_name_len);
766 p += pe_name_len;
767 }
768 }
769
770 spin_unlock_bh(&ipvs->sync_buff_lock);
771
772 control:
773 /* synchronize its controller if it has */
774 cp = cp->control;
775 if (!cp)
776 return;
777 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
778 pkts = atomic_inc_return(&cp->in_pkts);
779 else
780 pkts = sysctl_sync_threshold(ipvs);
781 goto sloop;
782 }
783
784 /*
785 * fill_param used by version 1
786 */
787 static inline int
ip_vs_conn_fill_param_sync(struct netns_ipvs * ipvs,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)788 ip_vs_conn_fill_param_sync(struct netns_ipvs *ipvs, int af, union ip_vs_sync_conn *sc,
789 struct ip_vs_conn_param *p,
790 __u8 *pe_data, unsigned int pe_data_len,
791 __u8 *pe_name, unsigned int pe_name_len)
792 {
793 #ifdef CONFIG_IP_VS_IPV6
794 if (af == AF_INET6)
795 ip_vs_conn_fill_param(ipvs, af, sc->v6.protocol,
796 (const union nf_inet_addr *)&sc->v6.caddr,
797 sc->v6.cport,
798 (const union nf_inet_addr *)&sc->v6.vaddr,
799 sc->v6.vport, p);
800 else
801 #endif
802 ip_vs_conn_fill_param(ipvs, af, sc->v4.protocol,
803 (const union nf_inet_addr *)&sc->v4.caddr,
804 sc->v4.cport,
805 (const union nf_inet_addr *)&sc->v4.vaddr,
806 sc->v4.vport, p);
807 /* Handle pe data */
808 if (pe_data_len) {
809 if (pe_name_len) {
810 char buff[IP_VS_PENAME_MAXLEN+1];
811
812 memcpy(buff, pe_name, pe_name_len);
813 buff[pe_name_len]=0;
814 p->pe = __ip_vs_pe_getbyname(buff);
815 if (!p->pe) {
816 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n",
817 buff);
818 return 1;
819 }
820 } else {
821 IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n");
822 return 1;
823 }
824
825 p->pe_data = kmemdup(pe_data, pe_data_len, GFP_ATOMIC);
826 if (!p->pe_data) {
827 module_put(p->pe->module);
828 return -ENOMEM;
829 }
830 p->pe_data_len = pe_data_len;
831 }
832 return 0;
833 }
834
835 /*
836 * Connection Add / Update.
837 * Common for version 0 and 1 reception of backup sync_conns.
838 * Param: ...
839 * timeout is in sec.
840 */
ip_vs_proc_conn(struct netns_ipvs * ipvs,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)841 static void ip_vs_proc_conn(struct netns_ipvs *ipvs, struct ip_vs_conn_param *param,
842 unsigned int flags, unsigned int state,
843 unsigned int protocol, unsigned int type,
844 const union nf_inet_addr *daddr, __be16 dport,
845 unsigned long timeout, __u32 fwmark,
846 struct ip_vs_sync_conn_options *opt)
847 {
848 struct ip_vs_dest *dest;
849 struct ip_vs_conn *cp;
850
851 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
852 cp = ip_vs_conn_in_get(param);
853 if (cp && ((cp->dport != dport) ||
854 !ip_vs_addr_equal(cp->daf, &cp->daddr, daddr))) {
855 if (!(flags & IP_VS_CONN_F_INACTIVE)) {
856 ip_vs_conn_expire_now(cp);
857 __ip_vs_conn_put(cp);
858 cp = NULL;
859 } else {
860 /* This is the expiration message for the
861 * connection that was already replaced, so we
862 * just ignore it.
863 */
864 __ip_vs_conn_put(cp);
865 kfree(param->pe_data);
866 return;
867 }
868 }
869 } else {
870 cp = ip_vs_ct_in_get(param);
871 }
872
873 if (cp) {
874 /* Free pe_data */
875 kfree(param->pe_data);
876
877 dest = cp->dest;
878 spin_lock_bh(&cp->lock);
879 if ((cp->flags ^ flags) & IP_VS_CONN_F_INACTIVE &&
880 !(flags & IP_VS_CONN_F_TEMPLATE) && dest) {
881 if (flags & IP_VS_CONN_F_INACTIVE)
882 atomic_dec(&dest->activeconns);
883 else
884 atomic_inc(&dest->activeconns);
885 }
886 flags &= IP_VS_CONN_F_BACKUP_UPD_MASK;
887 flags |= cp->flags & ~IP_VS_CONN_F_BACKUP_UPD_MASK;
888 cp->flags = flags;
889 spin_unlock_bh(&cp->lock);
890 if (!dest)
891 ip_vs_try_bind_dest(cp);
892 } else {
893 /*
894 * Find the appropriate destination for the connection.
895 * If it is not found the connection will remain unbound
896 * but still handled.
897 */
898 rcu_read_lock();
899 /* This function is only invoked by the synchronization
900 * code. We do not currently support heterogeneous pools
901 * with synchronization, so we can make the assumption that
902 * the svc_af is the same as the dest_af
903 */
904 dest = ip_vs_find_dest(ipvs, type, type, daddr, dport,
905 param->vaddr, param->vport, protocol,
906 fwmark, flags);
907
908 cp = ip_vs_conn_new(param, type, daddr, dport, flags, dest,
909 fwmark);
910 rcu_read_unlock();
911 if (!cp) {
912 kfree(param->pe_data);
913 IP_VS_DBG(2, "BACKUP, add new conn. failed\n");
914 return;
915 }
916 if (!(flags & IP_VS_CONN_F_TEMPLATE))
917 kfree(param->pe_data);
918 }
919
920 if (opt) {
921 cp->in_seq = opt->in_seq;
922 cp->out_seq = opt->out_seq;
923 }
924 atomic_set(&cp->in_pkts, sysctl_sync_threshold(ipvs));
925 cp->state = state;
926 cp->old_state = cp->state;
927 /*
928 * For Ver 0 messages style
929 * - Not possible to recover the right timeout for templates
930 * - can not find the right fwmark
931 * virtual service. If needed, we can do it for
932 * non-fwmark persistent services.
933 * Ver 1 messages style.
934 * - No problem.
935 */
936 if (timeout) {
937 if (timeout > MAX_SCHEDULE_TIMEOUT / HZ)
938 timeout = MAX_SCHEDULE_TIMEOUT / HZ;
939 cp->timeout = timeout*HZ;
940 } else {
941 struct ip_vs_proto_data *pd;
942
943 pd = ip_vs_proto_data_get(ipvs, protocol);
944 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pd && pd->timeout_table)
945 cp->timeout = pd->timeout_table[state];
946 else
947 cp->timeout = (3*60*HZ);
948 }
949 ip_vs_conn_put(cp);
950 }
951
952 /*
953 * Process received multicast message for Version 0
954 */
ip_vs_process_message_v0(struct netns_ipvs * ipvs,const char * buffer,const size_t buflen)955 static void ip_vs_process_message_v0(struct netns_ipvs *ipvs, const char *buffer,
956 const size_t buflen)
957 {
958 struct ip_vs_sync_mesg_v0 *m = (struct ip_vs_sync_mesg_v0 *)buffer;
959 struct ip_vs_sync_conn_v0 *s;
960 struct ip_vs_sync_conn_options *opt;
961 struct ip_vs_protocol *pp;
962 struct ip_vs_conn_param param;
963 char *p;
964 int i;
965
966 p = (char *)buffer + sizeof(struct ip_vs_sync_mesg_v0);
967 for (i=0; i<m->nr_conns; i++) {
968 unsigned int flags, state;
969
970 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
971 IP_VS_ERR_RL("BACKUP v0, bogus conn\n");
972 return;
973 }
974 s = (struct ip_vs_sync_conn_v0 *) p;
975 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
976 flags &= ~IP_VS_CONN_F_HASHED;
977 if (flags & IP_VS_CONN_F_SEQ_MASK) {
978 opt = (struct ip_vs_sync_conn_options *)&s[1];
979 p += FULL_CONN_SIZE;
980 if (p > buffer+buflen) {
981 IP_VS_ERR_RL("BACKUP v0, Dropping buffer bogus conn options\n");
982 return;
983 }
984 } else {
985 opt = NULL;
986 p += SIMPLE_CONN_SIZE;
987 }
988
989 state = ntohs(s->state);
990 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
991 pp = ip_vs_proto_get(s->protocol);
992 if (!pp) {
993 IP_VS_DBG(2, "BACKUP v0, Unsupported protocol %u\n",
994 s->protocol);
995 continue;
996 }
997 if (state >= pp->num_states) {
998 IP_VS_DBG(2, "BACKUP v0, Invalid %s state %u\n",
999 pp->name, state);
1000 continue;
1001 }
1002 } else {
1003 if (state >= IP_VS_CTPL_S_LAST)
1004 IP_VS_DBG(7, "BACKUP v0, Invalid tpl state %u\n",
1005 state);
1006 }
1007
1008 ip_vs_conn_fill_param(ipvs, AF_INET, s->protocol,
1009 (const union nf_inet_addr *)&s->caddr,
1010 s->cport,
1011 (const union nf_inet_addr *)&s->vaddr,
1012 s->vport, ¶m);
1013
1014 /* Send timeout as Zero */
1015 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->protocol, AF_INET,
1016 (union nf_inet_addr *)&s->daddr, s->dport,
1017 0, 0, opt);
1018 }
1019 }
1020
1021 /*
1022 * Handle options
1023 */
ip_vs_proc_seqopt(__u8 * p,unsigned int plen,__u32 * opt_flags,struct ip_vs_sync_conn_options * opt)1024 static inline int ip_vs_proc_seqopt(__u8 *p, unsigned int plen,
1025 __u32 *opt_flags,
1026 struct ip_vs_sync_conn_options *opt)
1027 {
1028 struct ip_vs_sync_conn_options *topt;
1029
1030 topt = (struct ip_vs_sync_conn_options *)p;
1031
1032 if (plen != sizeof(struct ip_vs_sync_conn_options)) {
1033 IP_VS_DBG(2, "BACKUP, bogus conn options length\n");
1034 return -EINVAL;
1035 }
1036 if (*opt_flags & IPVS_OPT_F_SEQ_DATA) {
1037 IP_VS_DBG(2, "BACKUP, conn options found twice\n");
1038 return -EINVAL;
1039 }
1040 ntoh_seq(&topt->in_seq, &opt->in_seq);
1041 ntoh_seq(&topt->out_seq, &opt->out_seq);
1042 *opt_flags |= IPVS_OPT_F_SEQ_DATA;
1043 return 0;
1044 }
1045
ip_vs_proc_str(__u8 * p,unsigned int plen,unsigned int * data_len,__u8 ** data,unsigned int maxlen,__u32 * opt_flags,__u32 flag)1046 static int ip_vs_proc_str(__u8 *p, unsigned int plen, unsigned int *data_len,
1047 __u8 **data, unsigned int maxlen,
1048 __u32 *opt_flags, __u32 flag)
1049 {
1050 if (plen > maxlen) {
1051 IP_VS_DBG(2, "BACKUP, bogus par.data len > %d\n", maxlen);
1052 return -EINVAL;
1053 }
1054 if (*opt_flags & flag) {
1055 IP_VS_DBG(2, "BACKUP, Par.data found twice 0x%x\n", flag);
1056 return -EINVAL;
1057 }
1058 *data_len = plen;
1059 *data = p;
1060 *opt_flags |= flag;
1061 return 0;
1062 }
1063 /*
1064 * Process a Version 1 sync. connection
1065 */
ip_vs_proc_sync_conn(struct netns_ipvs * ipvs,__u8 * p,__u8 * msg_end)1066 static inline int ip_vs_proc_sync_conn(struct netns_ipvs *ipvs, __u8 *p, __u8 *msg_end)
1067 {
1068 struct ip_vs_sync_conn_options opt;
1069 union ip_vs_sync_conn *s;
1070 struct ip_vs_protocol *pp;
1071 struct ip_vs_conn_param param;
1072 __u32 flags;
1073 unsigned int af, state, pe_data_len=0, pe_name_len=0;
1074 __u8 *pe_data=NULL, *pe_name=NULL;
1075 __u32 opt_flags=0;
1076 int retc=0;
1077
1078 s = (union ip_vs_sync_conn *) p;
1079
1080 if (s->v6.type & STYPE_F_INET6) {
1081 #ifdef CONFIG_IP_VS_IPV6
1082 af = AF_INET6;
1083 p += sizeof(struct ip_vs_sync_v6);
1084 #else
1085 IP_VS_DBG(3,"BACKUP, IPv6 msg received, and IPVS is not compiled for IPv6\n");
1086 retc = 10;
1087 goto out;
1088 #endif
1089 } else if (!s->v4.type) {
1090 af = AF_INET;
1091 p += sizeof(struct ip_vs_sync_v4);
1092 } else {
1093 return -10;
1094 }
1095 if (p > msg_end)
1096 return -20;
1097
1098 /* Process optional params check Type & Len. */
1099 while (p < msg_end) {
1100 int ptype;
1101 int plen;
1102
1103 if (p+2 > msg_end)
1104 return -30;
1105 ptype = *(p++);
1106 plen = *(p++);
1107
1108 if (!plen || ((p + plen) > msg_end))
1109 return -40;
1110 /* Handle seq option p = param data */
1111 switch (ptype & ~IPVS_OPT_F_PARAM) {
1112 case IPVS_OPT_SEQ_DATA:
1113 if (ip_vs_proc_seqopt(p, plen, &opt_flags, &opt))
1114 return -50;
1115 break;
1116
1117 case IPVS_OPT_PE_DATA:
1118 if (ip_vs_proc_str(p, plen, &pe_data_len, &pe_data,
1119 IP_VS_PEDATA_MAXLEN, &opt_flags,
1120 IPVS_OPT_F_PE_DATA))
1121 return -60;
1122 break;
1123
1124 case IPVS_OPT_PE_NAME:
1125 if (ip_vs_proc_str(p, plen,&pe_name_len, &pe_name,
1126 IP_VS_PENAME_MAXLEN, &opt_flags,
1127 IPVS_OPT_F_PE_NAME))
1128 return -70;
1129 break;
1130
1131 default:
1132 /* Param data mandatory ? */
1133 if (!(ptype & IPVS_OPT_F_PARAM)) {
1134 IP_VS_DBG(3, "BACKUP, Unknown mandatory param %d found\n",
1135 ptype & ~IPVS_OPT_F_PARAM);
1136 retc = 20;
1137 goto out;
1138 }
1139 }
1140 p += plen; /* Next option */
1141 }
1142
1143 /* Get flags and Mask off unsupported */
1144 flags = ntohl(s->v4.flags) & IP_VS_CONN_F_BACKUP_MASK;
1145 flags |= IP_VS_CONN_F_SYNC;
1146 state = ntohs(s->v4.state);
1147
1148 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
1149 pp = ip_vs_proto_get(s->v4.protocol);
1150 if (!pp) {
1151 IP_VS_DBG(3,"BACKUP, Unsupported protocol %u\n",
1152 s->v4.protocol);
1153 retc = 30;
1154 goto out;
1155 }
1156 if (state >= pp->num_states) {
1157 IP_VS_DBG(3, "BACKUP, Invalid %s state %u\n",
1158 pp->name, state);
1159 retc = 40;
1160 goto out;
1161 }
1162 } else {
1163 if (state >= IP_VS_CTPL_S_LAST)
1164 IP_VS_DBG(7, "BACKUP, Invalid tpl state %u\n",
1165 state);
1166 }
1167 if (ip_vs_conn_fill_param_sync(ipvs, af, s, ¶m, pe_data,
1168 pe_data_len, pe_name, pe_name_len)) {
1169 retc = 50;
1170 goto out;
1171 }
1172 /* If only IPv4, just silent skip IPv6 */
1173 if (af == AF_INET)
1174 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->v4.protocol, af,
1175 (union nf_inet_addr *)&s->v4.daddr, s->v4.dport,
1176 ntohl(s->v4.timeout), ntohl(s->v4.fwmark),
1177 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1178 );
1179 #ifdef CONFIG_IP_VS_IPV6
1180 else
1181 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->v6.protocol, af,
1182 (union nf_inet_addr *)&s->v6.daddr, s->v6.dport,
1183 ntohl(s->v6.timeout), ntohl(s->v6.fwmark),
1184 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1185 );
1186 #endif
1187 ip_vs_pe_put(param.pe);
1188 return 0;
1189 /* Error exit */
1190 out:
1191 IP_VS_DBG(2, "BACKUP, Single msg dropped err:%d\n", retc);
1192 return retc;
1193
1194 }
1195 /*
1196 * Process received multicast message and create the corresponding
1197 * ip_vs_conn entries.
1198 * Handles Version 0 & 1
1199 */
ip_vs_process_message(struct netns_ipvs * ipvs,__u8 * buffer,const size_t buflen)1200 static void ip_vs_process_message(struct netns_ipvs *ipvs, __u8 *buffer,
1201 const size_t buflen)
1202 {
1203 struct ip_vs_sync_mesg *m2 = (struct ip_vs_sync_mesg *)buffer;
1204 __u8 *p, *msg_end;
1205 int i, nr_conns;
1206
1207 if (buflen < sizeof(struct ip_vs_sync_mesg_v0)) {
1208 IP_VS_DBG(2, "BACKUP, message header too short\n");
1209 return;
1210 }
1211
1212 if (buflen != ntohs(m2->size)) {
1213 IP_VS_DBG(2, "BACKUP, bogus message size\n");
1214 return;
1215 }
1216 /* SyncID sanity check */
1217 if (ipvs->bcfg.syncid != 0 && m2->syncid != ipvs->bcfg.syncid) {
1218 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2->syncid);
1219 return;
1220 }
1221 /* Handle version 1 message */
1222 if ((m2->version == SYNC_PROTO_VER) && (m2->reserved == 0)
1223 && (m2->spare == 0)) {
1224
1225 msg_end = buffer + sizeof(struct ip_vs_sync_mesg);
1226 nr_conns = m2->nr_conns;
1227
1228 for (i=0; i<nr_conns; i++) {
1229 union ip_vs_sync_conn *s;
1230 unsigned int size;
1231 int retc;
1232
1233 p = msg_end;
1234 if (p + sizeof(s->v4) > buffer+buflen) {
1235 IP_VS_ERR_RL("BACKUP, Dropping buffer, too small\n");
1236 return;
1237 }
1238 s = (union ip_vs_sync_conn *)p;
1239 size = ntohs(s->v4.ver_size) & SVER_MASK;
1240 msg_end = p + size;
1241 /* Basic sanity checks */
1242 if (msg_end > buffer+buflen) {
1243 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n");
1244 return;
1245 }
1246 if (ntohs(s->v4.ver_size) >> SVER_SHIFT) {
1247 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n",
1248 ntohs(s->v4.ver_size) >> SVER_SHIFT);
1249 return;
1250 }
1251 /* Process a single sync_conn */
1252 retc = ip_vs_proc_sync_conn(ipvs, p, msg_end);
1253 if (retc < 0) {
1254 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n",
1255 retc);
1256 return;
1257 }
1258 /* Make sure we have 32 bit alignment */
1259 msg_end = p + ((size + 3) & ~3);
1260 }
1261 } else {
1262 /* Old type of message */
1263 ip_vs_process_message_v0(ipvs, buffer, buflen);
1264 return;
1265 }
1266 }
1267
1268
1269 /*
1270 * Setup sndbuf (mode=1) or rcvbuf (mode=0)
1271 */
set_sock_size(struct sock * sk,int mode,int val)1272 static void set_sock_size(struct sock *sk, int mode, int val)
1273 {
1274 /* setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &val, sizeof(val)); */
1275 /* setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)); */
1276 lock_sock(sk);
1277 if (mode) {
1278 val = clamp_t(int, val, (SOCK_MIN_SNDBUF + 1) / 2,
1279 READ_ONCE(sysctl_wmem_max));
1280 sk->sk_sndbuf = val * 2;
1281 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1282 } else {
1283 val = clamp_t(int, val, (SOCK_MIN_RCVBUF + 1) / 2,
1284 READ_ONCE(sysctl_rmem_max));
1285 sk->sk_rcvbuf = val * 2;
1286 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1287 }
1288 release_sock(sk);
1289 }
1290
1291 /*
1292 * Setup loopback of outgoing multicasts on a sending socket
1293 */
set_mcast_loop(struct sock * sk,u_char loop)1294 static void set_mcast_loop(struct sock *sk, u_char loop)
1295 {
1296 /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
1297 inet_assign_bit(MC_LOOP, sk, loop);
1298 #ifdef CONFIG_IP_VS_IPV6
1299 if (READ_ONCE(sk->sk_family) == AF_INET6) {
1300 /* IPV6_MULTICAST_LOOP */
1301 inet6_assign_bit(MC6_LOOP, sk, loop);
1302 }
1303 #endif
1304 }
1305
1306 /*
1307 * Specify TTL for outgoing multicasts on a sending socket
1308 */
set_mcast_ttl(struct sock * sk,u_char ttl)1309 static void set_mcast_ttl(struct sock *sk, u_char ttl)
1310 {
1311 struct inet_sock *inet = inet_sk(sk);
1312
1313 /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
1314 lock_sock(sk);
1315 WRITE_ONCE(inet->mc_ttl, ttl);
1316 #ifdef CONFIG_IP_VS_IPV6
1317 if (sk->sk_family == AF_INET6) {
1318 struct ipv6_pinfo *np = inet6_sk(sk);
1319
1320 /* IPV6_MULTICAST_HOPS */
1321 WRITE_ONCE(np->mcast_hops, ttl);
1322 }
1323 #endif
1324 release_sock(sk);
1325 }
1326
1327 /* Control fragmentation of messages */
set_mcast_pmtudisc(struct sock * sk,int val)1328 static void set_mcast_pmtudisc(struct sock *sk, int val)
1329 {
1330 struct inet_sock *inet = inet_sk(sk);
1331
1332 /* setsockopt(sock, SOL_IP, IP_MTU_DISCOVER, &val, sizeof(val)); */
1333 lock_sock(sk);
1334 WRITE_ONCE(inet->pmtudisc, val);
1335 #ifdef CONFIG_IP_VS_IPV6
1336 if (sk->sk_family == AF_INET6) {
1337 struct ipv6_pinfo *np = inet6_sk(sk);
1338
1339 /* IPV6_MTU_DISCOVER */
1340 WRITE_ONCE(np->pmtudisc, val);
1341 }
1342 #endif
1343 release_sock(sk);
1344 }
1345
1346 /*
1347 * Specifiy default interface for outgoing multicasts
1348 */
set_mcast_if(struct sock * sk,struct net_device * dev)1349 static int set_mcast_if(struct sock *sk, struct net_device *dev)
1350 {
1351 struct inet_sock *inet = inet_sk(sk);
1352
1353 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1354 return -EINVAL;
1355
1356 lock_sock(sk);
1357 inet->mc_index = dev->ifindex;
1358 /* inet->mc_addr = 0; */
1359 #ifdef CONFIG_IP_VS_IPV6
1360 if (sk->sk_family == AF_INET6) {
1361 struct ipv6_pinfo *np = inet6_sk(sk);
1362
1363 /* IPV6_MULTICAST_IF */
1364 WRITE_ONCE(np->mcast_oif, dev->ifindex);
1365 }
1366 #endif
1367 release_sock(sk);
1368
1369 return 0;
1370 }
1371
1372
1373 /*
1374 * Join a multicast group.
1375 * the group is specified by a class D multicast address 224.0.0.0/8
1376 * in the in_addr structure passed in as a parameter.
1377 */
1378 static int
join_mcast_group(struct sock * sk,struct in_addr * addr,struct net_device * dev)1379 join_mcast_group(struct sock *sk, struct in_addr *addr, struct net_device *dev)
1380 {
1381 struct ip_mreqn mreq;
1382 int ret;
1383
1384 memset(&mreq, 0, sizeof(mreq));
1385 memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
1386
1387 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1388 return -EINVAL;
1389
1390 mreq.imr_ifindex = dev->ifindex;
1391
1392 lock_sock(sk);
1393 ret = ip_mc_join_group(sk, &mreq);
1394 release_sock(sk);
1395
1396 return ret;
1397 }
1398
1399 #ifdef CONFIG_IP_VS_IPV6
join_mcast_group6(struct sock * sk,struct in6_addr * addr,struct net_device * dev)1400 static int join_mcast_group6(struct sock *sk, struct in6_addr *addr,
1401 struct net_device *dev)
1402 {
1403 int ret;
1404
1405 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1406 return -EINVAL;
1407
1408 lock_sock(sk);
1409 ret = ipv6_sock_mc_join(sk, dev->ifindex, addr);
1410 release_sock(sk);
1411
1412 return ret;
1413 }
1414 #endif
1415
bind_mcastif_addr(struct socket * sock,struct net_device * dev)1416 static int bind_mcastif_addr(struct socket *sock, struct net_device *dev)
1417 {
1418 __be32 addr;
1419 struct sockaddr_in sin;
1420
1421 addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
1422 if (!addr)
1423 pr_err("You probably need to specify IP address on "
1424 "multicast interface.\n");
1425
1426 IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
1427 dev->name, &addr);
1428
1429 /* Now bind the socket with the address of multicast interface */
1430 sin.sin_family = AF_INET;
1431 sin.sin_addr.s_addr = addr;
1432 sin.sin_port = 0;
1433
1434 return kernel_bind(sock, (struct sockaddr_unsized *)&sin, sizeof(sin));
1435 }
1436
get_mcast_sockaddr(union ipvs_sockaddr * sa,int * salen,struct ipvs_sync_daemon_cfg * c,int id)1437 static void get_mcast_sockaddr(union ipvs_sockaddr *sa, int *salen,
1438 struct ipvs_sync_daemon_cfg *c, int id)
1439 {
1440 if (AF_INET6 == c->mcast_af) {
1441 sa->in6 = (struct sockaddr_in6) {
1442 .sin6_family = AF_INET6,
1443 .sin6_port = htons(c->mcast_port + id),
1444 };
1445 sa->in6.sin6_addr = c->mcast_group.in6;
1446 *salen = sizeof(sa->in6);
1447 } else {
1448 sa->in = (struct sockaddr_in) {
1449 .sin_family = AF_INET,
1450 .sin_port = htons(c->mcast_port + id),
1451 };
1452 sa->in.sin_addr = c->mcast_group.in;
1453 *salen = sizeof(sa->in);
1454 }
1455 }
1456
1457 /*
1458 * Set up sending multicast socket over UDP
1459 */
make_send_sock(struct netns_ipvs * ipvs,int id,struct net_device * dev,struct socket ** sock_ret)1460 static int make_send_sock(struct netns_ipvs *ipvs, int id,
1461 struct net_device *dev, struct socket **sock_ret)
1462 {
1463 /* multicast addr */
1464 union ipvs_sockaddr mcast_addr;
1465 struct socket *sock;
1466 int result, salen;
1467
1468 /* First create a socket */
1469 result = sock_create_kern(ipvs->net, ipvs->mcfg.mcast_af, SOCK_DGRAM,
1470 IPPROTO_UDP, &sock);
1471 if (result < 0) {
1472 pr_err("Error during creation of socket; terminating\n");
1473 goto error;
1474 }
1475 *sock_ret = sock;
1476 result = set_mcast_if(sock->sk, dev);
1477 if (result < 0) {
1478 pr_err("Error setting outbound mcast interface\n");
1479 goto error;
1480 }
1481
1482 set_mcast_loop(sock->sk, 0);
1483 set_mcast_ttl(sock->sk, ipvs->mcfg.mcast_ttl);
1484 /* Allow fragmentation if MTU changes */
1485 set_mcast_pmtudisc(sock->sk, IP_PMTUDISC_DONT);
1486 result = sysctl_sync_sock_size(ipvs);
1487 if (result > 0)
1488 set_sock_size(sock->sk, 1, result);
1489
1490 if (AF_INET == ipvs->mcfg.mcast_af)
1491 result = bind_mcastif_addr(sock, dev);
1492 else
1493 result = 0;
1494 if (result < 0) {
1495 pr_err("Error binding address of the mcast interface\n");
1496 goto error;
1497 }
1498
1499 get_mcast_sockaddr(&mcast_addr, &salen, &ipvs->mcfg, id);
1500 result = kernel_connect(sock, (struct sockaddr_unsized *)&mcast_addr,
1501 salen, 0);
1502 if (result < 0) {
1503 pr_err("Error connecting to the multicast addr\n");
1504 goto error;
1505 }
1506
1507 return 0;
1508
1509 error:
1510 return result;
1511 }
1512
1513
1514 /*
1515 * Set up receiving multicast socket over UDP
1516 */
make_receive_sock(struct netns_ipvs * ipvs,int id,struct net_device * dev,struct socket ** sock_ret)1517 static int make_receive_sock(struct netns_ipvs *ipvs, int id,
1518 struct net_device *dev, struct socket **sock_ret)
1519 {
1520 /* multicast addr */
1521 union ipvs_sockaddr mcast_addr;
1522 struct socket *sock;
1523 int result, salen;
1524
1525 /* First create a socket */
1526 result = sock_create_kern(ipvs->net, ipvs->bcfg.mcast_af, SOCK_DGRAM,
1527 IPPROTO_UDP, &sock);
1528 if (result < 0) {
1529 pr_err("Error during creation of socket; terminating\n");
1530 goto error;
1531 }
1532 *sock_ret = sock;
1533 /* it is equivalent to the REUSEADDR option in user-space */
1534 sock->sk->sk_reuse = SK_CAN_REUSE;
1535 result = sysctl_sync_sock_size(ipvs);
1536 if (result > 0)
1537 set_sock_size(sock->sk, 0, result);
1538
1539 get_mcast_sockaddr(&mcast_addr, &salen, &ipvs->bcfg, id);
1540 sock->sk->sk_bound_dev_if = dev->ifindex;
1541 result = kernel_bind(sock, (struct sockaddr_unsized *)&mcast_addr, salen);
1542 if (result < 0) {
1543 pr_err("Error binding to the multicast addr\n");
1544 goto error;
1545 }
1546
1547 /* join the multicast group */
1548 #ifdef CONFIG_IP_VS_IPV6
1549 if (ipvs->bcfg.mcast_af == AF_INET6)
1550 result = join_mcast_group6(sock->sk, &mcast_addr.in6.sin6_addr,
1551 dev);
1552 else
1553 #endif
1554 result = join_mcast_group(sock->sk, &mcast_addr.in.sin_addr,
1555 dev);
1556 if (result < 0) {
1557 pr_err("Error joining to the multicast group\n");
1558 goto error;
1559 }
1560
1561 return 0;
1562
1563 error:
1564 return result;
1565 }
1566
1567
1568 static int
ip_vs_send_async(struct socket * sock,const char * buffer,const size_t length)1569 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
1570 {
1571 struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
1572 struct kvec iov;
1573 int len;
1574
1575 iov.iov_base = (void *)buffer;
1576 iov.iov_len = length;
1577
1578 len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
1579
1580 return len;
1581 }
1582
1583 static int
ip_vs_send_sync_msg(struct socket * sock,struct ip_vs_sync_mesg * msg)1584 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
1585 {
1586 int msize;
1587 int ret;
1588
1589 msize = ntohs(msg->size);
1590
1591 ret = ip_vs_send_async(sock, (char *)msg, msize);
1592 if (ret >= 0 || ret == -EAGAIN)
1593 return ret;
1594 pr_err("ip_vs_send_async error %d\n", ret);
1595 return 0;
1596 }
1597
1598 static int
ip_vs_receive(struct socket * sock,char * buffer,const size_t buflen)1599 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
1600 {
1601 struct msghdr msg = {NULL,};
1602 struct kvec iov = {buffer, buflen};
1603 int len;
1604
1605 /* Receive a packet */
1606 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &iov, 1, buflen);
1607 len = sock_recvmsg(sock, &msg, MSG_DONTWAIT);
1608 if (len < 0)
1609 return len;
1610
1611 return len;
1612 }
1613
1614 /* Wakeup the master thread for sending */
master_wakeup_work_handler(struct work_struct * work)1615 static void master_wakeup_work_handler(struct work_struct *work)
1616 {
1617 struct ipvs_master_sync_state *ms =
1618 container_of(work, struct ipvs_master_sync_state,
1619 master_wakeup_work.work);
1620 struct netns_ipvs *ipvs = ms->ipvs;
1621
1622 spin_lock_bh(&ipvs->sync_lock);
1623 if (ms->sync_queue_len &&
1624 ms->sync_queue_delay < IPVS_SYNC_WAKEUP_RATE) {
1625 int id = (int)(ms - ipvs->ms);
1626
1627 ms->sync_queue_delay = IPVS_SYNC_WAKEUP_RATE;
1628 wake_up_process(ipvs->master_tinfo[id].task);
1629 }
1630 spin_unlock_bh(&ipvs->sync_lock);
1631 }
1632
1633 /* Get next buffer to send */
1634 static inline struct ip_vs_sync_buff *
next_sync_buff(struct netns_ipvs * ipvs,struct ipvs_master_sync_state * ms)1635 next_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
1636 {
1637 struct ip_vs_sync_buff *sb;
1638
1639 sb = sb_dequeue(ipvs, ms);
1640 if (sb)
1641 return sb;
1642 /* Do not delay entries in buffer for more than 2 seconds */
1643 return get_curr_sync_buff(ipvs, ms, IPVS_SYNC_FLUSH_TIME);
1644 }
1645
sync_thread_master(void * data)1646 static int sync_thread_master(void *data)
1647 {
1648 struct ip_vs_sync_thread_data *tinfo = data;
1649 struct netns_ipvs *ipvs = tinfo->ipvs;
1650 struct ipvs_master_sync_state *ms = &ipvs->ms[tinfo->id];
1651 struct sock *sk = tinfo->sock->sk;
1652 struct ip_vs_sync_buff *sb;
1653
1654 pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
1655 "syncid = %d, id = %d\n",
1656 ipvs->mcfg.mcast_ifn, ipvs->mcfg.syncid, tinfo->id);
1657
1658 for (;;) {
1659 sb = next_sync_buff(ipvs, ms);
1660 if (unlikely(kthread_should_stop()))
1661 break;
1662 if (!sb) {
1663 schedule_timeout(IPVS_SYNC_CHECK_PERIOD);
1664 continue;
1665 }
1666 while (ip_vs_send_sync_msg(tinfo->sock, sb->mesg) < 0) {
1667 /* (Ab)use interruptible sleep to avoid increasing
1668 * the load avg.
1669 */
1670 __wait_event_interruptible(*sk_sleep(sk),
1671 sock_writeable(sk) ||
1672 kthread_should_stop());
1673 if (unlikely(kthread_should_stop()))
1674 goto done;
1675 }
1676 ip_vs_sync_buff_release(sb);
1677 }
1678
1679 done:
1680 __set_current_state(TASK_RUNNING);
1681 if (sb)
1682 ip_vs_sync_buff_release(sb);
1683
1684 /* clean up the sync_buff queue */
1685 while ((sb = sb_dequeue(ipvs, ms)))
1686 ip_vs_sync_buff_release(sb);
1687 __set_current_state(TASK_RUNNING);
1688
1689 /* clean up the current sync_buff */
1690 sb = get_curr_sync_buff(ipvs, ms, 0);
1691 if (sb)
1692 ip_vs_sync_buff_release(sb);
1693
1694 return 0;
1695 }
1696
1697
sync_thread_backup(void * data)1698 static int sync_thread_backup(void *data)
1699 {
1700 struct ip_vs_sync_thread_data *tinfo = data;
1701 struct netns_ipvs *ipvs = tinfo->ipvs;
1702 struct sock *sk = tinfo->sock->sk;
1703 struct udp_sock *up = udp_sk(sk);
1704 int len;
1705
1706 pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
1707 "syncid = %d, id = %d\n",
1708 ipvs->bcfg.mcast_ifn, ipvs->bcfg.syncid, tinfo->id);
1709
1710 while (!kthread_should_stop()) {
1711 wait_event_interruptible(*sk_sleep(sk),
1712 !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
1713 !skb_queue_empty_lockless(&up->reader_queue) ||
1714 kthread_should_stop());
1715
1716 /* do we have data now? */
1717 while (!skb_queue_empty_lockless(&sk->sk_receive_queue) ||
1718 !skb_queue_empty_lockless(&up->reader_queue)) {
1719 len = ip_vs_receive(tinfo->sock, tinfo->buf,
1720 ipvs->bcfg.sync_maxlen);
1721 if (len <= 0) {
1722 if (len != -EAGAIN)
1723 pr_err("receiving message error\n");
1724 break;
1725 }
1726
1727 ip_vs_process_message(ipvs, tinfo->buf, len);
1728 }
1729 }
1730
1731 return 0;
1732 }
1733
1734
start_sync_thread(struct netns_ipvs * ipvs,struct ipvs_sync_daemon_cfg * c,int state)1735 int start_sync_thread(struct netns_ipvs *ipvs, struct ipvs_sync_daemon_cfg *c,
1736 int state)
1737 {
1738 struct ip_vs_sync_thread_data *ti = NULL, *tinfo;
1739 struct task_struct *task;
1740 struct net_device *dev;
1741 char *name;
1742 int (*threadfn)(void *data);
1743 int id = 0, count, hlen;
1744 int result = -ENOMEM;
1745 u16 mtu, min_mtu;
1746
1747 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1748 IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %zd bytes\n",
1749 sizeof(struct ip_vs_sync_conn_v0));
1750
1751 /* increase the module use count */
1752 if (!ip_vs_use_count_inc())
1753 return -ENOPROTOOPT;
1754
1755 /* Backup server can be started without services just to sync conns,
1756 * make sure conn_tab is created even if ipvs->enable is 0.
1757 */
1758 if (state == IP_VS_STATE_BACKUP) {
1759 mutex_lock(&ipvs->service_mutex);
1760 if (!rcu_dereference_protected(ipvs->conn_tab, 1)) {
1761 int lfactor = sysctl_conn_lfactor(ipvs);
1762 int new_size = ip_vs_conn_desired_size(ipvs, NULL,
1763 lfactor);
1764 struct ip_vs_rht *tc_new;
1765
1766 tc_new = ip_vs_conn_tab_alloc(ipvs, new_size, lfactor);
1767 if (!tc_new) {
1768 mutex_unlock(&ipvs->service_mutex);
1769 result = -ENOMEM;
1770 goto out_module;
1771 }
1772 rcu_assign_pointer(ipvs->conn_tab, tc_new);
1773 }
1774 mutex_unlock(&ipvs->service_mutex);
1775 }
1776
1777 /* Do not hold one mutex and then to block on another */
1778 for (;;) {
1779 rtnl_lock();
1780 if (mutex_trylock(&ipvs->sync_mutex))
1781 break;
1782 rtnl_unlock();
1783 mutex_lock(&ipvs->sync_mutex);
1784 if (rtnl_trylock())
1785 break;
1786 mutex_unlock(&ipvs->sync_mutex);
1787 }
1788
1789 if (!ipvs->sync_state) {
1790 count = clamp(sysctl_sync_ports(ipvs), 1, IPVS_SYNC_PORTS_MAX);
1791 ipvs->threads_mask = count - 1;
1792 } else
1793 count = ipvs->threads_mask + 1;
1794
1795 if (c->mcast_af == AF_UNSPEC) {
1796 c->mcast_af = AF_INET;
1797 c->mcast_group.ip = cpu_to_be32(IP_VS_SYNC_GROUP);
1798 }
1799 if (!c->mcast_port)
1800 c->mcast_port = IP_VS_SYNC_PORT;
1801 if (!c->mcast_ttl)
1802 c->mcast_ttl = 1;
1803
1804 dev = __dev_get_by_name(ipvs->net, c->mcast_ifn);
1805 if (!dev) {
1806 pr_err("Unknown mcast interface: %s\n", c->mcast_ifn);
1807 result = -ENODEV;
1808 goto out_early;
1809 }
1810 hlen = (AF_INET6 == c->mcast_af) ?
1811 sizeof(struct ipv6hdr) + sizeof(struct udphdr) :
1812 sizeof(struct iphdr) + sizeof(struct udphdr);
1813 mtu = (state == IP_VS_STATE_BACKUP) ?
1814 clamp(dev->mtu, 1500U, 65535U) : 1500U;
1815 min_mtu = (state == IP_VS_STATE_BACKUP) ? 1024 : 1;
1816
1817 if (c->sync_maxlen)
1818 c->sync_maxlen = clamp_t(unsigned int,
1819 c->sync_maxlen, min_mtu,
1820 65535 - hlen);
1821 else
1822 c->sync_maxlen = mtu - hlen;
1823
1824 if (state == IP_VS_STATE_MASTER) {
1825 result = -EEXIST;
1826 if (ipvs->ms)
1827 goto out_early;
1828
1829 ipvs->mcfg = *c;
1830 name = "ipvs-m:%d:%d";
1831 threadfn = sync_thread_master;
1832 } else if (state == IP_VS_STATE_BACKUP) {
1833 result = -EEXIST;
1834 if (ipvs->backup_tinfo)
1835 goto out_early;
1836
1837 ipvs->bcfg = *c;
1838 name = "ipvs-b:%d:%d";
1839 threadfn = sync_thread_backup;
1840 } else {
1841 result = -EINVAL;
1842 goto out_early;
1843 }
1844
1845 if (state == IP_VS_STATE_MASTER) {
1846 struct ipvs_master_sync_state *ms;
1847
1848 result = -ENOMEM;
1849 ipvs->ms = kzalloc_objs(ipvs->ms[0], count);
1850 if (!ipvs->ms)
1851 goto out;
1852 ms = ipvs->ms;
1853 for (id = 0; id < count; id++, ms++) {
1854 INIT_LIST_HEAD(&ms->sync_queue);
1855 ms->sync_queue_len = 0;
1856 ms->sync_queue_delay = 0;
1857 INIT_DELAYED_WORK(&ms->master_wakeup_work,
1858 master_wakeup_work_handler);
1859 ms->ipvs = ipvs;
1860 }
1861 }
1862 result = -ENOMEM;
1863 ti = kzalloc_objs(struct ip_vs_sync_thread_data, count);
1864 if (!ti)
1865 goto out;
1866
1867 for (id = 0; id < count; id++) {
1868 tinfo = &ti[id];
1869 tinfo->ipvs = ipvs;
1870 if (state == IP_VS_STATE_BACKUP) {
1871 result = -ENOMEM;
1872 tinfo->buf = kmalloc(ipvs->bcfg.sync_maxlen,
1873 GFP_KERNEL);
1874 if (!tinfo->buf)
1875 goto out;
1876 }
1877 tinfo->id = id;
1878 if (state == IP_VS_STATE_MASTER)
1879 result = make_send_sock(ipvs, id, dev, &tinfo->sock);
1880 else
1881 result = make_receive_sock(ipvs, id, dev, &tinfo->sock);
1882 if (result < 0)
1883 goto out;
1884
1885 task = kthread_run(threadfn, tinfo, name, ipvs->gen, id);
1886 if (IS_ERR(task)) {
1887 result = PTR_ERR(task);
1888 goto out;
1889 }
1890 tinfo->task = task;
1891 }
1892
1893 /* mark as active */
1894
1895 if (state == IP_VS_STATE_MASTER)
1896 ipvs->master_tinfo = ti;
1897 else
1898 ipvs->backup_tinfo = ti;
1899 spin_lock_bh(&ipvs->sync_buff_lock);
1900 ipvs->sync_state |= state;
1901 spin_unlock_bh(&ipvs->sync_buff_lock);
1902
1903 mutex_unlock(&ipvs->sync_mutex);
1904 rtnl_unlock();
1905
1906 return 0;
1907
1908 out:
1909 /* We do not need RTNL lock anymore, release it here so that
1910 * sock_release below can use rtnl_lock to leave the mcast group.
1911 */
1912 rtnl_unlock();
1913 id = min(id, count - 1);
1914 if (ti) {
1915 for (tinfo = ti + id; tinfo >= ti; tinfo--) {
1916 if (tinfo->task)
1917 kthread_stop(tinfo->task);
1918 }
1919 }
1920 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
1921 kfree(ipvs->ms);
1922 ipvs->ms = NULL;
1923 }
1924 mutex_unlock(&ipvs->sync_mutex);
1925
1926 /* No more mutexes, release socks */
1927 if (ti) {
1928 for (tinfo = ti + id; tinfo >= ti; tinfo--) {
1929 if (tinfo->sock)
1930 sock_release(tinfo->sock);
1931 kfree(tinfo->buf);
1932 }
1933 kfree(ti);
1934 }
1935
1936 /* decrease the module use count */
1937 ip_vs_use_count_dec();
1938 return result;
1939
1940 out_early:
1941 mutex_unlock(&ipvs->sync_mutex);
1942 rtnl_unlock();
1943
1944 out_module:
1945 /* decrease the module use count */
1946 ip_vs_use_count_dec();
1947 return result;
1948 }
1949
1950
stop_sync_thread(struct netns_ipvs * ipvs,int state)1951 int stop_sync_thread(struct netns_ipvs *ipvs, int state)
1952 {
1953 struct ip_vs_sync_thread_data *ti, *tinfo;
1954 int id;
1955 int retc = -EINVAL;
1956
1957 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1958
1959 mutex_lock(&ipvs->sync_mutex);
1960 if (state == IP_VS_STATE_MASTER) {
1961 retc = -ESRCH;
1962 if (!ipvs->ms)
1963 goto err;
1964 ti = ipvs->master_tinfo;
1965
1966 /*
1967 * The lock synchronizes with sb_queue_tail(), so that we don't
1968 * add sync buffers to the queue, when we are already in
1969 * progress of stopping the master sync daemon.
1970 */
1971
1972 spin_lock_bh(&ipvs->sync_buff_lock);
1973 spin_lock(&ipvs->sync_lock);
1974 ipvs->sync_state &= ~IP_VS_STATE_MASTER;
1975 spin_unlock(&ipvs->sync_lock);
1976 spin_unlock_bh(&ipvs->sync_buff_lock);
1977
1978 retc = 0;
1979 for (id = ipvs->threads_mask; id >= 0; id--) {
1980 struct ipvs_master_sync_state *ms = &ipvs->ms[id];
1981 int ret;
1982
1983 tinfo = &ti[id];
1984 pr_info("stopping master sync thread %d ...\n",
1985 task_pid_nr(tinfo->task));
1986 cancel_delayed_work_sync(&ms->master_wakeup_work);
1987 ret = kthread_stop(tinfo->task);
1988 if (retc >= 0)
1989 retc = ret;
1990 }
1991 kfree(ipvs->ms);
1992 ipvs->ms = NULL;
1993 ipvs->master_tinfo = NULL;
1994 } else if (state == IP_VS_STATE_BACKUP) {
1995 retc = -ESRCH;
1996 if (!ipvs->backup_tinfo)
1997 goto err;
1998 ti = ipvs->backup_tinfo;
1999
2000 ipvs->sync_state &= ~IP_VS_STATE_BACKUP;
2001 retc = 0;
2002 for (id = ipvs->threads_mask; id >= 0; id--) {
2003 int ret;
2004
2005 tinfo = &ti[id];
2006 pr_info("stopping backup sync thread %d ...\n",
2007 task_pid_nr(tinfo->task));
2008 ret = kthread_stop(tinfo->task);
2009 if (retc >= 0)
2010 retc = ret;
2011 }
2012 ipvs->backup_tinfo = NULL;
2013 } else {
2014 goto err;
2015 }
2016 id = ipvs->threads_mask;
2017 mutex_unlock(&ipvs->sync_mutex);
2018
2019 /* No more mutexes, release socks */
2020 for (tinfo = ti + id; tinfo >= ti; tinfo--) {
2021 if (tinfo->sock)
2022 sock_release(tinfo->sock);
2023 kfree(tinfo->buf);
2024 }
2025 kfree(ti);
2026
2027 /* decrease the module use count */
2028 ip_vs_use_count_dec();
2029 return retc;
2030
2031 err:
2032 mutex_unlock(&ipvs->sync_mutex);
2033 return retc;
2034 }
2035
2036 /*
2037 * Initialize data struct for each netns
2038 */
ip_vs_sync_net_init(struct netns_ipvs * ipvs)2039 int __net_init ip_vs_sync_net_init(struct netns_ipvs *ipvs)
2040 {
2041 __mutex_init(&ipvs->sync_mutex, "ipvs->sync_mutex", &__ipvs_sync_key);
2042 spin_lock_init(&ipvs->sync_lock);
2043 spin_lock_init(&ipvs->sync_buff_lock);
2044 return 0;
2045 }
2046
ip_vs_sync_net_cleanup(struct netns_ipvs * ipvs)2047 void ip_vs_sync_net_cleanup(struct netns_ipvs *ipvs)
2048 {
2049 int retc;
2050
2051 retc = stop_sync_thread(ipvs, IP_VS_STATE_MASTER);
2052 if (retc && retc != -ESRCH)
2053 pr_err("Failed to stop Master Daemon\n");
2054
2055 retc = stop_sync_thread(ipvs, IP_VS_STATE_BACKUP);
2056 if (retc && retc != -ESRCH)
2057 pr_err("Failed to stop Backup Daemon\n");
2058 }
2059