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 KMSG_COMPONENT "IPVS"
36 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
37
38 #include <linux/module.h>
39 #include <linux/slab.h>
40 #include <linux/inetdevice.h>
41 #include <linux/net.h>
42 #include <linux/completion.h>
43 #include <linux/delay.h>
44 #include <linux/skbuff.h>
45 #include <linux/in.h>
46 #include <linux/igmp.h> /* for ip_mc_join_group */
47 #include <linux/udp.h>
48 #include <linux/err.h>
49 #include <linux/kthread.h>
50 #include <linux/wait.h>
51 #include <linux/kernel.h>
52 #include <linux/sched/signal.h>
53
54 #include <asm/unaligned.h> /* Used for ntoh_seq and hton_seq */
55
56 #include <net/ip.h>
57 #include <net/sock.h>
58
59 #include <net/ip_vs.h>
60
61 #define IP_VS_SYNC_GROUP 0xe0000051 /* multicast addr - 224.0.0.81 */
62 #define IP_VS_SYNC_PORT 8848 /* multicast port */
63
64 #define SYNC_PROTO_VER 1 /* Protocol version in header */
65
66 static struct lock_class_key __ipvs_sync_key;
67 /*
68 * IPVS sync connection entry
69 * Version 0, i.e. original version.
70 */
71 struct ip_vs_sync_conn_v0 {
72 __u8 reserved;
73
74 /* Protocol, addresses and port numbers */
75 __u8 protocol; /* Which protocol (TCP/UDP) */
76 __be16 cport;
77 __be16 vport;
78 __be16 dport;
79 __be32 caddr; /* client address */
80 __be32 vaddr; /* virtual address */
81 __be32 daddr; /* destination address */
82
83 /* Flags and state transition */
84 __be16 flags; /* status flags */
85 __be16 state; /* state info */
86
87 /* The sequence options start here */
88 };
89
90 struct ip_vs_sync_conn_options {
91 struct ip_vs_seq in_seq; /* incoming seq. struct */
92 struct ip_vs_seq out_seq; /* outgoing seq. struct */
93 };
94
95 /*
96 Sync Connection format (sync_conn)
97
98 0 1 2 3
99 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
100 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
101 | Type | Protocol | Ver. | Size |
102 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
103 | Flags |
104 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
105 | State | cport |
106 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
107 | vport | dport |
108 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
109 | fwmark |
110 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
111 | timeout (in sec.) |
112 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
113 | ... |
114 | IP-Addresses (v4 or v6) |
115 | ... |
116 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
117 Optional Parameters.
118 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
119 | Param. Type | Param. Length | Param. data |
120 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
121 | ... |
122 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
123 | | Param Type | Param. Length |
124 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
125 | Param data |
126 | Last Param data should be padded for 32 bit alignment |
127 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
128 */
129
130 /*
131 * Type 0, IPv4 sync connection format
132 */
133 struct ip_vs_sync_v4 {
134 __u8 type;
135 __u8 protocol; /* Which protocol (TCP/UDP) */
136 __be16 ver_size; /* Version msb 4 bits */
137 /* Flags and state transition */
138 __be32 flags; /* status flags */
139 __be16 state; /* state info */
140 /* Protocol, addresses and port numbers */
141 __be16 cport;
142 __be16 vport;
143 __be16 dport;
144 __be32 fwmark; /* Firewall mark from skb */
145 __be32 timeout; /* cp timeout */
146 __be32 caddr; /* client address */
147 __be32 vaddr; /* virtual address */
148 __be32 daddr; /* destination address */
149 /* The sequence options start here */
150 /* PE data padded to 32bit alignment after seq. options */
151 };
152 /*
153 * Type 2 messages IPv6
154 */
155 struct ip_vs_sync_v6 {
156 __u8 type;
157 __u8 protocol; /* Which protocol (TCP/UDP) */
158 __be16 ver_size; /* Version msb 4 bits */
159 /* Flags and state transition */
160 __be32 flags; /* status flags */
161 __be16 state; /* state info */
162 /* Protocol, addresses and port numbers */
163 __be16 cport;
164 __be16 vport;
165 __be16 dport;
166 __be32 fwmark; /* Firewall mark from skb */
167 __be32 timeout; /* cp timeout */
168 struct in6_addr caddr; /* client address */
169 struct in6_addr vaddr; /* virtual address */
170 struct in6_addr daddr; /* destination address */
171 /* The sequence options start here */
172 /* PE data padded to 32bit alignment after seq. options */
173 };
174
175 union ip_vs_sync_conn {
176 struct ip_vs_sync_v4 v4;
177 struct ip_vs_sync_v6 v6;
178 };
179
180 /* Bits in Type field in above */
181 #define STYPE_INET6 0
182 #define STYPE_F_INET6 (1 << STYPE_INET6)
183
184 #define SVER_SHIFT 12 /* Shift to get version */
185 #define SVER_MASK 0x0fff /* Mask to strip version */
186
187 #define IPVS_OPT_SEQ_DATA 1
188 #define IPVS_OPT_PE_DATA 2
189 #define IPVS_OPT_PE_NAME 3
190 #define IPVS_OPT_PARAM 7
191
192 #define IPVS_OPT_F_SEQ_DATA (1 << (IPVS_OPT_SEQ_DATA-1))
193 #define IPVS_OPT_F_PE_DATA (1 << (IPVS_OPT_PE_DATA-1))
194 #define IPVS_OPT_F_PE_NAME (1 << (IPVS_OPT_PE_NAME-1))
195 #define IPVS_OPT_F_PARAM (1 << (IPVS_OPT_PARAM-1))
196
197 struct ip_vs_sync_thread_data {
198 struct task_struct *task;
199 struct netns_ipvs *ipvs;
200 struct socket *sock;
201 char *buf;
202 int id;
203 };
204
205 /* Version 0 definition of packet sizes */
206 #define SIMPLE_CONN_SIZE (sizeof(struct ip_vs_sync_conn_v0))
207 #define FULL_CONN_SIZE \
208 (sizeof(struct ip_vs_sync_conn_v0) + sizeof(struct ip_vs_sync_conn_options))
209
210
211 /*
212 The master mulitcasts messages (Datagrams) to the backup load balancers
213 in the following format.
214
215 Version 1:
216 Note, first byte should be Zero, so ver 0 receivers will drop the packet.
217
218 0 1 2 3
219 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
220 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
221 | 0 | SyncID | Size |
222 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
223 | Count Conns | Version | Reserved, set to Zero |
224 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
225 | |
226 | IPVS Sync Connection (1) |
227 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
228 | . |
229 ~ . ~
230 | . |
231 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
232 | |
233 | IPVS Sync Connection (n) |
234 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
235
236 Version 0 Header
237 0 1 2 3
238 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
239 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
240 | Count Conns | SyncID | Size |
241 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
242 | IPVS Sync Connection (1) |
243 */
244
245 #define SYNC_MESG_HEADER_LEN 4
246 #define MAX_CONNS_PER_SYNCBUFF 255 /* nr_conns in ip_vs_sync_mesg is 8 bit */
247
248 /* Version 0 header */
249 struct ip_vs_sync_mesg_v0 {
250 __u8 nr_conns;
251 __u8 syncid;
252 __be16 size;
253
254 /* ip_vs_sync_conn entries start here */
255 };
256
257 /* Version 1 header */
258 struct ip_vs_sync_mesg {
259 __u8 reserved; /* must be zero */
260 __u8 syncid;
261 __be16 size;
262 __u8 nr_conns;
263 __s8 version; /* SYNC_PROTO_VER */
264 __u16 spare;
265 /* ip_vs_sync_conn entries start here */
266 };
267
268 union ipvs_sockaddr {
269 struct sockaddr_in in;
270 struct sockaddr_in6 in6;
271 };
272
273 struct ip_vs_sync_buff {
274 struct list_head list;
275 unsigned long firstuse;
276
277 /* pointers for the message data */
278 struct ip_vs_sync_mesg *mesg;
279 unsigned char *head;
280 unsigned char *end;
281 };
282
283 /*
284 * Copy of struct ip_vs_seq
285 * From unaligned network order to aligned host order
286 */
ntoh_seq(struct ip_vs_seq * no,struct ip_vs_seq * ho)287 static void ntoh_seq(struct ip_vs_seq *no, struct ip_vs_seq *ho)
288 {
289 memset(ho, 0, sizeof(*ho));
290 ho->init_seq = get_unaligned_be32(&no->init_seq);
291 ho->delta = get_unaligned_be32(&no->delta);
292 ho->previous_delta = get_unaligned_be32(&no->previous_delta);
293 }
294
295 /*
296 * Copy of struct ip_vs_seq
297 * From Aligned host order to unaligned network order
298 */
hton_seq(struct ip_vs_seq * ho,struct ip_vs_seq * no)299 static void hton_seq(struct ip_vs_seq *ho, struct ip_vs_seq *no)
300 {
301 put_unaligned_be32(ho->init_seq, &no->init_seq);
302 put_unaligned_be32(ho->delta, &no->delta);
303 put_unaligned_be32(ho->previous_delta, &no->previous_delta);
304 }
305
306 static inline struct ip_vs_sync_buff *
sb_dequeue(struct netns_ipvs * ipvs,struct ipvs_master_sync_state * ms)307 sb_dequeue(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
308 {
309 struct ip_vs_sync_buff *sb;
310
311 spin_lock_bh(&ipvs->sync_lock);
312 if (list_empty(&ms->sync_queue)) {
313 sb = NULL;
314 __set_current_state(TASK_INTERRUPTIBLE);
315 } else {
316 sb = list_entry(ms->sync_queue.next, struct ip_vs_sync_buff,
317 list);
318 list_del(&sb->list);
319 ms->sync_queue_len--;
320 if (!ms->sync_queue_len)
321 ms->sync_queue_delay = 0;
322 }
323 spin_unlock_bh(&ipvs->sync_lock);
324
325 return sb;
326 }
327
328 /*
329 * Create a new sync buffer for Version 1 proto.
330 */
331 static inline struct ip_vs_sync_buff *
ip_vs_sync_buff_create(struct netns_ipvs * ipvs,unsigned int len)332 ip_vs_sync_buff_create(struct netns_ipvs *ipvs, unsigned int len)
333 {
334 struct ip_vs_sync_buff *sb;
335
336 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
337 return NULL;
338
339 len = max_t(unsigned int, len + sizeof(struct ip_vs_sync_mesg),
340 ipvs->mcfg.sync_maxlen);
341 sb->mesg = kmalloc(len, GFP_ATOMIC);
342 if (!sb->mesg) {
343 kfree(sb);
344 return NULL;
345 }
346 sb->mesg->reserved = 0; /* old nr_conns i.e. must be zero now */
347 sb->mesg->version = SYNC_PROTO_VER;
348 sb->mesg->syncid = ipvs->mcfg.syncid;
349 sb->mesg->size = htons(sizeof(struct ip_vs_sync_mesg));
350 sb->mesg->nr_conns = 0;
351 sb->mesg->spare = 0;
352 sb->head = (unsigned char *)sb->mesg + sizeof(struct ip_vs_sync_mesg);
353 sb->end = (unsigned char *)sb->mesg + len;
354
355 sb->firstuse = jiffies;
356 return sb;
357 }
358
ip_vs_sync_buff_release(struct ip_vs_sync_buff * sb)359 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
360 {
361 kfree(sb->mesg);
362 kfree(sb);
363 }
364
sb_queue_tail(struct netns_ipvs * ipvs,struct ipvs_master_sync_state * ms)365 static inline void sb_queue_tail(struct netns_ipvs *ipvs,
366 struct ipvs_master_sync_state *ms)
367 {
368 struct ip_vs_sync_buff *sb = ms->sync_buff;
369
370 spin_lock(&ipvs->sync_lock);
371 if (ipvs->sync_state & IP_VS_STATE_MASTER &&
372 ms->sync_queue_len < sysctl_sync_qlen_max(ipvs)) {
373 if (!ms->sync_queue_len)
374 schedule_delayed_work(&ms->master_wakeup_work,
375 max(IPVS_SYNC_SEND_DELAY, 1));
376 ms->sync_queue_len++;
377 list_add_tail(&sb->list, &ms->sync_queue);
378 if ((++ms->sync_queue_delay) == IPVS_SYNC_WAKEUP_RATE) {
379 int id = (int)(ms - ipvs->ms);
380
381 wake_up_process(ipvs->master_tinfo[id].task);
382 }
383 } else
384 ip_vs_sync_buff_release(sb);
385 spin_unlock(&ipvs->sync_lock);
386 }
387
388 /*
389 * Get the current sync buffer if it has been created for more
390 * than the specified time or the specified time is zero.
391 */
392 static inline struct ip_vs_sync_buff *
get_curr_sync_buff(struct netns_ipvs * ipvs,struct ipvs_master_sync_state * ms,unsigned long time)393 get_curr_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms,
394 unsigned long time)
395 {
396 struct ip_vs_sync_buff *sb;
397
398 spin_lock_bh(&ipvs->sync_buff_lock);
399 sb = ms->sync_buff;
400 if (sb && time_after_eq(jiffies - sb->firstuse, time)) {
401 ms->sync_buff = NULL;
402 __set_current_state(TASK_RUNNING);
403 } else
404 sb = NULL;
405 spin_unlock_bh(&ipvs->sync_buff_lock);
406 return sb;
407 }
408
409 static inline int
select_master_thread_id(struct netns_ipvs * ipvs,struct ip_vs_conn * cp)410 select_master_thread_id(struct netns_ipvs *ipvs, struct ip_vs_conn *cp)
411 {
412 return ((long) cp >> (1 + ilog2(sizeof(*cp)))) & ipvs->threads_mask;
413 }
414
415 /*
416 * Create a new sync buffer for Version 0 proto.
417 */
418 static inline struct ip_vs_sync_buff *
ip_vs_sync_buff_create_v0(struct netns_ipvs * ipvs,unsigned int len)419 ip_vs_sync_buff_create_v0(struct netns_ipvs *ipvs, unsigned int len)
420 {
421 struct ip_vs_sync_buff *sb;
422 struct ip_vs_sync_mesg_v0 *mesg;
423
424 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
425 return NULL;
426
427 len = max_t(unsigned int, len + sizeof(struct ip_vs_sync_mesg_v0),
428 ipvs->mcfg.sync_maxlen);
429 sb->mesg = kmalloc(len, GFP_ATOMIC);
430 if (!sb->mesg) {
431 kfree(sb);
432 return NULL;
433 }
434 mesg = (struct ip_vs_sync_mesg_v0 *)sb->mesg;
435 mesg->nr_conns = 0;
436 mesg->syncid = ipvs->mcfg.syncid;
437 mesg->size = htons(sizeof(struct ip_vs_sync_mesg_v0));
438 sb->head = (unsigned char *)mesg + sizeof(struct ip_vs_sync_mesg_v0);
439 sb->end = (unsigned char *)mesg + len;
440 sb->firstuse = jiffies;
441 return sb;
442 }
443
444 /* Check if connection is controlled by persistence */
in_persistence(struct ip_vs_conn * cp)445 static inline bool in_persistence(struct ip_vs_conn *cp)
446 {
447 for (cp = cp->control; cp; cp = cp->control) {
448 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
449 return true;
450 }
451 return false;
452 }
453
454 /* Check if conn should be synced.
455 * pkts: conn packets, use sysctl_sync_threshold to avoid packet check
456 * - (1) sync_refresh_period: reduce sync rate. Additionally, retry
457 * sync_retries times with period of sync_refresh_period/8
458 * - (2) if both sync_refresh_period and sync_period are 0 send sync only
459 * for state changes or only once when pkts matches sync_threshold
460 * - (3) templates: rate can be reduced only with sync_refresh_period or
461 * with (2)
462 */
ip_vs_sync_conn_needed(struct netns_ipvs * ipvs,struct ip_vs_conn * cp,int pkts)463 static int ip_vs_sync_conn_needed(struct netns_ipvs *ipvs,
464 struct ip_vs_conn *cp, int pkts)
465 {
466 unsigned long orig = READ_ONCE(cp->sync_endtime);
467 unsigned long now = jiffies;
468 unsigned long n = (now + cp->timeout) & ~3UL;
469 unsigned int sync_refresh_period;
470 int sync_period;
471 int force;
472
473 /* Check if we sync in current state */
474 if (unlikely(cp->flags & IP_VS_CONN_F_TEMPLATE))
475 force = 0;
476 else if (unlikely(sysctl_sync_persist_mode(ipvs) && in_persistence(cp)))
477 return 0;
478 else if (likely(cp->protocol == IPPROTO_TCP)) {
479 if (!((1 << cp->state) &
480 ((1 << IP_VS_TCP_S_ESTABLISHED) |
481 (1 << IP_VS_TCP_S_FIN_WAIT) |
482 (1 << IP_VS_TCP_S_CLOSE) |
483 (1 << IP_VS_TCP_S_CLOSE_WAIT) |
484 (1 << IP_VS_TCP_S_TIME_WAIT))))
485 return 0;
486 force = cp->state != cp->old_state;
487 if (force && cp->state != IP_VS_TCP_S_ESTABLISHED)
488 goto set;
489 } else if (unlikely(cp->protocol == IPPROTO_SCTP)) {
490 if (!((1 << cp->state) &
491 ((1 << IP_VS_SCTP_S_ESTABLISHED) |
492 (1 << IP_VS_SCTP_S_SHUTDOWN_SENT) |
493 (1 << IP_VS_SCTP_S_SHUTDOWN_RECEIVED) |
494 (1 << IP_VS_SCTP_S_SHUTDOWN_ACK_SENT) |
495 (1 << IP_VS_SCTP_S_CLOSED))))
496 return 0;
497 force = cp->state != cp->old_state;
498 if (force && cp->state != IP_VS_SCTP_S_ESTABLISHED)
499 goto set;
500 } else {
501 /* UDP or another protocol with single state */
502 force = 0;
503 }
504
505 sync_refresh_period = sysctl_sync_refresh_period(ipvs);
506 if (sync_refresh_period > 0) {
507 long diff = n - orig;
508 long min_diff = max(cp->timeout >> 1, 10UL * HZ);
509
510 /* Avoid sync if difference is below sync_refresh_period
511 * and below the half timeout.
512 */
513 if (abs(diff) < min_t(long, sync_refresh_period, min_diff)) {
514 int retries = orig & 3;
515
516 if (retries >= sysctl_sync_retries(ipvs))
517 return 0;
518 if (time_before(now, orig - cp->timeout +
519 (sync_refresh_period >> 3)))
520 return 0;
521 n |= retries + 1;
522 }
523 }
524 sync_period = sysctl_sync_period(ipvs);
525 if (sync_period > 0) {
526 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE) &&
527 pkts % sync_period != sysctl_sync_threshold(ipvs))
528 return 0;
529 } else if (!sync_refresh_period &&
530 pkts != sysctl_sync_threshold(ipvs))
531 return 0;
532
533 set:
534 cp->old_state = cp->state;
535 n = cmpxchg(&cp->sync_endtime, orig, n);
536 return n == orig || force;
537 }
538
539 /*
540 * Version 0 , could be switched in by sys_ctl.
541 * Add an ip_vs_conn information into the current sync_buff.
542 */
ip_vs_sync_conn_v0(struct netns_ipvs * ipvs,struct ip_vs_conn * cp,int pkts)543 static void ip_vs_sync_conn_v0(struct netns_ipvs *ipvs, struct ip_vs_conn *cp,
544 int pkts)
545 {
546 struct ip_vs_sync_mesg_v0 *m;
547 struct ip_vs_sync_conn_v0 *s;
548 struct ip_vs_sync_buff *buff;
549 struct ipvs_master_sync_state *ms;
550 int id;
551 unsigned int len;
552
553 if (unlikely(cp->af != AF_INET))
554 return;
555 /* Do not sync ONE PACKET */
556 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
557 return;
558
559 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
560 return;
561
562 spin_lock_bh(&ipvs->sync_buff_lock);
563 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
564 spin_unlock_bh(&ipvs->sync_buff_lock);
565 return;
566 }
567
568 id = select_master_thread_id(ipvs, cp);
569 ms = &ipvs->ms[id];
570 buff = ms->sync_buff;
571 len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
572 SIMPLE_CONN_SIZE;
573 if (buff) {
574 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
575 /* Send buffer if it is for v1 */
576 if (buff->head + len > buff->end || !m->nr_conns) {
577 sb_queue_tail(ipvs, ms);
578 ms->sync_buff = NULL;
579 buff = NULL;
580 }
581 }
582 if (!buff) {
583 buff = ip_vs_sync_buff_create_v0(ipvs, len);
584 if (!buff) {
585 spin_unlock_bh(&ipvs->sync_buff_lock);
586 pr_err("ip_vs_sync_buff_create failed.\n");
587 return;
588 }
589 ms->sync_buff = buff;
590 }
591
592 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
593 s = (struct ip_vs_sync_conn_v0 *) buff->head;
594
595 /* copy members */
596 s->reserved = 0;
597 s->protocol = cp->protocol;
598 s->cport = cp->cport;
599 s->vport = cp->vport;
600 s->dport = cp->dport;
601 s->caddr = cp->caddr.ip;
602 s->vaddr = cp->vaddr.ip;
603 s->daddr = cp->daddr.ip;
604 s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
605 s->state = htons(cp->state);
606 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
607 struct ip_vs_sync_conn_options *opt =
608 (struct ip_vs_sync_conn_options *)&s[1];
609 memcpy(opt, &cp->in_seq, sizeof(*opt));
610 }
611
612 m->nr_conns++;
613 m->size = htons(ntohs(m->size) + len);
614 buff->head += len;
615 spin_unlock_bh(&ipvs->sync_buff_lock);
616
617 /* synchronize its controller if it has */
618 cp = cp->control;
619 if (cp) {
620 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
621 pkts = atomic_add_return(1, &cp->in_pkts);
622 else
623 pkts = sysctl_sync_threshold(ipvs);
624 ip_vs_sync_conn(ipvs, cp, pkts);
625 }
626 }
627
628 /*
629 * Add an ip_vs_conn information into the current sync_buff.
630 * Called by ip_vs_in.
631 * Sending Version 1 messages
632 */
ip_vs_sync_conn(struct netns_ipvs * ipvs,struct ip_vs_conn * cp,int pkts)633 void ip_vs_sync_conn(struct netns_ipvs *ipvs, struct ip_vs_conn *cp, int pkts)
634 {
635 struct ip_vs_sync_mesg *m;
636 union ip_vs_sync_conn *s;
637 struct ip_vs_sync_buff *buff;
638 struct ipvs_master_sync_state *ms;
639 int id;
640 __u8 *p;
641 unsigned int len, pe_name_len, pad;
642
643 /* Handle old version of the protocol */
644 if (sysctl_sync_ver(ipvs) == 0) {
645 ip_vs_sync_conn_v0(ipvs, cp, pkts);
646 return;
647 }
648 /* Do not sync ONE PACKET */
649 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
650 goto control;
651 sloop:
652 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
653 goto control;
654
655 /* Sanity checks */
656 pe_name_len = 0;
657 if (cp->pe_data_len) {
658 if (!cp->pe_data || !cp->dest) {
659 IP_VS_ERR_RL("SYNC, connection pe_data invalid\n");
660 return;
661 }
662 pe_name_len = strnlen(cp->pe->name, IP_VS_PENAME_MAXLEN);
663 }
664
665 spin_lock_bh(&ipvs->sync_buff_lock);
666 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
667 spin_unlock_bh(&ipvs->sync_buff_lock);
668 return;
669 }
670
671 id = select_master_thread_id(ipvs, cp);
672 ms = &ipvs->ms[id];
673
674 #ifdef CONFIG_IP_VS_IPV6
675 if (cp->af == AF_INET6)
676 len = sizeof(struct ip_vs_sync_v6);
677 else
678 #endif
679 len = sizeof(struct ip_vs_sync_v4);
680
681 if (cp->flags & IP_VS_CONN_F_SEQ_MASK)
682 len += sizeof(struct ip_vs_sync_conn_options) + 2;
683
684 if (cp->pe_data_len)
685 len += cp->pe_data_len + 2; /* + Param hdr field */
686 if (pe_name_len)
687 len += pe_name_len + 2;
688
689 /* check if there is a space for this one */
690 pad = 0;
691 buff = ms->sync_buff;
692 if (buff) {
693 m = buff->mesg;
694 pad = (4 - (size_t) buff->head) & 3;
695 /* Send buffer if it is for v0 */
696 if (buff->head + len + pad > buff->end || m->reserved) {
697 sb_queue_tail(ipvs, ms);
698 ms->sync_buff = NULL;
699 buff = NULL;
700 pad = 0;
701 }
702 }
703
704 if (!buff) {
705 buff = ip_vs_sync_buff_create(ipvs, len);
706 if (!buff) {
707 spin_unlock_bh(&ipvs->sync_buff_lock);
708 pr_err("ip_vs_sync_buff_create failed.\n");
709 return;
710 }
711 ms->sync_buff = buff;
712 m = buff->mesg;
713 }
714
715 p = buff->head;
716 buff->head += pad + len;
717 m->size = htons(ntohs(m->size) + pad + len);
718 /* Add ev. padding from prev. sync_conn */
719 while (pad--)
720 *(p++) = 0;
721
722 s = (union ip_vs_sync_conn *)p;
723
724 /* Set message type & copy members */
725 s->v4.type = (cp->af == AF_INET6 ? STYPE_F_INET6 : 0);
726 s->v4.ver_size = htons(len & SVER_MASK); /* Version 0 */
727 s->v4.flags = htonl(cp->flags & ~IP_VS_CONN_F_HASHED);
728 s->v4.state = htons(cp->state);
729 s->v4.protocol = cp->protocol;
730 s->v4.cport = cp->cport;
731 s->v4.vport = cp->vport;
732 s->v4.dport = cp->dport;
733 s->v4.fwmark = htonl(cp->fwmark);
734 s->v4.timeout = htonl(cp->timeout / HZ);
735 m->nr_conns++;
736
737 #ifdef CONFIG_IP_VS_IPV6
738 if (cp->af == AF_INET6) {
739 p += sizeof(struct ip_vs_sync_v6);
740 s->v6.caddr = cp->caddr.in6;
741 s->v6.vaddr = cp->vaddr.in6;
742 s->v6.daddr = cp->daddr.in6;
743 } else
744 #endif
745 {
746 p += sizeof(struct ip_vs_sync_v4); /* options ptr */
747 s->v4.caddr = cp->caddr.ip;
748 s->v4.vaddr = cp->vaddr.ip;
749 s->v4.daddr = cp->daddr.ip;
750 }
751 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
752 *(p++) = IPVS_OPT_SEQ_DATA;
753 *(p++) = sizeof(struct ip_vs_sync_conn_options);
754 hton_seq((struct ip_vs_seq *)p, &cp->in_seq);
755 p += sizeof(struct ip_vs_seq);
756 hton_seq((struct ip_vs_seq *)p, &cp->out_seq);
757 p += sizeof(struct ip_vs_seq);
758 }
759 /* Handle pe data */
760 if (cp->pe_data_len && cp->pe_data) {
761 *(p++) = IPVS_OPT_PE_DATA;
762 *(p++) = cp->pe_data_len;
763 memcpy(p, cp->pe_data, cp->pe_data_len);
764 p += cp->pe_data_len;
765 if (pe_name_len) {
766 /* Add PE_NAME */
767 *(p++) = IPVS_OPT_PE_NAME;
768 *(p++) = pe_name_len;
769 memcpy(p, cp->pe->name, pe_name_len);
770 p += pe_name_len;
771 }
772 }
773
774 spin_unlock_bh(&ipvs->sync_buff_lock);
775
776 control:
777 /* synchronize its controller if it has */
778 cp = cp->control;
779 if (!cp)
780 return;
781 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
782 pkts = atomic_add_return(1, &cp->in_pkts);
783 else
784 pkts = sysctl_sync_threshold(ipvs);
785 goto sloop;
786 }
787
788 /*
789 * fill_param used by version 1
790 */
791 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)792 ip_vs_conn_fill_param_sync(struct netns_ipvs *ipvs, int af, union ip_vs_sync_conn *sc,
793 struct ip_vs_conn_param *p,
794 __u8 *pe_data, unsigned int pe_data_len,
795 __u8 *pe_name, unsigned int pe_name_len)
796 {
797 #ifdef CONFIG_IP_VS_IPV6
798 if (af == AF_INET6)
799 ip_vs_conn_fill_param(ipvs, af, sc->v6.protocol,
800 (const union nf_inet_addr *)&sc->v6.caddr,
801 sc->v6.cport,
802 (const union nf_inet_addr *)&sc->v6.vaddr,
803 sc->v6.vport, p);
804 else
805 #endif
806 ip_vs_conn_fill_param(ipvs, af, sc->v4.protocol,
807 (const union nf_inet_addr *)&sc->v4.caddr,
808 sc->v4.cport,
809 (const union nf_inet_addr *)&sc->v4.vaddr,
810 sc->v4.vport, p);
811 /* Handle pe data */
812 if (pe_data_len) {
813 if (pe_name_len) {
814 char buff[IP_VS_PENAME_MAXLEN+1];
815
816 memcpy(buff, pe_name, pe_name_len);
817 buff[pe_name_len]=0;
818 p->pe = __ip_vs_pe_getbyname(buff);
819 if (!p->pe) {
820 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n",
821 buff);
822 return 1;
823 }
824 } else {
825 IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n");
826 return 1;
827 }
828
829 p->pe_data = kmemdup(pe_data, pe_data_len, GFP_ATOMIC);
830 if (!p->pe_data) {
831 module_put(p->pe->module);
832 return -ENOMEM;
833 }
834 p->pe_data_len = pe_data_len;
835 }
836 return 0;
837 }
838
839 /*
840 * Connection Add / Update.
841 * Common for version 0 and 1 reception of backup sync_conns.
842 * Param: ...
843 * timeout is in sec.
844 */
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)845 static void ip_vs_proc_conn(struct netns_ipvs *ipvs, struct ip_vs_conn_param *param,
846 unsigned int flags, unsigned int state,
847 unsigned int protocol, unsigned int type,
848 const union nf_inet_addr *daddr, __be16 dport,
849 unsigned long timeout, __u32 fwmark,
850 struct ip_vs_sync_conn_options *opt)
851 {
852 struct ip_vs_dest *dest;
853 struct ip_vs_conn *cp;
854
855 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
856 cp = ip_vs_conn_in_get(param);
857 if (cp && ((cp->dport != dport) ||
858 !ip_vs_addr_equal(cp->daf, &cp->daddr, daddr))) {
859 if (!(flags & IP_VS_CONN_F_INACTIVE)) {
860 ip_vs_conn_expire_now(cp);
861 __ip_vs_conn_put(cp);
862 cp = NULL;
863 } else {
864 /* This is the expiration message for the
865 * connection that was already replaced, so we
866 * just ignore it.
867 */
868 __ip_vs_conn_put(cp);
869 kfree(param->pe_data);
870 return;
871 }
872 }
873 } else {
874 cp = ip_vs_ct_in_get(param);
875 }
876
877 if (cp) {
878 /* Free pe_data */
879 kfree(param->pe_data);
880
881 dest = cp->dest;
882 spin_lock_bh(&cp->lock);
883 if ((cp->flags ^ flags) & IP_VS_CONN_F_INACTIVE &&
884 !(flags & IP_VS_CONN_F_TEMPLATE) && dest) {
885 if (flags & IP_VS_CONN_F_INACTIVE) {
886 atomic_dec(&dest->activeconns);
887 atomic_inc(&dest->inactconns);
888 } else {
889 atomic_inc(&dest->activeconns);
890 atomic_dec(&dest->inactconns);
891 }
892 }
893 flags &= IP_VS_CONN_F_BACKUP_UPD_MASK;
894 flags |= cp->flags & ~IP_VS_CONN_F_BACKUP_UPD_MASK;
895 cp->flags = flags;
896 spin_unlock_bh(&cp->lock);
897 if (!dest)
898 ip_vs_try_bind_dest(cp);
899 } else {
900 /*
901 * Find the appropriate destination for the connection.
902 * If it is not found the connection will remain unbound
903 * but still handled.
904 */
905 rcu_read_lock();
906 /* This function is only invoked by the synchronization
907 * code. We do not currently support heterogeneous pools
908 * with synchronization, so we can make the assumption that
909 * the svc_af is the same as the dest_af
910 */
911 dest = ip_vs_find_dest(ipvs, type, type, daddr, dport,
912 param->vaddr, param->vport, protocol,
913 fwmark, flags);
914
915 cp = ip_vs_conn_new(param, type, daddr, dport, flags, dest,
916 fwmark);
917 rcu_read_unlock();
918 if (!cp) {
919 kfree(param->pe_data);
920 IP_VS_DBG(2, "BACKUP, add new conn. failed\n");
921 return;
922 }
923 if (!(flags & IP_VS_CONN_F_TEMPLATE))
924 kfree(param->pe_data);
925 }
926
927 if (opt) {
928 cp->in_seq = opt->in_seq;
929 cp->out_seq = opt->out_seq;
930 }
931 atomic_set(&cp->in_pkts, sysctl_sync_threshold(ipvs));
932 cp->state = state;
933 cp->old_state = cp->state;
934 /*
935 * For Ver 0 messages style
936 * - Not possible to recover the right timeout for templates
937 * - can not find the right fwmark
938 * virtual service. If needed, we can do it for
939 * non-fwmark persistent services.
940 * Ver 1 messages style.
941 * - No problem.
942 */
943 if (timeout) {
944 if (timeout > MAX_SCHEDULE_TIMEOUT / HZ)
945 timeout = MAX_SCHEDULE_TIMEOUT / HZ;
946 cp->timeout = timeout*HZ;
947 } else {
948 struct ip_vs_proto_data *pd;
949
950 pd = ip_vs_proto_data_get(ipvs, protocol);
951 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pd && pd->timeout_table)
952 cp->timeout = pd->timeout_table[state];
953 else
954 cp->timeout = (3*60*HZ);
955 }
956 ip_vs_conn_put(cp);
957 }
958
959 /*
960 * Process received multicast message for Version 0
961 */
ip_vs_process_message_v0(struct netns_ipvs * ipvs,const char * buffer,const size_t buflen)962 static void ip_vs_process_message_v0(struct netns_ipvs *ipvs, const char *buffer,
963 const size_t buflen)
964 {
965 struct ip_vs_sync_mesg_v0 *m = (struct ip_vs_sync_mesg_v0 *)buffer;
966 struct ip_vs_sync_conn_v0 *s;
967 struct ip_vs_sync_conn_options *opt;
968 struct ip_vs_protocol *pp;
969 struct ip_vs_conn_param param;
970 char *p;
971 int i;
972
973 p = (char *)buffer + sizeof(struct ip_vs_sync_mesg_v0);
974 for (i=0; i<m->nr_conns; i++) {
975 unsigned int flags, state;
976
977 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
978 IP_VS_ERR_RL("BACKUP v0, bogus conn\n");
979 return;
980 }
981 s = (struct ip_vs_sync_conn_v0 *) p;
982 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
983 flags &= ~IP_VS_CONN_F_HASHED;
984 if (flags & IP_VS_CONN_F_SEQ_MASK) {
985 opt = (struct ip_vs_sync_conn_options *)&s[1];
986 p += FULL_CONN_SIZE;
987 if (p > buffer+buflen) {
988 IP_VS_ERR_RL("BACKUP v0, Dropping buffer bogus conn options\n");
989 return;
990 }
991 } else {
992 opt = NULL;
993 p += SIMPLE_CONN_SIZE;
994 }
995
996 state = ntohs(s->state);
997 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
998 pp = ip_vs_proto_get(s->protocol);
999 if (!pp) {
1000 IP_VS_DBG(2, "BACKUP v0, Unsupported protocol %u\n",
1001 s->protocol);
1002 continue;
1003 }
1004 if (state >= pp->num_states) {
1005 IP_VS_DBG(2, "BACKUP v0, Invalid %s state %u\n",
1006 pp->name, state);
1007 continue;
1008 }
1009 } else {
1010 if (state >= IP_VS_CTPL_S_LAST)
1011 IP_VS_DBG(7, "BACKUP v0, Invalid tpl state %u\n",
1012 state);
1013 }
1014
1015 ip_vs_conn_fill_param(ipvs, AF_INET, s->protocol,
1016 (const union nf_inet_addr *)&s->caddr,
1017 s->cport,
1018 (const union nf_inet_addr *)&s->vaddr,
1019 s->vport, ¶m);
1020
1021 /* Send timeout as Zero */
1022 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->protocol, AF_INET,
1023 (union nf_inet_addr *)&s->daddr, s->dport,
1024 0, 0, opt);
1025 }
1026 }
1027
1028 /*
1029 * Handle options
1030 */
ip_vs_proc_seqopt(__u8 * p,unsigned int plen,__u32 * opt_flags,struct ip_vs_sync_conn_options * opt)1031 static inline int ip_vs_proc_seqopt(__u8 *p, unsigned int plen,
1032 __u32 *opt_flags,
1033 struct ip_vs_sync_conn_options *opt)
1034 {
1035 struct ip_vs_sync_conn_options *topt;
1036
1037 topt = (struct ip_vs_sync_conn_options *)p;
1038
1039 if (plen != sizeof(struct ip_vs_sync_conn_options)) {
1040 IP_VS_DBG(2, "BACKUP, bogus conn options length\n");
1041 return -EINVAL;
1042 }
1043 if (*opt_flags & IPVS_OPT_F_SEQ_DATA) {
1044 IP_VS_DBG(2, "BACKUP, conn options found twice\n");
1045 return -EINVAL;
1046 }
1047 ntoh_seq(&topt->in_seq, &opt->in_seq);
1048 ntoh_seq(&topt->out_seq, &opt->out_seq);
1049 *opt_flags |= IPVS_OPT_F_SEQ_DATA;
1050 return 0;
1051 }
1052
ip_vs_proc_str(__u8 * p,unsigned int plen,unsigned int * data_len,__u8 ** data,unsigned int maxlen,__u32 * opt_flags,__u32 flag)1053 static int ip_vs_proc_str(__u8 *p, unsigned int plen, unsigned int *data_len,
1054 __u8 **data, unsigned int maxlen,
1055 __u32 *opt_flags, __u32 flag)
1056 {
1057 if (plen > maxlen) {
1058 IP_VS_DBG(2, "BACKUP, bogus par.data len > %d\n", maxlen);
1059 return -EINVAL;
1060 }
1061 if (*opt_flags & flag) {
1062 IP_VS_DBG(2, "BACKUP, Par.data found twice 0x%x\n", flag);
1063 return -EINVAL;
1064 }
1065 *data_len = plen;
1066 *data = p;
1067 *opt_flags |= flag;
1068 return 0;
1069 }
1070 /*
1071 * Process a Version 1 sync. connection
1072 */
ip_vs_proc_sync_conn(struct netns_ipvs * ipvs,__u8 * p,__u8 * msg_end)1073 static inline int ip_vs_proc_sync_conn(struct netns_ipvs *ipvs, __u8 *p, __u8 *msg_end)
1074 {
1075 struct ip_vs_sync_conn_options opt;
1076 union ip_vs_sync_conn *s;
1077 struct ip_vs_protocol *pp;
1078 struct ip_vs_conn_param param;
1079 __u32 flags;
1080 unsigned int af, state, pe_data_len=0, pe_name_len=0;
1081 __u8 *pe_data=NULL, *pe_name=NULL;
1082 __u32 opt_flags=0;
1083 int retc=0;
1084
1085 s = (union ip_vs_sync_conn *) p;
1086
1087 if (s->v6.type & STYPE_F_INET6) {
1088 #ifdef CONFIG_IP_VS_IPV6
1089 af = AF_INET6;
1090 p += sizeof(struct ip_vs_sync_v6);
1091 #else
1092 IP_VS_DBG(3,"BACKUP, IPv6 msg received, and IPVS is not compiled for IPv6\n");
1093 retc = 10;
1094 goto out;
1095 #endif
1096 } else if (!s->v4.type) {
1097 af = AF_INET;
1098 p += sizeof(struct ip_vs_sync_v4);
1099 } else {
1100 return -10;
1101 }
1102 if (p > msg_end)
1103 return -20;
1104
1105 /* Process optional params check Type & Len. */
1106 while (p < msg_end) {
1107 int ptype;
1108 int plen;
1109
1110 if (p+2 > msg_end)
1111 return -30;
1112 ptype = *(p++);
1113 plen = *(p++);
1114
1115 if (!plen || ((p + plen) > msg_end))
1116 return -40;
1117 /* Handle seq option p = param data */
1118 switch (ptype & ~IPVS_OPT_F_PARAM) {
1119 case IPVS_OPT_SEQ_DATA:
1120 if (ip_vs_proc_seqopt(p, plen, &opt_flags, &opt))
1121 return -50;
1122 break;
1123
1124 case IPVS_OPT_PE_DATA:
1125 if (ip_vs_proc_str(p, plen, &pe_data_len, &pe_data,
1126 IP_VS_PEDATA_MAXLEN, &opt_flags,
1127 IPVS_OPT_F_PE_DATA))
1128 return -60;
1129 break;
1130
1131 case IPVS_OPT_PE_NAME:
1132 if (ip_vs_proc_str(p, plen,&pe_name_len, &pe_name,
1133 IP_VS_PENAME_MAXLEN, &opt_flags,
1134 IPVS_OPT_F_PE_NAME))
1135 return -70;
1136 break;
1137
1138 default:
1139 /* Param data mandatory ? */
1140 if (!(ptype & IPVS_OPT_F_PARAM)) {
1141 IP_VS_DBG(3, "BACKUP, Unknown mandatory param %d found\n",
1142 ptype & ~IPVS_OPT_F_PARAM);
1143 retc = 20;
1144 goto out;
1145 }
1146 }
1147 p += plen; /* Next option */
1148 }
1149
1150 /* Get flags and Mask off unsupported */
1151 flags = ntohl(s->v4.flags) & IP_VS_CONN_F_BACKUP_MASK;
1152 flags |= IP_VS_CONN_F_SYNC;
1153 state = ntohs(s->v4.state);
1154
1155 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
1156 pp = ip_vs_proto_get(s->v4.protocol);
1157 if (!pp) {
1158 IP_VS_DBG(3,"BACKUP, Unsupported protocol %u\n",
1159 s->v4.protocol);
1160 retc = 30;
1161 goto out;
1162 }
1163 if (state >= pp->num_states) {
1164 IP_VS_DBG(3, "BACKUP, Invalid %s state %u\n",
1165 pp->name, state);
1166 retc = 40;
1167 goto out;
1168 }
1169 } else {
1170 if (state >= IP_VS_CTPL_S_LAST)
1171 IP_VS_DBG(7, "BACKUP, Invalid tpl state %u\n",
1172 state);
1173 }
1174 if (ip_vs_conn_fill_param_sync(ipvs, af, s, ¶m, pe_data,
1175 pe_data_len, pe_name, pe_name_len)) {
1176 retc = 50;
1177 goto out;
1178 }
1179 /* If only IPv4, just silent skip IPv6 */
1180 if (af == AF_INET)
1181 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->v4.protocol, af,
1182 (union nf_inet_addr *)&s->v4.daddr, s->v4.dport,
1183 ntohl(s->v4.timeout), ntohl(s->v4.fwmark),
1184 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1185 );
1186 #ifdef CONFIG_IP_VS_IPV6
1187 else
1188 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->v6.protocol, af,
1189 (union nf_inet_addr *)&s->v6.daddr, s->v6.dport,
1190 ntohl(s->v6.timeout), ntohl(s->v6.fwmark),
1191 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1192 );
1193 #endif
1194 ip_vs_pe_put(param.pe);
1195 return 0;
1196 /* Error exit */
1197 out:
1198 IP_VS_DBG(2, "BACKUP, Single msg dropped err:%d\n", retc);
1199 return retc;
1200
1201 }
1202 /*
1203 * Process received multicast message and create the corresponding
1204 * ip_vs_conn entries.
1205 * Handles Version 0 & 1
1206 */
ip_vs_process_message(struct netns_ipvs * ipvs,__u8 * buffer,const size_t buflen)1207 static void ip_vs_process_message(struct netns_ipvs *ipvs, __u8 *buffer,
1208 const size_t buflen)
1209 {
1210 struct ip_vs_sync_mesg *m2 = (struct ip_vs_sync_mesg *)buffer;
1211 __u8 *p, *msg_end;
1212 int i, nr_conns;
1213
1214 if (buflen < sizeof(struct ip_vs_sync_mesg_v0)) {
1215 IP_VS_DBG(2, "BACKUP, message header too short\n");
1216 return;
1217 }
1218
1219 if (buflen != ntohs(m2->size)) {
1220 IP_VS_DBG(2, "BACKUP, bogus message size\n");
1221 return;
1222 }
1223 /* SyncID sanity check */
1224 if (ipvs->bcfg.syncid != 0 && m2->syncid != ipvs->bcfg.syncid) {
1225 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2->syncid);
1226 return;
1227 }
1228 /* Handle version 1 message */
1229 if ((m2->version == SYNC_PROTO_VER) && (m2->reserved == 0)
1230 && (m2->spare == 0)) {
1231
1232 msg_end = buffer + sizeof(struct ip_vs_sync_mesg);
1233 nr_conns = m2->nr_conns;
1234
1235 for (i=0; i<nr_conns; i++) {
1236 union ip_vs_sync_conn *s;
1237 unsigned int size;
1238 int retc;
1239
1240 p = msg_end;
1241 if (p + sizeof(s->v4) > buffer+buflen) {
1242 IP_VS_ERR_RL("BACKUP, Dropping buffer, to small\n");
1243 return;
1244 }
1245 s = (union ip_vs_sync_conn *)p;
1246 size = ntohs(s->v4.ver_size) & SVER_MASK;
1247 msg_end = p + size;
1248 /* Basic sanity checks */
1249 if (msg_end > buffer+buflen) {
1250 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n");
1251 return;
1252 }
1253 if (ntohs(s->v4.ver_size) >> SVER_SHIFT) {
1254 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n",
1255 ntohs(s->v4.ver_size) >> SVER_SHIFT);
1256 return;
1257 }
1258 /* Process a single sync_conn */
1259 retc = ip_vs_proc_sync_conn(ipvs, p, msg_end);
1260 if (retc < 0) {
1261 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n",
1262 retc);
1263 return;
1264 }
1265 /* Make sure we have 32 bit alignment */
1266 msg_end = p + ((size + 3) & ~3);
1267 }
1268 } else {
1269 /* Old type of message */
1270 ip_vs_process_message_v0(ipvs, buffer, buflen);
1271 return;
1272 }
1273 }
1274
1275
1276 /*
1277 * Setup sndbuf (mode=1) or rcvbuf (mode=0)
1278 */
set_sock_size(struct sock * sk,int mode,int val)1279 static void set_sock_size(struct sock *sk, int mode, int val)
1280 {
1281 /* setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &val, sizeof(val)); */
1282 /* setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)); */
1283 lock_sock(sk);
1284 if (mode) {
1285 val = clamp_t(int, val, (SOCK_MIN_SNDBUF + 1) / 2,
1286 sysctl_wmem_max);
1287 sk->sk_sndbuf = val * 2;
1288 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1289 } else {
1290 val = clamp_t(int, val, (SOCK_MIN_RCVBUF + 1) / 2,
1291 sysctl_rmem_max);
1292 sk->sk_rcvbuf = val * 2;
1293 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1294 }
1295 release_sock(sk);
1296 }
1297
1298 /*
1299 * Setup loopback of outgoing multicasts on a sending socket
1300 */
set_mcast_loop(struct sock * sk,u_char loop)1301 static void set_mcast_loop(struct sock *sk, u_char loop)
1302 {
1303 struct inet_sock *inet = inet_sk(sk);
1304
1305 /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
1306 lock_sock(sk);
1307 inet->mc_loop = loop ? 1 : 0;
1308 #ifdef CONFIG_IP_VS_IPV6
1309 if (sk->sk_family == AF_INET6) {
1310 struct ipv6_pinfo *np = inet6_sk(sk);
1311
1312 /* IPV6_MULTICAST_LOOP */
1313 np->mc_loop = loop ? 1 : 0;
1314 }
1315 #endif
1316 release_sock(sk);
1317 }
1318
1319 /*
1320 * Specify TTL for outgoing multicasts on a sending socket
1321 */
set_mcast_ttl(struct sock * sk,u_char ttl)1322 static void set_mcast_ttl(struct sock *sk, u_char ttl)
1323 {
1324 struct inet_sock *inet = inet_sk(sk);
1325
1326 /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
1327 lock_sock(sk);
1328 inet->mc_ttl = ttl;
1329 #ifdef CONFIG_IP_VS_IPV6
1330 if (sk->sk_family == AF_INET6) {
1331 struct ipv6_pinfo *np = inet6_sk(sk);
1332
1333 /* IPV6_MULTICAST_HOPS */
1334 np->mcast_hops = ttl;
1335 }
1336 #endif
1337 release_sock(sk);
1338 }
1339
1340 /* Control fragmentation of messages */
set_mcast_pmtudisc(struct sock * sk,int val)1341 static void set_mcast_pmtudisc(struct sock *sk, int val)
1342 {
1343 struct inet_sock *inet = inet_sk(sk);
1344
1345 /* setsockopt(sock, SOL_IP, IP_MTU_DISCOVER, &val, sizeof(val)); */
1346 lock_sock(sk);
1347 inet->pmtudisc = val;
1348 #ifdef CONFIG_IP_VS_IPV6
1349 if (sk->sk_family == AF_INET6) {
1350 struct ipv6_pinfo *np = inet6_sk(sk);
1351
1352 /* IPV6_MTU_DISCOVER */
1353 np->pmtudisc = val;
1354 }
1355 #endif
1356 release_sock(sk);
1357 }
1358
1359 /*
1360 * Specifiy default interface for outgoing multicasts
1361 */
set_mcast_if(struct sock * sk,struct net_device * dev)1362 static int set_mcast_if(struct sock *sk, struct net_device *dev)
1363 {
1364 struct inet_sock *inet = inet_sk(sk);
1365
1366 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1367 return -EINVAL;
1368
1369 lock_sock(sk);
1370 inet->mc_index = dev->ifindex;
1371 /* inet->mc_addr = 0; */
1372 #ifdef CONFIG_IP_VS_IPV6
1373 if (sk->sk_family == AF_INET6) {
1374 struct ipv6_pinfo *np = inet6_sk(sk);
1375
1376 /* IPV6_MULTICAST_IF */
1377 np->mcast_oif = dev->ifindex;
1378 }
1379 #endif
1380 release_sock(sk);
1381
1382 return 0;
1383 }
1384
1385
1386 /*
1387 * Join a multicast group.
1388 * the group is specified by a class D multicast address 224.0.0.0/8
1389 * in the in_addr structure passed in as a parameter.
1390 */
1391 static int
join_mcast_group(struct sock * sk,struct in_addr * addr,struct net_device * dev)1392 join_mcast_group(struct sock *sk, struct in_addr *addr, struct net_device *dev)
1393 {
1394 struct ip_mreqn mreq;
1395 int ret;
1396
1397 memset(&mreq, 0, sizeof(mreq));
1398 memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
1399
1400 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1401 return -EINVAL;
1402
1403 mreq.imr_ifindex = dev->ifindex;
1404
1405 lock_sock(sk);
1406 ret = ip_mc_join_group(sk, &mreq);
1407 release_sock(sk);
1408
1409 return ret;
1410 }
1411
1412 #ifdef CONFIG_IP_VS_IPV6
join_mcast_group6(struct sock * sk,struct in6_addr * addr,struct net_device * dev)1413 static int join_mcast_group6(struct sock *sk, struct in6_addr *addr,
1414 struct net_device *dev)
1415 {
1416 int ret;
1417
1418 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1419 return -EINVAL;
1420
1421 lock_sock(sk);
1422 ret = ipv6_sock_mc_join(sk, dev->ifindex, addr);
1423 release_sock(sk);
1424
1425 return ret;
1426 }
1427 #endif
1428
bind_mcastif_addr(struct socket * sock,struct net_device * dev)1429 static int bind_mcastif_addr(struct socket *sock, struct net_device *dev)
1430 {
1431 __be32 addr;
1432 struct sockaddr_in sin;
1433
1434 addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
1435 if (!addr)
1436 pr_err("You probably need to specify IP address on "
1437 "multicast interface.\n");
1438
1439 IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
1440 dev->name, &addr);
1441
1442 /* Now bind the socket with the address of multicast interface */
1443 sin.sin_family = AF_INET;
1444 sin.sin_addr.s_addr = addr;
1445 sin.sin_port = 0;
1446
1447 return kernel_bind(sock, (struct sockaddr *)&sin, sizeof(sin));
1448 }
1449
get_mcast_sockaddr(union ipvs_sockaddr * sa,int * salen,struct ipvs_sync_daemon_cfg * c,int id)1450 static void get_mcast_sockaddr(union ipvs_sockaddr *sa, int *salen,
1451 struct ipvs_sync_daemon_cfg *c, int id)
1452 {
1453 if (AF_INET6 == c->mcast_af) {
1454 sa->in6 = (struct sockaddr_in6) {
1455 .sin6_family = AF_INET6,
1456 .sin6_port = htons(c->mcast_port + id),
1457 };
1458 sa->in6.sin6_addr = c->mcast_group.in6;
1459 *salen = sizeof(sa->in6);
1460 } else {
1461 sa->in = (struct sockaddr_in) {
1462 .sin_family = AF_INET,
1463 .sin_port = htons(c->mcast_port + id),
1464 };
1465 sa->in.sin_addr = c->mcast_group.in;
1466 *salen = sizeof(sa->in);
1467 }
1468 }
1469
1470 /*
1471 * Set up sending multicast socket over UDP
1472 */
make_send_sock(struct netns_ipvs * ipvs,int id,struct net_device * dev,struct socket ** sock_ret)1473 static int make_send_sock(struct netns_ipvs *ipvs, int id,
1474 struct net_device *dev, struct socket **sock_ret)
1475 {
1476 /* multicast addr */
1477 union ipvs_sockaddr mcast_addr;
1478 struct socket *sock;
1479 int result, salen;
1480
1481 /* First create a socket */
1482 result = sock_create_kern(ipvs->net, ipvs->mcfg.mcast_af, SOCK_DGRAM,
1483 IPPROTO_UDP, &sock);
1484 if (result < 0) {
1485 pr_err("Error during creation of socket; terminating\n");
1486 goto error;
1487 }
1488 *sock_ret = sock;
1489 result = set_mcast_if(sock->sk, dev);
1490 if (result < 0) {
1491 pr_err("Error setting outbound mcast interface\n");
1492 goto error;
1493 }
1494
1495 set_mcast_loop(sock->sk, 0);
1496 set_mcast_ttl(sock->sk, ipvs->mcfg.mcast_ttl);
1497 /* Allow fragmentation if MTU changes */
1498 set_mcast_pmtudisc(sock->sk, IP_PMTUDISC_DONT);
1499 result = sysctl_sync_sock_size(ipvs);
1500 if (result > 0)
1501 set_sock_size(sock->sk, 1, result);
1502
1503 if (AF_INET == ipvs->mcfg.mcast_af)
1504 result = bind_mcastif_addr(sock, dev);
1505 else
1506 result = 0;
1507 if (result < 0) {
1508 pr_err("Error binding address of the mcast interface\n");
1509 goto error;
1510 }
1511
1512 get_mcast_sockaddr(&mcast_addr, &salen, &ipvs->mcfg, id);
1513 result = kernel_connect(sock, (struct sockaddr *)&mcast_addr,
1514 salen, 0);
1515 if (result < 0) {
1516 pr_err("Error connecting to the multicast addr\n");
1517 goto error;
1518 }
1519
1520 return 0;
1521
1522 error:
1523 return result;
1524 }
1525
1526
1527 /*
1528 * Set up receiving multicast socket over UDP
1529 */
make_receive_sock(struct netns_ipvs * ipvs,int id,struct net_device * dev,struct socket ** sock_ret)1530 static int make_receive_sock(struct netns_ipvs *ipvs, int id,
1531 struct net_device *dev, struct socket **sock_ret)
1532 {
1533 /* multicast addr */
1534 union ipvs_sockaddr mcast_addr;
1535 struct socket *sock;
1536 int result, salen;
1537
1538 /* First create a socket */
1539 result = sock_create_kern(ipvs->net, ipvs->bcfg.mcast_af, SOCK_DGRAM,
1540 IPPROTO_UDP, &sock);
1541 if (result < 0) {
1542 pr_err("Error during creation of socket; terminating\n");
1543 goto error;
1544 }
1545 *sock_ret = sock;
1546 /* it is equivalent to the REUSEADDR option in user-space */
1547 sock->sk->sk_reuse = SK_CAN_REUSE;
1548 result = sysctl_sync_sock_size(ipvs);
1549 if (result > 0)
1550 set_sock_size(sock->sk, 0, result);
1551
1552 get_mcast_sockaddr(&mcast_addr, &salen, &ipvs->bcfg, id);
1553 sock->sk->sk_bound_dev_if = dev->ifindex;
1554 result = kernel_bind(sock, (struct sockaddr *)&mcast_addr, salen);
1555 if (result < 0) {
1556 pr_err("Error binding to the multicast addr\n");
1557 goto error;
1558 }
1559
1560 /* join the multicast group */
1561 #ifdef CONFIG_IP_VS_IPV6
1562 if (ipvs->bcfg.mcast_af == AF_INET6)
1563 result = join_mcast_group6(sock->sk, &mcast_addr.in6.sin6_addr,
1564 dev);
1565 else
1566 #endif
1567 result = join_mcast_group(sock->sk, &mcast_addr.in.sin_addr,
1568 dev);
1569 if (result < 0) {
1570 pr_err("Error joining to the multicast group\n");
1571 goto error;
1572 }
1573
1574 return 0;
1575
1576 error:
1577 return result;
1578 }
1579
1580
1581 static int
ip_vs_send_async(struct socket * sock,const char * buffer,const size_t length)1582 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
1583 {
1584 struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
1585 struct kvec iov;
1586 int len;
1587
1588 EnterFunction(7);
1589 iov.iov_base = (void *)buffer;
1590 iov.iov_len = length;
1591
1592 len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
1593
1594 LeaveFunction(7);
1595 return len;
1596 }
1597
1598 static int
ip_vs_send_sync_msg(struct socket * sock,struct ip_vs_sync_mesg * msg)1599 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
1600 {
1601 int msize;
1602 int ret;
1603
1604 msize = ntohs(msg->size);
1605
1606 ret = ip_vs_send_async(sock, (char *)msg, msize);
1607 if (ret >= 0 || ret == -EAGAIN)
1608 return ret;
1609 pr_err("ip_vs_send_async error %d\n", ret);
1610 return 0;
1611 }
1612
1613 static int
ip_vs_receive(struct socket * sock,char * buffer,const size_t buflen)1614 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
1615 {
1616 struct msghdr msg = {NULL,};
1617 struct kvec iov = {buffer, buflen};
1618 int len;
1619
1620 EnterFunction(7);
1621
1622 /* Receive a packet */
1623 iov_iter_kvec(&msg.msg_iter, READ, &iov, 1, buflen);
1624 len = sock_recvmsg(sock, &msg, MSG_DONTWAIT);
1625 if (len < 0)
1626 return len;
1627
1628 LeaveFunction(7);
1629 return len;
1630 }
1631
1632 /* Wakeup the master thread for sending */
master_wakeup_work_handler(struct work_struct * work)1633 static void master_wakeup_work_handler(struct work_struct *work)
1634 {
1635 struct ipvs_master_sync_state *ms =
1636 container_of(work, struct ipvs_master_sync_state,
1637 master_wakeup_work.work);
1638 struct netns_ipvs *ipvs = ms->ipvs;
1639
1640 spin_lock_bh(&ipvs->sync_lock);
1641 if (ms->sync_queue_len &&
1642 ms->sync_queue_delay < IPVS_SYNC_WAKEUP_RATE) {
1643 int id = (int)(ms - ipvs->ms);
1644
1645 ms->sync_queue_delay = IPVS_SYNC_WAKEUP_RATE;
1646 wake_up_process(ipvs->master_tinfo[id].task);
1647 }
1648 spin_unlock_bh(&ipvs->sync_lock);
1649 }
1650
1651 /* Get next buffer to send */
1652 static inline struct ip_vs_sync_buff *
next_sync_buff(struct netns_ipvs * ipvs,struct ipvs_master_sync_state * ms)1653 next_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
1654 {
1655 struct ip_vs_sync_buff *sb;
1656
1657 sb = sb_dequeue(ipvs, ms);
1658 if (sb)
1659 return sb;
1660 /* Do not delay entries in buffer for more than 2 seconds */
1661 return get_curr_sync_buff(ipvs, ms, IPVS_SYNC_FLUSH_TIME);
1662 }
1663
sync_thread_master(void * data)1664 static int sync_thread_master(void *data)
1665 {
1666 struct ip_vs_sync_thread_data *tinfo = data;
1667 struct netns_ipvs *ipvs = tinfo->ipvs;
1668 struct ipvs_master_sync_state *ms = &ipvs->ms[tinfo->id];
1669 struct sock *sk = tinfo->sock->sk;
1670 struct ip_vs_sync_buff *sb;
1671
1672 pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
1673 "syncid = %d, id = %d\n",
1674 ipvs->mcfg.mcast_ifn, ipvs->mcfg.syncid, tinfo->id);
1675
1676 for (;;) {
1677 sb = next_sync_buff(ipvs, ms);
1678 if (unlikely(kthread_should_stop()))
1679 break;
1680 if (!sb) {
1681 schedule_timeout(IPVS_SYNC_CHECK_PERIOD);
1682 continue;
1683 }
1684 while (ip_vs_send_sync_msg(tinfo->sock, sb->mesg) < 0) {
1685 /* (Ab)use interruptible sleep to avoid increasing
1686 * the load avg.
1687 */
1688 __wait_event_interruptible(*sk_sleep(sk),
1689 sock_writeable(sk) ||
1690 kthread_should_stop());
1691 if (unlikely(kthread_should_stop()))
1692 goto done;
1693 }
1694 ip_vs_sync_buff_release(sb);
1695 }
1696
1697 done:
1698 __set_current_state(TASK_RUNNING);
1699 if (sb)
1700 ip_vs_sync_buff_release(sb);
1701
1702 /* clean up the sync_buff queue */
1703 while ((sb = sb_dequeue(ipvs, ms)))
1704 ip_vs_sync_buff_release(sb);
1705 __set_current_state(TASK_RUNNING);
1706
1707 /* clean up the current sync_buff */
1708 sb = get_curr_sync_buff(ipvs, ms, 0);
1709 if (sb)
1710 ip_vs_sync_buff_release(sb);
1711
1712 return 0;
1713 }
1714
1715
sync_thread_backup(void * data)1716 static int sync_thread_backup(void *data)
1717 {
1718 struct ip_vs_sync_thread_data *tinfo = data;
1719 struct netns_ipvs *ipvs = tinfo->ipvs;
1720 struct sock *sk = tinfo->sock->sk;
1721 struct udp_sock *up = udp_sk(sk);
1722 int len;
1723
1724 pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
1725 "syncid = %d, id = %d\n",
1726 ipvs->bcfg.mcast_ifn, ipvs->bcfg.syncid, tinfo->id);
1727
1728 while (!kthread_should_stop()) {
1729 wait_event_interruptible(*sk_sleep(sk),
1730 !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
1731 !skb_queue_empty_lockless(&up->reader_queue) ||
1732 kthread_should_stop());
1733
1734 /* do we have data now? */
1735 while (!skb_queue_empty_lockless(&sk->sk_receive_queue) ||
1736 !skb_queue_empty_lockless(&up->reader_queue)) {
1737 len = ip_vs_receive(tinfo->sock, tinfo->buf,
1738 ipvs->bcfg.sync_maxlen);
1739 if (len <= 0) {
1740 if (len != -EAGAIN)
1741 pr_err("receiving message error\n");
1742 break;
1743 }
1744
1745 ip_vs_process_message(ipvs, tinfo->buf, len);
1746 }
1747 }
1748
1749 return 0;
1750 }
1751
1752
start_sync_thread(struct netns_ipvs * ipvs,struct ipvs_sync_daemon_cfg * c,int state)1753 int start_sync_thread(struct netns_ipvs *ipvs, struct ipvs_sync_daemon_cfg *c,
1754 int state)
1755 {
1756 struct ip_vs_sync_thread_data *ti = NULL, *tinfo;
1757 struct task_struct *task;
1758 struct net_device *dev;
1759 char *name;
1760 int (*threadfn)(void *data);
1761 int id = 0, count, hlen;
1762 int result = -ENOMEM;
1763 u16 mtu, min_mtu;
1764
1765 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1766 IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %zd bytes\n",
1767 sizeof(struct ip_vs_sync_conn_v0));
1768
1769 /* increase the module use count */
1770 if (!ip_vs_use_count_inc())
1771 return -ENOPROTOOPT;
1772
1773 /* Do not hold one mutex and then to block on another */
1774 for (;;) {
1775 rtnl_lock();
1776 if (mutex_trylock(&ipvs->sync_mutex))
1777 break;
1778 rtnl_unlock();
1779 mutex_lock(&ipvs->sync_mutex);
1780 if (rtnl_trylock())
1781 break;
1782 mutex_unlock(&ipvs->sync_mutex);
1783 }
1784
1785 if (!ipvs->sync_state) {
1786 count = clamp(sysctl_sync_ports(ipvs), 1, IPVS_SYNC_PORTS_MAX);
1787 ipvs->threads_mask = count - 1;
1788 } else
1789 count = ipvs->threads_mask + 1;
1790
1791 if (c->mcast_af == AF_UNSPEC) {
1792 c->mcast_af = AF_INET;
1793 c->mcast_group.ip = cpu_to_be32(IP_VS_SYNC_GROUP);
1794 }
1795 if (!c->mcast_port)
1796 c->mcast_port = IP_VS_SYNC_PORT;
1797 if (!c->mcast_ttl)
1798 c->mcast_ttl = 1;
1799
1800 dev = __dev_get_by_name(ipvs->net, c->mcast_ifn);
1801 if (!dev) {
1802 pr_err("Unknown mcast interface: %s\n", c->mcast_ifn);
1803 result = -ENODEV;
1804 goto out_early;
1805 }
1806 hlen = (AF_INET6 == c->mcast_af) ?
1807 sizeof(struct ipv6hdr) + sizeof(struct udphdr) :
1808 sizeof(struct iphdr) + sizeof(struct udphdr);
1809 mtu = (state == IP_VS_STATE_BACKUP) ?
1810 clamp(dev->mtu, 1500U, 65535U) : 1500U;
1811 min_mtu = (state == IP_VS_STATE_BACKUP) ? 1024 : 1;
1812
1813 if (c->sync_maxlen)
1814 c->sync_maxlen = clamp_t(unsigned int,
1815 c->sync_maxlen, min_mtu,
1816 65535 - hlen);
1817 else
1818 c->sync_maxlen = mtu - hlen;
1819
1820 if (state == IP_VS_STATE_MASTER) {
1821 result = -EEXIST;
1822 if (ipvs->ms)
1823 goto out_early;
1824
1825 ipvs->mcfg = *c;
1826 name = "ipvs-m:%d:%d";
1827 threadfn = sync_thread_master;
1828 } else if (state == IP_VS_STATE_BACKUP) {
1829 result = -EEXIST;
1830 if (ipvs->backup_tinfo)
1831 goto out_early;
1832
1833 ipvs->bcfg = *c;
1834 name = "ipvs-b:%d:%d";
1835 threadfn = sync_thread_backup;
1836 } else {
1837 result = -EINVAL;
1838 goto out_early;
1839 }
1840
1841 if (state == IP_VS_STATE_MASTER) {
1842 struct ipvs_master_sync_state *ms;
1843
1844 result = -ENOMEM;
1845 ipvs->ms = kcalloc(count, sizeof(ipvs->ms[0]), GFP_KERNEL);
1846 if (!ipvs->ms)
1847 goto out;
1848 ms = ipvs->ms;
1849 for (id = 0; id < count; id++, ms++) {
1850 INIT_LIST_HEAD(&ms->sync_queue);
1851 ms->sync_queue_len = 0;
1852 ms->sync_queue_delay = 0;
1853 INIT_DELAYED_WORK(&ms->master_wakeup_work,
1854 master_wakeup_work_handler);
1855 ms->ipvs = ipvs;
1856 }
1857 }
1858 result = -ENOMEM;
1859 ti = kcalloc(count, sizeof(struct ip_vs_sync_thread_data),
1860 GFP_KERNEL);
1861 if (!ti)
1862 goto out;
1863
1864 for (id = 0; id < count; id++) {
1865 tinfo = &ti[id];
1866 tinfo->ipvs = ipvs;
1867 if (state == IP_VS_STATE_BACKUP) {
1868 result = -ENOMEM;
1869 tinfo->buf = kmalloc(ipvs->bcfg.sync_maxlen,
1870 GFP_KERNEL);
1871 if (!tinfo->buf)
1872 goto out;
1873 }
1874 tinfo->id = id;
1875 if (state == IP_VS_STATE_MASTER)
1876 result = make_send_sock(ipvs, id, dev, &tinfo->sock);
1877 else
1878 result = make_receive_sock(ipvs, id, dev, &tinfo->sock);
1879 if (result < 0)
1880 goto out;
1881
1882 task = kthread_run(threadfn, tinfo, name, ipvs->gen, id);
1883 if (IS_ERR(task)) {
1884 result = PTR_ERR(task);
1885 goto out;
1886 }
1887 tinfo->task = task;
1888 }
1889
1890 /* mark as active */
1891
1892 if (state == IP_VS_STATE_MASTER)
1893 ipvs->master_tinfo = ti;
1894 else
1895 ipvs->backup_tinfo = ti;
1896 spin_lock_bh(&ipvs->sync_buff_lock);
1897 ipvs->sync_state |= state;
1898 spin_unlock_bh(&ipvs->sync_buff_lock);
1899
1900 mutex_unlock(&ipvs->sync_mutex);
1901 rtnl_unlock();
1902
1903 return 0;
1904
1905 out:
1906 /* We do not need RTNL lock anymore, release it here so that
1907 * sock_release below can use rtnl_lock to leave the mcast group.
1908 */
1909 rtnl_unlock();
1910 id = min(id, count - 1);
1911 if (ti) {
1912 for (tinfo = ti + id; tinfo >= ti; tinfo--) {
1913 if (tinfo->task)
1914 kthread_stop(tinfo->task);
1915 }
1916 }
1917 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
1918 kfree(ipvs->ms);
1919 ipvs->ms = NULL;
1920 }
1921 mutex_unlock(&ipvs->sync_mutex);
1922
1923 /* No more mutexes, release socks */
1924 if (ti) {
1925 for (tinfo = ti + id; tinfo >= ti; tinfo--) {
1926 if (tinfo->sock)
1927 sock_release(tinfo->sock);
1928 kfree(tinfo->buf);
1929 }
1930 kfree(ti);
1931 }
1932
1933 /* decrease the module use count */
1934 ip_vs_use_count_dec();
1935 return result;
1936
1937 out_early:
1938 mutex_unlock(&ipvs->sync_mutex);
1939 rtnl_unlock();
1940
1941 /* decrease the module use count */
1942 ip_vs_use_count_dec();
1943 return result;
1944 }
1945
1946
stop_sync_thread(struct netns_ipvs * ipvs,int state)1947 int stop_sync_thread(struct netns_ipvs *ipvs, int state)
1948 {
1949 struct ip_vs_sync_thread_data *ti, *tinfo;
1950 int id;
1951 int retc = -EINVAL;
1952
1953 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1954
1955 mutex_lock(&ipvs->sync_mutex);
1956 if (state == IP_VS_STATE_MASTER) {
1957 retc = -ESRCH;
1958 if (!ipvs->ms)
1959 goto err;
1960 ti = ipvs->master_tinfo;
1961
1962 /*
1963 * The lock synchronizes with sb_queue_tail(), so that we don't
1964 * add sync buffers to the queue, when we are already in
1965 * progress of stopping the master sync daemon.
1966 */
1967
1968 spin_lock_bh(&ipvs->sync_buff_lock);
1969 spin_lock(&ipvs->sync_lock);
1970 ipvs->sync_state &= ~IP_VS_STATE_MASTER;
1971 spin_unlock(&ipvs->sync_lock);
1972 spin_unlock_bh(&ipvs->sync_buff_lock);
1973
1974 retc = 0;
1975 for (id = ipvs->threads_mask; id >= 0; id--) {
1976 struct ipvs_master_sync_state *ms = &ipvs->ms[id];
1977 int ret;
1978
1979 tinfo = &ti[id];
1980 pr_info("stopping master sync thread %d ...\n",
1981 task_pid_nr(tinfo->task));
1982 cancel_delayed_work_sync(&ms->master_wakeup_work);
1983 ret = kthread_stop(tinfo->task);
1984 if (retc >= 0)
1985 retc = ret;
1986 }
1987 kfree(ipvs->ms);
1988 ipvs->ms = NULL;
1989 ipvs->master_tinfo = NULL;
1990 } else if (state == IP_VS_STATE_BACKUP) {
1991 retc = -ESRCH;
1992 if (!ipvs->backup_tinfo)
1993 goto err;
1994 ti = ipvs->backup_tinfo;
1995
1996 ipvs->sync_state &= ~IP_VS_STATE_BACKUP;
1997 retc = 0;
1998 for (id = ipvs->threads_mask; id >= 0; id--) {
1999 int ret;
2000
2001 tinfo = &ti[id];
2002 pr_info("stopping backup sync thread %d ...\n",
2003 task_pid_nr(tinfo->task));
2004 ret = kthread_stop(tinfo->task);
2005 if (retc >= 0)
2006 retc = ret;
2007 }
2008 ipvs->backup_tinfo = NULL;
2009 } else {
2010 goto err;
2011 }
2012 id = ipvs->threads_mask;
2013 mutex_unlock(&ipvs->sync_mutex);
2014
2015 /* No more mutexes, release socks */
2016 for (tinfo = ti + id; tinfo >= ti; tinfo--) {
2017 if (tinfo->sock)
2018 sock_release(tinfo->sock);
2019 kfree(tinfo->buf);
2020 }
2021 kfree(ti);
2022
2023 /* decrease the module use count */
2024 ip_vs_use_count_dec();
2025 return retc;
2026
2027 err:
2028 mutex_unlock(&ipvs->sync_mutex);
2029 return retc;
2030 }
2031
2032 /*
2033 * Initialize data struct for each netns
2034 */
ip_vs_sync_net_init(struct netns_ipvs * ipvs)2035 int __net_init ip_vs_sync_net_init(struct netns_ipvs *ipvs)
2036 {
2037 __mutex_init(&ipvs->sync_mutex, "ipvs->sync_mutex", &__ipvs_sync_key);
2038 spin_lock_init(&ipvs->sync_lock);
2039 spin_lock_init(&ipvs->sync_buff_lock);
2040 return 0;
2041 }
2042
ip_vs_sync_net_cleanup(struct netns_ipvs * ipvs)2043 void ip_vs_sync_net_cleanup(struct netns_ipvs *ipvs)
2044 {
2045 int retc;
2046
2047 retc = stop_sync_thread(ipvs, IP_VS_STATE_MASTER);
2048 if (retc && retc != -ESRCH)
2049 pr_err("Failed to stop Master Daemon\n");
2050
2051 retc = stop_sync_thread(ipvs, IP_VS_STATE_BACKUP);
2052 if (retc && retc != -ESRCH)
2053 pr_err("Failed to stop Backup Daemon\n");
2054 }
2055