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