Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * IPVS An implementation of the IP virtual server support for the | |
3 | * LINUX operating system. IPVS is now implemented as a module | |
4 | * over the Netfilter framework. IPVS can be used to build a | |
5 | * high-performance and highly available server based on a | |
6 | * cluster of servers. | |
7 | * | |
1da177e4 LT |
8 | * Authors: Wensong Zhang <wensong@linuxvirtualserver.org> |
9 | * Peter Kese <peter.kese@ijs.si> | |
10 | * Julian Anastasov <ja@ssi.bg> | |
11 | * | |
12 | * This program is free software; you can redistribute it and/or | |
13 | * modify it under the terms of the GNU General Public License | |
14 | * as published by the Free Software Foundation; either version | |
15 | * 2 of the License, or (at your option) any later version. | |
16 | * | |
17 | * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese, | |
18 | * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms | |
19 | * and others. Many code here is taken from IP MASQ code of kernel 2.2. | |
20 | * | |
21 | * Changes: | |
22 | * | |
23 | */ | |
24 | ||
e924283b | 25 | #include <linux/interrupt.h> |
14c85021 | 26 | #include <linux/in.h> |
f190055f | 27 | #include <linux/net.h> |
1da177e4 | 28 | #include <linux/kernel.h> |
14c85021 | 29 | #include <linux/module.h> |
1da177e4 LT |
30 | #include <linux/vmalloc.h> |
31 | #include <linux/proc_fs.h> /* for proc_net_* */ | |
32 | #include <linux/seq_file.h> | |
33 | #include <linux/jhash.h> | |
34 | #include <linux/random.h> | |
35 | ||
457c4cbc | 36 | #include <net/net_namespace.h> |
1da177e4 LT |
37 | #include <net/ip_vs.h> |
38 | ||
39 | ||
40 | /* | |
41 | * Connection hash table: for input and output packets lookups of IPVS | |
42 | */ | |
43 | static struct list_head *ip_vs_conn_tab; | |
44 | ||
45 | /* SLAB cache for IPVS connections */ | |
e18b890b | 46 | static struct kmem_cache *ip_vs_conn_cachep __read_mostly; |
1da177e4 LT |
47 | |
48 | /* counter for current IPVS connections */ | |
49 | static atomic_t ip_vs_conn_count = ATOMIC_INIT(0); | |
50 | ||
51 | /* counter for no client port connections */ | |
52 | static atomic_t ip_vs_conn_no_cport_cnt = ATOMIC_INIT(0); | |
53 | ||
54 | /* random value for IPVS connection hash */ | |
55 | static unsigned int ip_vs_conn_rnd; | |
56 | ||
57 | /* | |
58 | * Fine locking granularity for big connection hash table | |
59 | */ | |
60 | #define CT_LOCKARRAY_BITS 4 | |
61 | #define CT_LOCKARRAY_SIZE (1<<CT_LOCKARRAY_BITS) | |
62 | #define CT_LOCKARRAY_MASK (CT_LOCKARRAY_SIZE-1) | |
63 | ||
64 | struct ip_vs_aligned_lock | |
65 | { | |
66 | rwlock_t l; | |
67 | } __attribute__((__aligned__(SMP_CACHE_BYTES))); | |
68 | ||
69 | /* lock array for conn table */ | |
70 | static struct ip_vs_aligned_lock | |
71 | __ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned; | |
72 | ||
73 | static inline void ct_read_lock(unsigned key) | |
74 | { | |
75 | read_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l); | |
76 | } | |
77 | ||
78 | static inline void ct_read_unlock(unsigned key) | |
79 | { | |
80 | read_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l); | |
81 | } | |
82 | ||
83 | static inline void ct_write_lock(unsigned key) | |
84 | { | |
85 | write_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l); | |
86 | } | |
87 | ||
88 | static inline void ct_write_unlock(unsigned key) | |
89 | { | |
90 | write_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l); | |
91 | } | |
92 | ||
93 | static inline void ct_read_lock_bh(unsigned key) | |
94 | { | |
95 | read_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l); | |
96 | } | |
97 | ||
98 | static inline void ct_read_unlock_bh(unsigned key) | |
99 | { | |
100 | read_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l); | |
101 | } | |
102 | ||
103 | static inline void ct_write_lock_bh(unsigned key) | |
104 | { | |
105 | write_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l); | |
106 | } | |
107 | ||
108 | static inline void ct_write_unlock_bh(unsigned key) | |
109 | { | |
110 | write_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l); | |
111 | } | |
112 | ||
113 | ||
114 | /* | |
115 | * Returns hash value for IPVS connection entry | |
116 | */ | |
014d730d | 117 | static unsigned int ip_vs_conn_hashkey(unsigned proto, __be32 addr, __be16 port) |
1da177e4 | 118 | { |
014d730d | 119 | return jhash_3words((__force u32)addr, (__force u32)port, proto, ip_vs_conn_rnd) |
1da177e4 LT |
120 | & IP_VS_CONN_TAB_MASK; |
121 | } | |
122 | ||
123 | ||
124 | /* | |
125 | * Hashes ip_vs_conn in ip_vs_conn_tab by proto,addr,port. | |
126 | * returns bool success. | |
127 | */ | |
128 | static inline int ip_vs_conn_hash(struct ip_vs_conn *cp) | |
129 | { | |
130 | unsigned hash; | |
131 | int ret; | |
132 | ||
133 | /* Hash by protocol, client address and port */ | |
134 | hash = ip_vs_conn_hashkey(cp->protocol, cp->caddr, cp->cport); | |
135 | ||
136 | ct_write_lock(hash); | |
137 | ||
138 | if (!(cp->flags & IP_VS_CONN_F_HASHED)) { | |
139 | list_add(&cp->c_list, &ip_vs_conn_tab[hash]); | |
140 | cp->flags |= IP_VS_CONN_F_HASHED; | |
141 | atomic_inc(&cp->refcnt); | |
142 | ret = 1; | |
143 | } else { | |
144 | IP_VS_ERR("ip_vs_conn_hash(): request for already hashed, " | |
145 | "called from %p\n", __builtin_return_address(0)); | |
146 | ret = 0; | |
147 | } | |
148 | ||
149 | ct_write_unlock(hash); | |
150 | ||
151 | return ret; | |
152 | } | |
153 | ||
154 | ||
155 | /* | |
156 | * UNhashes ip_vs_conn from ip_vs_conn_tab. | |
157 | * returns bool success. | |
158 | */ | |
159 | static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp) | |
160 | { | |
161 | unsigned hash; | |
162 | int ret; | |
163 | ||
164 | /* unhash it and decrease its reference counter */ | |
165 | hash = ip_vs_conn_hashkey(cp->protocol, cp->caddr, cp->cport); | |
166 | ||
167 | ct_write_lock(hash); | |
168 | ||
169 | if (cp->flags & IP_VS_CONN_F_HASHED) { | |
170 | list_del(&cp->c_list); | |
171 | cp->flags &= ~IP_VS_CONN_F_HASHED; | |
172 | atomic_dec(&cp->refcnt); | |
173 | ret = 1; | |
174 | } else | |
175 | ret = 0; | |
176 | ||
177 | ct_write_unlock(hash); | |
178 | ||
179 | return ret; | |
180 | } | |
181 | ||
182 | ||
183 | /* | |
184 | * Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab. | |
185 | * Called for pkts coming from OUTside-to-INside. | |
186 | * s_addr, s_port: pkt source address (foreign host) | |
187 | * d_addr, d_port: pkt dest address (load balancer) | |
188 | */ | |
189 | static inline struct ip_vs_conn *__ip_vs_conn_in_get | |
014d730d | 190 | (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port) |
1da177e4 LT |
191 | { |
192 | unsigned hash; | |
193 | struct ip_vs_conn *cp; | |
194 | ||
195 | hash = ip_vs_conn_hashkey(protocol, s_addr, s_port); | |
196 | ||
197 | ct_read_lock(hash); | |
198 | ||
199 | list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) { | |
200 | if (s_addr==cp->caddr && s_port==cp->cport && | |
201 | d_port==cp->vport && d_addr==cp->vaddr && | |
87375ab4 | 202 | ((!s_port) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) && |
1da177e4 LT |
203 | protocol==cp->protocol) { |
204 | /* HIT */ | |
205 | atomic_inc(&cp->refcnt); | |
206 | ct_read_unlock(hash); | |
207 | return cp; | |
208 | } | |
209 | } | |
210 | ||
211 | ct_read_unlock(hash); | |
212 | ||
213 | return NULL; | |
214 | } | |
215 | ||
216 | struct ip_vs_conn *ip_vs_conn_in_get | |
014d730d | 217 | (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port) |
1da177e4 LT |
218 | { |
219 | struct ip_vs_conn *cp; | |
220 | ||
221 | cp = __ip_vs_conn_in_get(protocol, s_addr, s_port, d_addr, d_port); | |
222 | if (!cp && atomic_read(&ip_vs_conn_no_cport_cnt)) | |
223 | cp = __ip_vs_conn_in_get(protocol, s_addr, 0, d_addr, d_port); | |
224 | ||
4b5bdf5c | 225 | IP_VS_DBG(9, "lookup/in %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n", |
1da177e4 LT |
226 | ip_vs_proto_name(protocol), |
227 | NIPQUAD(s_addr), ntohs(s_port), | |
228 | NIPQUAD(d_addr), ntohs(d_port), | |
229 | cp?"hit":"not hit"); | |
230 | ||
231 | return cp; | |
232 | } | |
233 | ||
87375ab4 JA |
234 | /* Get reference to connection template */ |
235 | struct ip_vs_conn *ip_vs_ct_in_get | |
014d730d | 236 | (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port) |
87375ab4 JA |
237 | { |
238 | unsigned hash; | |
239 | struct ip_vs_conn *cp; | |
240 | ||
241 | hash = ip_vs_conn_hashkey(protocol, s_addr, s_port); | |
242 | ||
243 | ct_read_lock(hash); | |
244 | ||
245 | list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) { | |
246 | if (s_addr==cp->caddr && s_port==cp->cport && | |
247 | d_port==cp->vport && d_addr==cp->vaddr && | |
248 | cp->flags & IP_VS_CONN_F_TEMPLATE && | |
249 | protocol==cp->protocol) { | |
250 | /* HIT */ | |
251 | atomic_inc(&cp->refcnt); | |
252 | goto out; | |
253 | } | |
254 | } | |
255 | cp = NULL; | |
256 | ||
257 | out: | |
258 | ct_read_unlock(hash); | |
259 | ||
4b5bdf5c | 260 | IP_VS_DBG(9, "template lookup/in %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n", |
87375ab4 JA |
261 | ip_vs_proto_name(protocol), |
262 | NIPQUAD(s_addr), ntohs(s_port), | |
263 | NIPQUAD(d_addr), ntohs(d_port), | |
264 | cp?"hit":"not hit"); | |
265 | ||
266 | return cp; | |
267 | } | |
1da177e4 LT |
268 | |
269 | /* | |
270 | * Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab. | |
271 | * Called for pkts coming from inside-to-OUTside. | |
272 | * s_addr, s_port: pkt source address (inside host) | |
273 | * d_addr, d_port: pkt dest address (foreign host) | |
274 | */ | |
275 | struct ip_vs_conn *ip_vs_conn_out_get | |
014d730d | 276 | (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port) |
1da177e4 LT |
277 | { |
278 | unsigned hash; | |
279 | struct ip_vs_conn *cp, *ret=NULL; | |
280 | ||
281 | /* | |
282 | * Check for "full" addressed entries | |
283 | */ | |
284 | hash = ip_vs_conn_hashkey(protocol, d_addr, d_port); | |
285 | ||
286 | ct_read_lock(hash); | |
287 | ||
288 | list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) { | |
289 | if (d_addr == cp->caddr && d_port == cp->cport && | |
290 | s_port == cp->dport && s_addr == cp->daddr && | |
291 | protocol == cp->protocol) { | |
292 | /* HIT */ | |
293 | atomic_inc(&cp->refcnt); | |
294 | ret = cp; | |
295 | break; | |
296 | } | |
297 | } | |
298 | ||
299 | ct_read_unlock(hash); | |
300 | ||
4b5bdf5c | 301 | IP_VS_DBG(9, "lookup/out %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n", |
1da177e4 LT |
302 | ip_vs_proto_name(protocol), |
303 | NIPQUAD(s_addr), ntohs(s_port), | |
304 | NIPQUAD(d_addr), ntohs(d_port), | |
305 | ret?"hit":"not hit"); | |
306 | ||
307 | return ret; | |
308 | } | |
309 | ||
310 | ||
311 | /* | |
312 | * Put back the conn and restart its timer with its timeout | |
313 | */ | |
314 | void ip_vs_conn_put(struct ip_vs_conn *cp) | |
315 | { | |
316 | /* reset it expire in its timeout */ | |
317 | mod_timer(&cp->timer, jiffies+cp->timeout); | |
318 | ||
319 | __ip_vs_conn_put(cp); | |
320 | } | |
321 | ||
322 | ||
323 | /* | |
324 | * Fill a no_client_port connection with a client port number | |
325 | */ | |
014d730d | 326 | void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport) |
1da177e4 LT |
327 | { |
328 | if (ip_vs_conn_unhash(cp)) { | |
329 | spin_lock(&cp->lock); | |
330 | if (cp->flags & IP_VS_CONN_F_NO_CPORT) { | |
331 | atomic_dec(&ip_vs_conn_no_cport_cnt); | |
332 | cp->flags &= ~IP_VS_CONN_F_NO_CPORT; | |
333 | cp->cport = cport; | |
334 | } | |
335 | spin_unlock(&cp->lock); | |
336 | ||
337 | /* hash on new dport */ | |
338 | ip_vs_conn_hash(cp); | |
339 | } | |
340 | } | |
341 | ||
342 | ||
343 | /* | |
344 | * Bind a connection entry with the corresponding packet_xmit. | |
345 | * Called by ip_vs_conn_new. | |
346 | */ | |
347 | static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp) | |
348 | { | |
349 | switch (IP_VS_FWD_METHOD(cp)) { | |
350 | case IP_VS_CONN_F_MASQ: | |
351 | cp->packet_xmit = ip_vs_nat_xmit; | |
352 | break; | |
353 | ||
354 | case IP_VS_CONN_F_TUNNEL: | |
355 | cp->packet_xmit = ip_vs_tunnel_xmit; | |
356 | break; | |
357 | ||
358 | case IP_VS_CONN_F_DROUTE: | |
359 | cp->packet_xmit = ip_vs_dr_xmit; | |
360 | break; | |
361 | ||
362 | case IP_VS_CONN_F_LOCALNODE: | |
363 | cp->packet_xmit = ip_vs_null_xmit; | |
364 | break; | |
365 | ||
366 | case IP_VS_CONN_F_BYPASS: | |
367 | cp->packet_xmit = ip_vs_bypass_xmit; | |
368 | break; | |
369 | } | |
370 | } | |
371 | ||
372 | ||
373 | static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest) | |
374 | { | |
375 | return atomic_read(&dest->activeconns) | |
376 | + atomic_read(&dest->inactconns); | |
377 | } | |
378 | ||
379 | /* | |
380 | * Bind a connection entry with a virtual service destination | |
381 | * Called just after a new connection entry is created. | |
382 | */ | |
383 | static inline void | |
384 | ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest) | |
385 | { | |
386 | /* if dest is NULL, then return directly */ | |
387 | if (!dest) | |
388 | return; | |
389 | ||
390 | /* Increase the refcnt counter of the dest */ | |
391 | atomic_inc(&dest->refcnt); | |
392 | ||
393 | /* Bind with the destination and its corresponding transmitter */ | |
b209639e RB |
394 | if ((cp->flags & IP_VS_CONN_F_SYNC) && |
395 | (!(cp->flags & IP_VS_CONN_F_TEMPLATE))) | |
396 | /* if the connection is not template and is created | |
397 | * by sync, preserve the activity flag. | |
398 | */ | |
399 | cp->flags |= atomic_read(&dest->conn_flags) & | |
400 | (~IP_VS_CONN_F_INACTIVE); | |
401 | else | |
402 | cp->flags |= atomic_read(&dest->conn_flags); | |
1da177e4 LT |
403 | cp->dest = dest; |
404 | ||
4b5bdf5c RN |
405 | IP_VS_DBG(7, "Bind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d " |
406 | "d:%u.%u.%u.%u:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d " | |
407 | "dest->refcnt:%d\n", | |
1da177e4 LT |
408 | ip_vs_proto_name(cp->protocol), |
409 | NIPQUAD(cp->caddr), ntohs(cp->cport), | |
410 | NIPQUAD(cp->vaddr), ntohs(cp->vport), | |
411 | NIPQUAD(cp->daddr), ntohs(cp->dport), | |
412 | ip_vs_fwd_tag(cp), cp->state, | |
413 | cp->flags, atomic_read(&cp->refcnt), | |
414 | atomic_read(&dest->refcnt)); | |
415 | ||
416 | /* Update the connection counters */ | |
87375ab4 | 417 | if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) { |
1da177e4 LT |
418 | /* It is a normal connection, so increase the inactive |
419 | connection counter because it is in TCP SYNRECV | |
420 | state (inactive) or other protocol inacive state */ | |
b209639e RB |
421 | if ((cp->flags & IP_VS_CONN_F_SYNC) && |
422 | (!(cp->flags & IP_VS_CONN_F_INACTIVE))) | |
423 | atomic_inc(&dest->activeconns); | |
424 | else | |
425 | atomic_inc(&dest->inactconns); | |
1da177e4 LT |
426 | } else { |
427 | /* It is a persistent connection/template, so increase | |
428 | the peristent connection counter */ | |
429 | atomic_inc(&dest->persistconns); | |
430 | } | |
431 | ||
432 | if (dest->u_threshold != 0 && | |
433 | ip_vs_dest_totalconns(dest) >= dest->u_threshold) | |
434 | dest->flags |= IP_VS_DEST_F_OVERLOAD; | |
435 | } | |
436 | ||
437 | ||
1e356f9c RB |
438 | /* |
439 | * Check if there is a destination for the connection, if so | |
440 | * bind the connection to the destination. | |
441 | */ | |
442 | struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp) | |
443 | { | |
444 | struct ip_vs_dest *dest; | |
445 | ||
446 | if ((cp) && (!cp->dest)) { | |
447 | dest = ip_vs_find_dest(cp->daddr, cp->dport, | |
448 | cp->vaddr, cp->vport, cp->protocol); | |
449 | ip_vs_bind_dest(cp, dest); | |
450 | return dest; | |
451 | } else | |
452 | return NULL; | |
453 | } | |
1e356f9c RB |
454 | |
455 | ||
1da177e4 LT |
456 | /* |
457 | * Unbind a connection entry with its VS destination | |
458 | * Called by the ip_vs_conn_expire function. | |
459 | */ | |
460 | static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp) | |
461 | { | |
462 | struct ip_vs_dest *dest = cp->dest; | |
463 | ||
464 | if (!dest) | |
465 | return; | |
466 | ||
4b5bdf5c RN |
467 | IP_VS_DBG(7, "Unbind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d " |
468 | "d:%u.%u.%u.%u:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d " | |
469 | "dest->refcnt:%d\n", | |
1da177e4 LT |
470 | ip_vs_proto_name(cp->protocol), |
471 | NIPQUAD(cp->caddr), ntohs(cp->cport), | |
472 | NIPQUAD(cp->vaddr), ntohs(cp->vport), | |
473 | NIPQUAD(cp->daddr), ntohs(cp->dport), | |
474 | ip_vs_fwd_tag(cp), cp->state, | |
475 | cp->flags, atomic_read(&cp->refcnt), | |
476 | atomic_read(&dest->refcnt)); | |
477 | ||
478 | /* Update the connection counters */ | |
87375ab4 | 479 | if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) { |
1da177e4 LT |
480 | /* It is a normal connection, so decrease the inactconns |
481 | or activeconns counter */ | |
482 | if (cp->flags & IP_VS_CONN_F_INACTIVE) { | |
483 | atomic_dec(&dest->inactconns); | |
484 | } else { | |
485 | atomic_dec(&dest->activeconns); | |
486 | } | |
487 | } else { | |
488 | /* It is a persistent connection/template, so decrease | |
489 | the peristent connection counter */ | |
490 | atomic_dec(&dest->persistconns); | |
491 | } | |
492 | ||
493 | if (dest->l_threshold != 0) { | |
494 | if (ip_vs_dest_totalconns(dest) < dest->l_threshold) | |
495 | dest->flags &= ~IP_VS_DEST_F_OVERLOAD; | |
496 | } else if (dest->u_threshold != 0) { | |
497 | if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3) | |
498 | dest->flags &= ~IP_VS_DEST_F_OVERLOAD; | |
499 | } else { | |
500 | if (dest->flags & IP_VS_DEST_F_OVERLOAD) | |
501 | dest->flags &= ~IP_VS_DEST_F_OVERLOAD; | |
502 | } | |
503 | ||
504 | /* | |
505 | * Simply decrease the refcnt of the dest, because the | |
506 | * dest will be either in service's destination list | |
507 | * or in the trash. | |
508 | */ | |
509 | atomic_dec(&dest->refcnt); | |
510 | } | |
511 | ||
512 | ||
513 | /* | |
514 | * Checking if the destination of a connection template is available. | |
515 | * If available, return 1, otherwise invalidate this connection | |
516 | * template and return 0. | |
517 | */ | |
518 | int ip_vs_check_template(struct ip_vs_conn *ct) | |
519 | { | |
520 | struct ip_vs_dest *dest = ct->dest; | |
521 | ||
522 | /* | |
523 | * Checking the dest server status. | |
524 | */ | |
525 | if ((dest == NULL) || | |
e905a9ed YH |
526 | !(dest->flags & IP_VS_DEST_F_AVAILABLE) || |
527 | (sysctl_ip_vs_expire_quiescent_template && | |
1da177e4 LT |
528 | (atomic_read(&dest->weight) == 0))) { |
529 | IP_VS_DBG(9, "check_template: dest not available for " | |
530 | "protocol %s s:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d " | |
531 | "-> d:%u.%u.%u.%u:%d\n", | |
532 | ip_vs_proto_name(ct->protocol), | |
533 | NIPQUAD(ct->caddr), ntohs(ct->cport), | |
534 | NIPQUAD(ct->vaddr), ntohs(ct->vport), | |
535 | NIPQUAD(ct->daddr), ntohs(ct->dport)); | |
536 | ||
537 | /* | |
538 | * Invalidate the connection template | |
539 | */ | |
014d730d | 540 | if (ct->vport != htons(0xffff)) { |
1da177e4 | 541 | if (ip_vs_conn_unhash(ct)) { |
014d730d AV |
542 | ct->dport = htons(0xffff); |
543 | ct->vport = htons(0xffff); | |
1da177e4 LT |
544 | ct->cport = 0; |
545 | ip_vs_conn_hash(ct); | |
546 | } | |
547 | } | |
548 | ||
549 | /* | |
550 | * Simply decrease the refcnt of the template, | |
551 | * don't restart its timer. | |
552 | */ | |
553 | atomic_dec(&ct->refcnt); | |
554 | return 0; | |
555 | } | |
556 | return 1; | |
557 | } | |
558 | ||
559 | static void ip_vs_conn_expire(unsigned long data) | |
560 | { | |
561 | struct ip_vs_conn *cp = (struct ip_vs_conn *)data; | |
562 | ||
563 | cp->timeout = 60*HZ; | |
564 | ||
565 | /* | |
566 | * hey, I'm using it | |
567 | */ | |
568 | atomic_inc(&cp->refcnt); | |
569 | ||
570 | /* | |
571 | * do I control anybody? | |
572 | */ | |
573 | if (atomic_read(&cp->n_control)) | |
574 | goto expire_later; | |
575 | ||
576 | /* | |
577 | * unhash it if it is hashed in the conn table | |
578 | */ | |
579 | if (!ip_vs_conn_unhash(cp)) | |
580 | goto expire_later; | |
581 | ||
582 | /* | |
583 | * refcnt==1 implies I'm the only one referrer | |
584 | */ | |
585 | if (likely(atomic_read(&cp->refcnt) == 1)) { | |
586 | /* delete the timer if it is activated by other users */ | |
587 | if (timer_pending(&cp->timer)) | |
588 | del_timer(&cp->timer); | |
589 | ||
590 | /* does anybody control me? */ | |
591 | if (cp->control) | |
592 | ip_vs_control_del(cp); | |
593 | ||
594 | if (unlikely(cp->app != NULL)) | |
595 | ip_vs_unbind_app(cp); | |
596 | ip_vs_unbind_dest(cp); | |
597 | if (cp->flags & IP_VS_CONN_F_NO_CPORT) | |
598 | atomic_dec(&ip_vs_conn_no_cport_cnt); | |
599 | atomic_dec(&ip_vs_conn_count); | |
600 | ||
601 | kmem_cache_free(ip_vs_conn_cachep, cp); | |
602 | return; | |
603 | } | |
604 | ||
605 | /* hash it back to the table */ | |
606 | ip_vs_conn_hash(cp); | |
607 | ||
608 | expire_later: | |
4b5bdf5c | 609 | IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n", |
1da177e4 LT |
610 | atomic_read(&cp->refcnt)-1, |
611 | atomic_read(&cp->n_control)); | |
612 | ||
613 | ip_vs_conn_put(cp); | |
614 | } | |
615 | ||
616 | ||
617 | void ip_vs_conn_expire_now(struct ip_vs_conn *cp) | |
618 | { | |
619 | if (del_timer(&cp->timer)) | |
620 | mod_timer(&cp->timer, jiffies); | |
1da177e4 LT |
621 | } |
622 | ||
623 | ||
624 | /* | |
625 | * Create a new connection entry and hash it into the ip_vs_conn_tab | |
626 | */ | |
627 | struct ip_vs_conn * | |
014d730d AV |
628 | ip_vs_conn_new(int proto, __be32 caddr, __be16 cport, __be32 vaddr, __be16 vport, |
629 | __be32 daddr, __be16 dport, unsigned flags, | |
1da177e4 LT |
630 | struct ip_vs_dest *dest) |
631 | { | |
632 | struct ip_vs_conn *cp; | |
633 | struct ip_vs_protocol *pp = ip_vs_proto_get(proto); | |
634 | ||
c3762229 | 635 | cp = kmem_cache_zalloc(ip_vs_conn_cachep, GFP_ATOMIC); |
1da177e4 LT |
636 | if (cp == NULL) { |
637 | IP_VS_ERR_RL("ip_vs_conn_new: no memory available.\n"); | |
638 | return NULL; | |
639 | } | |
640 | ||
1da177e4 | 641 | INIT_LIST_HEAD(&cp->c_list); |
b24b8a24 | 642 | setup_timer(&cp->timer, ip_vs_conn_expire, (unsigned long)cp); |
1da177e4 LT |
643 | cp->protocol = proto; |
644 | cp->caddr = caddr; | |
645 | cp->cport = cport; | |
646 | cp->vaddr = vaddr; | |
647 | cp->vport = vport; | |
648 | cp->daddr = daddr; | |
649 | cp->dport = dport; | |
650 | cp->flags = flags; | |
651 | spin_lock_init(&cp->lock); | |
652 | ||
653 | /* | |
654 | * Set the entry is referenced by the current thread before hashing | |
655 | * it in the table, so that other thread run ip_vs_random_dropentry | |
656 | * but cannot drop this entry. | |
657 | */ | |
658 | atomic_set(&cp->refcnt, 1); | |
659 | ||
660 | atomic_set(&cp->n_control, 0); | |
661 | atomic_set(&cp->in_pkts, 0); | |
662 | ||
663 | atomic_inc(&ip_vs_conn_count); | |
664 | if (flags & IP_VS_CONN_F_NO_CPORT) | |
665 | atomic_inc(&ip_vs_conn_no_cport_cnt); | |
666 | ||
667 | /* Bind the connection with a destination server */ | |
668 | ip_vs_bind_dest(cp, dest); | |
669 | ||
670 | /* Set its state and timeout */ | |
671 | cp->state = 0; | |
672 | cp->timeout = 3*HZ; | |
673 | ||
674 | /* Bind its packet transmitter */ | |
675 | ip_vs_bind_xmit(cp); | |
676 | ||
677 | if (unlikely(pp && atomic_read(&pp->appcnt))) | |
678 | ip_vs_bind_app(cp, pp); | |
679 | ||
680 | /* Hash it in the ip_vs_conn_tab finally */ | |
681 | ip_vs_conn_hash(cp); | |
682 | ||
683 | return cp; | |
684 | } | |
685 | ||
686 | ||
687 | /* | |
688 | * /proc/net/ip_vs_conn entries | |
689 | */ | |
690 | #ifdef CONFIG_PROC_FS | |
691 | ||
692 | static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos) | |
693 | { | |
694 | int idx; | |
695 | struct ip_vs_conn *cp; | |
e905a9ed | 696 | |
1da177e4 LT |
697 | for(idx = 0; idx < IP_VS_CONN_TAB_SIZE; idx++) { |
698 | ct_read_lock_bh(idx); | |
699 | list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) { | |
700 | if (pos-- == 0) { | |
701 | seq->private = &ip_vs_conn_tab[idx]; | |
702 | return cp; | |
703 | } | |
704 | } | |
705 | ct_read_unlock_bh(idx); | |
706 | } | |
707 | ||
708 | return NULL; | |
709 | } | |
710 | ||
711 | static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos) | |
712 | { | |
713 | seq->private = NULL; | |
714 | return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN; | |
715 | } | |
716 | ||
717 | static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
718 | { | |
719 | struct ip_vs_conn *cp = v; | |
720 | struct list_head *e, *l = seq->private; | |
721 | int idx; | |
722 | ||
723 | ++*pos; | |
e905a9ed | 724 | if (v == SEQ_START_TOKEN) |
1da177e4 LT |
725 | return ip_vs_conn_array(seq, 0); |
726 | ||
727 | /* more on same hash chain? */ | |
728 | if ((e = cp->c_list.next) != l) | |
729 | return list_entry(e, struct ip_vs_conn, c_list); | |
730 | ||
731 | idx = l - ip_vs_conn_tab; | |
732 | ct_read_unlock_bh(idx); | |
733 | ||
734 | while (++idx < IP_VS_CONN_TAB_SIZE) { | |
735 | ct_read_lock_bh(idx); | |
736 | list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) { | |
737 | seq->private = &ip_vs_conn_tab[idx]; | |
738 | return cp; | |
e905a9ed | 739 | } |
1da177e4 LT |
740 | ct_read_unlock_bh(idx); |
741 | } | |
742 | seq->private = NULL; | |
743 | return NULL; | |
744 | } | |
745 | ||
746 | static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v) | |
747 | { | |
748 | struct list_head *l = seq->private; | |
749 | ||
750 | if (l) | |
751 | ct_read_unlock_bh(l - ip_vs_conn_tab); | |
752 | } | |
753 | ||
754 | static int ip_vs_conn_seq_show(struct seq_file *seq, void *v) | |
755 | { | |
756 | ||
757 | if (v == SEQ_START_TOKEN) | |
758 | seq_puts(seq, | |
759 | "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Expires\n"); | |
760 | else { | |
761 | const struct ip_vs_conn *cp = v; | |
762 | ||
763 | seq_printf(seq, | |
764 | "%-3s %08X %04X %08X %04X %08X %04X %-11s %7lu\n", | |
765 | ip_vs_proto_name(cp->protocol), | |
766 | ntohl(cp->caddr), ntohs(cp->cport), | |
767 | ntohl(cp->vaddr), ntohs(cp->vport), | |
768 | ntohl(cp->daddr), ntohs(cp->dport), | |
769 | ip_vs_state_name(cp->protocol, cp->state), | |
770 | (cp->timer.expires-jiffies)/HZ); | |
771 | } | |
772 | return 0; | |
773 | } | |
774 | ||
56b3d975 | 775 | static const struct seq_operations ip_vs_conn_seq_ops = { |
1da177e4 LT |
776 | .start = ip_vs_conn_seq_start, |
777 | .next = ip_vs_conn_seq_next, | |
778 | .stop = ip_vs_conn_seq_stop, | |
779 | .show = ip_vs_conn_seq_show, | |
780 | }; | |
781 | ||
782 | static int ip_vs_conn_open(struct inode *inode, struct file *file) | |
783 | { | |
784 | return seq_open(file, &ip_vs_conn_seq_ops); | |
785 | } | |
786 | ||
9a32144e | 787 | static const struct file_operations ip_vs_conn_fops = { |
1da177e4 LT |
788 | .owner = THIS_MODULE, |
789 | .open = ip_vs_conn_open, | |
790 | .read = seq_read, | |
791 | .llseek = seq_lseek, | |
792 | .release = seq_release, | |
793 | }; | |
7a4fbb1f RB |
794 | |
795 | static const char *ip_vs_origin_name(unsigned flags) | |
796 | { | |
797 | if (flags & IP_VS_CONN_F_SYNC) | |
798 | return "SYNC"; | |
799 | else | |
800 | return "LOCAL"; | |
801 | } | |
802 | ||
803 | static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v) | |
804 | { | |
805 | ||
806 | if (v == SEQ_START_TOKEN) | |
807 | seq_puts(seq, | |
808 | "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Origin Expires\n"); | |
809 | else { | |
810 | const struct ip_vs_conn *cp = v; | |
811 | ||
812 | seq_printf(seq, | |
813 | "%-3s %08X %04X %08X %04X %08X %04X %-11s %-6s %7lu\n", | |
814 | ip_vs_proto_name(cp->protocol), | |
815 | ntohl(cp->caddr), ntohs(cp->cport), | |
816 | ntohl(cp->vaddr), ntohs(cp->vport), | |
817 | ntohl(cp->daddr), ntohs(cp->dport), | |
818 | ip_vs_state_name(cp->protocol, cp->state), | |
819 | ip_vs_origin_name(cp->flags), | |
820 | (cp->timer.expires-jiffies)/HZ); | |
821 | } | |
822 | return 0; | |
823 | } | |
824 | ||
825 | static const struct seq_operations ip_vs_conn_sync_seq_ops = { | |
826 | .start = ip_vs_conn_seq_start, | |
827 | .next = ip_vs_conn_seq_next, | |
828 | .stop = ip_vs_conn_seq_stop, | |
829 | .show = ip_vs_conn_sync_seq_show, | |
830 | }; | |
831 | ||
832 | static int ip_vs_conn_sync_open(struct inode *inode, struct file *file) | |
833 | { | |
834 | return seq_open(file, &ip_vs_conn_sync_seq_ops); | |
835 | } | |
836 | ||
837 | static const struct file_operations ip_vs_conn_sync_fops = { | |
838 | .owner = THIS_MODULE, | |
839 | .open = ip_vs_conn_sync_open, | |
840 | .read = seq_read, | |
841 | .llseek = seq_lseek, | |
842 | .release = seq_release, | |
843 | }; | |
844 | ||
1da177e4 LT |
845 | #endif |
846 | ||
847 | ||
848 | /* | |
849 | * Randomly drop connection entries before running out of memory | |
850 | */ | |
851 | static inline int todrop_entry(struct ip_vs_conn *cp) | |
852 | { | |
853 | /* | |
854 | * The drop rate array needs tuning for real environments. | |
855 | * Called from timer bh only => no locking | |
856 | */ | |
9b5b5cff | 857 | static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8}; |
1da177e4 LT |
858 | static char todrop_counter[9] = {0}; |
859 | int i; | |
860 | ||
861 | /* if the conn entry hasn't lasted for 60 seconds, don't drop it. | |
862 | This will leave enough time for normal connection to get | |
863 | through. */ | |
864 | if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ)) | |
865 | return 0; | |
866 | ||
867 | /* Don't drop the entry if its number of incoming packets is not | |
868 | located in [0, 8] */ | |
869 | i = atomic_read(&cp->in_pkts); | |
870 | if (i > 8 || i < 0) return 0; | |
871 | ||
872 | if (!todrop_rate[i]) return 0; | |
873 | if (--todrop_counter[i] > 0) return 0; | |
874 | ||
875 | todrop_counter[i] = todrop_rate[i]; | |
876 | return 1; | |
877 | } | |
878 | ||
af9debd4 | 879 | /* Called from keventd and must protect itself from softirqs */ |
1da177e4 LT |
880 | void ip_vs_random_dropentry(void) |
881 | { | |
882 | int idx; | |
883 | struct ip_vs_conn *cp; | |
1da177e4 LT |
884 | |
885 | /* | |
886 | * Randomly scan 1/32 of the whole table every second | |
887 | */ | |
888 | for (idx = 0; idx < (IP_VS_CONN_TAB_SIZE>>5); idx++) { | |
889 | unsigned hash = net_random() & IP_VS_CONN_TAB_MASK; | |
890 | ||
891 | /* | |
892 | * Lock is actually needed in this loop. | |
893 | */ | |
af9debd4 | 894 | ct_write_lock_bh(hash); |
1da177e4 LT |
895 | |
896 | list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) { | |
87375ab4 | 897 | if (cp->flags & IP_VS_CONN_F_TEMPLATE) |
1da177e4 LT |
898 | /* connection template */ |
899 | continue; | |
900 | ||
901 | if (cp->protocol == IPPROTO_TCP) { | |
902 | switch(cp->state) { | |
903 | case IP_VS_TCP_S_SYN_RECV: | |
904 | case IP_VS_TCP_S_SYNACK: | |
905 | break; | |
906 | ||
907 | case IP_VS_TCP_S_ESTABLISHED: | |
908 | if (todrop_entry(cp)) | |
909 | break; | |
910 | continue; | |
911 | ||
912 | default: | |
913 | continue; | |
914 | } | |
915 | } else { | |
916 | if (!todrop_entry(cp)) | |
917 | continue; | |
918 | } | |
919 | ||
1da177e4 LT |
920 | IP_VS_DBG(4, "del connection\n"); |
921 | ip_vs_conn_expire_now(cp); | |
fb3d8949 | 922 | if (cp->control) { |
1da177e4 | 923 | IP_VS_DBG(4, "del conn template\n"); |
fb3d8949 | 924 | ip_vs_conn_expire_now(cp->control); |
1da177e4 | 925 | } |
1da177e4 | 926 | } |
af9debd4 | 927 | ct_write_unlock_bh(hash); |
1da177e4 LT |
928 | } |
929 | } | |
930 | ||
931 | ||
932 | /* | |
933 | * Flush all the connection entries in the ip_vs_conn_tab | |
934 | */ | |
935 | static void ip_vs_conn_flush(void) | |
936 | { | |
937 | int idx; | |
938 | struct ip_vs_conn *cp; | |
1da177e4 LT |
939 | |
940 | flush_again: | |
941 | for (idx=0; idx<IP_VS_CONN_TAB_SIZE; idx++) { | |
942 | /* | |
943 | * Lock is actually needed in this loop. | |
944 | */ | |
945 | ct_write_lock_bh(idx); | |
946 | ||
947 | list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) { | |
1da177e4 | 948 | |
1da177e4 LT |
949 | IP_VS_DBG(4, "del connection\n"); |
950 | ip_vs_conn_expire_now(cp); | |
fb3d8949 | 951 | if (cp->control) { |
1da177e4 | 952 | IP_VS_DBG(4, "del conn template\n"); |
fb3d8949 | 953 | ip_vs_conn_expire_now(cp->control); |
1da177e4 | 954 | } |
1da177e4 LT |
955 | } |
956 | ct_write_unlock_bh(idx); | |
957 | } | |
958 | ||
959 | /* the counter may be not NULL, because maybe some conn entries | |
960 | are run by slow timer handler or unhashed but still referred */ | |
961 | if (atomic_read(&ip_vs_conn_count) != 0) { | |
962 | schedule(); | |
963 | goto flush_again; | |
964 | } | |
965 | } | |
966 | ||
967 | ||
968 | int ip_vs_conn_init(void) | |
969 | { | |
970 | int idx; | |
971 | ||
972 | /* | |
973 | * Allocate the connection hash table and initialize its list heads | |
974 | */ | |
975 | ip_vs_conn_tab = vmalloc(IP_VS_CONN_TAB_SIZE*sizeof(struct list_head)); | |
976 | if (!ip_vs_conn_tab) | |
977 | return -ENOMEM; | |
978 | ||
979 | /* Allocate ip_vs_conn slab cache */ | |
980 | ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn", | |
981 | sizeof(struct ip_vs_conn), 0, | |
20c2df83 | 982 | SLAB_HWCACHE_ALIGN, NULL); |
1da177e4 LT |
983 | if (!ip_vs_conn_cachep) { |
984 | vfree(ip_vs_conn_tab); | |
985 | return -ENOMEM; | |
986 | } | |
987 | ||
988 | IP_VS_INFO("Connection hash table configured " | |
989 | "(size=%d, memory=%ldKbytes)\n", | |
990 | IP_VS_CONN_TAB_SIZE, | |
991 | (long)(IP_VS_CONN_TAB_SIZE*sizeof(struct list_head))/1024); | |
992 | IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n", | |
993 | sizeof(struct ip_vs_conn)); | |
994 | ||
995 | for (idx = 0; idx < IP_VS_CONN_TAB_SIZE; idx++) { | |
996 | INIT_LIST_HEAD(&ip_vs_conn_tab[idx]); | |
997 | } | |
998 | ||
999 | for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++) { | |
1000 | rwlock_init(&__ip_vs_conntbl_lock_array[idx].l); | |
1001 | } | |
1002 | ||
457c4cbc | 1003 | proc_net_fops_create(&init_net, "ip_vs_conn", 0, &ip_vs_conn_fops); |
7a4fbb1f | 1004 | proc_net_fops_create(&init_net, "ip_vs_conn_sync", 0, &ip_vs_conn_sync_fops); |
1da177e4 LT |
1005 | |
1006 | /* calculate the random value for connection hash */ | |
1007 | get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd)); | |
1008 | ||
1009 | return 0; | |
1010 | } | |
1011 | ||
1012 | ||
1013 | void ip_vs_conn_cleanup(void) | |
1014 | { | |
1015 | /* flush all the connection entries first */ | |
1016 | ip_vs_conn_flush(); | |
1017 | ||
1018 | /* Release the empty cache */ | |
1019 | kmem_cache_destroy(ip_vs_conn_cachep); | |
457c4cbc | 1020 | proc_net_remove(&init_net, "ip_vs_conn"); |
7a4fbb1f | 1021 | proc_net_remove(&init_net, "ip_vs_conn_sync"); |
1da177e4 LT |
1022 | vfree(ip_vs_conn_tab); |
1023 | } |