]>
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 | * | |
8 | * Version: $Id: ip_vs_conn.c,v 1.31 2003/04/18 09:03:16 wensong Exp $ | |
9 | * | |
10 | * Authors: Wensong Zhang <wensong@linuxvirtualserver.org> | |
11 | * Peter Kese <peter.kese@ijs.si> | |
12 | * Julian Anastasov <ja@ssi.bg> | |
13 | * | |
14 | * This program is free software; you can redistribute it and/or | |
15 | * modify it under the terms of the GNU General Public License | |
16 | * as published by the Free Software Foundation; either version | |
17 | * 2 of the License, or (at your option) any later version. | |
18 | * | |
19 | * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese, | |
20 | * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms | |
21 | * and others. Many code here is taken from IP MASQ code of kernel 2.2. | |
22 | * | |
23 | * Changes: | |
24 | * | |
25 | */ | |
26 | ||
e924283b | 27 | #include <linux/interrupt.h> |
14c85021 | 28 | #include <linux/in.h> |
f190055f | 29 | #include <linux/net.h> |
1da177e4 | 30 | #include <linux/kernel.h> |
14c85021 | 31 | #include <linux/module.h> |
1da177e4 LT |
32 | #include <linux/vmalloc.h> |
33 | #include <linux/proc_fs.h> /* for proc_net_* */ | |
34 | #include <linux/seq_file.h> | |
35 | #include <linux/jhash.h> | |
36 | #include <linux/random.h> | |
37 | ||
457c4cbc | 38 | #include <net/net_namespace.h> |
1da177e4 LT |
39 | #include <net/ip_vs.h> |
40 | ||
41 | ||
42 | /* | |
43 | * Connection hash table: for input and output packets lookups of IPVS | |
44 | */ | |
45 | static struct list_head *ip_vs_conn_tab; | |
46 | ||
47 | /* SLAB cache for IPVS connections */ | |
e18b890b | 48 | static struct kmem_cache *ip_vs_conn_cachep __read_mostly; |
1da177e4 LT |
49 | |
50 | /* counter for current IPVS connections */ | |
51 | static atomic_t ip_vs_conn_count = ATOMIC_INIT(0); | |
52 | ||
53 | /* counter for no client port connections */ | |
54 | static atomic_t ip_vs_conn_no_cport_cnt = ATOMIC_INIT(0); | |
55 | ||
56 | /* random value for IPVS connection hash */ | |
57 | static unsigned int ip_vs_conn_rnd; | |
58 | ||
59 | /* | |
60 | * Fine locking granularity for big connection hash table | |
61 | */ | |
62 | #define CT_LOCKARRAY_BITS 4 | |
63 | #define CT_LOCKARRAY_SIZE (1<<CT_LOCKARRAY_BITS) | |
64 | #define CT_LOCKARRAY_MASK (CT_LOCKARRAY_SIZE-1) | |
65 | ||
66 | struct ip_vs_aligned_lock | |
67 | { | |
68 | rwlock_t l; | |
69 | } __attribute__((__aligned__(SMP_CACHE_BYTES))); | |
70 | ||
71 | /* lock array for conn table */ | |
72 | static struct ip_vs_aligned_lock | |
73 | __ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned; | |
74 | ||
75 | static inline void ct_read_lock(unsigned key) | |
76 | { | |
77 | read_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l); | |
78 | } | |
79 | ||
80 | static inline void ct_read_unlock(unsigned key) | |
81 | { | |
82 | read_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l); | |
83 | } | |
84 | ||
85 | static inline void ct_write_lock(unsigned key) | |
86 | { | |
87 | write_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l); | |
88 | } | |
89 | ||
90 | static inline void ct_write_unlock(unsigned key) | |
91 | { | |
92 | write_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l); | |
93 | } | |
94 | ||
95 | static inline void ct_read_lock_bh(unsigned key) | |
96 | { | |
97 | read_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l); | |
98 | } | |
99 | ||
100 | static inline void ct_read_unlock_bh(unsigned key) | |
101 | { | |
102 | read_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l); | |
103 | } | |
104 | ||
105 | static inline void ct_write_lock_bh(unsigned key) | |
106 | { | |
107 | write_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l); | |
108 | } | |
109 | ||
110 | static inline void ct_write_unlock_bh(unsigned key) | |
111 | { | |
112 | write_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l); | |
113 | } | |
114 | ||
115 | ||
116 | /* | |
117 | * Returns hash value for IPVS connection entry | |
118 | */ | |
014d730d | 119 | static unsigned int ip_vs_conn_hashkey(unsigned proto, __be32 addr, __be16 port) |
1da177e4 | 120 | { |
014d730d | 121 | return jhash_3words((__force u32)addr, (__force u32)port, proto, ip_vs_conn_rnd) |
1da177e4 LT |
122 | & IP_VS_CONN_TAB_MASK; |
123 | } | |
124 | ||
125 | ||
126 | /* | |
127 | * Hashes ip_vs_conn in ip_vs_conn_tab by proto,addr,port. | |
128 | * returns bool success. | |
129 | */ | |
130 | static inline int ip_vs_conn_hash(struct ip_vs_conn *cp) | |
131 | { | |
132 | unsigned hash; | |
133 | int ret; | |
134 | ||
135 | /* Hash by protocol, client address and port */ | |
136 | hash = ip_vs_conn_hashkey(cp->protocol, cp->caddr, cp->cport); | |
137 | ||
138 | ct_write_lock(hash); | |
139 | ||
140 | if (!(cp->flags & IP_VS_CONN_F_HASHED)) { | |
141 | list_add(&cp->c_list, &ip_vs_conn_tab[hash]); | |
142 | cp->flags |= IP_VS_CONN_F_HASHED; | |
143 | atomic_inc(&cp->refcnt); | |
144 | ret = 1; | |
145 | } else { | |
146 | IP_VS_ERR("ip_vs_conn_hash(): request for already hashed, " | |
147 | "called from %p\n", __builtin_return_address(0)); | |
148 | ret = 0; | |
149 | } | |
150 | ||
151 | ct_write_unlock(hash); | |
152 | ||
153 | return ret; | |
154 | } | |
155 | ||
156 | ||
157 | /* | |
158 | * UNhashes ip_vs_conn from ip_vs_conn_tab. | |
159 | * returns bool success. | |
160 | */ | |
161 | static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp) | |
162 | { | |
163 | unsigned hash; | |
164 | int ret; | |
165 | ||
166 | /* unhash it and decrease its reference counter */ | |
167 | hash = ip_vs_conn_hashkey(cp->protocol, cp->caddr, cp->cport); | |
168 | ||
169 | ct_write_lock(hash); | |
170 | ||
171 | if (cp->flags & IP_VS_CONN_F_HASHED) { | |
172 | list_del(&cp->c_list); | |
173 | cp->flags &= ~IP_VS_CONN_F_HASHED; | |
174 | atomic_dec(&cp->refcnt); | |
175 | ret = 1; | |
176 | } else | |
177 | ret = 0; | |
178 | ||
179 | ct_write_unlock(hash); | |
180 | ||
181 | return ret; | |
182 | } | |
183 | ||
184 | ||
185 | /* | |
186 | * Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab. | |
187 | * Called for pkts coming from OUTside-to-INside. | |
188 | * s_addr, s_port: pkt source address (foreign host) | |
189 | * d_addr, d_port: pkt dest address (load balancer) | |
190 | */ | |
191 | static inline struct ip_vs_conn *__ip_vs_conn_in_get | |
014d730d | 192 | (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port) |
1da177e4 LT |
193 | { |
194 | unsigned hash; | |
195 | struct ip_vs_conn *cp; | |
196 | ||
197 | hash = ip_vs_conn_hashkey(protocol, s_addr, s_port); | |
198 | ||
199 | ct_read_lock(hash); | |
200 | ||
201 | list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) { | |
202 | if (s_addr==cp->caddr && s_port==cp->cport && | |
203 | d_port==cp->vport && d_addr==cp->vaddr && | |
87375ab4 | 204 | ((!s_port) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) && |
1da177e4 LT |
205 | protocol==cp->protocol) { |
206 | /* HIT */ | |
207 | atomic_inc(&cp->refcnt); | |
208 | ct_read_unlock(hash); | |
209 | return cp; | |
210 | } | |
211 | } | |
212 | ||
213 | ct_read_unlock(hash); | |
214 | ||
215 | return NULL; | |
216 | } | |
217 | ||
218 | struct ip_vs_conn *ip_vs_conn_in_get | |
014d730d | 219 | (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port) |
1da177e4 LT |
220 | { |
221 | struct ip_vs_conn *cp; | |
222 | ||
223 | cp = __ip_vs_conn_in_get(protocol, s_addr, s_port, d_addr, d_port); | |
224 | if (!cp && atomic_read(&ip_vs_conn_no_cport_cnt)) | |
225 | cp = __ip_vs_conn_in_get(protocol, s_addr, 0, d_addr, d_port); | |
226 | ||
4b5bdf5c | 227 | IP_VS_DBG(9, "lookup/in %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n", |
1da177e4 LT |
228 | ip_vs_proto_name(protocol), |
229 | NIPQUAD(s_addr), ntohs(s_port), | |
230 | NIPQUAD(d_addr), ntohs(d_port), | |
231 | cp?"hit":"not hit"); | |
232 | ||
233 | return cp; | |
234 | } | |
235 | ||
87375ab4 JA |
236 | /* Get reference to connection template */ |
237 | struct ip_vs_conn *ip_vs_ct_in_get | |
014d730d | 238 | (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port) |
87375ab4 JA |
239 | { |
240 | unsigned hash; | |
241 | struct ip_vs_conn *cp; | |
242 | ||
243 | hash = ip_vs_conn_hashkey(protocol, s_addr, s_port); | |
244 | ||
245 | ct_read_lock(hash); | |
246 | ||
247 | list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) { | |
248 | if (s_addr==cp->caddr && s_port==cp->cport && | |
249 | d_port==cp->vport && d_addr==cp->vaddr && | |
250 | cp->flags & IP_VS_CONN_F_TEMPLATE && | |
251 | protocol==cp->protocol) { | |
252 | /* HIT */ | |
253 | atomic_inc(&cp->refcnt); | |
254 | goto out; | |
255 | } | |
256 | } | |
257 | cp = NULL; | |
258 | ||
259 | out: | |
260 | ct_read_unlock(hash); | |
261 | ||
4b5bdf5c | 262 | IP_VS_DBG(9, "template lookup/in %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n", |
87375ab4 JA |
263 | ip_vs_proto_name(protocol), |
264 | NIPQUAD(s_addr), ntohs(s_port), | |
265 | NIPQUAD(d_addr), ntohs(d_port), | |
266 | cp?"hit":"not hit"); | |
267 | ||
268 | return cp; | |
269 | } | |
1da177e4 LT |
270 | |
271 | /* | |
272 | * Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab. | |
273 | * Called for pkts coming from inside-to-OUTside. | |
274 | * s_addr, s_port: pkt source address (inside host) | |
275 | * d_addr, d_port: pkt dest address (foreign host) | |
276 | */ | |
277 | struct ip_vs_conn *ip_vs_conn_out_get | |
014d730d | 278 | (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port) |
1da177e4 LT |
279 | { |
280 | unsigned hash; | |
281 | struct ip_vs_conn *cp, *ret=NULL; | |
282 | ||
283 | /* | |
284 | * Check for "full" addressed entries | |
285 | */ | |
286 | hash = ip_vs_conn_hashkey(protocol, d_addr, d_port); | |
287 | ||
288 | ct_read_lock(hash); | |
289 | ||
290 | list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) { | |
291 | if (d_addr == cp->caddr && d_port == cp->cport && | |
292 | s_port == cp->dport && s_addr == cp->daddr && | |
293 | protocol == cp->protocol) { | |
294 | /* HIT */ | |
295 | atomic_inc(&cp->refcnt); | |
296 | ret = cp; | |
297 | break; | |
298 | } | |
299 | } | |
300 | ||
301 | ct_read_unlock(hash); | |
302 | ||
4b5bdf5c | 303 | IP_VS_DBG(9, "lookup/out %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n", |
1da177e4 LT |
304 | ip_vs_proto_name(protocol), |
305 | NIPQUAD(s_addr), ntohs(s_port), | |
306 | NIPQUAD(d_addr), ntohs(d_port), | |
307 | ret?"hit":"not hit"); | |
308 | ||
309 | return ret; | |
310 | } | |
311 | ||
312 | ||
313 | /* | |
314 | * Put back the conn and restart its timer with its timeout | |
315 | */ | |
316 | void ip_vs_conn_put(struct ip_vs_conn *cp) | |
317 | { | |
318 | /* reset it expire in its timeout */ | |
319 | mod_timer(&cp->timer, jiffies+cp->timeout); | |
320 | ||
321 | __ip_vs_conn_put(cp); | |
322 | } | |
323 | ||
324 | ||
325 | /* | |
326 | * Fill a no_client_port connection with a client port number | |
327 | */ | |
014d730d | 328 | void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport) |
1da177e4 LT |
329 | { |
330 | if (ip_vs_conn_unhash(cp)) { | |
331 | spin_lock(&cp->lock); | |
332 | if (cp->flags & IP_VS_CONN_F_NO_CPORT) { | |
333 | atomic_dec(&ip_vs_conn_no_cport_cnt); | |
334 | cp->flags &= ~IP_VS_CONN_F_NO_CPORT; | |
335 | cp->cport = cport; | |
336 | } | |
337 | spin_unlock(&cp->lock); | |
338 | ||
339 | /* hash on new dport */ | |
340 | ip_vs_conn_hash(cp); | |
341 | } | |
342 | } | |
343 | ||
344 | ||
345 | /* | |
346 | * Bind a connection entry with the corresponding packet_xmit. | |
347 | * Called by ip_vs_conn_new. | |
348 | */ | |
349 | static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp) | |
350 | { | |
351 | switch (IP_VS_FWD_METHOD(cp)) { | |
352 | case IP_VS_CONN_F_MASQ: | |
353 | cp->packet_xmit = ip_vs_nat_xmit; | |
354 | break; | |
355 | ||
356 | case IP_VS_CONN_F_TUNNEL: | |
357 | cp->packet_xmit = ip_vs_tunnel_xmit; | |
358 | break; | |
359 | ||
360 | case IP_VS_CONN_F_DROUTE: | |
361 | cp->packet_xmit = ip_vs_dr_xmit; | |
362 | break; | |
363 | ||
364 | case IP_VS_CONN_F_LOCALNODE: | |
365 | cp->packet_xmit = ip_vs_null_xmit; | |
366 | break; | |
367 | ||
368 | case IP_VS_CONN_F_BYPASS: | |
369 | cp->packet_xmit = ip_vs_bypass_xmit; | |
370 | break; | |
371 | } | |
372 | } | |
373 | ||
374 | ||
375 | static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest) | |
376 | { | |
377 | return atomic_read(&dest->activeconns) | |
378 | + atomic_read(&dest->inactconns); | |
379 | } | |
380 | ||
381 | /* | |
382 | * Bind a connection entry with a virtual service destination | |
383 | * Called just after a new connection entry is created. | |
384 | */ | |
385 | static inline void | |
386 | ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest) | |
387 | { | |
388 | /* if dest is NULL, then return directly */ | |
389 | if (!dest) | |
390 | return; | |
391 | ||
392 | /* Increase the refcnt counter of the dest */ | |
393 | atomic_inc(&dest->refcnt); | |
394 | ||
395 | /* Bind with the destination and its corresponding transmitter */ | |
396 | cp->flags |= atomic_read(&dest->conn_flags); | |
397 | cp->dest = dest; | |
398 | ||
4b5bdf5c RN |
399 | IP_VS_DBG(7, "Bind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d " |
400 | "d:%u.%u.%u.%u:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d " | |
401 | "dest->refcnt:%d\n", | |
1da177e4 LT |
402 | ip_vs_proto_name(cp->protocol), |
403 | NIPQUAD(cp->caddr), ntohs(cp->cport), | |
404 | NIPQUAD(cp->vaddr), ntohs(cp->vport), | |
405 | NIPQUAD(cp->daddr), ntohs(cp->dport), | |
406 | ip_vs_fwd_tag(cp), cp->state, | |
407 | cp->flags, atomic_read(&cp->refcnt), | |
408 | atomic_read(&dest->refcnt)); | |
409 | ||
410 | /* Update the connection counters */ | |
87375ab4 | 411 | if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) { |
1da177e4 LT |
412 | /* It is a normal connection, so increase the inactive |
413 | connection counter because it is in TCP SYNRECV | |
414 | state (inactive) or other protocol inacive state */ | |
415 | atomic_inc(&dest->inactconns); | |
416 | } else { | |
417 | /* It is a persistent connection/template, so increase | |
418 | the peristent connection counter */ | |
419 | atomic_inc(&dest->persistconns); | |
420 | } | |
421 | ||
422 | if (dest->u_threshold != 0 && | |
423 | ip_vs_dest_totalconns(dest) >= dest->u_threshold) | |
424 | dest->flags |= IP_VS_DEST_F_OVERLOAD; | |
425 | } | |
426 | ||
427 | ||
428 | /* | |
429 | * Unbind a connection entry with its VS destination | |
430 | * Called by the ip_vs_conn_expire function. | |
431 | */ | |
432 | static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp) | |
433 | { | |
434 | struct ip_vs_dest *dest = cp->dest; | |
435 | ||
436 | if (!dest) | |
437 | return; | |
438 | ||
4b5bdf5c RN |
439 | IP_VS_DBG(7, "Unbind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d " |
440 | "d:%u.%u.%u.%u:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d " | |
441 | "dest->refcnt:%d\n", | |
1da177e4 LT |
442 | ip_vs_proto_name(cp->protocol), |
443 | NIPQUAD(cp->caddr), ntohs(cp->cport), | |
444 | NIPQUAD(cp->vaddr), ntohs(cp->vport), | |
445 | NIPQUAD(cp->daddr), ntohs(cp->dport), | |
446 | ip_vs_fwd_tag(cp), cp->state, | |
447 | cp->flags, atomic_read(&cp->refcnt), | |
448 | atomic_read(&dest->refcnt)); | |
449 | ||
450 | /* Update the connection counters */ | |
87375ab4 | 451 | if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) { |
1da177e4 LT |
452 | /* It is a normal connection, so decrease the inactconns |
453 | or activeconns counter */ | |
454 | if (cp->flags & IP_VS_CONN_F_INACTIVE) { | |
455 | atomic_dec(&dest->inactconns); | |
456 | } else { | |
457 | atomic_dec(&dest->activeconns); | |
458 | } | |
459 | } else { | |
460 | /* It is a persistent connection/template, so decrease | |
461 | the peristent connection counter */ | |
462 | atomic_dec(&dest->persistconns); | |
463 | } | |
464 | ||
465 | if (dest->l_threshold != 0) { | |
466 | if (ip_vs_dest_totalconns(dest) < dest->l_threshold) | |
467 | dest->flags &= ~IP_VS_DEST_F_OVERLOAD; | |
468 | } else if (dest->u_threshold != 0) { | |
469 | if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3) | |
470 | dest->flags &= ~IP_VS_DEST_F_OVERLOAD; | |
471 | } else { | |
472 | if (dest->flags & IP_VS_DEST_F_OVERLOAD) | |
473 | dest->flags &= ~IP_VS_DEST_F_OVERLOAD; | |
474 | } | |
475 | ||
476 | /* | |
477 | * Simply decrease the refcnt of the dest, because the | |
478 | * dest will be either in service's destination list | |
479 | * or in the trash. | |
480 | */ | |
481 | atomic_dec(&dest->refcnt); | |
482 | } | |
483 | ||
484 | ||
485 | /* | |
486 | * Checking if the destination of a connection template is available. | |
487 | * If available, return 1, otherwise invalidate this connection | |
488 | * template and return 0. | |
489 | */ | |
490 | int ip_vs_check_template(struct ip_vs_conn *ct) | |
491 | { | |
492 | struct ip_vs_dest *dest = ct->dest; | |
493 | ||
494 | /* | |
495 | * Checking the dest server status. | |
496 | */ | |
497 | if ((dest == NULL) || | |
e905a9ed YH |
498 | !(dest->flags & IP_VS_DEST_F_AVAILABLE) || |
499 | (sysctl_ip_vs_expire_quiescent_template && | |
1da177e4 LT |
500 | (atomic_read(&dest->weight) == 0))) { |
501 | IP_VS_DBG(9, "check_template: dest not available for " | |
502 | "protocol %s s:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d " | |
503 | "-> d:%u.%u.%u.%u:%d\n", | |
504 | ip_vs_proto_name(ct->protocol), | |
505 | NIPQUAD(ct->caddr), ntohs(ct->cport), | |
506 | NIPQUAD(ct->vaddr), ntohs(ct->vport), | |
507 | NIPQUAD(ct->daddr), ntohs(ct->dport)); | |
508 | ||
509 | /* | |
510 | * Invalidate the connection template | |
511 | */ | |
014d730d | 512 | if (ct->vport != htons(0xffff)) { |
1da177e4 | 513 | if (ip_vs_conn_unhash(ct)) { |
014d730d AV |
514 | ct->dport = htons(0xffff); |
515 | ct->vport = htons(0xffff); | |
1da177e4 LT |
516 | ct->cport = 0; |
517 | ip_vs_conn_hash(ct); | |
518 | } | |
519 | } | |
520 | ||
521 | /* | |
522 | * Simply decrease the refcnt of the template, | |
523 | * don't restart its timer. | |
524 | */ | |
525 | atomic_dec(&ct->refcnt); | |
526 | return 0; | |
527 | } | |
528 | return 1; | |
529 | } | |
530 | ||
531 | static void ip_vs_conn_expire(unsigned long data) | |
532 | { | |
533 | struct ip_vs_conn *cp = (struct ip_vs_conn *)data; | |
534 | ||
535 | cp->timeout = 60*HZ; | |
536 | ||
537 | /* | |
538 | * hey, I'm using it | |
539 | */ | |
540 | atomic_inc(&cp->refcnt); | |
541 | ||
542 | /* | |
543 | * do I control anybody? | |
544 | */ | |
545 | if (atomic_read(&cp->n_control)) | |
546 | goto expire_later; | |
547 | ||
548 | /* | |
549 | * unhash it if it is hashed in the conn table | |
550 | */ | |
551 | if (!ip_vs_conn_unhash(cp)) | |
552 | goto expire_later; | |
553 | ||
554 | /* | |
555 | * refcnt==1 implies I'm the only one referrer | |
556 | */ | |
557 | if (likely(atomic_read(&cp->refcnt) == 1)) { | |
558 | /* delete the timer if it is activated by other users */ | |
559 | if (timer_pending(&cp->timer)) | |
560 | del_timer(&cp->timer); | |
561 | ||
562 | /* does anybody control me? */ | |
563 | if (cp->control) | |
564 | ip_vs_control_del(cp); | |
565 | ||
566 | if (unlikely(cp->app != NULL)) | |
567 | ip_vs_unbind_app(cp); | |
568 | ip_vs_unbind_dest(cp); | |
569 | if (cp->flags & IP_VS_CONN_F_NO_CPORT) | |
570 | atomic_dec(&ip_vs_conn_no_cport_cnt); | |
571 | atomic_dec(&ip_vs_conn_count); | |
572 | ||
573 | kmem_cache_free(ip_vs_conn_cachep, cp); | |
574 | return; | |
575 | } | |
576 | ||
577 | /* hash it back to the table */ | |
578 | ip_vs_conn_hash(cp); | |
579 | ||
580 | expire_later: | |
4b5bdf5c | 581 | IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n", |
1da177e4 LT |
582 | atomic_read(&cp->refcnt)-1, |
583 | atomic_read(&cp->n_control)); | |
584 | ||
585 | ip_vs_conn_put(cp); | |
586 | } | |
587 | ||
588 | ||
589 | void ip_vs_conn_expire_now(struct ip_vs_conn *cp) | |
590 | { | |
591 | if (del_timer(&cp->timer)) | |
592 | mod_timer(&cp->timer, jiffies); | |
1da177e4 LT |
593 | } |
594 | ||
595 | ||
596 | /* | |
597 | * Create a new connection entry and hash it into the ip_vs_conn_tab | |
598 | */ | |
599 | struct ip_vs_conn * | |
014d730d AV |
600 | ip_vs_conn_new(int proto, __be32 caddr, __be16 cport, __be32 vaddr, __be16 vport, |
601 | __be32 daddr, __be16 dport, unsigned flags, | |
1da177e4 LT |
602 | struct ip_vs_dest *dest) |
603 | { | |
604 | struct ip_vs_conn *cp; | |
605 | struct ip_vs_protocol *pp = ip_vs_proto_get(proto); | |
606 | ||
c3762229 | 607 | cp = kmem_cache_zalloc(ip_vs_conn_cachep, GFP_ATOMIC); |
1da177e4 LT |
608 | if (cp == NULL) { |
609 | IP_VS_ERR_RL("ip_vs_conn_new: no memory available.\n"); | |
610 | return NULL; | |
611 | } | |
612 | ||
1da177e4 LT |
613 | INIT_LIST_HEAD(&cp->c_list); |
614 | init_timer(&cp->timer); | |
615 | cp->timer.data = (unsigned long)cp; | |
616 | cp->timer.function = ip_vs_conn_expire; | |
617 | cp->protocol = proto; | |
618 | cp->caddr = caddr; | |
619 | cp->cport = cport; | |
620 | cp->vaddr = vaddr; | |
621 | cp->vport = vport; | |
622 | cp->daddr = daddr; | |
623 | cp->dport = dport; | |
624 | cp->flags = flags; | |
625 | spin_lock_init(&cp->lock); | |
626 | ||
627 | /* | |
628 | * Set the entry is referenced by the current thread before hashing | |
629 | * it in the table, so that other thread run ip_vs_random_dropentry | |
630 | * but cannot drop this entry. | |
631 | */ | |
632 | atomic_set(&cp->refcnt, 1); | |
633 | ||
634 | atomic_set(&cp->n_control, 0); | |
635 | atomic_set(&cp->in_pkts, 0); | |
636 | ||
637 | atomic_inc(&ip_vs_conn_count); | |
638 | if (flags & IP_VS_CONN_F_NO_CPORT) | |
639 | atomic_inc(&ip_vs_conn_no_cport_cnt); | |
640 | ||
641 | /* Bind the connection with a destination server */ | |
642 | ip_vs_bind_dest(cp, dest); | |
643 | ||
644 | /* Set its state and timeout */ | |
645 | cp->state = 0; | |
646 | cp->timeout = 3*HZ; | |
647 | ||
648 | /* Bind its packet transmitter */ | |
649 | ip_vs_bind_xmit(cp); | |
650 | ||
651 | if (unlikely(pp && atomic_read(&pp->appcnt))) | |
652 | ip_vs_bind_app(cp, pp); | |
653 | ||
654 | /* Hash it in the ip_vs_conn_tab finally */ | |
655 | ip_vs_conn_hash(cp); | |
656 | ||
657 | return cp; | |
658 | } | |
659 | ||
660 | ||
661 | /* | |
662 | * /proc/net/ip_vs_conn entries | |
663 | */ | |
664 | #ifdef CONFIG_PROC_FS | |
665 | ||
666 | static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos) | |
667 | { | |
668 | int idx; | |
669 | struct ip_vs_conn *cp; | |
e905a9ed | 670 | |
1da177e4 LT |
671 | for(idx = 0; idx < IP_VS_CONN_TAB_SIZE; idx++) { |
672 | ct_read_lock_bh(idx); | |
673 | list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) { | |
674 | if (pos-- == 0) { | |
675 | seq->private = &ip_vs_conn_tab[idx]; | |
676 | return cp; | |
677 | } | |
678 | } | |
679 | ct_read_unlock_bh(idx); | |
680 | } | |
681 | ||
682 | return NULL; | |
683 | } | |
684 | ||
685 | static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos) | |
686 | { | |
687 | seq->private = NULL; | |
688 | return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN; | |
689 | } | |
690 | ||
691 | static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
692 | { | |
693 | struct ip_vs_conn *cp = v; | |
694 | struct list_head *e, *l = seq->private; | |
695 | int idx; | |
696 | ||
697 | ++*pos; | |
e905a9ed | 698 | if (v == SEQ_START_TOKEN) |
1da177e4 LT |
699 | return ip_vs_conn_array(seq, 0); |
700 | ||
701 | /* more on same hash chain? */ | |
702 | if ((e = cp->c_list.next) != l) | |
703 | return list_entry(e, struct ip_vs_conn, c_list); | |
704 | ||
705 | idx = l - ip_vs_conn_tab; | |
706 | ct_read_unlock_bh(idx); | |
707 | ||
708 | while (++idx < IP_VS_CONN_TAB_SIZE) { | |
709 | ct_read_lock_bh(idx); | |
710 | list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) { | |
711 | seq->private = &ip_vs_conn_tab[idx]; | |
712 | return cp; | |
e905a9ed | 713 | } |
1da177e4 LT |
714 | ct_read_unlock_bh(idx); |
715 | } | |
716 | seq->private = NULL; | |
717 | return NULL; | |
718 | } | |
719 | ||
720 | static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v) | |
721 | { | |
722 | struct list_head *l = seq->private; | |
723 | ||
724 | if (l) | |
725 | ct_read_unlock_bh(l - ip_vs_conn_tab); | |
726 | } | |
727 | ||
728 | static int ip_vs_conn_seq_show(struct seq_file *seq, void *v) | |
729 | { | |
730 | ||
731 | if (v == SEQ_START_TOKEN) | |
732 | seq_puts(seq, | |
733 | "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Expires\n"); | |
734 | else { | |
735 | const struct ip_vs_conn *cp = v; | |
736 | ||
737 | seq_printf(seq, | |
738 | "%-3s %08X %04X %08X %04X %08X %04X %-11s %7lu\n", | |
739 | ip_vs_proto_name(cp->protocol), | |
740 | ntohl(cp->caddr), ntohs(cp->cport), | |
741 | ntohl(cp->vaddr), ntohs(cp->vport), | |
742 | ntohl(cp->daddr), ntohs(cp->dport), | |
743 | ip_vs_state_name(cp->protocol, cp->state), | |
744 | (cp->timer.expires-jiffies)/HZ); | |
745 | } | |
746 | return 0; | |
747 | } | |
748 | ||
56b3d975 | 749 | static const struct seq_operations ip_vs_conn_seq_ops = { |
1da177e4 LT |
750 | .start = ip_vs_conn_seq_start, |
751 | .next = ip_vs_conn_seq_next, | |
752 | .stop = ip_vs_conn_seq_stop, | |
753 | .show = ip_vs_conn_seq_show, | |
754 | }; | |
755 | ||
756 | static int ip_vs_conn_open(struct inode *inode, struct file *file) | |
757 | { | |
758 | return seq_open(file, &ip_vs_conn_seq_ops); | |
759 | } | |
760 | ||
9a32144e | 761 | static const struct file_operations ip_vs_conn_fops = { |
1da177e4 LT |
762 | .owner = THIS_MODULE, |
763 | .open = ip_vs_conn_open, | |
764 | .read = seq_read, | |
765 | .llseek = seq_lseek, | |
766 | .release = seq_release, | |
767 | }; | |
768 | #endif | |
769 | ||
770 | ||
771 | /* | |
772 | * Randomly drop connection entries before running out of memory | |
773 | */ | |
774 | static inline int todrop_entry(struct ip_vs_conn *cp) | |
775 | { | |
776 | /* | |
777 | * The drop rate array needs tuning for real environments. | |
778 | * Called from timer bh only => no locking | |
779 | */ | |
9b5b5cff | 780 | static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8}; |
1da177e4 LT |
781 | static char todrop_counter[9] = {0}; |
782 | int i; | |
783 | ||
784 | /* if the conn entry hasn't lasted for 60 seconds, don't drop it. | |
785 | This will leave enough time for normal connection to get | |
786 | through. */ | |
787 | if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ)) | |
788 | return 0; | |
789 | ||
790 | /* Don't drop the entry if its number of incoming packets is not | |
791 | located in [0, 8] */ | |
792 | i = atomic_read(&cp->in_pkts); | |
793 | if (i > 8 || i < 0) return 0; | |
794 | ||
795 | if (!todrop_rate[i]) return 0; | |
796 | if (--todrop_counter[i] > 0) return 0; | |
797 | ||
798 | todrop_counter[i] = todrop_rate[i]; | |
799 | return 1; | |
800 | } | |
801 | ||
af9debd4 | 802 | /* Called from keventd and must protect itself from softirqs */ |
1da177e4 LT |
803 | void ip_vs_random_dropentry(void) |
804 | { | |
805 | int idx; | |
806 | struct ip_vs_conn *cp; | |
1da177e4 LT |
807 | |
808 | /* | |
809 | * Randomly scan 1/32 of the whole table every second | |
810 | */ | |
811 | for (idx = 0; idx < (IP_VS_CONN_TAB_SIZE>>5); idx++) { | |
812 | unsigned hash = net_random() & IP_VS_CONN_TAB_MASK; | |
813 | ||
814 | /* | |
815 | * Lock is actually needed in this loop. | |
816 | */ | |
af9debd4 | 817 | ct_write_lock_bh(hash); |
1da177e4 LT |
818 | |
819 | list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) { | |
87375ab4 | 820 | if (cp->flags & IP_VS_CONN_F_TEMPLATE) |
1da177e4 LT |
821 | /* connection template */ |
822 | continue; | |
823 | ||
824 | if (cp->protocol == IPPROTO_TCP) { | |
825 | switch(cp->state) { | |
826 | case IP_VS_TCP_S_SYN_RECV: | |
827 | case IP_VS_TCP_S_SYNACK: | |
828 | break; | |
829 | ||
830 | case IP_VS_TCP_S_ESTABLISHED: | |
831 | if (todrop_entry(cp)) | |
832 | break; | |
833 | continue; | |
834 | ||
835 | default: | |
836 | continue; | |
837 | } | |
838 | } else { | |
839 | if (!todrop_entry(cp)) | |
840 | continue; | |
841 | } | |
842 | ||
1da177e4 LT |
843 | IP_VS_DBG(4, "del connection\n"); |
844 | ip_vs_conn_expire_now(cp); | |
fb3d8949 | 845 | if (cp->control) { |
1da177e4 | 846 | IP_VS_DBG(4, "del conn template\n"); |
fb3d8949 | 847 | ip_vs_conn_expire_now(cp->control); |
1da177e4 | 848 | } |
1da177e4 | 849 | } |
af9debd4 | 850 | ct_write_unlock_bh(hash); |
1da177e4 LT |
851 | } |
852 | } | |
853 | ||
854 | ||
855 | /* | |
856 | * Flush all the connection entries in the ip_vs_conn_tab | |
857 | */ | |
858 | static void ip_vs_conn_flush(void) | |
859 | { | |
860 | int idx; | |
861 | struct ip_vs_conn *cp; | |
1da177e4 LT |
862 | |
863 | flush_again: | |
864 | for (idx=0; idx<IP_VS_CONN_TAB_SIZE; idx++) { | |
865 | /* | |
866 | * Lock is actually needed in this loop. | |
867 | */ | |
868 | ct_write_lock_bh(idx); | |
869 | ||
870 | list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) { | |
1da177e4 | 871 | |
1da177e4 LT |
872 | IP_VS_DBG(4, "del connection\n"); |
873 | ip_vs_conn_expire_now(cp); | |
fb3d8949 | 874 | if (cp->control) { |
1da177e4 | 875 | IP_VS_DBG(4, "del conn template\n"); |
fb3d8949 | 876 | ip_vs_conn_expire_now(cp->control); |
1da177e4 | 877 | } |
1da177e4 LT |
878 | } |
879 | ct_write_unlock_bh(idx); | |
880 | } | |
881 | ||
882 | /* the counter may be not NULL, because maybe some conn entries | |
883 | are run by slow timer handler or unhashed but still referred */ | |
884 | if (atomic_read(&ip_vs_conn_count) != 0) { | |
885 | schedule(); | |
886 | goto flush_again; | |
887 | } | |
888 | } | |
889 | ||
890 | ||
891 | int ip_vs_conn_init(void) | |
892 | { | |
893 | int idx; | |
894 | ||
895 | /* | |
896 | * Allocate the connection hash table and initialize its list heads | |
897 | */ | |
898 | ip_vs_conn_tab = vmalloc(IP_VS_CONN_TAB_SIZE*sizeof(struct list_head)); | |
899 | if (!ip_vs_conn_tab) | |
900 | return -ENOMEM; | |
901 | ||
902 | /* Allocate ip_vs_conn slab cache */ | |
903 | ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn", | |
904 | sizeof(struct ip_vs_conn), 0, | |
20c2df83 | 905 | SLAB_HWCACHE_ALIGN, NULL); |
1da177e4 LT |
906 | if (!ip_vs_conn_cachep) { |
907 | vfree(ip_vs_conn_tab); | |
908 | return -ENOMEM; | |
909 | } | |
910 | ||
911 | IP_VS_INFO("Connection hash table configured " | |
912 | "(size=%d, memory=%ldKbytes)\n", | |
913 | IP_VS_CONN_TAB_SIZE, | |
914 | (long)(IP_VS_CONN_TAB_SIZE*sizeof(struct list_head))/1024); | |
915 | IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n", | |
916 | sizeof(struct ip_vs_conn)); | |
917 | ||
918 | for (idx = 0; idx < IP_VS_CONN_TAB_SIZE; idx++) { | |
919 | INIT_LIST_HEAD(&ip_vs_conn_tab[idx]); | |
920 | } | |
921 | ||
922 | for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++) { | |
923 | rwlock_init(&__ip_vs_conntbl_lock_array[idx].l); | |
924 | } | |
925 | ||
457c4cbc | 926 | proc_net_fops_create(&init_net, "ip_vs_conn", 0, &ip_vs_conn_fops); |
1da177e4 LT |
927 | |
928 | /* calculate the random value for connection hash */ | |
929 | get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd)); | |
930 | ||
931 | return 0; | |
932 | } | |
933 | ||
934 | ||
935 | void ip_vs_conn_cleanup(void) | |
936 | { | |
937 | /* flush all the connection entries first */ | |
938 | ip_vs_conn_flush(); | |
939 | ||
940 | /* Release the empty cache */ | |
941 | kmem_cache_destroy(ip_vs_conn_cachep); | |
457c4cbc | 942 | proc_net_remove(&init_net, "ip_vs_conn"); |
1da177e4 LT |
943 | vfree(ip_vs_conn_tab); |
944 | } |