Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * ip_vs_proto_tcp.c: TCP load balancing support for IPVS | |
3 | * | |
4 | * Version: $Id: ip_vs_proto_tcp.c,v 1.3 2002/11/30 01:50:35 wensong Exp $ | |
5 | * | |
6 | * Authors: Wensong Zhang <wensong@linuxvirtualserver.org> | |
7 | * Julian Anastasov <ja@ssi.bg> | |
8 | * | |
9 | * This program is free software; you can redistribute it and/or | |
10 | * modify it under the terms of the GNU General Public License | |
11 | * as published by the Free Software Foundation; either version | |
12 | * 2 of the License, or (at your option) any later version. | |
13 | * | |
14 | * Changes: | |
15 | * | |
16 | */ | |
17 | ||
18 | #include <linux/kernel.h> | |
19 | #include <linux/ip.h> | |
20 | #include <linux/tcp.h> /* for tcphdr */ | |
21 | #include <net/ip.h> | |
22 | #include <net/tcp.h> /* for csum_tcpudp_magic */ | |
23 | #include <linux/netfilter_ipv4.h> | |
24 | ||
25 | #include <net/ip_vs.h> | |
26 | ||
27 | ||
28 | static struct ip_vs_conn * | |
29 | tcp_conn_in_get(const struct sk_buff *skb, struct ip_vs_protocol *pp, | |
30 | const struct iphdr *iph, unsigned int proto_off, int inverse) | |
31 | { | |
32 | __u16 _ports[2], *pptr; | |
33 | ||
34 | pptr = skb_header_pointer(skb, proto_off, sizeof(_ports), _ports); | |
35 | if (pptr == NULL) | |
36 | return NULL; | |
37 | ||
38 | if (likely(!inverse)) { | |
39 | return ip_vs_conn_in_get(iph->protocol, | |
40 | iph->saddr, pptr[0], | |
41 | iph->daddr, pptr[1]); | |
42 | } else { | |
43 | return ip_vs_conn_in_get(iph->protocol, | |
44 | iph->daddr, pptr[1], | |
45 | iph->saddr, pptr[0]); | |
46 | } | |
47 | } | |
48 | ||
49 | static struct ip_vs_conn * | |
50 | tcp_conn_out_get(const struct sk_buff *skb, struct ip_vs_protocol *pp, | |
51 | const struct iphdr *iph, unsigned int proto_off, int inverse) | |
52 | { | |
53 | __u16 _ports[2], *pptr; | |
54 | ||
55 | pptr = skb_header_pointer(skb, proto_off, sizeof(_ports), _ports); | |
56 | if (pptr == NULL) | |
57 | return NULL; | |
58 | ||
59 | if (likely(!inverse)) { | |
60 | return ip_vs_conn_out_get(iph->protocol, | |
61 | iph->saddr, pptr[0], | |
62 | iph->daddr, pptr[1]); | |
63 | } else { | |
64 | return ip_vs_conn_out_get(iph->protocol, | |
65 | iph->daddr, pptr[1], | |
66 | iph->saddr, pptr[0]); | |
67 | } | |
68 | } | |
69 | ||
70 | ||
71 | static int | |
72 | tcp_conn_schedule(struct sk_buff *skb, | |
73 | struct ip_vs_protocol *pp, | |
74 | int *verdict, struct ip_vs_conn **cpp) | |
75 | { | |
76 | struct ip_vs_service *svc; | |
77 | struct tcphdr _tcph, *th; | |
78 | ||
79 | th = skb_header_pointer(skb, skb->nh.iph->ihl*4, | |
80 | sizeof(_tcph), &_tcph); | |
81 | if (th == NULL) { | |
82 | *verdict = NF_DROP; | |
83 | return 0; | |
84 | } | |
85 | ||
86 | if (th->syn && | |
87 | (svc = ip_vs_service_get(skb->nfmark, skb->nh.iph->protocol, | |
88 | skb->nh.iph->daddr, th->dest))) { | |
89 | if (ip_vs_todrop()) { | |
90 | /* | |
91 | * It seems that we are very loaded. | |
92 | * We have to drop this packet :( | |
93 | */ | |
94 | ip_vs_service_put(svc); | |
95 | *verdict = NF_DROP; | |
96 | return 0; | |
97 | } | |
98 | ||
99 | /* | |
100 | * Let the virtual server select a real server for the | |
101 | * incoming connection, and create a connection entry. | |
102 | */ | |
103 | *cpp = ip_vs_schedule(svc, skb); | |
104 | if (!*cpp) { | |
105 | *verdict = ip_vs_leave(svc, skb, pp); | |
106 | return 0; | |
107 | } | |
108 | ip_vs_service_put(svc); | |
109 | } | |
110 | return 1; | |
111 | } | |
112 | ||
113 | ||
114 | static inline void | |
115 | tcp_fast_csum_update(struct tcphdr *tcph, u32 oldip, u32 newip, | |
116 | u16 oldport, u16 newport) | |
117 | { | |
118 | tcph->check = | |
119 | ip_vs_check_diff(~oldip, newip, | |
120 | ip_vs_check_diff(oldport ^ 0xFFFF, | |
121 | newport, tcph->check)); | |
122 | } | |
123 | ||
124 | ||
125 | static int | |
126 | tcp_snat_handler(struct sk_buff **pskb, | |
127 | struct ip_vs_protocol *pp, struct ip_vs_conn *cp) | |
128 | { | |
129 | struct tcphdr *tcph; | |
130 | unsigned int tcphoff = (*pskb)->nh.iph->ihl * 4; | |
131 | ||
132 | /* csum_check requires unshared skb */ | |
133 | if (!ip_vs_make_skb_writable(pskb, tcphoff+sizeof(*tcph))) | |
134 | return 0; | |
135 | ||
136 | if (unlikely(cp->app != NULL)) { | |
137 | /* Some checks before mangling */ | |
138 | if (pp->csum_check && !pp->csum_check(*pskb, pp)) | |
139 | return 0; | |
140 | ||
141 | /* Call application helper if needed */ | |
142 | if (!ip_vs_app_pkt_out(cp, pskb)) | |
143 | return 0; | |
144 | } | |
145 | ||
146 | tcph = (void *)(*pskb)->nh.iph + tcphoff; | |
147 | tcph->source = cp->vport; | |
148 | ||
149 | /* Adjust TCP checksums */ | |
150 | if (!cp->app) { | |
151 | /* Only port and addr are changed, do fast csum update */ | |
152 | tcp_fast_csum_update(tcph, cp->daddr, cp->vaddr, | |
153 | cp->dport, cp->vport); | |
154 | if ((*pskb)->ip_summed == CHECKSUM_HW) | |
155 | (*pskb)->ip_summed = CHECKSUM_NONE; | |
156 | } else { | |
157 | /* full checksum calculation */ | |
158 | tcph->check = 0; | |
159 | (*pskb)->csum = skb_checksum(*pskb, tcphoff, | |
160 | (*pskb)->len - tcphoff, 0); | |
161 | tcph->check = csum_tcpudp_magic(cp->vaddr, cp->caddr, | |
162 | (*pskb)->len - tcphoff, | |
163 | cp->protocol, | |
164 | (*pskb)->csum); | |
165 | IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%zd)\n", | |
166 | pp->name, tcph->check, | |
167 | (char*)&(tcph->check) - (char*)tcph); | |
168 | } | |
169 | return 1; | |
170 | } | |
171 | ||
172 | ||
173 | static int | |
174 | tcp_dnat_handler(struct sk_buff **pskb, | |
175 | struct ip_vs_protocol *pp, struct ip_vs_conn *cp) | |
176 | { | |
177 | struct tcphdr *tcph; | |
178 | unsigned int tcphoff = (*pskb)->nh.iph->ihl * 4; | |
179 | ||
180 | /* csum_check requires unshared skb */ | |
181 | if (!ip_vs_make_skb_writable(pskb, tcphoff+sizeof(*tcph))) | |
182 | return 0; | |
183 | ||
184 | if (unlikely(cp->app != NULL)) { | |
185 | /* Some checks before mangling */ | |
186 | if (pp->csum_check && !pp->csum_check(*pskb, pp)) | |
187 | return 0; | |
188 | ||
189 | /* | |
190 | * Attempt ip_vs_app call. | |
191 | * It will fix ip_vs_conn and iph ack_seq stuff | |
192 | */ | |
193 | if (!ip_vs_app_pkt_in(cp, pskb)) | |
194 | return 0; | |
195 | } | |
196 | ||
197 | tcph = (void *)(*pskb)->nh.iph + tcphoff; | |
198 | tcph->dest = cp->dport; | |
199 | ||
200 | /* | |
201 | * Adjust TCP checksums | |
202 | */ | |
203 | if (!cp->app) { | |
204 | /* Only port and addr are changed, do fast csum update */ | |
205 | tcp_fast_csum_update(tcph, cp->vaddr, cp->daddr, | |
206 | cp->vport, cp->dport); | |
207 | if ((*pskb)->ip_summed == CHECKSUM_HW) | |
208 | (*pskb)->ip_summed = CHECKSUM_NONE; | |
209 | } else { | |
210 | /* full checksum calculation */ | |
211 | tcph->check = 0; | |
212 | (*pskb)->csum = skb_checksum(*pskb, tcphoff, | |
213 | (*pskb)->len - tcphoff, 0); | |
214 | tcph->check = csum_tcpudp_magic(cp->caddr, cp->daddr, | |
215 | (*pskb)->len - tcphoff, | |
216 | cp->protocol, | |
217 | (*pskb)->csum); | |
218 | (*pskb)->ip_summed = CHECKSUM_UNNECESSARY; | |
219 | } | |
220 | return 1; | |
221 | } | |
222 | ||
223 | ||
224 | static int | |
225 | tcp_csum_check(struct sk_buff *skb, struct ip_vs_protocol *pp) | |
226 | { | |
227 | unsigned int tcphoff = skb->nh.iph->ihl*4; | |
228 | ||
229 | switch (skb->ip_summed) { | |
230 | case CHECKSUM_NONE: | |
231 | skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0); | |
232 | case CHECKSUM_HW: | |
233 | if (csum_tcpudp_magic(skb->nh.iph->saddr, skb->nh.iph->daddr, | |
234 | skb->len - tcphoff, | |
235 | skb->nh.iph->protocol, skb->csum)) { | |
236 | IP_VS_DBG_RL_PKT(0, pp, skb, 0, | |
237 | "Failed checksum for"); | |
238 | return 0; | |
239 | } | |
240 | break; | |
241 | default: | |
242 | /* CHECKSUM_UNNECESSARY */ | |
243 | break; | |
244 | } | |
245 | ||
246 | return 1; | |
247 | } | |
248 | ||
249 | ||
250 | #define TCP_DIR_INPUT 0 | |
251 | #define TCP_DIR_OUTPUT 4 | |
252 | #define TCP_DIR_INPUT_ONLY 8 | |
253 | ||
254 | static int tcp_state_off[IP_VS_DIR_LAST] = { | |
255 | [IP_VS_DIR_INPUT] = TCP_DIR_INPUT, | |
256 | [IP_VS_DIR_OUTPUT] = TCP_DIR_OUTPUT, | |
257 | [IP_VS_DIR_INPUT_ONLY] = TCP_DIR_INPUT_ONLY, | |
258 | }; | |
259 | ||
260 | /* | |
261 | * Timeout table[state] | |
262 | */ | |
263 | static int tcp_timeouts[IP_VS_TCP_S_LAST+1] = { | |
264 | [IP_VS_TCP_S_NONE] = 2*HZ, | |
265 | [IP_VS_TCP_S_ESTABLISHED] = 15*60*HZ, | |
266 | [IP_VS_TCP_S_SYN_SENT] = 2*60*HZ, | |
267 | [IP_VS_TCP_S_SYN_RECV] = 1*60*HZ, | |
268 | [IP_VS_TCP_S_FIN_WAIT] = 2*60*HZ, | |
269 | [IP_VS_TCP_S_TIME_WAIT] = 2*60*HZ, | |
270 | [IP_VS_TCP_S_CLOSE] = 10*HZ, | |
271 | [IP_VS_TCP_S_CLOSE_WAIT] = 60*HZ, | |
272 | [IP_VS_TCP_S_LAST_ACK] = 30*HZ, | |
273 | [IP_VS_TCP_S_LISTEN] = 2*60*HZ, | |
274 | [IP_VS_TCP_S_SYNACK] = 120*HZ, | |
275 | [IP_VS_TCP_S_LAST] = 2*HZ, | |
276 | }; | |
277 | ||
278 | ||
279 | #if 0 | |
280 | ||
281 | /* FIXME: This is going to die */ | |
282 | ||
283 | static int tcp_timeouts_dos[IP_VS_TCP_S_LAST+1] = { | |
284 | [IP_VS_TCP_S_NONE] = 2*HZ, | |
285 | [IP_VS_TCP_S_ESTABLISHED] = 8*60*HZ, | |
286 | [IP_VS_TCP_S_SYN_SENT] = 60*HZ, | |
287 | [IP_VS_TCP_S_SYN_RECV] = 10*HZ, | |
288 | [IP_VS_TCP_S_FIN_WAIT] = 60*HZ, | |
289 | [IP_VS_TCP_S_TIME_WAIT] = 60*HZ, | |
290 | [IP_VS_TCP_S_CLOSE] = 10*HZ, | |
291 | [IP_VS_TCP_S_CLOSE_WAIT] = 60*HZ, | |
292 | [IP_VS_TCP_S_LAST_ACK] = 30*HZ, | |
293 | [IP_VS_TCP_S_LISTEN] = 2*60*HZ, | |
294 | [IP_VS_TCP_S_SYNACK] = 100*HZ, | |
295 | [IP_VS_TCP_S_LAST] = 2*HZ, | |
296 | }; | |
297 | ||
298 | #endif | |
299 | ||
300 | static char * tcp_state_name_table[IP_VS_TCP_S_LAST+1] = { | |
301 | [IP_VS_TCP_S_NONE] = "NONE", | |
302 | [IP_VS_TCP_S_ESTABLISHED] = "ESTABLISHED", | |
303 | [IP_VS_TCP_S_SYN_SENT] = "SYN_SENT", | |
304 | [IP_VS_TCP_S_SYN_RECV] = "SYN_RECV", | |
305 | [IP_VS_TCP_S_FIN_WAIT] = "FIN_WAIT", | |
306 | [IP_VS_TCP_S_TIME_WAIT] = "TIME_WAIT", | |
307 | [IP_VS_TCP_S_CLOSE] = "CLOSE", | |
308 | [IP_VS_TCP_S_CLOSE_WAIT] = "CLOSE_WAIT", | |
309 | [IP_VS_TCP_S_LAST_ACK] = "LAST_ACK", | |
310 | [IP_VS_TCP_S_LISTEN] = "LISTEN", | |
311 | [IP_VS_TCP_S_SYNACK] = "SYNACK", | |
312 | [IP_VS_TCP_S_LAST] = "BUG!", | |
313 | }; | |
314 | ||
315 | #define sNO IP_VS_TCP_S_NONE | |
316 | #define sES IP_VS_TCP_S_ESTABLISHED | |
317 | #define sSS IP_VS_TCP_S_SYN_SENT | |
318 | #define sSR IP_VS_TCP_S_SYN_RECV | |
319 | #define sFW IP_VS_TCP_S_FIN_WAIT | |
320 | #define sTW IP_VS_TCP_S_TIME_WAIT | |
321 | #define sCL IP_VS_TCP_S_CLOSE | |
322 | #define sCW IP_VS_TCP_S_CLOSE_WAIT | |
323 | #define sLA IP_VS_TCP_S_LAST_ACK | |
324 | #define sLI IP_VS_TCP_S_LISTEN | |
325 | #define sSA IP_VS_TCP_S_SYNACK | |
326 | ||
327 | struct tcp_states_t { | |
328 | int next_state[IP_VS_TCP_S_LAST]; | |
329 | }; | |
330 | ||
331 | static const char * tcp_state_name(int state) | |
332 | { | |
333 | if (state >= IP_VS_TCP_S_LAST) | |
334 | return "ERR!"; | |
335 | return tcp_state_name_table[state] ? tcp_state_name_table[state] : "?"; | |
336 | } | |
337 | ||
338 | static struct tcp_states_t tcp_states [] = { | |
339 | /* INPUT */ | |
340 | /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */ | |
341 | /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }}, | |
342 | /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sTW }}, | |
343 | /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }}, | |
344 | /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sSR }}, | |
345 | ||
346 | /* OUTPUT */ | |
347 | /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */ | |
348 | /*syn*/ {{sSS, sES, sSS, sSR, sSS, sSS, sSS, sSS, sSS, sLI, sSR }}, | |
349 | /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }}, | |
350 | /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }}, | |
351 | /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }}, | |
352 | ||
353 | /* INPUT-ONLY */ | |
354 | /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */ | |
355 | /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }}, | |
356 | /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }}, | |
357 | /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }}, | |
358 | /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }}, | |
359 | }; | |
360 | ||
361 | static struct tcp_states_t tcp_states_dos [] = { | |
362 | /* INPUT */ | |
363 | /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */ | |
364 | /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSA }}, | |
365 | /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sSA }}, | |
366 | /*ack*/ {{sCL, sES, sSS, sSR, sFW, sTW, sCL, sCW, sCL, sLI, sSA }}, | |
367 | /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }}, | |
368 | ||
369 | /* OUTPUT */ | |
370 | /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */ | |
371 | /*syn*/ {{sSS, sES, sSS, sSA, sSS, sSS, sSS, sSS, sSS, sLI, sSA }}, | |
372 | /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }}, | |
373 | /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }}, | |
374 | /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }}, | |
375 | ||
376 | /* INPUT-ONLY */ | |
377 | /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */ | |
378 | /*syn*/ {{sSA, sES, sES, sSR, sSA, sSA, sSA, sSA, sSA, sSA, sSA }}, | |
379 | /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }}, | |
380 | /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }}, | |
381 | /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }}, | |
382 | }; | |
383 | ||
384 | static struct tcp_states_t *tcp_state_table = tcp_states; | |
385 | ||
386 | ||
387 | static void tcp_timeout_change(struct ip_vs_protocol *pp, int flags) | |
388 | { | |
389 | int on = (flags & 1); /* secure_tcp */ | |
390 | ||
391 | /* | |
392 | ** FIXME: change secure_tcp to independent sysctl var | |
393 | ** or make it per-service or per-app because it is valid | |
394 | ** for most if not for all of the applications. Something | |
395 | ** like "capabilities" (flags) for each object. | |
396 | */ | |
397 | tcp_state_table = (on? tcp_states_dos : tcp_states); | |
398 | } | |
399 | ||
400 | static int | |
401 | tcp_set_state_timeout(struct ip_vs_protocol *pp, char *sname, int to) | |
402 | { | |
403 | return ip_vs_set_state_timeout(pp->timeout_table, IP_VS_TCP_S_LAST, | |
404 | tcp_state_name_table, sname, to); | |
405 | } | |
406 | ||
407 | static inline int tcp_state_idx(struct tcphdr *th) | |
408 | { | |
409 | if (th->rst) | |
410 | return 3; | |
411 | if (th->syn) | |
412 | return 0; | |
413 | if (th->fin) | |
414 | return 1; | |
415 | if (th->ack) | |
416 | return 2; | |
417 | return -1; | |
418 | } | |
419 | ||
420 | static inline void | |
421 | set_tcp_state(struct ip_vs_protocol *pp, struct ip_vs_conn *cp, | |
422 | int direction, struct tcphdr *th) | |
423 | { | |
424 | int state_idx; | |
425 | int new_state = IP_VS_TCP_S_CLOSE; | |
426 | int state_off = tcp_state_off[direction]; | |
427 | ||
428 | /* | |
429 | * Update state offset to INPUT_ONLY if necessary | |
430 | * or delete NO_OUTPUT flag if output packet detected | |
431 | */ | |
432 | if (cp->flags & IP_VS_CONN_F_NOOUTPUT) { | |
433 | if (state_off == TCP_DIR_OUTPUT) | |
434 | cp->flags &= ~IP_VS_CONN_F_NOOUTPUT; | |
435 | else | |
436 | state_off = TCP_DIR_INPUT_ONLY; | |
437 | } | |
438 | ||
439 | if ((state_idx = tcp_state_idx(th)) < 0) { | |
440 | IP_VS_DBG(8, "tcp_state_idx=%d!!!\n", state_idx); | |
441 | goto tcp_state_out; | |
442 | } | |
443 | ||
444 | new_state = tcp_state_table[state_off+state_idx].next_state[cp->state]; | |
445 | ||
446 | tcp_state_out: | |
447 | if (new_state != cp->state) { | |
448 | struct ip_vs_dest *dest = cp->dest; | |
449 | ||
450 | IP_VS_DBG(8, "%s %s [%c%c%c%c] %u.%u.%u.%u:%d->" | |
451 | "%u.%u.%u.%u:%d state: %s->%s cnt:%d\n", | |
452 | pp->name, | |
453 | (state_off==TCP_DIR_OUTPUT)?"output ":"input ", | |
454 | th->syn? 'S' : '.', | |
455 | th->fin? 'F' : '.', | |
456 | th->ack? 'A' : '.', | |
457 | th->rst? 'R' : '.', | |
458 | NIPQUAD(cp->daddr), ntohs(cp->dport), | |
459 | NIPQUAD(cp->caddr), ntohs(cp->cport), | |
460 | tcp_state_name(cp->state), | |
461 | tcp_state_name(new_state), | |
462 | atomic_read(&cp->refcnt)); | |
463 | if (dest) { | |
464 | if (!(cp->flags & IP_VS_CONN_F_INACTIVE) && | |
465 | (new_state != IP_VS_TCP_S_ESTABLISHED)) { | |
466 | atomic_dec(&dest->activeconns); | |
467 | atomic_inc(&dest->inactconns); | |
468 | cp->flags |= IP_VS_CONN_F_INACTIVE; | |
469 | } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) && | |
470 | (new_state == IP_VS_TCP_S_ESTABLISHED)) { | |
471 | atomic_inc(&dest->activeconns); | |
472 | atomic_dec(&dest->inactconns); | |
473 | cp->flags &= ~IP_VS_CONN_F_INACTIVE; | |
474 | } | |
475 | } | |
476 | } | |
477 | ||
478 | cp->timeout = pp->timeout_table[cp->state = new_state]; | |
479 | } | |
480 | ||
481 | ||
482 | /* | |
483 | * Handle state transitions | |
484 | */ | |
485 | static int | |
486 | tcp_state_transition(struct ip_vs_conn *cp, int direction, | |
487 | const struct sk_buff *skb, | |
488 | struct ip_vs_protocol *pp) | |
489 | { | |
490 | struct tcphdr _tcph, *th; | |
491 | ||
492 | th = skb_header_pointer(skb, skb->nh.iph->ihl*4, | |
493 | sizeof(_tcph), &_tcph); | |
494 | if (th == NULL) | |
495 | return 0; | |
496 | ||
497 | spin_lock(&cp->lock); | |
498 | set_tcp_state(pp, cp, direction, th); | |
499 | spin_unlock(&cp->lock); | |
500 | ||
501 | return 1; | |
502 | } | |
503 | ||
504 | ||
505 | /* | |
506 | * Hash table for TCP application incarnations | |
507 | */ | |
508 | #define TCP_APP_TAB_BITS 4 | |
509 | #define TCP_APP_TAB_SIZE (1 << TCP_APP_TAB_BITS) | |
510 | #define TCP_APP_TAB_MASK (TCP_APP_TAB_SIZE - 1) | |
511 | ||
512 | static struct list_head tcp_apps[TCP_APP_TAB_SIZE]; | |
513 | static DEFINE_SPINLOCK(tcp_app_lock); | |
514 | ||
515 | static inline __u16 tcp_app_hashkey(__u16 port) | |
516 | { | |
517 | return ((port >> TCP_APP_TAB_BITS) ^ port) & TCP_APP_TAB_MASK; | |
518 | } | |
519 | ||
520 | ||
521 | static int tcp_register_app(struct ip_vs_app *inc) | |
522 | { | |
523 | struct ip_vs_app *i; | |
524 | __u16 hash, port = inc->port; | |
525 | int ret = 0; | |
526 | ||
527 | hash = tcp_app_hashkey(port); | |
528 | ||
529 | spin_lock_bh(&tcp_app_lock); | |
530 | list_for_each_entry(i, &tcp_apps[hash], p_list) { | |
531 | if (i->port == port) { | |
532 | ret = -EEXIST; | |
533 | goto out; | |
534 | } | |
535 | } | |
536 | list_add(&inc->p_list, &tcp_apps[hash]); | |
537 | atomic_inc(&ip_vs_protocol_tcp.appcnt); | |
538 | ||
539 | out: | |
540 | spin_unlock_bh(&tcp_app_lock); | |
541 | return ret; | |
542 | } | |
543 | ||
544 | ||
545 | static void | |
546 | tcp_unregister_app(struct ip_vs_app *inc) | |
547 | { | |
548 | spin_lock_bh(&tcp_app_lock); | |
549 | atomic_dec(&ip_vs_protocol_tcp.appcnt); | |
550 | list_del(&inc->p_list); | |
551 | spin_unlock_bh(&tcp_app_lock); | |
552 | } | |
553 | ||
554 | ||
555 | static int | |
556 | tcp_app_conn_bind(struct ip_vs_conn *cp) | |
557 | { | |
558 | int hash; | |
559 | struct ip_vs_app *inc; | |
560 | int result = 0; | |
561 | ||
562 | /* Default binding: bind app only for NAT */ | |
563 | if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) | |
564 | return 0; | |
565 | ||
566 | /* Lookup application incarnations and bind the right one */ | |
567 | hash = tcp_app_hashkey(cp->vport); | |
568 | ||
569 | spin_lock(&tcp_app_lock); | |
570 | list_for_each_entry(inc, &tcp_apps[hash], p_list) { | |
571 | if (inc->port == cp->vport) { | |
572 | if (unlikely(!ip_vs_app_inc_get(inc))) | |
573 | break; | |
574 | spin_unlock(&tcp_app_lock); | |
575 | ||
576 | IP_VS_DBG(9, "%s: Binding conn %u.%u.%u.%u:%u->" | |
577 | "%u.%u.%u.%u:%u to app %s on port %u\n", | |
578 | __FUNCTION__, | |
579 | NIPQUAD(cp->caddr), ntohs(cp->cport), | |
580 | NIPQUAD(cp->vaddr), ntohs(cp->vport), | |
581 | inc->name, ntohs(inc->port)); | |
582 | cp->app = inc; | |
583 | if (inc->init_conn) | |
584 | result = inc->init_conn(inc, cp); | |
585 | goto out; | |
586 | } | |
587 | } | |
588 | spin_unlock(&tcp_app_lock); | |
589 | ||
590 | out: | |
591 | return result; | |
592 | } | |
593 | ||
594 | ||
595 | /* | |
596 | * Set LISTEN timeout. (ip_vs_conn_put will setup timer) | |
597 | */ | |
598 | void ip_vs_tcp_conn_listen(struct ip_vs_conn *cp) | |
599 | { | |
600 | spin_lock(&cp->lock); | |
601 | cp->state = IP_VS_TCP_S_LISTEN; | |
602 | cp->timeout = ip_vs_protocol_tcp.timeout_table[IP_VS_TCP_S_LISTEN]; | |
603 | spin_unlock(&cp->lock); | |
604 | } | |
605 | ||
606 | ||
ba602a81 | 607 | static void ip_vs_tcp_init(struct ip_vs_protocol *pp) |
1da177e4 LT |
608 | { |
609 | IP_VS_INIT_HASH_TABLE(tcp_apps); | |
610 | pp->timeout_table = tcp_timeouts; | |
611 | } | |
612 | ||
613 | ||
ba602a81 | 614 | static void ip_vs_tcp_exit(struct ip_vs_protocol *pp) |
1da177e4 LT |
615 | { |
616 | } | |
617 | ||
618 | ||
619 | struct ip_vs_protocol ip_vs_protocol_tcp = { | |
620 | .name = "TCP", | |
621 | .protocol = IPPROTO_TCP, | |
622 | .dont_defrag = 0, | |
623 | .appcnt = ATOMIC_INIT(0), | |
ba602a81 DM |
624 | .init = ip_vs_tcp_init, |
625 | .exit = ip_vs_tcp_exit, | |
1da177e4 LT |
626 | .register_app = tcp_register_app, |
627 | .unregister_app = tcp_unregister_app, | |
628 | .conn_schedule = tcp_conn_schedule, | |
629 | .conn_in_get = tcp_conn_in_get, | |
630 | .conn_out_get = tcp_conn_out_get, | |
631 | .snat_handler = tcp_snat_handler, | |
632 | .dnat_handler = tcp_dnat_handler, | |
633 | .csum_check = tcp_csum_check, | |
634 | .state_name = tcp_state_name, | |
635 | .state_transition = tcp_state_transition, | |
636 | .app_conn_bind = tcp_app_conn_bind, | |
637 | .debug_packet = ip_vs_tcpudp_debug_packet, | |
638 | .timeout_change = tcp_timeout_change, | |
639 | .set_state_timeout = tcp_set_state_timeout, | |
640 | }; |