Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * ip_vs_proto_tcp.c: TCP load balancing support for IPVS | |
3 | * | |
1da177e4 LT |
4 | * Authors: Wensong Zhang <wensong@linuxvirtualserver.org> |
5 | * Julian Anastasov <ja@ssi.bg> | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or | |
8 | * modify it under the terms of the GNU General Public License | |
9 | * as published by the Free Software Foundation; either version | |
10 | * 2 of the License, or (at your option) any later version. | |
11 | * | |
4a85b96c | 12 | * Changes: Hans Schillstrom <hans.schillstrom@ericsson.com> |
1da177e4 | 13 | * |
4a85b96c HS |
14 | * Network name space (netns) aware. |
15 | * Global data moved to netns i.e struct netns_ipvs | |
16 | * tcp_timeouts table has copy per netns in a hash table per | |
17 | * protocol ip_vs_proto_data and is handled by netns | |
1da177e4 LT |
18 | */ |
19 | ||
9aada7ac HE |
20 | #define KMSG_COMPONENT "IPVS" |
21 | #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt | |
22 | ||
1da177e4 LT |
23 | #include <linux/kernel.h> |
24 | #include <linux/ip.h> | |
25 | #include <linux/tcp.h> /* for tcphdr */ | |
26 | #include <net/ip.h> | |
27 | #include <net/tcp.h> /* for csum_tcpudp_magic */ | |
63f2c046 | 28 | #include <net/ip6_checksum.h> |
af1e1cf0 | 29 | #include <linux/netfilter.h> |
1da177e4 LT |
30 | #include <linux/netfilter_ipv4.h> |
31 | ||
32 | #include <net/ip_vs.h> | |
33 | ||
1da177e4 | 34 | static int |
9330419d | 35 | tcp_conn_schedule(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd, |
1da177e4 LT |
36 | int *verdict, struct ip_vs_conn **cpp) |
37 | { | |
fc723250 | 38 | struct net *net; |
1da177e4 LT |
39 | struct ip_vs_service *svc; |
40 | struct tcphdr _tcph, *th; | |
3c2e0505 | 41 | struct ip_vs_iphdr iph; |
1da177e4 | 42 | |
51ef348b | 43 | ip_vs_fill_iphdr(af, skb_network_header(skb), &iph); |
3c2e0505 JV |
44 | |
45 | th = skb_header_pointer(skb, iph.len, sizeof(_tcph), &_tcph); | |
1da177e4 LT |
46 | if (th == NULL) { |
47 | *verdict = NF_DROP; | |
48 | return 0; | |
49 | } | |
fc723250 | 50 | net = skb_net(skb); |
190ecd27 | 51 | /* No !th->ack check to allow scheduling on SYN+ACK for Active FTP */ |
1da177e4 | 52 | if (th->syn && |
fc723250 HS |
53 | (svc = ip_vs_service_get(net, af, skb->mark, iph.protocol, |
54 | &iph.daddr, th->dest))) { | |
190ecd27 JA |
55 | int ignored; |
56 | ||
a0840e2e | 57 | if (ip_vs_todrop(net_ipvs(net))) { |
1da177e4 LT |
58 | /* |
59 | * It seems that we are very loaded. | |
60 | * We have to drop this packet :( | |
61 | */ | |
62 | ip_vs_service_put(svc); | |
63 | *verdict = NF_DROP; | |
64 | return 0; | |
65 | } | |
66 | ||
67 | /* | |
68 | * Let the virtual server select a real server for the | |
69 | * incoming connection, and create a connection entry. | |
70 | */ | |
9330419d | 71 | *cpp = ip_vs_schedule(svc, skb, pd, &ignored); |
a5959d53 HS |
72 | if (!*cpp && ignored <= 0) { |
73 | if (!ignored) | |
9330419d | 74 | *verdict = ip_vs_leave(svc, skb, pd); |
a5959d53 HS |
75 | else { |
76 | ip_vs_service_put(svc); | |
77 | *verdict = NF_DROP; | |
78 | } | |
1da177e4 LT |
79 | return 0; |
80 | } | |
81 | ip_vs_service_put(svc); | |
82 | } | |
a5959d53 | 83 | /* NF_ACCEPT */ |
1da177e4 LT |
84 | return 1; |
85 | } | |
86 | ||
87 | ||
88 | static inline void | |
0bbdd42b JV |
89 | tcp_fast_csum_update(int af, struct tcphdr *tcph, |
90 | const union nf_inet_addr *oldip, | |
91 | const union nf_inet_addr *newip, | |
014d730d | 92 | __be16 oldport, __be16 newport) |
1da177e4 | 93 | { |
0bbdd42b JV |
94 | #ifdef CONFIG_IP_VS_IPV6 |
95 | if (af == AF_INET6) | |
96 | tcph->check = | |
97 | csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6, | |
98 | ip_vs_check_diff2(oldport, newport, | |
99 | ~csum_unfold(tcph->check)))); | |
100 | else | |
101 | #endif | |
1da177e4 | 102 | tcph->check = |
0bbdd42b | 103 | csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip, |
f9214b26 AV |
104 | ip_vs_check_diff2(oldport, newport, |
105 | ~csum_unfold(tcph->check)))); | |
1da177e4 LT |
106 | } |
107 | ||
108 | ||
503e81f6 SH |
109 | static inline void |
110 | tcp_partial_csum_update(int af, struct tcphdr *tcph, | |
111 | const union nf_inet_addr *oldip, | |
112 | const union nf_inet_addr *newip, | |
113 | __be16 oldlen, __be16 newlen) | |
114 | { | |
115 | #ifdef CONFIG_IP_VS_IPV6 | |
116 | if (af == AF_INET6) | |
117 | tcph->check = | |
5bc9068e | 118 | ~csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6, |
503e81f6 | 119 | ip_vs_check_diff2(oldlen, newlen, |
5bc9068e | 120 | csum_unfold(tcph->check)))); |
503e81f6 SH |
121 | else |
122 | #endif | |
123 | tcph->check = | |
5bc9068e | 124 | ~csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip, |
503e81f6 | 125 | ip_vs_check_diff2(oldlen, newlen, |
5bc9068e | 126 | csum_unfold(tcph->check)))); |
503e81f6 SH |
127 | } |
128 | ||
129 | ||
1da177e4 | 130 | static int |
3db05fea | 131 | tcp_snat_handler(struct sk_buff *skb, |
1da177e4 LT |
132 | struct ip_vs_protocol *pp, struct ip_vs_conn *cp) |
133 | { | |
134 | struct tcphdr *tcph; | |
0bbdd42b | 135 | unsigned int tcphoff; |
503e81f6 | 136 | int oldlen; |
8b27b10f | 137 | int payload_csum = 0; |
0bbdd42b JV |
138 | |
139 | #ifdef CONFIG_IP_VS_IPV6 | |
140 | if (cp->af == AF_INET6) | |
141 | tcphoff = sizeof(struct ipv6hdr); | |
142 | else | |
143 | #endif | |
144 | tcphoff = ip_hdrlen(skb); | |
503e81f6 | 145 | oldlen = skb->len - tcphoff; |
1da177e4 LT |
146 | |
147 | /* csum_check requires unshared skb */ | |
3db05fea | 148 | if (!skb_make_writable(skb, tcphoff+sizeof(*tcph))) |
1da177e4 LT |
149 | return 0; |
150 | ||
151 | if (unlikely(cp->app != NULL)) { | |
8b27b10f JA |
152 | int ret; |
153 | ||
1da177e4 | 154 | /* Some checks before mangling */ |
0bbdd42b | 155 | if (pp->csum_check && !pp->csum_check(cp->af, skb, pp)) |
1da177e4 LT |
156 | return 0; |
157 | ||
158 | /* Call application helper if needed */ | |
8b27b10f | 159 | if (!(ret = ip_vs_app_pkt_out(cp, skb))) |
1da177e4 | 160 | return 0; |
8b27b10f JA |
161 | /* ret=2: csum update is needed after payload mangling */ |
162 | if (ret == 1) | |
163 | oldlen = skb->len - tcphoff; | |
164 | else | |
165 | payload_csum = 1; | |
1da177e4 LT |
166 | } |
167 | ||
0bbdd42b | 168 | tcph = (void *)skb_network_header(skb) + tcphoff; |
1da177e4 LT |
169 | tcph->source = cp->vport; |
170 | ||
171 | /* Adjust TCP checksums */ | |
503e81f6 SH |
172 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
173 | tcp_partial_csum_update(cp->af, tcph, &cp->daddr, &cp->vaddr, | |
ca62059b HH |
174 | htons(oldlen), |
175 | htons(skb->len - tcphoff)); | |
8b27b10f | 176 | } else if (!payload_csum) { |
1da177e4 | 177 | /* Only port and addr are changed, do fast csum update */ |
0bbdd42b | 178 | tcp_fast_csum_update(cp->af, tcph, &cp->daddr, &cp->vaddr, |
1da177e4 | 179 | cp->dport, cp->vport); |
3db05fea | 180 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
8b27b10f JA |
181 | skb->ip_summed = (cp->app && pp->csum_check) ? |
182 | CHECKSUM_UNNECESSARY : CHECKSUM_NONE; | |
1da177e4 LT |
183 | } else { |
184 | /* full checksum calculation */ | |
185 | tcph->check = 0; | |
3db05fea | 186 | skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0); |
0bbdd42b JV |
187 | #ifdef CONFIG_IP_VS_IPV6 |
188 | if (cp->af == AF_INET6) | |
189 | tcph->check = csum_ipv6_magic(&cp->vaddr.in6, | |
190 | &cp->caddr.in6, | |
191 | skb->len - tcphoff, | |
192 | cp->protocol, skb->csum); | |
193 | else | |
194 | #endif | |
195 | tcph->check = csum_tcpudp_magic(cp->vaddr.ip, | |
196 | cp->caddr.ip, | |
197 | skb->len - tcphoff, | |
198 | cp->protocol, | |
199 | skb->csum); | |
8b27b10f | 200 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
0bbdd42b | 201 | |
1da177e4 LT |
202 | IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%zd)\n", |
203 | pp->name, tcph->check, | |
204 | (char*)&(tcph->check) - (char*)tcph); | |
205 | } | |
206 | return 1; | |
207 | } | |
208 | ||
209 | ||
210 | static int | |
3db05fea | 211 | tcp_dnat_handler(struct sk_buff *skb, |
1da177e4 LT |
212 | struct ip_vs_protocol *pp, struct ip_vs_conn *cp) |
213 | { | |
214 | struct tcphdr *tcph; | |
0bbdd42b | 215 | unsigned int tcphoff; |
503e81f6 | 216 | int oldlen; |
8b27b10f | 217 | int payload_csum = 0; |
0bbdd42b JV |
218 | |
219 | #ifdef CONFIG_IP_VS_IPV6 | |
220 | if (cp->af == AF_INET6) | |
221 | tcphoff = sizeof(struct ipv6hdr); | |
222 | else | |
223 | #endif | |
224 | tcphoff = ip_hdrlen(skb); | |
503e81f6 | 225 | oldlen = skb->len - tcphoff; |
1da177e4 LT |
226 | |
227 | /* csum_check requires unshared skb */ | |
3db05fea | 228 | if (!skb_make_writable(skb, tcphoff+sizeof(*tcph))) |
1da177e4 LT |
229 | return 0; |
230 | ||
231 | if (unlikely(cp->app != NULL)) { | |
8b27b10f JA |
232 | int ret; |
233 | ||
1da177e4 | 234 | /* Some checks before mangling */ |
0bbdd42b | 235 | if (pp->csum_check && !pp->csum_check(cp->af, skb, pp)) |
1da177e4 LT |
236 | return 0; |
237 | ||
238 | /* | |
239 | * Attempt ip_vs_app call. | |
240 | * It will fix ip_vs_conn and iph ack_seq stuff | |
241 | */ | |
8b27b10f | 242 | if (!(ret = ip_vs_app_pkt_in(cp, skb))) |
1da177e4 | 243 | return 0; |
8b27b10f JA |
244 | /* ret=2: csum update is needed after payload mangling */ |
245 | if (ret == 1) | |
246 | oldlen = skb->len - tcphoff; | |
247 | else | |
248 | payload_csum = 1; | |
1da177e4 LT |
249 | } |
250 | ||
0bbdd42b | 251 | tcph = (void *)skb_network_header(skb) + tcphoff; |
1da177e4 LT |
252 | tcph->dest = cp->dport; |
253 | ||
254 | /* | |
255 | * Adjust TCP checksums | |
256 | */ | |
503e81f6 | 257 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
5bc9068e | 258 | tcp_partial_csum_update(cp->af, tcph, &cp->vaddr, &cp->daddr, |
ca62059b HH |
259 | htons(oldlen), |
260 | htons(skb->len - tcphoff)); | |
8b27b10f | 261 | } else if (!payload_csum) { |
1da177e4 | 262 | /* Only port and addr are changed, do fast csum update */ |
0bbdd42b | 263 | tcp_fast_csum_update(cp->af, tcph, &cp->vaddr, &cp->daddr, |
1da177e4 | 264 | cp->vport, cp->dport); |
3db05fea | 265 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
8b27b10f JA |
266 | skb->ip_summed = (cp->app && pp->csum_check) ? |
267 | CHECKSUM_UNNECESSARY : CHECKSUM_NONE; | |
1da177e4 LT |
268 | } else { |
269 | /* full checksum calculation */ | |
270 | tcph->check = 0; | |
3db05fea | 271 | skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0); |
0bbdd42b JV |
272 | #ifdef CONFIG_IP_VS_IPV6 |
273 | if (cp->af == AF_INET6) | |
274 | tcph->check = csum_ipv6_magic(&cp->caddr.in6, | |
275 | &cp->daddr.in6, | |
276 | skb->len - tcphoff, | |
277 | cp->protocol, skb->csum); | |
278 | else | |
279 | #endif | |
280 | tcph->check = csum_tcpudp_magic(cp->caddr.ip, | |
281 | cp->daddr.ip, | |
282 | skb->len - tcphoff, | |
283 | cp->protocol, | |
284 | skb->csum); | |
3db05fea | 285 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
1da177e4 LT |
286 | } |
287 | return 1; | |
288 | } | |
289 | ||
290 | ||
291 | static int | |
51ef348b | 292 | tcp_csum_check(int af, struct sk_buff *skb, struct ip_vs_protocol *pp) |
1da177e4 | 293 | { |
51ef348b JV |
294 | unsigned int tcphoff; |
295 | ||
296 | #ifdef CONFIG_IP_VS_IPV6 | |
297 | if (af == AF_INET6) | |
298 | tcphoff = sizeof(struct ipv6hdr); | |
299 | else | |
300 | #endif | |
301 | tcphoff = ip_hdrlen(skb); | |
1da177e4 LT |
302 | |
303 | switch (skb->ip_summed) { | |
304 | case CHECKSUM_NONE: | |
305 | skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0); | |
84fa7933 | 306 | case CHECKSUM_COMPLETE: |
51ef348b JV |
307 | #ifdef CONFIG_IP_VS_IPV6 |
308 | if (af == AF_INET6) { | |
309 | if (csum_ipv6_magic(&ipv6_hdr(skb)->saddr, | |
310 | &ipv6_hdr(skb)->daddr, | |
311 | skb->len - tcphoff, | |
312 | ipv6_hdr(skb)->nexthdr, | |
313 | skb->csum)) { | |
0d79641a | 314 | IP_VS_DBG_RL_PKT(0, af, pp, skb, 0, |
51ef348b JV |
315 | "Failed checksum for"); |
316 | return 0; | |
317 | } | |
318 | } else | |
319 | #endif | |
320 | if (csum_tcpudp_magic(ip_hdr(skb)->saddr, | |
321 | ip_hdr(skb)->daddr, | |
322 | skb->len - tcphoff, | |
323 | ip_hdr(skb)->protocol, | |
324 | skb->csum)) { | |
0d79641a | 325 | IP_VS_DBG_RL_PKT(0, af, pp, skb, 0, |
51ef348b JV |
326 | "Failed checksum for"); |
327 | return 0; | |
328 | } | |
1da177e4 LT |
329 | break; |
330 | default: | |
84fa7933 | 331 | /* No need to checksum. */ |
1da177e4 LT |
332 | break; |
333 | } | |
334 | ||
335 | return 1; | |
336 | } | |
337 | ||
338 | ||
339 | #define TCP_DIR_INPUT 0 | |
340 | #define TCP_DIR_OUTPUT 4 | |
341 | #define TCP_DIR_INPUT_ONLY 8 | |
342 | ||
9b5b5cff | 343 | static const int tcp_state_off[IP_VS_DIR_LAST] = { |
1da177e4 LT |
344 | [IP_VS_DIR_INPUT] = TCP_DIR_INPUT, |
345 | [IP_VS_DIR_OUTPUT] = TCP_DIR_OUTPUT, | |
346 | [IP_VS_DIR_INPUT_ONLY] = TCP_DIR_INPUT_ONLY, | |
347 | }; | |
348 | ||
349 | /* | |
350 | * Timeout table[state] | |
351 | */ | |
4a85b96c | 352 | static const int tcp_timeouts[IP_VS_TCP_S_LAST+1] = { |
1da177e4 LT |
353 | [IP_VS_TCP_S_NONE] = 2*HZ, |
354 | [IP_VS_TCP_S_ESTABLISHED] = 15*60*HZ, | |
355 | [IP_VS_TCP_S_SYN_SENT] = 2*60*HZ, | |
356 | [IP_VS_TCP_S_SYN_RECV] = 1*60*HZ, | |
357 | [IP_VS_TCP_S_FIN_WAIT] = 2*60*HZ, | |
358 | [IP_VS_TCP_S_TIME_WAIT] = 2*60*HZ, | |
359 | [IP_VS_TCP_S_CLOSE] = 10*HZ, | |
360 | [IP_VS_TCP_S_CLOSE_WAIT] = 60*HZ, | |
361 | [IP_VS_TCP_S_LAST_ACK] = 30*HZ, | |
362 | [IP_VS_TCP_S_LISTEN] = 2*60*HZ, | |
363 | [IP_VS_TCP_S_SYNACK] = 120*HZ, | |
364 | [IP_VS_TCP_S_LAST] = 2*HZ, | |
365 | }; | |
366 | ||
36cbd3dc | 367 | static const char *const tcp_state_name_table[IP_VS_TCP_S_LAST+1] = { |
1da177e4 LT |
368 | [IP_VS_TCP_S_NONE] = "NONE", |
369 | [IP_VS_TCP_S_ESTABLISHED] = "ESTABLISHED", | |
370 | [IP_VS_TCP_S_SYN_SENT] = "SYN_SENT", | |
371 | [IP_VS_TCP_S_SYN_RECV] = "SYN_RECV", | |
372 | [IP_VS_TCP_S_FIN_WAIT] = "FIN_WAIT", | |
373 | [IP_VS_TCP_S_TIME_WAIT] = "TIME_WAIT", | |
374 | [IP_VS_TCP_S_CLOSE] = "CLOSE", | |
375 | [IP_VS_TCP_S_CLOSE_WAIT] = "CLOSE_WAIT", | |
376 | [IP_VS_TCP_S_LAST_ACK] = "LAST_ACK", | |
377 | [IP_VS_TCP_S_LISTEN] = "LISTEN", | |
378 | [IP_VS_TCP_S_SYNACK] = "SYNACK", | |
379 | [IP_VS_TCP_S_LAST] = "BUG!", | |
380 | }; | |
381 | ||
382 | #define sNO IP_VS_TCP_S_NONE | |
383 | #define sES IP_VS_TCP_S_ESTABLISHED | |
384 | #define sSS IP_VS_TCP_S_SYN_SENT | |
385 | #define sSR IP_VS_TCP_S_SYN_RECV | |
386 | #define sFW IP_VS_TCP_S_FIN_WAIT | |
387 | #define sTW IP_VS_TCP_S_TIME_WAIT | |
388 | #define sCL IP_VS_TCP_S_CLOSE | |
389 | #define sCW IP_VS_TCP_S_CLOSE_WAIT | |
390 | #define sLA IP_VS_TCP_S_LAST_ACK | |
391 | #define sLI IP_VS_TCP_S_LISTEN | |
392 | #define sSA IP_VS_TCP_S_SYNACK | |
393 | ||
394 | struct tcp_states_t { | |
395 | int next_state[IP_VS_TCP_S_LAST]; | |
396 | }; | |
397 | ||
398 | static const char * tcp_state_name(int state) | |
399 | { | |
400 | if (state >= IP_VS_TCP_S_LAST) | |
401 | return "ERR!"; | |
402 | return tcp_state_name_table[state] ? tcp_state_name_table[state] : "?"; | |
403 | } | |
404 | ||
405 | static struct tcp_states_t tcp_states [] = { | |
406 | /* INPUT */ | |
407 | /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */ | |
408 | /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }}, | |
409 | /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sTW }}, | |
410 | /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }}, | |
411 | /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sSR }}, | |
412 | ||
413 | /* OUTPUT */ | |
414 | /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */ | |
415 | /*syn*/ {{sSS, sES, sSS, sSR, sSS, sSS, sSS, sSS, sSS, sLI, sSR }}, | |
416 | /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }}, | |
417 | /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }}, | |
418 | /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }}, | |
419 | ||
420 | /* INPUT-ONLY */ | |
421 | /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */ | |
422 | /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }}, | |
423 | /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }}, | |
424 | /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }}, | |
425 | /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }}, | |
426 | }; | |
427 | ||
428 | static struct tcp_states_t tcp_states_dos [] = { | |
429 | /* INPUT */ | |
430 | /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */ | |
431 | /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSA }}, | |
432 | /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sSA }}, | |
433 | /*ack*/ {{sCL, sES, sSS, sSR, sFW, sTW, sCL, sCW, sCL, sLI, sSA }}, | |
434 | /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }}, | |
435 | ||
436 | /* OUTPUT */ | |
437 | /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */ | |
438 | /*syn*/ {{sSS, sES, sSS, sSA, sSS, sSS, sSS, sSS, sSS, sLI, sSA }}, | |
439 | /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }}, | |
440 | /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }}, | |
441 | /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }}, | |
442 | ||
443 | /* INPUT-ONLY */ | |
444 | /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */ | |
445 | /*syn*/ {{sSA, sES, sES, sSR, sSA, sSA, sSA, sSA, sSA, sSA, sSA }}, | |
446 | /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }}, | |
447 | /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }}, | |
448 | /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }}, | |
449 | }; | |
450 | ||
9330419d | 451 | static void tcp_timeout_change(struct ip_vs_proto_data *pd, int flags) |
1da177e4 LT |
452 | { |
453 | int on = (flags & 1); /* secure_tcp */ | |
454 | ||
455 | /* | |
456 | ** FIXME: change secure_tcp to independent sysctl var | |
457 | ** or make it per-service or per-app because it is valid | |
458 | ** for most if not for all of the applications. Something | |
459 | ** like "capabilities" (flags) for each object. | |
460 | */ | |
9330419d | 461 | pd->tcp_state_table = (on ? tcp_states_dos : tcp_states); |
1da177e4 LT |
462 | } |
463 | ||
1da177e4 LT |
464 | static inline int tcp_state_idx(struct tcphdr *th) |
465 | { | |
466 | if (th->rst) | |
467 | return 3; | |
468 | if (th->syn) | |
469 | return 0; | |
470 | if (th->fin) | |
471 | return 1; | |
472 | if (th->ack) | |
473 | return 2; | |
474 | return -1; | |
475 | } | |
476 | ||
477 | static inline void | |
9330419d | 478 | set_tcp_state(struct ip_vs_proto_data *pd, struct ip_vs_conn *cp, |
1da177e4 LT |
479 | int direction, struct tcphdr *th) |
480 | { | |
481 | int state_idx; | |
482 | int new_state = IP_VS_TCP_S_CLOSE; | |
483 | int state_off = tcp_state_off[direction]; | |
484 | ||
485 | /* | |
486 | * Update state offset to INPUT_ONLY if necessary | |
487 | * or delete NO_OUTPUT flag if output packet detected | |
488 | */ | |
489 | if (cp->flags & IP_VS_CONN_F_NOOUTPUT) { | |
490 | if (state_off == TCP_DIR_OUTPUT) | |
491 | cp->flags &= ~IP_VS_CONN_F_NOOUTPUT; | |
492 | else | |
493 | state_off = TCP_DIR_INPUT_ONLY; | |
494 | } | |
495 | ||
496 | if ((state_idx = tcp_state_idx(th)) < 0) { | |
497 | IP_VS_DBG(8, "tcp_state_idx=%d!!!\n", state_idx); | |
498 | goto tcp_state_out; | |
499 | } | |
500 | ||
9330419d HS |
501 | new_state = |
502 | pd->tcp_state_table[state_off+state_idx].next_state[cp->state]; | |
1da177e4 LT |
503 | |
504 | tcp_state_out: | |
505 | if (new_state != cp->state) { | |
506 | struct ip_vs_dest *dest = cp->dest; | |
507 | ||
cfc78c5a JV |
508 | IP_VS_DBG_BUF(8, "%s %s [%c%c%c%c] %s:%d->" |
509 | "%s:%d state: %s->%s conn->refcnt:%d\n", | |
9330419d | 510 | pd->pp->name, |
cfc78c5a JV |
511 | ((state_off == TCP_DIR_OUTPUT) ? |
512 | "output " : "input "), | |
513 | th->syn ? 'S' : '.', | |
514 | th->fin ? 'F' : '.', | |
515 | th->ack ? 'A' : '.', | |
516 | th->rst ? 'R' : '.', | |
517 | IP_VS_DBG_ADDR(cp->af, &cp->daddr), | |
518 | ntohs(cp->dport), | |
519 | IP_VS_DBG_ADDR(cp->af, &cp->caddr), | |
520 | ntohs(cp->cport), | |
521 | tcp_state_name(cp->state), | |
522 | tcp_state_name(new_state), | |
523 | atomic_read(&cp->refcnt)); | |
524 | ||
1da177e4 LT |
525 | if (dest) { |
526 | if (!(cp->flags & IP_VS_CONN_F_INACTIVE) && | |
527 | (new_state != IP_VS_TCP_S_ESTABLISHED)) { | |
528 | atomic_dec(&dest->activeconns); | |
529 | atomic_inc(&dest->inactconns); | |
530 | cp->flags |= IP_VS_CONN_F_INACTIVE; | |
531 | } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) && | |
532 | (new_state == IP_VS_TCP_S_ESTABLISHED)) { | |
533 | atomic_inc(&dest->activeconns); | |
534 | atomic_dec(&dest->inactconns); | |
535 | cp->flags &= ~IP_VS_CONN_F_INACTIVE; | |
536 | } | |
537 | } | |
538 | } | |
539 | ||
4a85b96c HS |
540 | if (likely(pd)) |
541 | cp->timeout = pd->timeout_table[cp->state = new_state]; | |
542 | else /* What to do ? */ | |
543 | cp->timeout = tcp_timeouts[cp->state = new_state]; | |
1da177e4 LT |
544 | } |
545 | ||
1da177e4 LT |
546 | /* |
547 | * Handle state transitions | |
548 | */ | |
549 | static int | |
550 | tcp_state_transition(struct ip_vs_conn *cp, int direction, | |
551 | const struct sk_buff *skb, | |
9330419d | 552 | struct ip_vs_proto_data *pd) |
1da177e4 LT |
553 | { |
554 | struct tcphdr _tcph, *th; | |
555 | ||
0bbdd42b JV |
556 | #ifdef CONFIG_IP_VS_IPV6 |
557 | int ihl = cp->af == AF_INET ? ip_hdrlen(skb) : sizeof(struct ipv6hdr); | |
558 | #else | |
559 | int ihl = ip_hdrlen(skb); | |
560 | #endif | |
561 | ||
562 | th = skb_header_pointer(skb, ihl, sizeof(_tcph), &_tcph); | |
1da177e4 LT |
563 | if (th == NULL) |
564 | return 0; | |
565 | ||
566 | spin_lock(&cp->lock); | |
9330419d | 567 | set_tcp_state(pd, cp, direction, th); |
1da177e4 LT |
568 | spin_unlock(&cp->lock); |
569 | ||
570 | return 1; | |
571 | } | |
572 | ||
75e7ce66 | 573 | static inline __u16 tcp_app_hashkey(__be16 port) |
1da177e4 | 574 | { |
75e7ce66 AV |
575 | return (((__force u16)port >> TCP_APP_TAB_BITS) ^ (__force u16)port) |
576 | & TCP_APP_TAB_MASK; | |
1da177e4 LT |
577 | } |
578 | ||
579 | ||
ab8a5e84 | 580 | static int tcp_register_app(struct net *net, struct ip_vs_app *inc) |
1da177e4 LT |
581 | { |
582 | struct ip_vs_app *i; | |
75e7ce66 AV |
583 | __u16 hash; |
584 | __be16 port = inc->port; | |
1da177e4 | 585 | int ret = 0; |
ab8a5e84 HS |
586 | struct netns_ipvs *ipvs = net_ipvs(net); |
587 | struct ip_vs_proto_data *pd = ip_vs_proto_data_get(net, IPPROTO_TCP); | |
1da177e4 LT |
588 | |
589 | hash = tcp_app_hashkey(port); | |
590 | ||
4a85b96c HS |
591 | spin_lock_bh(&ipvs->tcp_app_lock); |
592 | list_for_each_entry(i, &ipvs->tcp_apps[hash], p_list) { | |
1da177e4 LT |
593 | if (i->port == port) { |
594 | ret = -EEXIST; | |
595 | goto out; | |
596 | } | |
597 | } | |
4a85b96c | 598 | list_add(&inc->p_list, &ipvs->tcp_apps[hash]); |
9bbac6a9 | 599 | atomic_inc(&pd->appcnt); |
1da177e4 LT |
600 | |
601 | out: | |
4a85b96c | 602 | spin_unlock_bh(&ipvs->tcp_app_lock); |
1da177e4 LT |
603 | return ret; |
604 | } | |
605 | ||
606 | ||
607 | static void | |
ab8a5e84 | 608 | tcp_unregister_app(struct net *net, struct ip_vs_app *inc) |
1da177e4 | 609 | { |
ab8a5e84 HS |
610 | struct netns_ipvs *ipvs = net_ipvs(net); |
611 | struct ip_vs_proto_data *pd = ip_vs_proto_data_get(net, IPPROTO_TCP); | |
4a85b96c HS |
612 | |
613 | spin_lock_bh(&ipvs->tcp_app_lock); | |
9bbac6a9 | 614 | atomic_dec(&pd->appcnt); |
1da177e4 | 615 | list_del(&inc->p_list); |
4a85b96c | 616 | spin_unlock_bh(&ipvs->tcp_app_lock); |
1da177e4 LT |
617 | } |
618 | ||
619 | ||
620 | static int | |
621 | tcp_app_conn_bind(struct ip_vs_conn *cp) | |
622 | { | |
6e67e586 | 623 | struct netns_ipvs *ipvs = net_ipvs(ip_vs_conn_net(cp)); |
1da177e4 LT |
624 | int hash; |
625 | struct ip_vs_app *inc; | |
626 | int result = 0; | |
627 | ||
628 | /* Default binding: bind app only for NAT */ | |
629 | if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) | |
630 | return 0; | |
631 | ||
632 | /* Lookup application incarnations and bind the right one */ | |
633 | hash = tcp_app_hashkey(cp->vport); | |
634 | ||
4a85b96c HS |
635 | spin_lock(&ipvs->tcp_app_lock); |
636 | list_for_each_entry(inc, &ipvs->tcp_apps[hash], p_list) { | |
1da177e4 LT |
637 | if (inc->port == cp->vport) { |
638 | if (unlikely(!ip_vs_app_inc_get(inc))) | |
639 | break; | |
4a85b96c | 640 | spin_unlock(&ipvs->tcp_app_lock); |
1da177e4 | 641 | |
1e3e238e | 642 | IP_VS_DBG_BUF(9, "%s(): Binding conn %s:%u->" |
cfc78c5a JV |
643 | "%s:%u to app %s on port %u\n", |
644 | __func__, | |
645 | IP_VS_DBG_ADDR(cp->af, &cp->caddr), | |
646 | ntohs(cp->cport), | |
647 | IP_VS_DBG_ADDR(cp->af, &cp->vaddr), | |
648 | ntohs(cp->vport), | |
649 | inc->name, ntohs(inc->port)); | |
650 | ||
1da177e4 LT |
651 | cp->app = inc; |
652 | if (inc->init_conn) | |
653 | result = inc->init_conn(inc, cp); | |
654 | goto out; | |
655 | } | |
656 | } | |
4a85b96c | 657 | spin_unlock(&ipvs->tcp_app_lock); |
1da177e4 LT |
658 | |
659 | out: | |
660 | return result; | |
661 | } | |
662 | ||
663 | ||
664 | /* | |
665 | * Set LISTEN timeout. (ip_vs_conn_put will setup timer) | |
666 | */ | |
4a85b96c | 667 | void ip_vs_tcp_conn_listen(struct net *net, struct ip_vs_conn *cp) |
1da177e4 | 668 | { |
4a85b96c HS |
669 | struct ip_vs_proto_data *pd = ip_vs_proto_data_get(net, IPPROTO_TCP); |
670 | ||
1da177e4 LT |
671 | spin_lock(&cp->lock); |
672 | cp->state = IP_VS_TCP_S_LISTEN; | |
4a85b96c HS |
673 | cp->timeout = (pd ? pd->timeout_table[IP_VS_TCP_S_LISTEN] |
674 | : tcp_timeouts[IP_VS_TCP_S_LISTEN]); | |
1da177e4 LT |
675 | spin_unlock(&cp->lock); |
676 | } | |
677 | ||
4a85b96c HS |
678 | /* --------------------------------------------- |
679 | * timeouts is netns related now. | |
680 | * --------------------------------------------- | |
681 | */ | |
682 | static void __ip_vs_tcp_init(struct net *net, struct ip_vs_proto_data *pd) | |
1da177e4 | 683 | { |
4a85b96c | 684 | struct netns_ipvs *ipvs = net_ipvs(net); |
1da177e4 | 685 | |
4a85b96c HS |
686 | ip_vs_init_hash_table(ipvs->tcp_apps, TCP_APP_TAB_SIZE); |
687 | spin_lock_init(&ipvs->tcp_app_lock); | |
688 | pd->timeout_table = ip_vs_create_timeout_table((int *)tcp_timeouts, | |
689 | sizeof(tcp_timeouts)); | |
9330419d | 690 | pd->tcp_state_table = tcp_states; |
4a85b96c | 691 | } |
1da177e4 | 692 | |
4a85b96c | 693 | static void __ip_vs_tcp_exit(struct net *net, struct ip_vs_proto_data *pd) |
1da177e4 | 694 | { |
4a85b96c | 695 | kfree(pd->timeout_table); |
1da177e4 LT |
696 | } |
697 | ||
698 | ||
699 | struct ip_vs_protocol ip_vs_protocol_tcp = { | |
700 | .name = "TCP", | |
701 | .protocol = IPPROTO_TCP, | |
2ad17def | 702 | .num_states = IP_VS_TCP_S_LAST, |
1da177e4 | 703 | .dont_defrag = 0, |
4a85b96c HS |
704 | .init = NULL, |
705 | .exit = NULL, | |
706 | .init_netns = __ip_vs_tcp_init, | |
707 | .exit_netns = __ip_vs_tcp_exit, | |
1da177e4 LT |
708 | .register_app = tcp_register_app, |
709 | .unregister_app = tcp_unregister_app, | |
710 | .conn_schedule = tcp_conn_schedule, | |
5c0d2374 SH |
711 | .conn_in_get = ip_vs_conn_in_get_proto, |
712 | .conn_out_get = ip_vs_conn_out_get_proto, | |
1da177e4 LT |
713 | .snat_handler = tcp_snat_handler, |
714 | .dnat_handler = tcp_dnat_handler, | |
715 | .csum_check = tcp_csum_check, | |
716 | .state_name = tcp_state_name, | |
717 | .state_transition = tcp_state_transition, | |
718 | .app_conn_bind = tcp_app_conn_bind, | |
719 | .debug_packet = ip_vs_tcpudp_debug_packet, | |
720 | .timeout_change = tcp_timeout_change, | |
1da177e4 | 721 | }; |