1 /*
2 * ip_vs_proto_tcp.c: TCP load balancing support for IPVS
3 *
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 *
12 * Changes: Hans Schillstrom <hans.schillstrom@ericsson.com>
13 *
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
18 */
19
20 #define KMSG_COMPONENT "IPVS"
21 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
22
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 */
28 #include <net/ip6_checksum.h>
29 #include <linux/netfilter.h>
30 #include <linux/netfilter_ipv4.h>
31
32 #include <net/ip_vs.h>
33
34 static int
tcp_conn_schedule(int af,struct sk_buff * skb,struct ip_vs_proto_data * pd,int * verdict,struct ip_vs_conn ** cpp)35 tcp_conn_schedule(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
36 int *verdict, struct ip_vs_conn **cpp)
37 {
38 struct net *net;
39 struct ip_vs_service *svc;
40 struct tcphdr _tcph, *th;
41 struct ip_vs_iphdr iph;
42
43 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
44
45 th = skb_header_pointer(skb, iph.len, sizeof(_tcph), &_tcph);
46 if (th == NULL) {
47 *verdict = NF_DROP;
48 return 0;
49 }
50 net = skb_net(skb);
51 /* No !th->ack check to allow scheduling on SYN+ACK for Active FTP */
52 if (th->syn &&
53 (svc = ip_vs_service_get(net, af, skb->mark, iph.protocol,
54 &iph.daddr, th->dest))) {
55 int ignored;
56
57 if (ip_vs_todrop(net_ipvs(net))) {
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 */
71 *cpp = ip_vs_schedule(svc, skb, pd, &ignored);
72 if (!*cpp && ignored <= 0) {
73 if (!ignored)
74 *verdict = ip_vs_leave(svc, skb, pd);
75 else {
76 ip_vs_service_put(svc);
77 *verdict = NF_DROP;
78 }
79 return 0;
80 }
81 ip_vs_service_put(svc);
82 }
83 /* NF_ACCEPT */
84 return 1;
85 }
86
87
88 static inline void
tcp_fast_csum_update(int af,struct tcphdr * tcph,const union nf_inet_addr * oldip,const union nf_inet_addr * newip,__be16 oldport,__be16 newport)89 tcp_fast_csum_update(int af, struct tcphdr *tcph,
90 const union nf_inet_addr *oldip,
91 const union nf_inet_addr *newip,
92 __be16 oldport, __be16 newport)
93 {
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
102 tcph->check =
103 csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
104 ip_vs_check_diff2(oldport, newport,
105 ~csum_unfold(tcph->check))));
106 }
107
108
109 static inline void
tcp_partial_csum_update(int af,struct tcphdr * tcph,const union nf_inet_addr * oldip,const union nf_inet_addr * newip,__be16 oldlen,__be16 newlen)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 =
118 ~csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
119 ip_vs_check_diff2(oldlen, newlen,
120 csum_unfold(tcph->check))));
121 else
122 #endif
123 tcph->check =
124 ~csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
125 ip_vs_check_diff2(oldlen, newlen,
126 csum_unfold(tcph->check))));
127 }
128
129
130 static int
tcp_snat_handler(struct sk_buff * skb,struct ip_vs_protocol * pp,struct ip_vs_conn * cp)131 tcp_snat_handler(struct sk_buff *skb,
132 struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
133 {
134 struct tcphdr *tcph;
135 unsigned int tcphoff;
136 int oldlen;
137 int payload_csum = 0;
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);
145 oldlen = skb->len - tcphoff;
146
147 /* csum_check requires unshared skb */
148 if (!skb_make_writable(skb, tcphoff+sizeof(*tcph)))
149 return 0;
150
151 if (unlikely(cp->app != NULL)) {
152 int ret;
153
154 /* Some checks before mangling */
155 if (pp->csum_check && !pp->csum_check(cp->af, skb, pp))
156 return 0;
157
158 /* Call application helper if needed */
159 if (!(ret = ip_vs_app_pkt_out(cp, skb)))
160 return 0;
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;
166 }
167
168 tcph = (void *)skb_network_header(skb) + tcphoff;
169 tcph->source = cp->vport;
170
171 /* Adjust TCP checksums */
172 if (skb->ip_summed == CHECKSUM_PARTIAL) {
173 tcp_partial_csum_update(cp->af, tcph, &cp->daddr, &cp->vaddr,
174 htons(oldlen),
175 htons(skb->len - tcphoff));
176 } else if (!payload_csum) {
177 /* Only port and addr are changed, do fast csum update */
178 tcp_fast_csum_update(cp->af, tcph, &cp->daddr, &cp->vaddr,
179 cp->dport, cp->vport);
180 if (skb->ip_summed == CHECKSUM_COMPLETE)
181 skb->ip_summed = (cp->app && pp->csum_check) ?
182 CHECKSUM_UNNECESSARY : CHECKSUM_NONE;
183 } else {
184 /* full checksum calculation */
185 tcph->check = 0;
186 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
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);
200 skb->ip_summed = CHECKSUM_UNNECESSARY;
201
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
tcp_dnat_handler(struct sk_buff * skb,struct ip_vs_protocol * pp,struct ip_vs_conn * cp)211 tcp_dnat_handler(struct sk_buff *skb,
212 struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
213 {
214 struct tcphdr *tcph;
215 unsigned int tcphoff;
216 int oldlen;
217 int payload_csum = 0;
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);
225 oldlen = skb->len - tcphoff;
226
227 /* csum_check requires unshared skb */
228 if (!skb_make_writable(skb, tcphoff+sizeof(*tcph)))
229 return 0;
230
231 if (unlikely(cp->app != NULL)) {
232 int ret;
233
234 /* Some checks before mangling */
235 if (pp->csum_check && !pp->csum_check(cp->af, skb, pp))
236 return 0;
237
238 /*
239 * Attempt ip_vs_app call.
240 * It will fix ip_vs_conn and iph ack_seq stuff
241 */
242 if (!(ret = ip_vs_app_pkt_in(cp, skb)))
243 return 0;
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;
249 }
250
251 tcph = (void *)skb_network_header(skb) + tcphoff;
252 tcph->dest = cp->dport;
253
254 /*
255 * Adjust TCP checksums
256 */
257 if (skb->ip_summed == CHECKSUM_PARTIAL) {
258 tcp_partial_csum_update(cp->af, tcph, &cp->vaddr, &cp->daddr,
259 htons(oldlen),
260 htons(skb->len - tcphoff));
261 } else if (!payload_csum) {
262 /* Only port and addr are changed, do fast csum update */
263 tcp_fast_csum_update(cp->af, tcph, &cp->vaddr, &cp->daddr,
264 cp->vport, cp->dport);
265 if (skb->ip_summed == CHECKSUM_COMPLETE)
266 skb->ip_summed = (cp->app && pp->csum_check) ?
267 CHECKSUM_UNNECESSARY : CHECKSUM_NONE;
268 } else {
269 /* full checksum calculation */
270 tcph->check = 0;
271 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
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);
285 skb->ip_summed = CHECKSUM_UNNECESSARY;
286 }
287 return 1;
288 }
289
290
291 static int
tcp_csum_check(int af,struct sk_buff * skb,struct ip_vs_protocol * pp)292 tcp_csum_check(int af, struct sk_buff *skb, struct ip_vs_protocol *pp)
293 {
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);
302
303 switch (skb->ip_summed) {
304 case CHECKSUM_NONE:
305 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
306 case CHECKSUM_COMPLETE:
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)) {
314 IP_VS_DBG_RL_PKT(0, af, pp, skb, 0,
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)) {
325 IP_VS_DBG_RL_PKT(0, af, pp, skb, 0,
326 "Failed checksum for");
327 return 0;
328 }
329 break;
330 default:
331 /* No need to checksum. */
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
343 static const int tcp_state_off[IP_VS_DIR_LAST] = {
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 */
352 static const int tcp_timeouts[IP_VS_TCP_S_LAST+1] = {
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
367 static const char *const tcp_state_name_table[IP_VS_TCP_S_LAST+1] = {
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
tcp_state_name(int state)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
tcp_timeout_change(struct ip_vs_proto_data * pd,int flags)451 static void tcp_timeout_change(struct ip_vs_proto_data *pd, int flags)
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 */
461 pd->tcp_state_table = (on ? tcp_states_dos : tcp_states);
462 }
463
tcp_state_idx(struct tcphdr * th)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
set_tcp_state(struct ip_vs_proto_data * pd,struct ip_vs_conn * cp,int direction,struct tcphdr * th)478 set_tcp_state(struct ip_vs_proto_data *pd, struct ip_vs_conn *cp,
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
501 new_state =
502 pd->tcp_state_table[state_off+state_idx].next_state[cp->state];
503
504 tcp_state_out:
505 if (new_state != cp->state) {
506 struct ip_vs_dest *dest = cp->dest;
507
508 IP_VS_DBG_BUF(8, "%s %s [%c%c%c%c] %s:%d->"
509 "%s:%d state: %s->%s conn->refcnt:%d\n",
510 pd->pp->name,
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
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
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];
544 }
545
546 /*
547 * Handle state transitions
548 */
549 static void
tcp_state_transition(struct ip_vs_conn * cp,int direction,const struct sk_buff * skb,struct ip_vs_proto_data * pd)550 tcp_state_transition(struct ip_vs_conn *cp, int direction,
551 const struct sk_buff *skb,
552 struct ip_vs_proto_data *pd)
553 {
554 struct tcphdr _tcph, *th;
555
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);
563 if (th == NULL)
564 return;
565
566 spin_lock(&cp->lock);
567 set_tcp_state(pd, cp, direction, th);
568 spin_unlock(&cp->lock);
569 }
570
tcp_app_hashkey(__be16 port)571 static inline __u16 tcp_app_hashkey(__be16 port)
572 {
573 return (((__force u16)port >> TCP_APP_TAB_BITS) ^ (__force u16)port)
574 & TCP_APP_TAB_MASK;
575 }
576
577
tcp_register_app(struct net * net,struct ip_vs_app * inc)578 static int tcp_register_app(struct net *net, struct ip_vs_app *inc)
579 {
580 struct ip_vs_app *i;
581 __u16 hash;
582 __be16 port = inc->port;
583 int ret = 0;
584 struct netns_ipvs *ipvs = net_ipvs(net);
585 struct ip_vs_proto_data *pd = ip_vs_proto_data_get(net, IPPROTO_TCP);
586
587 hash = tcp_app_hashkey(port);
588
589 spin_lock_bh(&ipvs->tcp_app_lock);
590 list_for_each_entry(i, &ipvs->tcp_apps[hash], p_list) {
591 if (i->port == port) {
592 ret = -EEXIST;
593 goto out;
594 }
595 }
596 list_add(&inc->p_list, &ipvs->tcp_apps[hash]);
597 atomic_inc(&pd->appcnt);
598
599 out:
600 spin_unlock_bh(&ipvs->tcp_app_lock);
601 return ret;
602 }
603
604
605 static void
tcp_unregister_app(struct net * net,struct ip_vs_app * inc)606 tcp_unregister_app(struct net *net, struct ip_vs_app *inc)
607 {
608 struct netns_ipvs *ipvs = net_ipvs(net);
609 struct ip_vs_proto_data *pd = ip_vs_proto_data_get(net, IPPROTO_TCP);
610
611 spin_lock_bh(&ipvs->tcp_app_lock);
612 atomic_dec(&pd->appcnt);
613 list_del(&inc->p_list);
614 spin_unlock_bh(&ipvs->tcp_app_lock);
615 }
616
617
618 static int
tcp_app_conn_bind(struct ip_vs_conn * cp)619 tcp_app_conn_bind(struct ip_vs_conn *cp)
620 {
621 struct netns_ipvs *ipvs = net_ipvs(ip_vs_conn_net(cp));
622 int hash;
623 struct ip_vs_app *inc;
624 int result = 0;
625
626 /* Default binding: bind app only for NAT */
627 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
628 return 0;
629
630 /* Lookup application incarnations and bind the right one */
631 hash = tcp_app_hashkey(cp->vport);
632
633 spin_lock(&ipvs->tcp_app_lock);
634 list_for_each_entry(inc, &ipvs->tcp_apps[hash], p_list) {
635 if (inc->port == cp->vport) {
636 if (unlikely(!ip_vs_app_inc_get(inc)))
637 break;
638 spin_unlock(&ipvs->tcp_app_lock);
639
640 IP_VS_DBG_BUF(9, "%s(): Binding conn %s:%u->"
641 "%s:%u to app %s on port %u\n",
642 __func__,
643 IP_VS_DBG_ADDR(cp->af, &cp->caddr),
644 ntohs(cp->cport),
645 IP_VS_DBG_ADDR(cp->af, &cp->vaddr),
646 ntohs(cp->vport),
647 inc->name, ntohs(inc->port));
648
649 cp->app = inc;
650 if (inc->init_conn)
651 result = inc->init_conn(inc, cp);
652 goto out;
653 }
654 }
655 spin_unlock(&ipvs->tcp_app_lock);
656
657 out:
658 return result;
659 }
660
661
662 /*
663 * Set LISTEN timeout. (ip_vs_conn_put will setup timer)
664 */
ip_vs_tcp_conn_listen(struct net * net,struct ip_vs_conn * cp)665 void ip_vs_tcp_conn_listen(struct net *net, struct ip_vs_conn *cp)
666 {
667 struct ip_vs_proto_data *pd = ip_vs_proto_data_get(net, IPPROTO_TCP);
668
669 spin_lock(&cp->lock);
670 cp->state = IP_VS_TCP_S_LISTEN;
671 cp->timeout = (pd ? pd->timeout_table[IP_VS_TCP_S_LISTEN]
672 : tcp_timeouts[IP_VS_TCP_S_LISTEN]);
673 spin_unlock(&cp->lock);
674 }
675
676 /* ---------------------------------------------
677 * timeouts is netns related now.
678 * ---------------------------------------------
679 */
__ip_vs_tcp_init(struct net * net,struct ip_vs_proto_data * pd)680 static void __ip_vs_tcp_init(struct net *net, struct ip_vs_proto_data *pd)
681 {
682 struct netns_ipvs *ipvs = net_ipvs(net);
683
684 ip_vs_init_hash_table(ipvs->tcp_apps, TCP_APP_TAB_SIZE);
685 spin_lock_init(&ipvs->tcp_app_lock);
686 pd->timeout_table = ip_vs_create_timeout_table((int *)tcp_timeouts,
687 sizeof(tcp_timeouts));
688 pd->tcp_state_table = tcp_states;
689 }
690
__ip_vs_tcp_exit(struct net * net,struct ip_vs_proto_data * pd)691 static void __ip_vs_tcp_exit(struct net *net, struct ip_vs_proto_data *pd)
692 {
693 kfree(pd->timeout_table);
694 }
695
696
697 struct ip_vs_protocol ip_vs_protocol_tcp = {
698 .name = "TCP",
699 .protocol = IPPROTO_TCP,
700 .num_states = IP_VS_TCP_S_LAST,
701 .dont_defrag = 0,
702 .init = NULL,
703 .exit = NULL,
704 .init_netns = __ip_vs_tcp_init,
705 .exit_netns = __ip_vs_tcp_exit,
706 .register_app = tcp_register_app,
707 .unregister_app = tcp_unregister_app,
708 .conn_schedule = tcp_conn_schedule,
709 .conn_in_get = ip_vs_conn_in_get_proto,
710 .conn_out_get = ip_vs_conn_out_get_proto,
711 .snat_handler = tcp_snat_handler,
712 .dnat_handler = tcp_dnat_handler,
713 .csum_check = tcp_csum_check,
714 .state_name = tcp_state_name,
715 .state_transition = tcp_state_transition,
716 .app_conn_bind = tcp_app_conn_bind,
717 .debug_packet = ip_vs_tcpudp_debug_packet,
718 .timeout_change = tcp_timeout_change,
719 };
720