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