aboutsummaryrefslogtreecommitdiffstats
path: root/net/dccp/ipv4.c
diff options
context:
space:
mode:
Diffstat (limited to 'net/dccp/ipv4.c')
-rw-r--r--net/dccp/ipv4.c1289
1 files changed, 1289 insertions, 0 deletions
diff --git a/net/dccp/ipv4.c b/net/dccp/ipv4.c
new file mode 100644
index 000000000000..083bacaecb3b
--- /dev/null
+++ b/net/dccp/ipv4.c
@@ -0,0 +1,1289 @@
1/*
2 * net/dccp/ipv4.c
3 *
4 * An implementation of the DCCP protocol
5 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>
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
13#include <linux/config.h>
14#include <linux/dccp.h>
15#include <linux/icmp.h>
16#include <linux/module.h>
17#include <linux/skbuff.h>
18#include <linux/random.h>
19
20#include <net/icmp.h>
21#include <net/inet_hashtables.h>
22#include <net/sock.h>
23#include <net/tcp_states.h>
24#include <net/xfrm.h>
25
26#include "ccid.h"
27#include "dccp.h"
28
29struct inet_hashinfo __cacheline_aligned dccp_hashinfo = {
30 .lhash_lock = RW_LOCK_UNLOCKED,
31 .lhash_users = ATOMIC_INIT(0),
32 .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(dccp_hashinfo.lhash_wait),
33 .portalloc_lock = SPIN_LOCK_UNLOCKED,
34 .port_rover = 1024 - 1,
35};
36
37static int dccp_v4_get_port(struct sock *sk, const unsigned short snum)
38{
39 return inet_csk_get_port(&dccp_hashinfo, sk, snum);
40}
41
42static void dccp_v4_hash(struct sock *sk)
43{
44 inet_hash(&dccp_hashinfo, sk);
45}
46
47static void dccp_v4_unhash(struct sock *sk)
48{
49 inet_unhash(&dccp_hashinfo, sk);
50}
51
52/* called with local bh disabled */
53static int __dccp_v4_check_established(struct sock *sk, const __u16 lport,
54 struct inet_timewait_sock **twp)
55{
56 struct inet_sock *inet = inet_sk(sk);
57 const u32 daddr = inet->rcv_saddr;
58 const u32 saddr = inet->daddr;
59 const int dif = sk->sk_bound_dev_if;
60 INET_ADDR_COOKIE(acookie, saddr, daddr)
61 const __u32 ports = INET_COMBINED_PORTS(inet->dport, lport);
62 const int hash = inet_ehashfn(daddr, lport, saddr, inet->dport, dccp_hashinfo.ehash_size);
63 struct inet_ehash_bucket *head = &dccp_hashinfo.ehash[hash];
64 const struct sock *sk2;
65 const struct hlist_node *node;
66 struct inet_timewait_sock *tw;
67
68 write_lock(&head->lock);
69
70 /* Check TIME-WAIT sockets first. */
71 sk_for_each(sk2, node, &(head + dccp_hashinfo.ehash_size)->chain) {
72 tw = inet_twsk(sk2);
73
74 if (INET_TW_MATCH(sk2, acookie, saddr, daddr, ports, dif))
75 goto not_unique;
76 }
77 tw = NULL;
78
79 /* And established part... */
80 sk_for_each(sk2, node, &head->chain) {
81 if (INET_MATCH(sk2, acookie, saddr, daddr, ports, dif))
82 goto not_unique;
83 }
84
85 /* Must record num and sport now. Otherwise we will see
86 * in hash table socket with a funny identity. */
87 inet->num = lport;
88 inet->sport = htons(lport);
89 sk->sk_hashent = hash;
90 BUG_TRAP(sk_unhashed(sk));
91 __sk_add_node(sk, &head->chain);
92 sock_prot_inc_use(sk->sk_prot);
93 write_unlock(&head->lock);
94
95 if (twp != NULL) {
96 *twp = tw;
97 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
98 } else if (tw != NULL) {
99 /* Silly. Should hash-dance instead... */
100 dccp_tw_deschedule(tw);
101 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
102
103 inet_twsk_put(tw);
104 }
105
106 return 0;
107
108not_unique:
109 write_unlock(&head->lock);
110 return -EADDRNOTAVAIL;
111}
112
113/*
114 * Bind a port for a connect operation and hash it.
115 */
116static int dccp_v4_hash_connect(struct sock *sk)
117{
118 const unsigned short snum = inet_sk(sk)->num;
119 struct inet_bind_hashbucket *head;
120 struct inet_bind_bucket *tb;
121 int ret;
122
123 if (snum == 0) {
124 int rover;
125 int low = sysctl_local_port_range[0];
126 int high = sysctl_local_port_range[1];
127 int remaining = (high - low) + 1;
128 struct hlist_node *node;
129 struct inet_timewait_sock *tw = NULL;
130
131 local_bh_disable();
132
133 /* TODO. Actually it is not so bad idea to remove
134 * dccp_hashinfo.portalloc_lock before next submission to Linus.
135 * As soon as we touch this place at all it is time to think.
136 *
137 * Now it protects single _advisory_ variable dccp_hashinfo.port_rover,
138 * hence it is mostly useless.
139 * Code will work nicely if we just delete it, but
140 * I am afraid in contented case it will work not better or
141 * even worse: another cpu just will hit the same bucket
142 * and spin there.
143 * So some cpu salt could remove both contention and
144 * memory pingpong. Any ideas how to do this in a nice way?
145 */
146 spin_lock(&dccp_hashinfo.portalloc_lock);
147 rover = dccp_hashinfo.port_rover;
148
149 do {
150 rover++;
151 if ((rover < low) || (rover > high))
152 rover = low;
153 head = &dccp_hashinfo.bhash[inet_bhashfn(rover, dccp_hashinfo.bhash_size)];
154 spin_lock(&head->lock);
155
156 /* Does not bother with rcv_saddr checks,
157 * because the established check is already
158 * unique enough.
159 */
160 inet_bind_bucket_for_each(tb, node, &head->chain) {
161 if (tb->port == rover) {
162 BUG_TRAP(!hlist_empty(&tb->owners));
163 if (tb->fastreuse >= 0)
164 goto next_port;
165 if (!__dccp_v4_check_established(sk,
166 rover,
167 &tw))
168 goto ok;
169 goto next_port;
170 }
171 }
172
173 tb = inet_bind_bucket_create(dccp_hashinfo.bind_bucket_cachep, head, rover);
174 if (tb == NULL) {
175 spin_unlock(&head->lock);
176 break;
177 }
178 tb->fastreuse = -1;
179 goto ok;
180
181 next_port:
182 spin_unlock(&head->lock);
183 } while (--remaining > 0);
184 dccp_hashinfo.port_rover = rover;
185 spin_unlock(&dccp_hashinfo.portalloc_lock);
186
187 local_bh_enable();
188
189 return -EADDRNOTAVAIL;
190
191ok:
192 /* All locks still held and bhs disabled */
193 dccp_hashinfo.port_rover = rover;
194 spin_unlock(&dccp_hashinfo.portalloc_lock);
195
196 inet_bind_hash(sk, tb, rover);
197 if (sk_unhashed(sk)) {
198 inet_sk(sk)->sport = htons(rover);
199 __inet_hash(&dccp_hashinfo, sk, 0);
200 }
201 spin_unlock(&head->lock);
202
203 if (tw != NULL) {
204 dccp_tw_deschedule(tw);
205 inet_twsk_put(tw);
206 }
207
208 ret = 0;
209 goto out;
210 }
211
212 head = &dccp_hashinfo.bhash[inet_bhashfn(snum, dccp_hashinfo.bhash_size)];
213 tb = inet_csk(sk)->icsk_bind_hash;
214 spin_lock_bh(&head->lock);
215 if (sk_head(&tb->owners) == sk && sk->sk_bind_node.next == NULL) {
216 __inet_hash(&dccp_hashinfo, sk, 0);
217 spin_unlock_bh(&head->lock);
218 return 0;
219 } else {
220 spin_unlock(&head->lock);
221 /* No definite answer... Walk to established hash table */
222 ret = __dccp_v4_check_established(sk, snum, NULL);
223out:
224 local_bh_enable();
225 return ret;
226 }
227}
228
229static int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
230 int addr_len)
231{
232 struct inet_sock *inet = inet_sk(sk);
233 struct dccp_sock *dp = dccp_sk(sk);
234 const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
235 struct rtable *rt;
236 u32 daddr, nexthop;
237 int tmp;
238 int err;
239
240 dp->dccps_role = DCCP_ROLE_CLIENT;
241
242 if (addr_len < sizeof(struct sockaddr_in))
243 return -EINVAL;
244
245 if (usin->sin_family != AF_INET)
246 return -EAFNOSUPPORT;
247
248 nexthop = daddr = usin->sin_addr.s_addr;
249 if (inet->opt != NULL && inet->opt->srr) {
250 if (daddr == 0)
251 return -EINVAL;
252 nexthop = inet->opt->faddr;
253 }
254
255 tmp = ip_route_connect(&rt, nexthop, inet->saddr,
256 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
257 IPPROTO_DCCP,
258 inet->sport, usin->sin_port, sk);
259 if (tmp < 0)
260 return tmp;
261
262 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
263 ip_rt_put(rt);
264 return -ENETUNREACH;
265 }
266
267 if (inet->opt == NULL || !inet->opt->srr)
268 daddr = rt->rt_dst;
269
270 if (inet->saddr == 0)
271 inet->saddr = rt->rt_src;
272 inet->rcv_saddr = inet->saddr;
273
274 inet->dport = usin->sin_port;
275 inet->daddr = daddr;
276
277 dp->dccps_ext_header_len = 0;
278 if (inet->opt != NULL)
279 dp->dccps_ext_header_len = inet->opt->optlen;
280 /*
281 * Socket identity is still unknown (sport may be zero).
282 * However we set state to DCCP_REQUESTING and not releasing socket
283 * lock select source port, enter ourselves into the hash tables and
284 * complete initialization after this.
285 */
286 dccp_set_state(sk, DCCP_REQUESTING);
287 err = dccp_v4_hash_connect(sk);
288 if (err != 0)
289 goto failure;
290
291 err = ip_route_newports(&rt, inet->sport, inet->dport, sk);
292 if (err != 0)
293 goto failure;
294
295 /* OK, now commit destination to socket. */
296 sk_setup_caps(sk, &rt->u.dst);
297
298 dp->dccps_gar =
299 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
300 inet->daddr,
301 inet->sport,
302 usin->sin_port);
303 dccp_update_gss(sk, dp->dccps_iss);
304
305 inet->id = dp->dccps_iss ^ jiffies;
306
307 err = dccp_connect(sk);
308 rt = NULL;
309 if (err != 0)
310 goto failure;
311out:
312 return err;
313failure:
314 /* This unhashes the socket and releases the local port, if necessary. */
315 dccp_set_state(sk, DCCP_CLOSED);
316 ip_rt_put(rt);
317 sk->sk_route_caps = 0;
318 inet->dport = 0;
319 goto out;
320}
321
322/*
323 * This routine does path mtu discovery as defined in RFC1191.
324 */
325static inline void dccp_do_pmtu_discovery(struct sock *sk,
326 const struct iphdr *iph,
327 u32 mtu)
328{
329 struct dst_entry *dst;
330 const struct inet_sock *inet = inet_sk(sk);
331 const struct dccp_sock *dp = dccp_sk(sk);
332
333 /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
334 * send out by Linux are always < 576bytes so they should go through
335 * unfragmented).
336 */
337 if (sk->sk_state == DCCP_LISTEN)
338 return;
339
340 /* We don't check in the destentry if pmtu discovery is forbidden
341 * on this route. We just assume that no packet_to_big packets
342 * are send back when pmtu discovery is not active.
343 * There is a small race when the user changes this flag in the
344 * route, but I think that's acceptable.
345 */
346 if ((dst = __sk_dst_check(sk, 0)) == NULL)
347 return;
348
349 dst->ops->update_pmtu(dst, mtu);
350
351 /* Something is about to be wrong... Remember soft error
352 * for the case, if this connection will not able to recover.
353 */
354 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
355 sk->sk_err_soft = EMSGSIZE;
356
357 mtu = dst_mtu(dst);
358
359 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
360 dp->dccps_pmtu_cookie > mtu) {
361 dccp_sync_mss(sk, mtu);
362
363 /*
364 * From: draft-ietf-dccp-spec-11.txt
365 *
366 * DCCP-Sync packets are the best choice for upward probing,
367 * since DCCP-Sync probes do not risk application data loss.
368 */
369 dccp_send_sync(sk, dp->dccps_gsr);
370 } /* else let the usual retransmit timer handle it */
371}
372
373static void dccp_v4_ctl_send_ack(struct sk_buff *rxskb)
374{
375 int err;
376 struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
377 const int dccp_hdr_ack_len = sizeof(struct dccp_hdr) +
378 sizeof(struct dccp_hdr_ext) +
379 sizeof(struct dccp_hdr_ack_bits);
380 struct sk_buff *skb;
381
382 if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
383 return;
384
385 skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
386 if (skb == NULL)
387 return;
388
389 /* Reserve space for headers. */
390 skb_reserve(skb, MAX_DCCP_HEADER);
391
392 skb->dst = dst_clone(rxskb->dst);
393
394 skb->h.raw = skb_push(skb, dccp_hdr_ack_len);
395 dh = dccp_hdr(skb);
396 memset(dh, 0, dccp_hdr_ack_len);
397
398 /* Build DCCP header and checksum it. */
399 dh->dccph_type = DCCP_PKT_ACK;
400 dh->dccph_sport = rxdh->dccph_dport;
401 dh->dccph_dport = rxdh->dccph_sport;
402 dh->dccph_doff = dccp_hdr_ack_len / 4;
403 dh->dccph_x = 1;
404
405 dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
406 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), DCCP_SKB_CB(rxskb)->dccpd_seq);
407
408 bh_lock_sock(dccp_ctl_socket->sk);
409 err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
410 rxskb->nh.iph->daddr, rxskb->nh.iph->saddr, NULL);
411 bh_unlock_sock(dccp_ctl_socket->sk);
412
413 if (err == NET_XMIT_CN || err == 0) {
414 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
415 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
416 }
417}
418
419static void dccp_v4_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req)
420{
421 dccp_v4_ctl_send_ack(skb);
422}
423
424static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
425 struct dst_entry *dst)
426{
427 int err = -1;
428 struct sk_buff *skb;
429
430 /* First, grab a route. */
431
432 if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
433 goto out;
434
435 skb = dccp_make_response(sk, dst, req);
436 if (skb != NULL) {
437 const struct inet_request_sock *ireq = inet_rsk(req);
438
439 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
440 ireq->rmt_addr,
441 ireq->opt);
442 if (err == NET_XMIT_CN)
443 err = 0;
444 }
445
446out:
447 dst_release(dst);
448 return err;
449}
450
451/*
452 * This routine is called by the ICMP module when it gets some sort of error
453 * condition. If err < 0 then the socket should be closed and the error
454 * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
455 * After adjustment header points to the first 8 bytes of the tcp header. We
456 * need to find the appropriate port.
457 *
458 * The locking strategy used here is very "optimistic". When someone else
459 * accesses the socket the ICMP is just dropped and for some paths there is no
460 * check at all. A more general error queue to queue errors for later handling
461 * is probably better.
462 */
463void dccp_v4_err(struct sk_buff *skb, u32 info)
464{
465 const struct iphdr *iph = (struct iphdr *)skb->data;
466 const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + (iph->ihl << 2));
467 struct dccp_sock *dp;
468 struct inet_sock *inet;
469 const int type = skb->h.icmph->type;
470 const int code = skb->h.icmph->code;
471 struct sock *sk;
472 __u64 seq;
473 int err;
474
475 if (skb->len < (iph->ihl << 2) + 8) {
476 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
477 return;
478 }
479
480 sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
481 iph->saddr, dh->dccph_sport, inet_iif(skb));
482 if (sk == NULL) {
483 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
484 return;
485 }
486
487 if (sk->sk_state == DCCP_TIME_WAIT) {
488 inet_twsk_put((struct inet_timewait_sock *)sk);
489 return;
490 }
491
492 bh_lock_sock(sk);
493 /* If too many ICMPs get dropped on busy
494 * servers this needs to be solved differently.
495 */
496 if (sock_owned_by_user(sk))
497 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
498
499 if (sk->sk_state == DCCP_CLOSED)
500 goto out;
501
502 dp = dccp_sk(sk);
503 seq = dccp_hdr_seq(skb);
504 if (sk->sk_state != DCCP_LISTEN &&
505 !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
506 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
507 goto out;
508 }
509
510 switch (type) {
511 case ICMP_SOURCE_QUENCH:
512 /* Just silently ignore these. */
513 goto out;
514 case ICMP_PARAMETERPROB:
515 err = EPROTO;
516 break;
517 case ICMP_DEST_UNREACH:
518 if (code > NR_ICMP_UNREACH)
519 goto out;
520
521 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
522 if (!sock_owned_by_user(sk))
523 dccp_do_pmtu_discovery(sk, iph, info);
524 goto out;
525 }
526
527 err = icmp_err_convert[code].errno;
528 break;
529 case ICMP_TIME_EXCEEDED:
530 err = EHOSTUNREACH;
531 break;
532 default:
533 goto out;
534 }
535
536 switch (sk->sk_state) {
537 struct request_sock *req , **prev;
538 case DCCP_LISTEN:
539 if (sock_owned_by_user(sk))
540 goto out;
541 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
542 iph->daddr, iph->saddr);
543 if (!req)
544 goto out;
545
546 /*
547 * ICMPs are not backlogged, hence we cannot get an established
548 * socket here.
549 */
550 BUG_TRAP(!req->sk);
551
552 if (seq != dccp_rsk(req)->dreq_iss) {
553 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
554 goto out;
555 }
556 /*
557 * Still in RESPOND, just remove it silently.
558 * There is no good way to pass the error to the newly
559 * created socket, and POSIX does not want network
560 * errors returned from accept().
561 */
562 inet_csk_reqsk_queue_drop(sk, req, prev);
563 goto out;
564
565 case DCCP_REQUESTING:
566 case DCCP_RESPOND:
567 if (!sock_owned_by_user(sk)) {
568 DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
569 sk->sk_err = err;
570
571 sk->sk_error_report(sk);
572
573 dccp_done(sk);
574 } else
575 sk->sk_err_soft = err;
576 goto out;
577 }
578
579 /* If we've already connected we will keep trying
580 * until we time out, or the user gives up.
581 *
582 * rfc1122 4.2.3.9 allows to consider as hard errors
583 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
584 * but it is obsoleted by pmtu discovery).
585 *
586 * Note, that in modern internet, where routing is unreliable
587 * and in each dark corner broken firewalls sit, sending random
588 * errors ordered by their masters even this two messages finally lose
589 * their original sense (even Linux sends invalid PORT_UNREACHs)
590 *
591 * Now we are in compliance with RFCs.
592 * --ANK (980905)
593 */
594
595 inet = inet_sk(sk);
596 if (!sock_owned_by_user(sk) && inet->recverr) {
597 sk->sk_err = err;
598 sk->sk_error_report(sk);
599 } else /* Only an error on timeout */
600 sk->sk_err_soft = err;
601out:
602 bh_unlock_sock(sk);
603 sock_put(sk);
604}
605
606extern struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst, enum dccp_reset_codes code);
607
608int dccp_v4_send_reset(struct sock *sk, enum dccp_reset_codes code)
609{
610 struct sk_buff *skb;
611 /*
612 * FIXME: what if rebuild_header fails?
613 * Should we be doing a rebuild_header here?
614 */
615 int err = inet_sk_rebuild_header(sk);
616
617 if (err != 0)
618 return err;
619
620 skb = dccp_make_reset(sk, sk->sk_dst_cache, code);
621 if (skb != NULL) {
622 const struct dccp_sock *dp = dccp_sk(sk);
623 const struct inet_sock *inet = inet_sk(sk);
624
625 err = ip_build_and_send_pkt(skb, sk,
626 inet->saddr, inet->daddr, NULL);
627 if (err == NET_XMIT_CN)
628 err = 0;
629
630 ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk);
631 ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk);
632 }
633
634 return err;
635}
636
637static inline u64 dccp_v4_init_sequence(const struct sock *sk,
638 const struct sk_buff *skb)
639{
640 return secure_dccp_sequence_number(skb->nh.iph->daddr,
641 skb->nh.iph->saddr,
642 dccp_hdr(skb)->dccph_dport,
643 dccp_hdr(skb)->dccph_sport);
644}
645
646int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
647{
648 struct inet_request_sock *ireq;
649 struct dccp_sock dp;
650 struct request_sock *req;
651 struct dccp_request_sock *dreq;
652 const __u32 saddr = skb->nh.iph->saddr;
653 const __u32 daddr = skb->nh.iph->daddr;
654 struct dst_entry *dst = NULL;
655
656 /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
657 if (((struct rtable *)skb->dst)->rt_flags &
658 (RTCF_BROADCAST | RTCF_MULTICAST))
659 goto drop;
660
661 /*
662 * TW buckets are converted to open requests without
663 * limitations, they conserve resources and peer is
664 * evidently real one.
665 */
666 if (inet_csk_reqsk_queue_is_full(sk))
667 goto drop;
668
669 /*
670 * Accept backlog is full. If we have already queued enough
671 * of warm entries in syn queue, drop request. It is better than
672 * clogging syn queue with openreqs with exponentially increasing
673 * timeout.
674 */
675 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
676 goto drop;
677
678 req = reqsk_alloc(sk->sk_prot->rsk_prot);
679 if (req == NULL)
680 goto drop;
681
682 /* FIXME: process options */
683
684 dccp_openreq_init(req, &dp, skb);
685
686 ireq = inet_rsk(req);
687 ireq->loc_addr = daddr;
688 ireq->rmt_addr = saddr;
689 /* FIXME: Merge Aristeu's option parsing code when ready */
690 req->rcv_wnd = 100; /* Fake, option parsing will get the right value */
691 ireq->opt = NULL;
692
693 /*
694 * Step 3: Process LISTEN state
695 *
696 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
697 *
698 * In fact we defer setting S.GSR, S.SWL, S.SWH to
699 * dccp_create_openreq_child.
700 */
701 dreq = dccp_rsk(req);
702 dreq->dreq_isr = DCCP_SKB_CB(skb)->dccpd_seq;
703 dreq->dreq_iss = dccp_v4_init_sequence(sk, skb);
704 dreq->dreq_service = dccp_hdr_request(skb)->dccph_req_service;
705
706 if (dccp_v4_send_response(sk, req, dst))
707 goto drop_and_free;
708
709 inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
710 return 0;
711
712drop_and_free:
713 /*
714 * FIXME: should be reqsk_free after implementing req->rsk_ops
715 */
716 __reqsk_free(req);
717drop:
718 DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
719 return -1;
720}
721
722/*
723 * The three way handshake has completed - we got a valid ACK or DATAACK -
724 * now create the new socket.
725 *
726 * This is the equivalent of TCP's tcp_v4_syn_recv_sock
727 */
728struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
729 struct request_sock *req,
730 struct dst_entry *dst)
731{
732 struct inet_request_sock *ireq;
733 struct inet_sock *newinet;
734 struct dccp_sock *newdp;
735 struct sock *newsk;
736
737 if (sk_acceptq_is_full(sk))
738 goto exit_overflow;
739
740 if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
741 goto exit;
742
743 newsk = dccp_create_openreq_child(sk, req, skb);
744 if (newsk == NULL)
745 goto exit;
746
747 sk_setup_caps(newsk, dst);
748
749 newdp = dccp_sk(newsk);
750 newinet = inet_sk(newsk);
751 ireq = inet_rsk(req);
752 newinet->daddr = ireq->rmt_addr;
753 newinet->rcv_saddr = ireq->loc_addr;
754 newinet->saddr = ireq->loc_addr;
755 newinet->opt = ireq->opt;
756 ireq->opt = NULL;
757 newinet->mc_index = inet_iif(skb);
758 newinet->mc_ttl = skb->nh.iph->ttl;
759 newinet->id = jiffies;
760
761 dccp_sync_mss(newsk, dst_mtu(dst));
762
763 __inet_hash(&dccp_hashinfo, newsk, 0);
764 __inet_inherit_port(&dccp_hashinfo, sk, newsk);
765
766 return newsk;
767
768exit_overflow:
769 NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
770exit:
771 NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
772 dst_release(dst);
773 return NULL;
774}
775
776static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
777{
778 const struct dccp_hdr *dh = dccp_hdr(skb);
779 const struct iphdr *iph = skb->nh.iph;
780 struct sock *nsk;
781 struct request_sock **prev;
782 /* Find possible connection requests. */
783 struct request_sock *req = inet_csk_search_req(sk, &prev,
784 dh->dccph_sport,
785 iph->saddr, iph->daddr);
786 if (req != NULL)
787 return dccp_check_req(sk, skb, req, prev);
788
789 nsk = __inet_lookup_established(&dccp_hashinfo,
790 iph->saddr, dh->dccph_sport,
791 iph->daddr, ntohs(dh->dccph_dport),
792 inet_iif(skb));
793 if (nsk != NULL) {
794 if (nsk->sk_state != DCCP_TIME_WAIT) {
795 bh_lock_sock(nsk);
796 return nsk;
797 }
798 inet_twsk_put((struct inet_timewait_sock *)nsk);
799 return NULL;
800 }
801
802 return sk;
803}
804
805int dccp_v4_checksum(struct sk_buff *skb)
806{
807 struct dccp_hdr* dh = dccp_hdr(skb);
808 int checksum_len;
809 u32 tmp;
810
811 if (dh->dccph_cscov == 0)
812 checksum_len = skb->len;
813 else {
814 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
815 checksum_len = checksum_len < skb->len ? checksum_len : skb->len;
816 }
817
818 tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
819 return csum_fold(tmp);
820}
821
822static int dccp_v4_verify_checksum(struct sk_buff *skb)
823{
824 struct dccp_hdr *th = dccp_hdr(skb);
825 const u16 remote_checksum = th->dccph_checksum;
826 u16 local_checksum;
827
828 /* FIXME: don't mess with skb payload */
829 th->dccph_checksum = 0; /* zero it for computation */
830
831 local_checksum = dccp_v4_checksum(skb);
832
833 /* FIXME: don't mess with skb payload */
834 th->dccph_checksum = remote_checksum; /* put it back */
835
836 return remote_checksum == local_checksum ? 0 : -1;
837}
838
839static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
840 struct sk_buff *skb)
841{
842 struct rtable *rt;
843 struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
844 .nl_u = { .ip4_u =
845 { .daddr = skb->nh.iph->saddr,
846 .saddr = skb->nh.iph->daddr,
847 .tos = RT_CONN_FLAGS(sk) } },
848 .proto = sk->sk_protocol,
849 .uli_u = { .ports =
850 { .sport = dccp_hdr(skb)->dccph_dport,
851 .dport = dccp_hdr(skb)->dccph_sport } } };
852
853 if (ip_route_output_flow(&rt, &fl, sk, 0)) {
854 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
855 return NULL;
856 }
857
858 return &rt->u.dst;
859}
860
861void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
862{
863 int err;
864 struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
865 const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
866 sizeof(struct dccp_hdr_ext) +
867 sizeof(struct dccp_hdr_reset);
868 struct sk_buff *skb;
869 struct dst_entry *dst;
870
871 /* Never send a reset in response to a reset. */
872 if (rxdh->dccph_type == DCCP_PKT_RESET)
873 return;
874
875 if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
876 return;
877
878 dst = dccp_v4_route_skb(dccp_ctl_socket->sk, rxskb);
879 if (dst == NULL)
880 return;
881
882 skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
883 if (skb == NULL)
884 goto out;
885
886 /* Reserve space for headers. */
887 skb_reserve(skb, MAX_DCCP_HEADER);
888 skb->dst = dst_clone(dst);
889
890 skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
891 dh = dccp_hdr(skb);
892 memset(dh, 0, dccp_hdr_reset_len);
893
894 /* Build DCCP header and checksum it. */
895 dh->dccph_type = DCCP_PKT_RESET;
896 dh->dccph_sport = rxdh->dccph_dport;
897 dh->dccph_dport = rxdh->dccph_sport;
898 dh->dccph_doff = dccp_hdr_reset_len / 4;
899 dh->dccph_x = 1;
900 dccp_hdr_reset(skb)->dccph_reset_code = DCCP_SKB_CB(rxskb)->dccpd_reset_code;
901
902 dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
903 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), DCCP_SKB_CB(rxskb)->dccpd_seq);
904
905 dh->dccph_checksum = dccp_v4_checksum(skb);
906
907 bh_lock_sock(dccp_ctl_socket->sk);
908 err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
909 rxskb->nh.iph->daddr, rxskb->nh.iph->saddr, NULL);
910 bh_unlock_sock(dccp_ctl_socket->sk);
911
912 if (err == NET_XMIT_CN || err == 0) {
913 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
914 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
915 }
916out:
917 dst_release(dst);
918}
919
920int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
921{
922 struct dccp_hdr *dh = dccp_hdr(skb);
923
924 if (sk->sk_state == DCCP_OPEN) { /* Fast path */
925 if (dccp_rcv_established(sk, skb, dh, skb->len))
926 goto reset;
927 return 0;
928 }
929
930 /*
931 * Step 3: Process LISTEN state
932 * If S.state == LISTEN,
933 * If P.type == Request or P contains a valid Init Cookie option,
934 * * Must scan the packet's options to check for an Init
935 * Cookie. Only the Init Cookie is processed here,
936 * however; other options are processed in Step 8. This
937 * scan need only be performed if the endpoint uses Init
938 * Cookies *
939 * * Generate a new socket and switch to that socket *
940 * Set S := new socket for this port pair
941 * S.state = RESPOND
942 * Choose S.ISS (initial seqno) or set from Init Cookie
943 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
944 * Continue with S.state == RESPOND
945 * * A Response packet will be generated in Step 11 *
946 * Otherwise,
947 * Generate Reset(No Connection) unless P.type == Reset
948 * Drop packet and return
949 *
950 * NOTE: the check for the packet types is done in dccp_rcv_state_process
951 */
952 if (sk->sk_state == DCCP_LISTEN) {
953 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
954
955 if (nsk == NULL)
956 goto discard;
957
958 if (nsk != sk) {
959 if (dccp_child_process(sk, nsk, skb))
960 goto reset;
961 return 0;
962 }
963 }
964
965 if (dccp_rcv_state_process(sk, skb, dh, skb->len))
966 goto reset;
967 return 0;
968
969reset:
970 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_NO_CONNECTION;
971 dccp_v4_ctl_send_reset(skb);
972discard:
973 kfree_skb(skb);
974 return 0;
975}
976
977static inline int dccp_invalid_packet(struct sk_buff *skb)
978{
979 const struct dccp_hdr *dh;
980
981 if (skb->pkt_type != PACKET_HOST)
982 return 1;
983
984 if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
985 dccp_pr_debug("pskb_may_pull failed\n");
986 return 1;
987 }
988
989 dh = dccp_hdr(skb);
990
991 /* If the packet type is not understood, drop packet and return */
992 if (dh->dccph_type >= DCCP_PKT_INVALID) {
993 dccp_pr_debug("invalid packet type\n");
994 return 1;
995 }
996
997 /*
998 * If P.Data Offset is too small for packet type, or too large for
999 * packet, drop packet and return
1000 */
1001 if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
1002 dccp_pr_debug("Offset(%u) too small 1\n", dh->dccph_doff);
1003 return 1;
1004 }
1005
1006 if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
1007 dccp_pr_debug("P.Data Offset(%u) too small 2\n", dh->dccph_doff);
1008 return 1;
1009 }
1010
1011 dh = dccp_hdr(skb);
1012
1013 /*
1014 * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
1015 * has short sequence numbers), drop packet and return
1016 */
1017 if (dh->dccph_x == 0 &&
1018 dh->dccph_type != DCCP_PKT_DATA &&
1019 dh->dccph_type != DCCP_PKT_ACK &&
1020 dh->dccph_type != DCCP_PKT_DATAACK) {
1021 dccp_pr_debug("P.type (%s) not Data, Ack nor DataAck and P.X == 0\n",
1022 dccp_packet_name(dh->dccph_type));
1023 return 1;
1024 }
1025
1026 /* If the header checksum is incorrect, drop packet and return */
1027 if (dccp_v4_verify_checksum(skb) < 0) {
1028 dccp_pr_debug("header checksum is incorrect\n");
1029 return 1;
1030 }
1031
1032 return 0;
1033}
1034
1035/* this is called when real data arrives */
1036int dccp_v4_rcv(struct sk_buff *skb)
1037{
1038 const struct dccp_hdr *dh;
1039 struct sock *sk;
1040 int rc;
1041
1042 /* Step 1: Check header basics: */
1043
1044 if (dccp_invalid_packet(skb))
1045 goto discard_it;
1046
1047 dh = dccp_hdr(skb);
1048#if 0
1049 /*
1050 * Use something like this to simulate some DATA/DATAACK loss to test
1051 * dccp_ackpkts_add, you'll get something like this on a session that
1052 * sends 10 DATA/DATAACK packets:
1053 *
1054 * dccp_ackpkts_print: 281473596467422 |0,0|3,0|0,0|3,0|0,0|3,0|0,0|3,0|0,1|
1055 *
1056 * 0, 0 means: DCCP_ACKPKTS_STATE_RECEIVED, RLE == just this packet
1057 * 0, 1 means: DCCP_ACKPKTS_STATE_RECEIVED, RLE == two adjacent packets with the same state
1058 * 3, 0 means: DCCP_ACKPKTS_STATE_NOT_RECEIVED, RLE == just this packet
1059 *
1060 * So...
1061 *
1062 * 281473596467422 was received
1063 * 281473596467421 was not received
1064 * 281473596467420 was received
1065 * 281473596467419 was not received
1066 * 281473596467418 was received
1067 * 281473596467417 was not received
1068 * 281473596467416 was received
1069 * 281473596467415 was not received
1070 * 281473596467414 was received
1071 * 281473596467413 was received (this one was the 3way handshake RESPONSE)
1072 *
1073 */
1074 if (dh->dccph_type == DCCP_PKT_DATA || dh->dccph_type == DCCP_PKT_DATAACK) {
1075 static int discard = 0;
1076
1077 if (discard) {
1078 discard = 0;
1079 goto discard_it;
1080 }
1081 discard = 1;
1082 }
1083#endif
1084 DCCP_SKB_CB(skb)->dccpd_seq = dccp_hdr_seq(skb);
1085 DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
1086
1087 dccp_pr_debug("%8.8s "
1088 "src=%u.%u.%u.%u@%-5d "
1089 "dst=%u.%u.%u.%u@%-5d seq=%llu",
1090 dccp_packet_name(dh->dccph_type),
1091 NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
1092 NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
1093 DCCP_SKB_CB(skb)->dccpd_seq);
1094
1095 if (dccp_packet_without_ack(skb)) {
1096 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
1097 dccp_pr_debug_cat("\n");
1098 } else {
1099 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
1100 dccp_pr_debug_cat(", ack=%llu\n", DCCP_SKB_CB(skb)->dccpd_ack_seq);
1101 }
1102
1103 /* Step 2:
1104 * Look up flow ID in table and get corresponding socket */
1105 sk = __inet_lookup(&dccp_hashinfo,
1106 skb->nh.iph->saddr, dh->dccph_sport,
1107 skb->nh.iph->daddr, ntohs(dh->dccph_dport),
1108 inet_iif(skb));
1109
1110 /*
1111 * Step 2:
1112 * If no socket ...
1113 * Generate Reset(No Connection) unless P.type == Reset
1114 * Drop packet and return
1115 */
1116 if (sk == NULL) {
1117 dccp_pr_debug("failed to look up flow ID in table and "
1118 "get corresponding socket\n");
1119 goto no_dccp_socket;
1120 }
1121
1122 /*
1123 * Step 2:
1124 * ... or S.state == TIMEWAIT,
1125 * Generate Reset(No Connection) unless P.type == Reset
1126 * Drop packet and return
1127 */
1128
1129 if (sk->sk_state == DCCP_TIME_WAIT) {
1130 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: discard_and_relse\n");
1131 goto discard_and_relse;
1132 }
1133
1134 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
1135 dccp_pr_debug("xfrm4_policy_check failed\n");
1136 goto discard_and_relse;
1137 }
1138
1139 if (sk_filter(sk, skb, 0)) {
1140 dccp_pr_debug("sk_filter failed\n");
1141 goto discard_and_relse;
1142 }
1143
1144 skb->dev = NULL;
1145
1146 bh_lock_sock(sk);
1147 rc = 0;
1148 if (!sock_owned_by_user(sk))
1149 rc = dccp_v4_do_rcv(sk, skb);
1150 else
1151 sk_add_backlog(sk, skb);
1152 bh_unlock_sock(sk);
1153
1154 sock_put(sk);
1155 return rc;
1156
1157no_dccp_socket:
1158 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1159 goto discard_it;
1160 /*
1161 * Step 2:
1162 * Generate Reset(No Connection) unless P.type == Reset
1163 * Drop packet and return
1164 */
1165 if (dh->dccph_type != DCCP_PKT_RESET) {
1166 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_NO_CONNECTION;
1167 dccp_v4_ctl_send_reset(skb);
1168 }
1169
1170discard_it:
1171 /* Discard frame. */
1172 kfree_skb(skb);
1173 return 0;
1174
1175discard_and_relse:
1176 sock_put(sk);
1177 goto discard_it;
1178}
1179
1180static int dccp_v4_init_sock(struct sock *sk)
1181{
1182 struct dccp_sock *dp = dccp_sk(sk);
1183 static int dccp_ctl_socket_init = 1;
1184
1185 dccp_options_init(&dp->dccps_options);
1186
1187 if (dp->dccps_options.dccpo_send_ack_vector) {
1188 dp->dccps_hc_rx_ackpkts = dccp_ackpkts_alloc(DCCP_MAX_ACK_VECTOR_LEN,
1189 GFP_KERNEL);
1190
1191 if (dp->dccps_hc_rx_ackpkts == NULL)
1192 return -ENOMEM;
1193 }
1194
1195 /*
1196 * FIXME: We're hardcoding the CCID, and doing this at this point makes
1197 * the listening (master) sock get CCID control blocks, which is not
1198 * necessary, but for now, to not mess with the test userspace apps,
1199 * lets leave it here, later the real solution is to do this in a
1200 * setsockopt(CCIDs-I-want/accept). -acme
1201 */
1202 if (likely(!dccp_ctl_socket_init)) {
1203 dp->dccps_hc_rx_ccid = ccid_init(dp->dccps_options.dccpo_ccid, sk);
1204 dp->dccps_hc_tx_ccid = ccid_init(dp->dccps_options.dccpo_ccid, sk);
1205 if (dp->dccps_hc_rx_ccid == NULL ||
1206 dp->dccps_hc_tx_ccid == NULL) {
1207 ccid_exit(dp->dccps_hc_rx_ccid, sk);
1208 ccid_exit(dp->dccps_hc_tx_ccid, sk);
1209 dccp_ackpkts_free(dp->dccps_hc_rx_ackpkts);
1210 dp->dccps_hc_rx_ackpkts = NULL;
1211 dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1212 return -ENOMEM;
1213 }
1214 } else
1215 dccp_ctl_socket_init = 0;
1216
1217 dccp_init_xmit_timers(sk);
1218 sk->sk_state = DCCP_CLOSED;
1219 dp->dccps_mss_cache = 536;
1220 dp->dccps_role = DCCP_ROLE_UNDEFINED;
1221
1222 return 0;
1223}
1224
1225int dccp_v4_destroy_sock(struct sock *sk)
1226{
1227 struct dccp_sock *dp = dccp_sk(sk);
1228
1229 /*
1230 * DCCP doesn't use sk_qrite_queue, just sk_send_head
1231 * for retransmissions
1232 */
1233 if (sk->sk_send_head != NULL) {
1234 kfree_skb(sk->sk_send_head);
1235 sk->sk_send_head = NULL;
1236 }
1237
1238 /* Clean up a referenced DCCP bind bucket. */
1239 if (inet_csk(sk)->icsk_bind_hash != NULL)
1240 inet_put_port(&dccp_hashinfo, sk);
1241
1242 dccp_ackpkts_free(dp->dccps_hc_rx_ackpkts);
1243 dp->dccps_hc_rx_ackpkts = NULL;
1244 ccid_exit(dp->dccps_hc_rx_ccid, sk);
1245 ccid_exit(dp->dccps_hc_tx_ccid, sk);
1246 dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1247
1248 return 0;
1249}
1250
1251static void dccp_v4_reqsk_destructor(struct request_sock *req)
1252{
1253 kfree(inet_rsk(req)->opt);
1254}
1255
1256static struct request_sock_ops dccp_request_sock_ops = {
1257 .family = PF_INET,
1258 .obj_size = sizeof(struct dccp_request_sock),
1259 .rtx_syn_ack = dccp_v4_send_response,
1260 .send_ack = dccp_v4_reqsk_send_ack,
1261 .destructor = dccp_v4_reqsk_destructor,
1262 .send_reset = dccp_v4_ctl_send_reset,
1263};
1264
1265struct proto dccp_v4_prot = {
1266 .name = "DCCP",
1267 .owner = THIS_MODULE,
1268 .close = dccp_close,
1269 .connect = dccp_v4_connect,
1270 .disconnect = dccp_disconnect,
1271 .ioctl = dccp_ioctl,
1272 .init = dccp_v4_init_sock,
1273 .setsockopt = dccp_setsockopt,
1274 .getsockopt = dccp_getsockopt,
1275 .sendmsg = dccp_sendmsg,
1276 .recvmsg = dccp_recvmsg,
1277 .backlog_rcv = dccp_v4_do_rcv,
1278 .hash = dccp_v4_hash,
1279 .unhash = dccp_v4_unhash,
1280 .accept = inet_csk_accept,
1281 .get_port = dccp_v4_get_port,
1282 .shutdown = dccp_shutdown,
1283 .destroy = dccp_v4_destroy_sock,
1284 .orphan_count = &dccp_orphan_count,
1285 .max_header = MAX_DCCP_HEADER,
1286 .obj_size = sizeof(struct dccp_sock),
1287 .rsk_prot = &dccp_request_sock_ops,
1288 .twsk_obj_size = sizeof(struct inet_timewait_sock), /* FIXME! create dccp_timewait_sock */
1289};
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-19 10:17:24 -0500