1/*
2 *  net/dccp/minisocks.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/dccp.h>
14#include <linux/gfp.h>
15#include <linux/kernel.h>
16#include <linux/skbuff.h>
17#include <linux/timer.h>
18
19#include <net/sock.h>
20#include <net/xfrm.h>
21#include <net/inet_timewait_sock.h>
22
23#include "ackvec.h"
24#include "ccid.h"
25#include "dccp.h"
26#include "feat.h"
27
28struct inet_timewait_death_row dccp_death_row = {
29	.sysctl_max_tw_buckets = NR_FILE * 2,
30	.period		= DCCP_TIMEWAIT_LEN / INET_TWDR_TWKILL_SLOTS,
31	.death_lock	= __SPIN_LOCK_UNLOCKED(dccp_death_row.death_lock),
32	.hashinfo	= &dccp_hashinfo,
33	.tw_timer	= TIMER_INITIALIZER(inet_twdr_hangman, 0,
34					    (unsigned long)&dccp_death_row),
35	.twkill_work	= __WORK_INITIALIZER(dccp_death_row.twkill_work,
36					     inet_twdr_twkill_work),
37/* Short-time timewait calendar */
38
39	.twcal_hand	= -1,
40	.twcal_timer	= TIMER_INITIALIZER(inet_twdr_twcal_tick, 0,
41					    (unsigned long)&dccp_death_row),
42};
43
44EXPORT_SYMBOL_GPL(dccp_death_row);
45
46void dccp_time_wait(struct sock *sk, int state, int timeo)
47{
48	struct inet_timewait_sock *tw = NULL;
49
50	if (dccp_death_row.tw_count < dccp_death_row.sysctl_max_tw_buckets)
51		tw = inet_twsk_alloc(sk, state);
52
53	if (tw != NULL) {
54		const struct inet_connection_sock *icsk = inet_csk(sk);
55		const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
56#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
57		if (tw->tw_family == PF_INET6) {
58			const struct ipv6_pinfo *np = inet6_sk(sk);
59			struct inet6_timewait_sock *tw6;
60
61			tw->tw_ipv6_offset = inet6_tw_offset(sk->sk_prot);
62			tw6 = inet6_twsk((struct sock *)tw);
63			ipv6_addr_copy(&tw6->tw_v6_daddr, &np->daddr);
64			ipv6_addr_copy(&tw6->tw_v6_rcv_saddr, &np->rcv_saddr);
65			tw->tw_ipv6only = np->ipv6only;
66		}
67#endif
68		/* Linkage updates. */
69		__inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
70
71		/* Get the TIME_WAIT timeout firing. */
72		if (timeo < rto)
73			timeo = rto;
74
75		tw->tw_timeout = DCCP_TIMEWAIT_LEN;
76		if (state == DCCP_TIME_WAIT)
77			timeo = DCCP_TIMEWAIT_LEN;
78
79		inet_twsk_schedule(tw, &dccp_death_row, timeo,
80				   DCCP_TIMEWAIT_LEN);
81		inet_twsk_put(tw);
82	} else {
83		/* Sorry, if we're out of memory, just CLOSE this
84		 * socket up.  We've got bigger problems than
85		 * non-graceful socket closings.
86		 */
87		DCCP_WARN("time wait bucket table overflow\n");
88	}
89
90	dccp_done(sk);
91}
92
93struct sock *dccp_create_openreq_child(struct sock *sk,
94				       const struct request_sock *req,
95				       const struct sk_buff *skb)
96{
97	/*
98	 * Step 3: Process LISTEN state
99	 *
100	 *   (* Generate a new socket and switch to that socket *)
101	 *   Set S := new socket for this port pair
102	 */
103	struct sock *newsk = inet_csk_clone(sk, req, GFP_ATOMIC);
104
105	if (newsk != NULL) {
106		struct dccp_request_sock *dreq = dccp_rsk(req);
107		struct inet_connection_sock *newicsk = inet_csk(newsk);
108		struct dccp_sock *newdp = dccp_sk(newsk);
109
110		newdp->dccps_role	    = DCCP_ROLE_SERVER;
111		newdp->dccps_hc_rx_ackvec   = NULL;
112		newdp->dccps_service_list   = NULL;
113		newdp->dccps_service	    = dreq->dreq_service;
114		newdp->dccps_timestamp_echo = dreq->dreq_timestamp_echo;
115		newdp->dccps_timestamp_time = dreq->dreq_timestamp_time;
116		newicsk->icsk_rto	    = DCCP_TIMEOUT_INIT;
117
118		INIT_LIST_HEAD(&newdp->dccps_featneg);
119		/*
120		 * Step 3: Process LISTEN state
121		 *
122		 *    Choose S.ISS (initial seqno) or set from Init Cookies
123		 *    Initialize S.GAR := S.ISS
124		 *    Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
125		 */
126		newdp->dccps_gar = newdp->dccps_iss = dreq->dreq_iss;
127		dccp_update_gss(newsk, dreq->dreq_iss);
128
129		newdp->dccps_isr = dreq->dreq_isr;
130		dccp_update_gsr(newsk, dreq->dreq_isr);
131
132		/*
133		 * SWL and AWL are initially adjusted so that they are not less than
134		 * the initial Sequence Numbers received and sent, respectively:
135		 *	SWL := max(GSR + 1 - floor(W/4), ISR),
136		 *	AWL := max(GSS - W' + 1, ISS).
137		 * These adjustments MUST be applied only at the beginning of the
138		 * connection.
139		 */
140		dccp_set_seqno(&newdp->dccps_swl,
141			       max48(newdp->dccps_swl, newdp->dccps_isr));
142		dccp_set_seqno(&newdp->dccps_awl,
143			       max48(newdp->dccps_awl, newdp->dccps_iss));
144
145		/*
146		 * Activate features after initialising the sequence numbers,
147		 * since CCID initialisation may depend on GSS, ISR, ISS etc.
148		 */
149		if (dccp_feat_activate_values(newsk, &dreq->dreq_featneg)) {
150			/* It is still raw copy of parent, so invalidate
151			 * destructor and make plain sk_free() */
152			newsk->sk_destruct = NULL;
153			sk_free(newsk);
154			return NULL;
155		}
156		dccp_init_xmit_timers(newsk);
157
158		DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS);
159	}
160	return newsk;
161}
162
163EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
164
165/*
166 * Process an incoming packet for RESPOND sockets represented
167 * as an request_sock.
168 */
169struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
170			    struct request_sock *req,
171			    struct request_sock **prev)
172{
173	struct sock *child = NULL;
174	struct dccp_request_sock *dreq = dccp_rsk(req);
175
176	/* Check for retransmitted REQUEST */
177	if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
178
179		if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_isr)) {
180			dccp_pr_debug("Retransmitted REQUEST\n");
181			dreq->dreq_isr = DCCP_SKB_CB(skb)->dccpd_seq;
182			/*
183			 * Send another RESPONSE packet
184			 * To protect against Request floods, increment retrans
185			 * counter (backoff, monitored by dccp_response_timer).
186			 */
187			req->retrans++;
188			req->rsk_ops->rtx_syn_ack(sk, req, NULL);
189		}
190		/* Network Duplicate, discard packet */
191		return NULL;
192	}
193
194	DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
195
196	if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
197	    dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
198		goto drop;
199
200	/* Invalid ACK */
201	if (DCCP_SKB_CB(skb)->dccpd_ack_seq != dreq->dreq_iss) {
202		dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
203			      "dreq_iss=%llu\n",
204			      (unsigned long long)
205			      DCCP_SKB_CB(skb)->dccpd_ack_seq,
206			      (unsigned long long) dreq->dreq_iss);
207		goto drop;
208	}
209
210	if (dccp_parse_options(sk, dreq, skb))
211		 goto drop;
212
213	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
214	if (child == NULL)
215		goto listen_overflow;
216
217	inet_csk_reqsk_queue_unlink(sk, req, prev);
218	inet_csk_reqsk_queue_removed(sk, req);
219	inet_csk_reqsk_queue_add(sk, req, child);
220out:
221	return child;
222listen_overflow:
223	dccp_pr_debug("listen_overflow!\n");
224	DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
225drop:
226	if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
227		req->rsk_ops->send_reset(sk, skb);
228
229	inet_csk_reqsk_queue_drop(sk, req, prev);
230	goto out;
231}
232
233EXPORT_SYMBOL_GPL(dccp_check_req);
234
235/*
236 *  Queue segment on the new socket if the new socket is active,
237 *  otherwise we just shortcircuit this and continue with
238 *  the new socket.
239 */
240int dccp_child_process(struct sock *parent, struct sock *child,
241		       struct sk_buff *skb)
242{
243	int ret = 0;
244	const int state = child->sk_state;
245
246	if (!sock_owned_by_user(child)) {
247		ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
248					     skb->len);
249
250		/* Wakeup parent, send SIGIO */
251		if (state == DCCP_RESPOND && child->sk_state != state)
252			parent->sk_data_ready(parent, 0);
253	} else {
254		/* Alas, it is possible again, because we do lookup
255		 * in main socket hash table and lock on listening
256		 * socket does not protect us more.
257		 */
258		__sk_add_backlog(child, skb);
259	}
260
261	bh_unlock_sock(child);
262	sock_put(child);
263	return ret;
264}
265
266EXPORT_SYMBOL_GPL(dccp_child_process);
267
268void dccp_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
269			 struct request_sock *rsk)
270{
271	DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state");
272}
273
274EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack);
275
276int dccp_reqsk_init(struct request_sock *req,
277		    struct dccp_sock const *dp, struct sk_buff const *skb)
278{
279	struct dccp_request_sock *dreq = dccp_rsk(req);
280
281	inet_rsk(req)->rmt_port	  = dccp_hdr(skb)->dccph_sport;
282	inet_rsk(req)->loc_port	  = dccp_hdr(skb)->dccph_dport;
283	inet_rsk(req)->acked	  = 0;
284	dreq->dreq_timestamp_echo = 0;
285
286	/* inherit feature negotiation options from listening socket */
287	return dccp_feat_clone_list(&dp->dccps_featneg, &dreq->dreq_featneg);
288}
289
290EXPORT_SYMBOL_GPL(dccp_reqsk_init);
291