xref: /linux/net/dccp/minisocks.c (revision 74ce1896c6c65b2f8cccbf59162d542988835835)
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 
28 struct inet_timewait_death_row dccp_death_row = {
29 	.sysctl_max_tw_buckets = NR_FILE * 2,
30 	.hashinfo	= &dccp_hashinfo,
31 };
32 
33 EXPORT_SYMBOL_GPL(dccp_death_row);
34 
35 void dccp_time_wait(struct sock *sk, int state, int timeo)
36 {
37 	struct inet_timewait_sock *tw;
38 
39 	tw = inet_twsk_alloc(sk, &dccp_death_row, state);
40 
41 	if (tw != NULL) {
42 		const struct inet_connection_sock *icsk = inet_csk(sk);
43 		const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
44 #if IS_ENABLED(CONFIG_IPV6)
45 		if (tw->tw_family == PF_INET6) {
46 			tw->tw_v6_daddr = sk->sk_v6_daddr;
47 			tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
48 			tw->tw_ipv6only = sk->sk_ipv6only;
49 		}
50 #endif
51 
52 		/* Get the TIME_WAIT timeout firing. */
53 		if (timeo < rto)
54 			timeo = rto;
55 
56 		tw->tw_timeout = DCCP_TIMEWAIT_LEN;
57 		if (state == DCCP_TIME_WAIT)
58 			timeo = DCCP_TIMEWAIT_LEN;
59 
60 		inet_twsk_schedule(tw, timeo);
61 		/* Linkage updates. */
62 		__inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
63 		inet_twsk_put(tw);
64 	} else {
65 		/* Sorry, if we're out of memory, just CLOSE this
66 		 * socket up.  We've got bigger problems than
67 		 * non-graceful socket closings.
68 		 */
69 		DCCP_WARN("time wait bucket table overflow\n");
70 	}
71 
72 	dccp_done(sk);
73 }
74 
75 struct sock *dccp_create_openreq_child(const struct sock *sk,
76 				       const struct request_sock *req,
77 				       const struct sk_buff *skb)
78 {
79 	/*
80 	 * Step 3: Process LISTEN state
81 	 *
82 	 *   (* Generate a new socket and switch to that socket *)
83 	 *   Set S := new socket for this port pair
84 	 */
85 	struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
86 
87 	if (newsk != NULL) {
88 		struct dccp_request_sock *dreq = dccp_rsk(req);
89 		struct inet_connection_sock *newicsk = inet_csk(newsk);
90 		struct dccp_sock *newdp = dccp_sk(newsk);
91 
92 		newdp->dccps_role	    = DCCP_ROLE_SERVER;
93 		newdp->dccps_hc_rx_ackvec   = NULL;
94 		newdp->dccps_service_list   = NULL;
95 		newdp->dccps_service	    = dreq->dreq_service;
96 		newdp->dccps_timestamp_echo = dreq->dreq_timestamp_echo;
97 		newdp->dccps_timestamp_time = dreq->dreq_timestamp_time;
98 		newicsk->icsk_rto	    = DCCP_TIMEOUT_INIT;
99 
100 		INIT_LIST_HEAD(&newdp->dccps_featneg);
101 		/*
102 		 * Step 3: Process LISTEN state
103 		 *
104 		 *    Choose S.ISS (initial seqno) or set from Init Cookies
105 		 *    Initialize S.GAR := S.ISS
106 		 *    Set S.ISR, S.GSR from packet (or Init Cookies)
107 		 *
108 		 *    Setting AWL/AWH and SWL/SWH happens as part of the feature
109 		 *    activation below, as these windows all depend on the local
110 		 *    and remote Sequence Window feature values (7.5.2).
111 		 */
112 		newdp->dccps_iss = dreq->dreq_iss;
113 		newdp->dccps_gss = dreq->dreq_gss;
114 		newdp->dccps_gar = newdp->dccps_iss;
115 		newdp->dccps_isr = dreq->dreq_isr;
116 		newdp->dccps_gsr = dreq->dreq_gsr;
117 
118 		/*
119 		 * Activate features: initialise CCIDs, sequence windows etc.
120 		 */
121 		if (dccp_feat_activate_values(newsk, &dreq->dreq_featneg)) {
122 			sk_free_unlock_clone(newsk);
123 			return NULL;
124 		}
125 		dccp_init_xmit_timers(newsk);
126 
127 		__DCCP_INC_STATS(DCCP_MIB_PASSIVEOPENS);
128 	}
129 	return newsk;
130 }
131 
132 EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
133 
134 /*
135  * Process an incoming packet for RESPOND sockets represented
136  * as an request_sock.
137  */
138 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
139 			    struct request_sock *req)
140 {
141 	struct sock *child = NULL;
142 	struct dccp_request_sock *dreq = dccp_rsk(req);
143 	bool own_req;
144 
145 	/* TCP/DCCP listeners became lockless.
146 	 * DCCP stores complex state in its request_sock, so we need
147 	 * a protection for them, now this code runs without being protected
148 	 * by the parent (listener) lock.
149 	 */
150 	spin_lock_bh(&dreq->dreq_lock);
151 
152 	/* Check for retransmitted REQUEST */
153 	if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
154 
155 		if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_gsr)) {
156 			dccp_pr_debug("Retransmitted REQUEST\n");
157 			dreq->dreq_gsr = DCCP_SKB_CB(skb)->dccpd_seq;
158 			/*
159 			 * Send another RESPONSE packet
160 			 * To protect against Request floods, increment retrans
161 			 * counter (backoff, monitored by dccp_response_timer).
162 			 */
163 			inet_rtx_syn_ack(sk, req);
164 		}
165 		/* Network Duplicate, discard packet */
166 		goto out;
167 	}
168 
169 	DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
170 
171 	if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
172 	    dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
173 		goto drop;
174 
175 	/* Invalid ACK */
176 	if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
177 				dreq->dreq_iss, dreq->dreq_gss)) {
178 		dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
179 			      "dreq_iss=%llu, dreq_gss=%llu\n",
180 			      (unsigned long long)
181 			      DCCP_SKB_CB(skb)->dccpd_ack_seq,
182 			      (unsigned long long) dreq->dreq_iss,
183 			      (unsigned long long) dreq->dreq_gss);
184 		goto drop;
185 	}
186 
187 	if (dccp_parse_options(sk, dreq, skb))
188 		 goto drop;
189 
190 	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
191 							 req, &own_req);
192 	if (child) {
193 		child = inet_csk_complete_hashdance(sk, child, req, own_req);
194 		goto out;
195 	}
196 
197 	DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
198 drop:
199 	if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
200 		req->rsk_ops->send_reset(sk, skb);
201 
202 	inet_csk_reqsk_queue_drop(sk, req);
203 out:
204 	spin_unlock_bh(&dreq->dreq_lock);
205 	return child;
206 }
207 
208 EXPORT_SYMBOL_GPL(dccp_check_req);
209 
210 /*
211  *  Queue segment on the new socket if the new socket is active,
212  *  otherwise we just shortcircuit this and continue with
213  *  the new socket.
214  */
215 int dccp_child_process(struct sock *parent, struct sock *child,
216 		       struct sk_buff *skb)
217 {
218 	int ret = 0;
219 	const int state = child->sk_state;
220 
221 	if (!sock_owned_by_user(child)) {
222 		ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
223 					     skb->len);
224 
225 		/* Wakeup parent, send SIGIO */
226 		if (state == DCCP_RESPOND && child->sk_state != state)
227 			parent->sk_data_ready(parent);
228 	} else {
229 		/* Alas, it is possible again, because we do lookup
230 		 * in main socket hash table and lock on listening
231 		 * socket does not protect us more.
232 		 */
233 		__sk_add_backlog(child, skb);
234 	}
235 
236 	bh_unlock_sock(child);
237 	sock_put(child);
238 	return ret;
239 }
240 
241 EXPORT_SYMBOL_GPL(dccp_child_process);
242 
243 void dccp_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
244 			 struct request_sock *rsk)
245 {
246 	DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state");
247 }
248 
249 EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack);
250 
251 int dccp_reqsk_init(struct request_sock *req,
252 		    struct dccp_sock const *dp, struct sk_buff const *skb)
253 {
254 	struct dccp_request_sock *dreq = dccp_rsk(req);
255 
256 	spin_lock_init(&dreq->dreq_lock);
257 	inet_rsk(req)->ir_rmt_port = dccp_hdr(skb)->dccph_sport;
258 	inet_rsk(req)->ir_num	   = ntohs(dccp_hdr(skb)->dccph_dport);
259 	inet_rsk(req)->acked	   = 0;
260 	dreq->dreq_timestamp_echo  = 0;
261 
262 	/* inherit feature negotiation options from listening socket */
263 	return dccp_feat_clone_list(&dp->dccps_featneg, &dreq->dreq_featneg);
264 }
265 
266 EXPORT_SYMBOL_GPL(dccp_reqsk_init);
267