xref: /openbmc/linux/net/dccp/minisocks.c (revision 3ac14b39)
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 		if (state == DCCP_TIME_WAIT)
57 			timeo = DCCP_TIMEWAIT_LEN;
58 
59 		/* tw_timer is pinned, so we need to make sure BH are disabled
60 		 * in following section, otherwise timer handler could run before
61 		 * we complete the initialization.
62 		 */
63 		local_bh_disable();
64 		inet_twsk_schedule(tw, timeo);
65 		/* Linkage updates.
66 		 * Note that access to tw after this point is illegal.
67 		 */
68 		inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
69 		local_bh_enable();
70 	} else {
71 		/* Sorry, if we're out of memory, just CLOSE this
72 		 * socket up.  We've got bigger problems than
73 		 * non-graceful socket closings.
74 		 */
75 		DCCP_WARN("time wait bucket table overflow\n");
76 	}
77 
78 	dccp_done(sk);
79 }
80 
81 struct sock *dccp_create_openreq_child(const struct sock *sk,
82 				       const struct request_sock *req,
83 				       const struct sk_buff *skb)
84 {
85 	/*
86 	 * Step 3: Process LISTEN state
87 	 *
88 	 *   (* Generate a new socket and switch to that socket *)
89 	 *   Set S := new socket for this port pair
90 	 */
91 	struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
92 
93 	if (newsk != NULL) {
94 		struct dccp_request_sock *dreq = dccp_rsk(req);
95 		struct inet_connection_sock *newicsk = inet_csk(newsk);
96 		struct dccp_sock *newdp = dccp_sk(newsk);
97 
98 		newdp->dccps_role	    = DCCP_ROLE_SERVER;
99 		newdp->dccps_hc_rx_ackvec   = NULL;
100 		newdp->dccps_service_list   = NULL;
101 		newdp->dccps_service	    = dreq->dreq_service;
102 		newdp->dccps_timestamp_echo = dreq->dreq_timestamp_echo;
103 		newdp->dccps_timestamp_time = dreq->dreq_timestamp_time;
104 		newicsk->icsk_rto	    = DCCP_TIMEOUT_INIT;
105 
106 		INIT_LIST_HEAD(&newdp->dccps_featneg);
107 		/*
108 		 * Step 3: Process LISTEN state
109 		 *
110 		 *    Choose S.ISS (initial seqno) or set from Init Cookies
111 		 *    Initialize S.GAR := S.ISS
112 		 *    Set S.ISR, S.GSR from packet (or Init Cookies)
113 		 *
114 		 *    Setting AWL/AWH and SWL/SWH happens as part of the feature
115 		 *    activation below, as these windows all depend on the local
116 		 *    and remote Sequence Window feature values (7.5.2).
117 		 */
118 		newdp->dccps_iss = dreq->dreq_iss;
119 		newdp->dccps_gss = dreq->dreq_gss;
120 		newdp->dccps_gar = newdp->dccps_iss;
121 		newdp->dccps_isr = dreq->dreq_isr;
122 		newdp->dccps_gsr = dreq->dreq_gsr;
123 
124 		/*
125 		 * Activate features: initialise CCIDs, sequence windows etc.
126 		 */
127 		if (dccp_feat_activate_values(newsk, &dreq->dreq_featneg)) {
128 			sk_free_unlock_clone(newsk);
129 			return NULL;
130 		}
131 		dccp_init_xmit_timers(newsk);
132 
133 		__DCCP_INC_STATS(DCCP_MIB_PASSIVEOPENS);
134 	}
135 	return newsk;
136 }
137 
138 EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
139 
140 /*
141  * Process an incoming packet for RESPOND sockets represented
142  * as an request_sock.
143  */
144 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
145 			    struct request_sock *req)
146 {
147 	struct sock *child = NULL;
148 	struct dccp_request_sock *dreq = dccp_rsk(req);
149 	bool own_req;
150 
151 	/* TCP/DCCP listeners became lockless.
152 	 * DCCP stores complex state in its request_sock, so we need
153 	 * a protection for them, now this code runs without being protected
154 	 * by the parent (listener) lock.
155 	 */
156 	spin_lock_bh(&dreq->dreq_lock);
157 
158 	/* Check for retransmitted REQUEST */
159 	if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
160 
161 		if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_gsr)) {
162 			dccp_pr_debug("Retransmitted REQUEST\n");
163 			dreq->dreq_gsr = DCCP_SKB_CB(skb)->dccpd_seq;
164 			/*
165 			 * Send another RESPONSE packet
166 			 * To protect against Request floods, increment retrans
167 			 * counter (backoff, monitored by dccp_response_timer).
168 			 */
169 			inet_rtx_syn_ack(sk, req);
170 		}
171 		/* Network Duplicate, discard packet */
172 		goto out;
173 	}
174 
175 	DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
176 
177 	if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
178 	    dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
179 		goto drop;
180 
181 	/* Invalid ACK */
182 	if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
183 				dreq->dreq_iss, dreq->dreq_gss)) {
184 		dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
185 			      "dreq_iss=%llu, dreq_gss=%llu\n",
186 			      (unsigned long long)
187 			      DCCP_SKB_CB(skb)->dccpd_ack_seq,
188 			      (unsigned long long) dreq->dreq_iss,
189 			      (unsigned long long) dreq->dreq_gss);
190 		goto drop;
191 	}
192 
193 	if (dccp_parse_options(sk, dreq, skb))
194 		 goto drop;
195 
196 	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
197 							 req, &own_req);
198 	if (child) {
199 		child = inet_csk_complete_hashdance(sk, child, req, own_req);
200 		goto out;
201 	}
202 
203 	DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
204 drop:
205 	if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
206 		req->rsk_ops->send_reset(sk, skb);
207 
208 	inet_csk_reqsk_queue_drop(sk, req);
209 out:
210 	spin_unlock_bh(&dreq->dreq_lock);
211 	return child;
212 }
213 
214 EXPORT_SYMBOL_GPL(dccp_check_req);
215 
216 /*
217  *  Queue segment on the new socket if the new socket is active,
218  *  otherwise we just shortcircuit this and continue with
219  *  the new socket.
220  */
221 int dccp_child_process(struct sock *parent, struct sock *child,
222 		       struct sk_buff *skb)
223 {
224 	int ret = 0;
225 	const int state = child->sk_state;
226 
227 	if (!sock_owned_by_user(child)) {
228 		ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
229 					     skb->len);
230 
231 		/* Wakeup parent, send SIGIO */
232 		if (state == DCCP_RESPOND && child->sk_state != state)
233 			parent->sk_data_ready(parent);
234 	} else {
235 		/* Alas, it is possible again, because we do lookup
236 		 * in main socket hash table and lock on listening
237 		 * socket does not protect us more.
238 		 */
239 		__sk_add_backlog(child, skb);
240 	}
241 
242 	bh_unlock_sock(child);
243 	sock_put(child);
244 	return ret;
245 }
246 
247 EXPORT_SYMBOL_GPL(dccp_child_process);
248 
249 void dccp_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
250 			 struct request_sock *rsk)
251 {
252 	DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state");
253 }
254 
255 EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack);
256 
257 int dccp_reqsk_init(struct request_sock *req,
258 		    struct dccp_sock const *dp, struct sk_buff const *skb)
259 {
260 	struct dccp_request_sock *dreq = dccp_rsk(req);
261 
262 	spin_lock_init(&dreq->dreq_lock);
263 	inet_rsk(req)->ir_rmt_port = dccp_hdr(skb)->dccph_sport;
264 	inet_rsk(req)->ir_num	   = ntohs(dccp_hdr(skb)->dccph_dport);
265 	inet_rsk(req)->acked	   = 0;
266 	dreq->dreq_timestamp_echo  = 0;
267 
268 	/* inherit feature negotiation options from listening socket */
269 	return dccp_feat_clone_list(&dp->dccps_featneg, &dreq->dreq_featneg);
270 }
271 
272 EXPORT_SYMBOL_GPL(dccp_reqsk_init);
273