xref: /openbmc/linux/net/dccp/minisocks.c (revision 12eb4683)
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 	.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 
44 EXPORT_SYMBOL_GPL(dccp_death_row);
45 
46 void 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 IS_ENABLED(CONFIG_IPV6)
57 		if (tw->tw_family == PF_INET6) {
58 			const struct ipv6_pinfo *np = inet6_sk(sk);
59 
60 			tw->tw_v6_daddr = sk->sk_v6_daddr;
61 			tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
62 			tw->tw_ipv6only = np->ipv6only;
63 		}
64 #endif
65 		/* Linkage updates. */
66 		__inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
67 
68 		/* Get the TIME_WAIT timeout firing. */
69 		if (timeo < rto)
70 			timeo = rto;
71 
72 		tw->tw_timeout = DCCP_TIMEWAIT_LEN;
73 		if (state == DCCP_TIME_WAIT)
74 			timeo = DCCP_TIMEWAIT_LEN;
75 
76 		inet_twsk_schedule(tw, &dccp_death_row, timeo,
77 				   DCCP_TIMEWAIT_LEN);
78 		inet_twsk_put(tw);
79 	} else {
80 		/* Sorry, if we're out of memory, just CLOSE this
81 		 * socket up.  We've got bigger problems than
82 		 * non-graceful socket closings.
83 		 */
84 		DCCP_WARN("time wait bucket table overflow\n");
85 	}
86 
87 	dccp_done(sk);
88 }
89 
90 struct sock *dccp_create_openreq_child(struct sock *sk,
91 				       const struct request_sock *req,
92 				       const struct sk_buff *skb)
93 {
94 	/*
95 	 * Step 3: Process LISTEN state
96 	 *
97 	 *   (* Generate a new socket and switch to that socket *)
98 	 *   Set S := new socket for this port pair
99 	 */
100 	struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
101 
102 	if (newsk != NULL) {
103 		struct dccp_request_sock *dreq = dccp_rsk(req);
104 		struct inet_connection_sock *newicsk = inet_csk(newsk);
105 		struct dccp_sock *newdp = dccp_sk(newsk);
106 
107 		newdp->dccps_role	    = DCCP_ROLE_SERVER;
108 		newdp->dccps_hc_rx_ackvec   = NULL;
109 		newdp->dccps_service_list   = NULL;
110 		newdp->dccps_service	    = dreq->dreq_service;
111 		newdp->dccps_timestamp_echo = dreq->dreq_timestamp_echo;
112 		newdp->dccps_timestamp_time = dreq->dreq_timestamp_time;
113 		newicsk->icsk_rto	    = DCCP_TIMEOUT_INIT;
114 
115 		INIT_LIST_HEAD(&newdp->dccps_featneg);
116 		/*
117 		 * Step 3: Process LISTEN state
118 		 *
119 		 *    Choose S.ISS (initial seqno) or set from Init Cookies
120 		 *    Initialize S.GAR := S.ISS
121 		 *    Set S.ISR, S.GSR from packet (or Init Cookies)
122 		 *
123 		 *    Setting AWL/AWH and SWL/SWH happens as part of the feature
124 		 *    activation below, as these windows all depend on the local
125 		 *    and remote Sequence Window feature values (7.5.2).
126 		 */
127 		newdp->dccps_iss = dreq->dreq_iss;
128 		newdp->dccps_gss = dreq->dreq_gss;
129 		newdp->dccps_gar = newdp->dccps_iss;
130 		newdp->dccps_isr = dreq->dreq_isr;
131 		newdp->dccps_gsr = dreq->dreq_gsr;
132 
133 		/*
134 		 * Activate features: initialise CCIDs, sequence windows etc.
135 		 */
136 		if (dccp_feat_activate_values(newsk, &dreq->dreq_featneg)) {
137 			/* It is still raw copy of parent, so invalidate
138 			 * destructor and make plain sk_free() */
139 			newsk->sk_destruct = NULL;
140 			sk_free(newsk);
141 			return NULL;
142 		}
143 		dccp_init_xmit_timers(newsk);
144 
145 		DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS);
146 	}
147 	return newsk;
148 }
149 
150 EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
151 
152 /*
153  * Process an incoming packet for RESPOND sockets represented
154  * as an request_sock.
155  */
156 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
157 			    struct request_sock *req,
158 			    struct request_sock **prev)
159 {
160 	struct sock *child = NULL;
161 	struct dccp_request_sock *dreq = dccp_rsk(req);
162 
163 	/* Check for retransmitted REQUEST */
164 	if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
165 
166 		if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_gsr)) {
167 			dccp_pr_debug("Retransmitted REQUEST\n");
168 			dreq->dreq_gsr = DCCP_SKB_CB(skb)->dccpd_seq;
169 			/*
170 			 * Send another RESPONSE packet
171 			 * To protect against Request floods, increment retrans
172 			 * counter (backoff, monitored by dccp_response_timer).
173 			 */
174 			inet_rtx_syn_ack(sk, req);
175 		}
176 		/* Network Duplicate, discard packet */
177 		return NULL;
178 	}
179 
180 	DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
181 
182 	if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
183 	    dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
184 		goto drop;
185 
186 	/* Invalid ACK */
187 	if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
188 				dreq->dreq_iss, dreq->dreq_gss)) {
189 		dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
190 			      "dreq_iss=%llu, dreq_gss=%llu\n",
191 			      (unsigned long long)
192 			      DCCP_SKB_CB(skb)->dccpd_ack_seq,
193 			      (unsigned long long) dreq->dreq_iss,
194 			      (unsigned long long) dreq->dreq_gss);
195 		goto drop;
196 	}
197 
198 	if (dccp_parse_options(sk, dreq, skb))
199 		 goto drop;
200 
201 	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
202 	if (child == NULL)
203 		goto listen_overflow;
204 
205 	inet_csk_reqsk_queue_unlink(sk, req, prev);
206 	inet_csk_reqsk_queue_removed(sk, req);
207 	inet_csk_reqsk_queue_add(sk, req, child);
208 out:
209 	return child;
210 listen_overflow:
211 	dccp_pr_debug("listen_overflow!\n");
212 	DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
213 drop:
214 	if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
215 		req->rsk_ops->send_reset(sk, skb);
216 
217 	inet_csk_reqsk_queue_drop(sk, req, prev);
218 	goto out;
219 }
220 
221 EXPORT_SYMBOL_GPL(dccp_check_req);
222 
223 /*
224  *  Queue segment on the new socket if the new socket is active,
225  *  otherwise we just shortcircuit this and continue with
226  *  the new socket.
227  */
228 int dccp_child_process(struct sock *parent, struct sock *child,
229 		       struct sk_buff *skb)
230 {
231 	int ret = 0;
232 	const int state = child->sk_state;
233 
234 	if (!sock_owned_by_user(child)) {
235 		ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
236 					     skb->len);
237 
238 		/* Wakeup parent, send SIGIO */
239 		if (state == DCCP_RESPOND && child->sk_state != state)
240 			parent->sk_data_ready(parent, 0);
241 	} else {
242 		/* Alas, it is possible again, because we do lookup
243 		 * in main socket hash table and lock on listening
244 		 * socket does not protect us more.
245 		 */
246 		__sk_add_backlog(child, skb);
247 	}
248 
249 	bh_unlock_sock(child);
250 	sock_put(child);
251 	return ret;
252 }
253 
254 EXPORT_SYMBOL_GPL(dccp_child_process);
255 
256 void dccp_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
257 			 struct request_sock *rsk)
258 {
259 	DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state");
260 }
261 
262 EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack);
263 
264 int dccp_reqsk_init(struct request_sock *req,
265 		    struct dccp_sock const *dp, struct sk_buff const *skb)
266 {
267 	struct dccp_request_sock *dreq = dccp_rsk(req);
268 
269 	inet_rsk(req)->ir_rmt_port = dccp_hdr(skb)->dccph_sport;
270 	inet_rsk(req)->ir_num	   = ntohs(dccp_hdr(skb)->dccph_dport);
271 	inet_rsk(req)->acked	   = 0;
272 	dreq->dreq_timestamp_echo  = 0;
273 
274 	/* inherit feature negotiation options from listening socket */
275 	return dccp_feat_clone_list(&dp->dccps_featneg, &dreq->dreq_featneg);
276 }
277 
278 EXPORT_SYMBOL_GPL(dccp_reqsk_init);
279