xref: /openbmc/linux/net/dccp/minisocks.c (revision 87c2ce3b)
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/config.h>
14 #include <linux/dccp.h>
15 #include <linux/skbuff.h>
16 #include <linux/timer.h>
17 
18 #include <net/sock.h>
19 #include <net/xfrm.h>
20 #include <net/inet_timewait_sock.h>
21 
22 #include "ackvec.h"
23 #include "ccid.h"
24 #include "dccp.h"
25 
26 struct inet_timewait_death_row dccp_death_row = {
27 	.sysctl_max_tw_buckets = NR_FILE * 2,
28 	.period		= DCCP_TIMEWAIT_LEN / INET_TWDR_TWKILL_SLOTS,
29 	.death_lock	= SPIN_LOCK_UNLOCKED,
30 	.hashinfo	= &dccp_hashinfo,
31 	.tw_timer	= TIMER_INITIALIZER(inet_twdr_hangman, 0,
32 					    (unsigned long)&dccp_death_row),
33 	.twkill_work	= __WORK_INITIALIZER(dccp_death_row.twkill_work,
34 					     inet_twdr_twkill_work,
35 					     &dccp_death_row),
36 /* Short-time timewait calendar */
37 
38 	.twcal_hand	= -1,
39 	.twcal_timer	= TIMER_INITIALIZER(inet_twdr_twcal_tick, 0,
40 					    (unsigned long)&dccp_death_row),
41 };
42 
43 EXPORT_SYMBOL_GPL(dccp_death_row);
44 
45 void dccp_time_wait(struct sock *sk, int state, int timeo)
46 {
47 	struct inet_timewait_sock *tw = NULL;
48 
49 	if (dccp_death_row.tw_count < dccp_death_row.sysctl_max_tw_buckets)
50 		tw = inet_twsk_alloc(sk, state);
51 
52 	if (tw != NULL) {
53 		const struct inet_connection_sock *icsk = inet_csk(sk);
54 		const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
55 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
56 		if (tw->tw_family == PF_INET6) {
57 			const struct ipv6_pinfo *np = inet6_sk(sk);
58 			struct inet6_timewait_sock *tw6;
59 
60 			tw->tw_ipv6_offset = inet6_tw_offset(sk->sk_prot);
61 			tw6 = inet6_twsk((struct sock *)tw);
62 			ipv6_addr_copy(&tw6->tw_v6_daddr, &np->daddr);
63 			ipv6_addr_copy(&tw6->tw_v6_rcv_saddr, &np->rcv_saddr);
64 			tw->tw_ipv6only = np->ipv6only;
65 		}
66 #endif
67 		/* Linkage updates. */
68 		__inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
69 
70 		/* Get the TIME_WAIT timeout firing. */
71 		if (timeo < rto)
72 			timeo = rto;
73 
74 		tw->tw_timeout = DCCP_TIMEWAIT_LEN;
75 		if (state == DCCP_TIME_WAIT)
76 			timeo = DCCP_TIMEWAIT_LEN;
77 
78 		inet_twsk_schedule(tw, &dccp_death_row, timeo,
79 				   DCCP_TIMEWAIT_LEN);
80 		inet_twsk_put(tw);
81 	} else {
82 		/* Sorry, if we're out of memory, just CLOSE this
83 		 * socket up.  We've got bigger problems than
84 		 * non-graceful socket closings.
85 		 */
86 		LIMIT_NETDEBUG(KERN_INFO "DCCP: time wait bucket "
87 					 "table overflow\n");
88 	}
89 
90 	dccp_done(sk);
91 }
92 
93 struct 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 		const struct dccp_request_sock *dreq = dccp_rsk(req);
107 		struct inet_connection_sock *newicsk = inet_csk(sk);
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 		newicsk->icsk_rto	   = DCCP_TIMEOUT_INIT;
115 		do_gettimeofday(&newdp->dccps_epoch);
116 
117 		if (newdp->dccps_options.dccpo_send_ack_vector) {
118 			newdp->dccps_hc_rx_ackvec =
119 				dccp_ackvec_alloc(DCCP_MAX_ACKVEC_LEN,
120 						  GFP_ATOMIC);
121 			/*
122 			 * XXX: We're using the same CCIDs set on the parent,
123 			 * i.e. sk_clone copied the master sock and left the
124 			 * CCID pointers for this child, that is why we do the
125 			 * __ccid_get calls.
126 			 */
127 			if (unlikely(newdp->dccps_hc_rx_ackvec == NULL))
128 				goto out_free;
129 		}
130 
131 		if (unlikely(ccid_hc_rx_init(newdp->dccps_hc_rx_ccid,
132 					     newsk) != 0 ||
133 			     ccid_hc_tx_init(newdp->dccps_hc_tx_ccid,
134 				     	     newsk) != 0)) {
135 			dccp_ackvec_free(newdp->dccps_hc_rx_ackvec);
136 			ccid_hc_rx_exit(newdp->dccps_hc_rx_ccid, newsk);
137 			ccid_hc_tx_exit(newdp->dccps_hc_tx_ccid, newsk);
138 out_free:
139 			/* It is still raw copy of parent, so invalidate
140 			 * destructor and make plain sk_free() */
141 			newsk->sk_destruct = NULL;
142 			sk_free(newsk);
143 			return NULL;
144 		}
145 
146 		__ccid_get(newdp->dccps_hc_rx_ccid);
147 		__ccid_get(newdp->dccps_hc_tx_ccid);
148 
149 		/*
150 		 * Step 3: Process LISTEN state
151 		 *
152 		 *	Choose S.ISS (initial seqno) or set from Init Cookie
153 		 *	Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init
154 		 *	Cookie
155 		 */
156 
157 		/* See dccp_v4_conn_request */
158 		newdp->dccps_options.dccpo_sequence_window = req->rcv_wnd;
159 
160 		newdp->dccps_gar = newdp->dccps_isr = dreq->dreq_isr;
161 		dccp_update_gsr(newsk, dreq->dreq_isr);
162 
163 		newdp->dccps_iss = dreq->dreq_iss;
164 		dccp_update_gss(newsk, dreq->dreq_iss);
165 
166 		/*
167 		 * SWL and AWL are initially adjusted so that they are not less than
168 		 * the initial Sequence Numbers received and sent, respectively:
169 		 *	SWL := max(GSR + 1 - floor(W/4), ISR),
170 		 *	AWL := max(GSS - W' + 1, ISS).
171 		 * These adjustments MUST be applied only at the beginning of the
172 		 * connection.
173 		 */
174 		dccp_set_seqno(&newdp->dccps_swl,
175 			       max48(newdp->dccps_swl, newdp->dccps_isr));
176 		dccp_set_seqno(&newdp->dccps_awl,
177 			       max48(newdp->dccps_awl, newdp->dccps_iss));
178 
179 		dccp_init_xmit_timers(newsk);
180 
181 		DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS);
182 	}
183 	return newsk;
184 }
185 
186 EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
187 
188 /*
189  * Process an incoming packet for RESPOND sockets represented
190  * as an request_sock.
191  */
192 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
193 			    struct request_sock *req,
194 			    struct request_sock **prev)
195 {
196 	struct sock *child = NULL;
197 
198 	/* Check for retransmitted REQUEST */
199 	if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
200 		if (after48(DCCP_SKB_CB(skb)->dccpd_seq,
201 			    dccp_rsk(req)->dreq_isr)) {
202 			struct dccp_request_sock *dreq = dccp_rsk(req);
203 
204 			dccp_pr_debug("Retransmitted REQUEST\n");
205 			/* Send another RESPONSE packet */
206 			dccp_set_seqno(&dreq->dreq_iss, dreq->dreq_iss + 1);
207 			dccp_set_seqno(&dreq->dreq_isr,
208 				       DCCP_SKB_CB(skb)->dccpd_seq);
209 			req->rsk_ops->rtx_syn_ack(sk, req, NULL);
210 		}
211 		/* Network Duplicate, discard packet */
212 		return NULL;
213 	}
214 
215 	DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
216 
217 	if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
218 	    dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
219 		goto drop;
220 
221 	/* Invalid ACK */
222 	if (DCCP_SKB_CB(skb)->dccpd_ack_seq != dccp_rsk(req)->dreq_iss) {
223 		dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
224 			      "dreq_iss=%llu\n",
225 			      (unsigned long long)
226 			      DCCP_SKB_CB(skb)->dccpd_ack_seq,
227 			      (unsigned long long)
228 			      dccp_rsk(req)->dreq_iss);
229 		goto drop;
230 	}
231 
232 	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
233 	if (child == NULL)
234 		goto listen_overflow;
235 
236 	/* FIXME: deal with options */
237 
238 	inet_csk_reqsk_queue_unlink(sk, req, prev);
239 	inet_csk_reqsk_queue_removed(sk, req);
240 	inet_csk_reqsk_queue_add(sk, req, child);
241 out:
242 	return child;
243 listen_overflow:
244 	dccp_pr_debug("listen_overflow!\n");
245 	DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
246 drop:
247 	if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
248 		req->rsk_ops->send_reset(skb);
249 
250 	inet_csk_reqsk_queue_drop(sk, req, prev);
251 	goto out;
252 }
253 
254 EXPORT_SYMBOL_GPL(dccp_check_req);
255 
256 /*
257  *  Queue segment on the new socket if the new socket is active,
258  *  otherwise we just shortcircuit this and continue with
259  *  the new socket.
260  */
261 int dccp_child_process(struct sock *parent, struct sock *child,
262 		       struct sk_buff *skb)
263 {
264 	int ret = 0;
265 	const int state = child->sk_state;
266 
267 	if (!sock_owned_by_user(child)) {
268 		ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
269 					     skb->len);
270 
271 		/* Wakeup parent, send SIGIO */
272 		if (state == DCCP_RESPOND && child->sk_state != state)
273 			parent->sk_data_ready(parent, 0);
274 	} else {
275 		/* Alas, it is possible again, because we do lookup
276 		 * in main socket hash table and lock on listening
277 		 * socket does not protect us more.
278 		 */
279 		sk_add_backlog(child, skb);
280 	}
281 
282 	bh_unlock_sock(child);
283 	sock_put(child);
284 	return ret;
285 }
286 
287 EXPORT_SYMBOL_GPL(dccp_child_process);
288