4 * An implementation of the DCCP protocol
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.
13 #include <linux/dccp.h>
14 #include <linux/skbuff.h>
15 #include <linux/timer.h>
19 #include <net/inet_timewait_sock.h>
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 /* Short-time timewait calendar */
38 .twcal_timer = TIMER_INITIALIZER(inet_twdr_twcal_tick, 0,
39 (unsigned long)&dccp_death_row),
42 EXPORT_SYMBOL_GPL(dccp_death_row);
44 void dccp_time_wait(struct sock *sk, int state, int timeo)
46 struct inet_timewait_sock *tw = NULL;
48 if (dccp_death_row.tw_count < dccp_death_row.sysctl_max_tw_buckets)
49 tw = inet_twsk_alloc(sk, state);
52 const struct inet_connection_sock *icsk = inet_csk(sk);
53 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
54 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
55 if (tw->tw_family == PF_INET6) {
56 const struct ipv6_pinfo *np = inet6_sk(sk);
57 struct inet6_timewait_sock *tw6;
59 tw->tw_ipv6_offset = inet6_tw_offset(sk->sk_prot);
60 tw6 = inet6_twsk((struct sock *)tw);
61 ipv6_addr_copy(&tw6->tw_v6_daddr, &np->daddr);
62 ipv6_addr_copy(&tw6->tw_v6_rcv_saddr, &np->rcv_saddr);
63 tw->tw_ipv6only = np->ipv6only;
66 /* Linkage updates. */
67 __inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
69 /* Get the TIME_WAIT timeout firing. */
73 tw->tw_timeout = DCCP_TIMEWAIT_LEN;
74 if (state == DCCP_TIME_WAIT)
75 timeo = DCCP_TIMEWAIT_LEN;
77 inet_twsk_schedule(tw, &dccp_death_row, timeo,
81 /* Sorry, if we're out of memory, just CLOSE this
82 * socket up. We've got bigger problems than
83 * non-graceful socket closings.
85 LIMIT_NETDEBUG(KERN_INFO "DCCP: time wait bucket "
92 struct sock *dccp_create_openreq_child(struct sock *sk,
93 const struct request_sock *req,
94 const struct sk_buff *skb)
97 * Step 3: Process LISTEN state
99 * // Generate a new socket and switch to that socket
100 * Set S := new socket for this port pair
102 struct sock *newsk = inet_csk_clone(sk, req, GFP_ATOMIC);
105 const struct dccp_request_sock *dreq = dccp_rsk(req);
106 struct inet_connection_sock *newicsk = inet_csk(sk);
107 struct dccp_sock *newdp = dccp_sk(newsk);
108 struct dccp_minisock *newdmsk = dccp_msk(newsk);
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);
117 if (dccp_feat_clone(sk, newsk))
120 if (newdmsk->dccpms_send_ack_vector) {
121 newdp->dccps_hc_rx_ackvec =
122 dccp_ackvec_alloc(GFP_ATOMIC);
123 if (unlikely(newdp->dccps_hc_rx_ackvec == NULL))
127 newdp->dccps_hc_rx_ccid =
128 ccid_hc_rx_new(newdmsk->dccpms_rx_ccid,
130 newdp->dccps_hc_tx_ccid =
131 ccid_hc_tx_new(newdmsk->dccpms_tx_ccid,
133 if (unlikely(newdp->dccps_hc_rx_ccid == NULL ||
134 newdp->dccps_hc_tx_ccid == NULL)) {
135 dccp_ackvec_free(newdp->dccps_hc_rx_ackvec);
136 ccid_hc_rx_delete(newdp->dccps_hc_rx_ccid, newsk);
137 ccid_hc_tx_delete(newdp->dccps_hc_tx_ccid, newsk);
139 /* It is still raw copy of parent, so invalidate
140 * destructor and make plain sk_free() */
141 newsk->sk_destruct = NULL;
147 * Step 3: Process LISTEN state
149 * Choose S.ISS (initial seqno) or set from Init Cookie
150 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init
154 /* See dccp_v4_conn_request */
155 newdmsk->dccpms_sequence_window = req->rcv_wnd;
157 newdp->dccps_gar = newdp->dccps_isr = dreq->dreq_isr;
158 dccp_update_gsr(newsk, dreq->dreq_isr);
160 newdp->dccps_iss = dreq->dreq_iss;
161 dccp_update_gss(newsk, dreq->dreq_iss);
164 * SWL and AWL are initially adjusted so that they are not less than
165 * the initial Sequence Numbers received and sent, respectively:
166 * SWL := max(GSR + 1 - floor(W/4), ISR),
167 * AWL := max(GSS - W' + 1, ISS).
168 * These adjustments MUST be applied only at the beginning of the
171 dccp_set_seqno(&newdp->dccps_swl,
172 max48(newdp->dccps_swl, newdp->dccps_isr));
173 dccp_set_seqno(&newdp->dccps_awl,
174 max48(newdp->dccps_awl, newdp->dccps_iss));
176 dccp_init_xmit_timers(newsk);
178 DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS);
183 EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
186 * Process an incoming packet for RESPOND sockets represented
187 * as an request_sock.
189 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
190 struct request_sock *req,
191 struct request_sock **prev)
193 struct sock *child = NULL;
195 /* Check for retransmitted REQUEST */
196 if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
197 if (after48(DCCP_SKB_CB(skb)->dccpd_seq,
198 dccp_rsk(req)->dreq_isr)) {
199 struct dccp_request_sock *dreq = dccp_rsk(req);
201 dccp_pr_debug("Retransmitted REQUEST\n");
202 /* Send another RESPONSE packet */
203 dccp_set_seqno(&dreq->dreq_iss, dreq->dreq_iss + 1);
204 dccp_set_seqno(&dreq->dreq_isr,
205 DCCP_SKB_CB(skb)->dccpd_seq);
206 req->rsk_ops->rtx_syn_ack(sk, req, NULL);
208 /* Network Duplicate, discard packet */
212 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
214 if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
215 dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
219 if (DCCP_SKB_CB(skb)->dccpd_ack_seq != dccp_rsk(req)->dreq_iss) {
220 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
223 DCCP_SKB_CB(skb)->dccpd_ack_seq,
225 dccp_rsk(req)->dreq_iss);
229 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
231 goto listen_overflow;
233 /* FIXME: deal with options */
235 inet_csk_reqsk_queue_unlink(sk, req, prev);
236 inet_csk_reqsk_queue_removed(sk, req);
237 inet_csk_reqsk_queue_add(sk, req, child);
241 dccp_pr_debug("listen_overflow!\n");
242 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
244 if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
245 req->rsk_ops->send_reset(skb);
247 inet_csk_reqsk_queue_drop(sk, req, prev);
251 EXPORT_SYMBOL_GPL(dccp_check_req);
254 * Queue segment on the new socket if the new socket is active,
255 * otherwise we just shortcircuit this and continue with
258 int dccp_child_process(struct sock *parent, struct sock *child,
262 const int state = child->sk_state;
264 if (!sock_owned_by_user(child)) {
265 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
268 /* Wakeup parent, send SIGIO */
269 if (state == DCCP_RESPOND && child->sk_state != state)
270 parent->sk_data_ready(parent, 0);
272 /* Alas, it is possible again, because we do lookup
273 * in main socket hash table and lock on listening
274 * socket does not protect us more.
276 sk_add_backlog(child, skb);
279 bh_unlock_sock(child);
284 EXPORT_SYMBOL_GPL(dccp_child_process);