4 * An implementation of the DCCP protocol
5 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>
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/config.h>
14 #include <linux/dccp.h>
15 #include <linux/skbuff.h>
16 #include <linux/timer.h>
20 #include <net/inet_timewait_sock.h>
27 struct inet_timewait_death_row dccp_death_row
= {
28 .sysctl_max_tw_buckets
= NR_FILE
* 2,
29 .period
= DCCP_TIMEWAIT_LEN
/ INET_TWDR_TWKILL_SLOTS
,
30 .death_lock
= SPIN_LOCK_UNLOCKED
,
31 .hashinfo
= &dccp_hashinfo
,
32 .tw_timer
= TIMER_INITIALIZER(inet_twdr_hangman
, 0,
33 (unsigned long)&dccp_death_row
),
34 .twkill_work
= __WORK_INITIALIZER(dccp_death_row
.twkill_work
,
35 inet_twdr_twkill_work
,
37 /* Short-time timewait calendar */
40 .twcal_timer
= TIMER_INITIALIZER(inet_twdr_twcal_tick
, 0,
41 (unsigned long)&dccp_death_row
),
44 EXPORT_SYMBOL_GPL(dccp_death_row
);
46 void dccp_time_wait(struct sock
*sk
, int state
, int timeo
)
48 struct inet_timewait_sock
*tw
= NULL
;
50 if (dccp_death_row
.tw_count
< dccp_death_row
.sysctl_max_tw_buckets
)
51 tw
= inet_twsk_alloc(sk
, state
);
54 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
55 const int rto
= (icsk
->icsk_rto
<< 2) - (icsk
->icsk_rto
>> 1);
56 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
57 if (tw
->tw_family
== PF_INET6
) {
58 const struct ipv6_pinfo
*np
= inet6_sk(sk
);
59 struct inet6_timewait_sock
*tw6
;
61 tw
->tw_ipv6_offset
= inet6_tw_offset(sk
->sk_prot
);
62 tw6
= inet6_twsk((struct sock
*)tw
);
63 ipv6_addr_copy(&tw6
->tw_v6_daddr
, &np
->daddr
);
64 ipv6_addr_copy(&tw6
->tw_v6_rcv_saddr
, &np
->rcv_saddr
);
65 tw
->tw_ipv6only
= np
->ipv6only
;
68 /* Linkage updates. */
69 __inet_twsk_hashdance(tw
, sk
, &dccp_hashinfo
);
71 /* Get the TIME_WAIT timeout firing. */
75 tw
->tw_timeout
= DCCP_TIMEWAIT_LEN
;
76 if (state
== DCCP_TIME_WAIT
)
77 timeo
= DCCP_TIMEWAIT_LEN
;
79 inet_twsk_schedule(tw
, &dccp_death_row
, timeo
,
83 /* Sorry, if we're out of memory, just CLOSE this
84 * socket up. We've got bigger problems than
85 * non-graceful socket closings.
87 LIMIT_NETDEBUG(KERN_INFO
"DCCP: time wait bucket "
94 struct sock
*dccp_create_openreq_child(struct sock
*sk
,
95 const struct request_sock
*req
,
96 const struct sk_buff
*skb
)
99 * Step 3: Process LISTEN state
101 * // Generate a new socket and switch to that socket
102 * Set S := new socket for this port pair
104 struct sock
*newsk
= inet_csk_clone(sk
, req
, GFP_ATOMIC
);
107 const struct dccp_request_sock
*dreq
= dccp_rsk(req
);
108 struct inet_connection_sock
*newicsk
= inet_csk(sk
);
109 struct dccp_sock
*newdp
= dccp_sk(newsk
);
111 newdp
->dccps_role
= DCCP_ROLE_SERVER
;
112 newdp
->dccps_hc_rx_ackvec
= NULL
;
113 newdp
->dccps_service_list
= NULL
;
114 newdp
->dccps_service
= dreq
->dreq_service
;
115 newicsk
->icsk_rto
= DCCP_TIMEOUT_INIT
;
116 do_gettimeofday(&newdp
->dccps_epoch
);
118 if (dccp_feat_clone(sk
, newsk
))
121 if (newdp
->dccps_options
.dccpo_send_ack_vector
) {
122 newdp
->dccps_hc_rx_ackvec
=
123 dccp_ackvec_alloc(GFP_ATOMIC
);
125 * XXX: We're using the same CCIDs set on the parent,
126 * i.e. sk_clone copied the master sock and left the
127 * CCID pointers for this child, that is why we do the
130 if (unlikely(newdp
->dccps_hc_rx_ackvec
== NULL
))
134 if (unlikely(ccid_hc_rx_init(newdp
->dccps_hc_rx_ccid
,
136 ccid_hc_tx_init(newdp
->dccps_hc_tx_ccid
,
138 dccp_ackvec_free(newdp
->dccps_hc_rx_ackvec
);
139 ccid_hc_rx_exit(newdp
->dccps_hc_rx_ccid
, newsk
);
140 ccid_hc_tx_exit(newdp
->dccps_hc_tx_ccid
, newsk
);
142 /* It is still raw copy of parent, so invalidate
143 * destructor and make plain sk_free() */
144 newsk
->sk_destruct
= NULL
;
149 __ccid_get(newdp
->dccps_hc_rx_ccid
);
150 __ccid_get(newdp
->dccps_hc_tx_ccid
);
153 * Step 3: Process LISTEN state
155 * Choose S.ISS (initial seqno) or set from Init Cookie
156 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init
160 /* See dccp_v4_conn_request */
161 newdp
->dccps_options
.dccpo_sequence_window
= req
->rcv_wnd
;
163 newdp
->dccps_gar
= newdp
->dccps_isr
= dreq
->dreq_isr
;
164 dccp_update_gsr(newsk
, dreq
->dreq_isr
);
166 newdp
->dccps_iss
= dreq
->dreq_iss
;
167 dccp_update_gss(newsk
, dreq
->dreq_iss
);
170 * SWL and AWL are initially adjusted so that they are not less than
171 * the initial Sequence Numbers received and sent, respectively:
172 * SWL := max(GSR + 1 - floor(W/4), ISR),
173 * AWL := max(GSS - W' + 1, ISS).
174 * These adjustments MUST be applied only at the beginning of the
177 dccp_set_seqno(&newdp
->dccps_swl
,
178 max48(newdp
->dccps_swl
, newdp
->dccps_isr
));
179 dccp_set_seqno(&newdp
->dccps_awl
,
180 max48(newdp
->dccps_awl
, newdp
->dccps_iss
));
182 dccp_init_xmit_timers(newsk
);
184 DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS
);
189 EXPORT_SYMBOL_GPL(dccp_create_openreq_child
);
192 * Process an incoming packet for RESPOND sockets represented
193 * as an request_sock.
195 struct sock
*dccp_check_req(struct sock
*sk
, struct sk_buff
*skb
,
196 struct request_sock
*req
,
197 struct request_sock
**prev
)
199 struct sock
*child
= NULL
;
201 /* Check for retransmitted REQUEST */
202 if (dccp_hdr(skb
)->dccph_type
== DCCP_PKT_REQUEST
) {
203 if (after48(DCCP_SKB_CB(skb
)->dccpd_seq
,
204 dccp_rsk(req
)->dreq_isr
)) {
205 struct dccp_request_sock
*dreq
= dccp_rsk(req
);
207 dccp_pr_debug("Retransmitted REQUEST\n");
208 /* Send another RESPONSE packet */
209 dccp_set_seqno(&dreq
->dreq_iss
, dreq
->dreq_iss
+ 1);
210 dccp_set_seqno(&dreq
->dreq_isr
,
211 DCCP_SKB_CB(skb
)->dccpd_seq
);
212 req
->rsk_ops
->rtx_syn_ack(sk
, req
, NULL
);
214 /* Network Duplicate, discard packet */
218 DCCP_SKB_CB(skb
)->dccpd_reset_code
= DCCP_RESET_CODE_PACKET_ERROR
;
220 if (dccp_hdr(skb
)->dccph_type
!= DCCP_PKT_ACK
&&
221 dccp_hdr(skb
)->dccph_type
!= DCCP_PKT_DATAACK
)
225 if (DCCP_SKB_CB(skb
)->dccpd_ack_seq
!= dccp_rsk(req
)->dreq_iss
) {
226 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
229 DCCP_SKB_CB(skb
)->dccpd_ack_seq
,
231 dccp_rsk(req
)->dreq_iss
);
235 child
= inet_csk(sk
)->icsk_af_ops
->syn_recv_sock(sk
, skb
, req
, NULL
);
237 goto listen_overflow
;
239 /* FIXME: deal with options */
241 inet_csk_reqsk_queue_unlink(sk
, req
, prev
);
242 inet_csk_reqsk_queue_removed(sk
, req
);
243 inet_csk_reqsk_queue_add(sk
, req
, child
);
247 dccp_pr_debug("listen_overflow!\n");
248 DCCP_SKB_CB(skb
)->dccpd_reset_code
= DCCP_RESET_CODE_TOO_BUSY
;
250 if (dccp_hdr(skb
)->dccph_type
!= DCCP_PKT_RESET
)
251 req
->rsk_ops
->send_reset(skb
);
253 inet_csk_reqsk_queue_drop(sk
, req
, prev
);
257 EXPORT_SYMBOL_GPL(dccp_check_req
);
260 * Queue segment on the new socket if the new socket is active,
261 * otherwise we just shortcircuit this and continue with
264 int dccp_child_process(struct sock
*parent
, struct sock
*child
,
268 const int state
= child
->sk_state
;
270 if (!sock_owned_by_user(child
)) {
271 ret
= dccp_rcv_state_process(child
, skb
, dccp_hdr(skb
),
274 /* Wakeup parent, send SIGIO */
275 if (state
== DCCP_RESPOND
&& child
->sk_state
!= state
)
276 parent
->sk_data_ready(parent
, 0);
278 /* Alas, it is possible again, because we do lookup
279 * in main socket hash table and lock on listening
280 * socket does not protect us more.
282 sk_add_backlog(child
, skb
);
285 bh_unlock_sock(child
);
290 EXPORT_SYMBOL_GPL(dccp_child_process
);