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>
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
,
36 /* Short-time timewait calendar */
39 .twcal_timer
= TIMER_INITIALIZER(inet_twdr_twcal_tick
, 0,
40 (unsigned long)&dccp_death_row
),
43 EXPORT_SYMBOL_GPL(dccp_death_row
);
45 void dccp_time_wait(struct sock
*sk
, int state
, int timeo
)
47 struct inet_timewait_sock
*tw
= NULL
;
49 if (dccp_death_row
.tw_count
< dccp_death_row
.sysctl_max_tw_buckets
)
50 tw
= inet_twsk_alloc(sk
, state
);
53 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
54 const int rto
= (icsk
->icsk_rto
<< 2) - (icsk
->icsk_rto
>> 1);
56 /* Linkage updates. */
57 __inet_twsk_hashdance(tw
, sk
, &dccp_hashinfo
);
59 /* Get the TIME_WAIT timeout firing. */
63 tw
->tw_timeout
= DCCP_TIMEWAIT_LEN
;
64 if (state
== DCCP_TIME_WAIT
)
65 timeo
= DCCP_TIMEWAIT_LEN
;
67 inet_twsk_schedule(tw
, &dccp_death_row
, timeo
,
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.
75 LIMIT_NETDEBUG(KERN_INFO
"DCCP: time wait bucket "
82 struct sock
*dccp_create_openreq_child(struct sock
*sk
,
83 const struct request_sock
*req
,
84 const struct sk_buff
*skb
)
87 * Step 3: Process LISTEN state
89 * // Generate a new socket and switch to that socket
90 * Set S := new socket for this port pair
92 struct sock
*newsk
= inet_csk_clone(sk
, req
, GFP_ATOMIC
);
95 const struct dccp_request_sock
*dreq
= dccp_rsk(req
);
96 struct inet_connection_sock
*newicsk
= inet_csk(sk
);
97 struct dccp_sock
*newdp
= dccp_sk(newsk
);
99 newdp
->dccps_role
= DCCP_ROLE_SERVER
;
100 newdp
->dccps_hc_rx_ackvec
= NULL
;
101 newdp
->dccps_service_list
= NULL
;
102 newdp
->dccps_service
= dreq
->dreq_service
;
103 newicsk
->icsk_rto
= DCCP_TIMEOUT_INIT
;
104 do_gettimeofday(&newdp
->dccps_epoch
);
106 if (newdp
->dccps_options
.dccpo_send_ack_vector
) {
107 newdp
->dccps_hc_rx_ackvec
=
108 dccp_ackvec_alloc(DCCP_MAX_ACKVEC_LEN
,
111 * XXX: We're using the same CCIDs set on the parent,
112 * i.e. sk_clone copied the master sock and left the
113 * CCID pointers for this child, that is why we do the
116 if (unlikely(newdp
->dccps_hc_rx_ackvec
== NULL
))
120 if (unlikely(ccid_hc_rx_init(newdp
->dccps_hc_rx_ccid
,
122 ccid_hc_tx_init(newdp
->dccps_hc_tx_ccid
,
124 dccp_ackvec_free(newdp
->dccps_hc_rx_ackvec
);
125 ccid_hc_rx_exit(newdp
->dccps_hc_rx_ccid
, newsk
);
126 ccid_hc_tx_exit(newdp
->dccps_hc_tx_ccid
, newsk
);
128 /* It is still raw copy of parent, so invalidate
129 * destructor and make plain sk_free() */
130 newsk
->sk_destruct
= NULL
;
135 __ccid_get(newdp
->dccps_hc_rx_ccid
);
136 __ccid_get(newdp
->dccps_hc_tx_ccid
);
139 * Step 3: Process LISTEN state
141 * Choose S.ISS (initial seqno) or set from Init Cookie
142 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init
146 /* See dccp_v4_conn_request */
147 newdp
->dccps_options
.dccpo_sequence_window
= req
->rcv_wnd
;
149 newdp
->dccps_gar
= newdp
->dccps_isr
= dreq
->dreq_isr
;
150 dccp_update_gsr(newsk
, dreq
->dreq_isr
);
152 newdp
->dccps_iss
= dreq
->dreq_iss
;
153 dccp_update_gss(newsk
, dreq
->dreq_iss
);
156 * SWL and AWL are initially adjusted so that they are not less than
157 * the initial Sequence Numbers received and sent, respectively:
158 * SWL := max(GSR + 1 - floor(W/4), ISR),
159 * AWL := max(GSS - W' + 1, ISS).
160 * These adjustments MUST be applied only at the beginning of the
163 dccp_set_seqno(&newdp
->dccps_swl
,
164 max48(newdp
->dccps_swl
, newdp
->dccps_isr
));
165 dccp_set_seqno(&newdp
->dccps_awl
,
166 max48(newdp
->dccps_awl
, newdp
->dccps_iss
));
168 dccp_init_xmit_timers(newsk
);
170 DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS
);
175 EXPORT_SYMBOL_GPL(dccp_create_openreq_child
);
178 * Process an incoming packet for RESPOND sockets represented
179 * as an request_sock.
181 struct sock
*dccp_check_req(struct sock
*sk
, struct sk_buff
*skb
,
182 struct request_sock
*req
,
183 struct request_sock
**prev
)
185 struct sock
*child
= NULL
;
187 /* Check for retransmitted REQUEST */
188 if (dccp_hdr(skb
)->dccph_type
== DCCP_PKT_REQUEST
) {
189 if (after48(DCCP_SKB_CB(skb
)->dccpd_seq
,
190 dccp_rsk(req
)->dreq_isr
)) {
191 struct dccp_request_sock
*dreq
= dccp_rsk(req
);
193 dccp_pr_debug("Retransmitted REQUEST\n");
194 /* Send another RESPONSE packet */
195 dccp_set_seqno(&dreq
->dreq_iss
, dreq
->dreq_iss
+ 1);
196 dccp_set_seqno(&dreq
->dreq_isr
,
197 DCCP_SKB_CB(skb
)->dccpd_seq
);
198 req
->rsk_ops
->rtx_syn_ack(sk
, req
, NULL
);
200 /* Network Duplicate, discard packet */
204 DCCP_SKB_CB(skb
)->dccpd_reset_code
= DCCP_RESET_CODE_PACKET_ERROR
;
206 if (dccp_hdr(skb
)->dccph_type
!= DCCP_PKT_ACK
&&
207 dccp_hdr(skb
)->dccph_type
!= DCCP_PKT_DATAACK
)
211 if (DCCP_SKB_CB(skb
)->dccpd_ack_seq
!= dccp_rsk(req
)->dreq_iss
) {
212 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
215 DCCP_SKB_CB(skb
)->dccpd_ack_seq
,
217 dccp_rsk(req
)->dreq_iss
);
221 child
= inet_csk(sk
)->icsk_af_ops
->syn_recv_sock(sk
, skb
, req
, NULL
);
223 goto listen_overflow
;
225 /* FIXME: deal with options */
227 inet_csk_reqsk_queue_unlink(sk
, req
, prev
);
228 inet_csk_reqsk_queue_removed(sk
, req
);
229 inet_csk_reqsk_queue_add(sk
, req
, child
);
233 dccp_pr_debug("listen_overflow!\n");
234 DCCP_SKB_CB(skb
)->dccpd_reset_code
= DCCP_RESET_CODE_TOO_BUSY
;
236 if (dccp_hdr(skb
)->dccph_type
!= DCCP_PKT_RESET
)
237 req
->rsk_ops
->send_reset(skb
);
239 inet_csk_reqsk_queue_drop(sk
, req
, prev
);
243 EXPORT_SYMBOL_GPL(dccp_check_req
);
246 * Queue segment on the new socket if the new socket is active,
247 * otherwise we just shortcircuit this and continue with
250 int dccp_child_process(struct sock
*parent
, struct sock
*child
,
254 const int state
= child
->sk_state
;
256 if (!sock_owned_by_user(child
)) {
257 ret
= dccp_rcv_state_process(child
, skb
, dccp_hdr(skb
),
260 /* Wakeup parent, send SIGIO */
261 if (state
== DCCP_RESPOND
&& child
->sk_state
!= state
)
262 parent
->sk_data_ready(parent
, 0);
264 /* Alas, it is possible again, because we do lookup
265 * in main socket hash table and lock on listening
266 * socket does not protect us more.
268 sk_add_backlog(child
, skb
);
271 bh_unlock_sock(child
);
276 EXPORT_SYMBOL_GPL(dccp_child_process
);