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/dccp.h>
14 #include <linux/kernel.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(dccp_death_row
.death_lock
),
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
),
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);
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
;
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
;
67 /* Linkage updates. */
68 __inet_twsk_hashdance(tw
, sk
, &dccp_hashinfo
);
70 /* Get the TIME_WAIT timeout firing. */
74 tw
->tw_timeout
= DCCP_TIMEWAIT_LEN
;
75 if (state
== DCCP_TIME_WAIT
)
76 timeo
= DCCP_TIMEWAIT_LEN
;
78 inet_twsk_schedule(tw
, &dccp_death_row
, timeo
,
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.
86 DCCP_WARN("time wait bucket table overflow\n");
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(newsk
);
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
;
116 if (dccp_feat_clone(sk
, newsk
))
119 if (newdmsk
->dccpms_send_ack_vector
) {
120 newdp
->dccps_hc_rx_ackvec
=
121 dccp_ackvec_alloc(GFP_ATOMIC
);
122 if (unlikely(newdp
->dccps_hc_rx_ackvec
== NULL
))
126 newdp
->dccps_hc_rx_ccid
=
127 ccid_hc_rx_new(newdmsk
->dccpms_rx_ccid
,
129 newdp
->dccps_hc_tx_ccid
=
130 ccid_hc_tx_new(newdmsk
->dccpms_tx_ccid
,
132 if (unlikely(newdp
->dccps_hc_rx_ccid
== NULL
||
133 newdp
->dccps_hc_tx_ccid
== NULL
)) {
134 dccp_ackvec_free(newdp
->dccps_hc_rx_ackvec
);
135 ccid_hc_rx_delete(newdp
->dccps_hc_rx_ccid
, newsk
);
136 ccid_hc_tx_delete(newdp
->dccps_hc_tx_ccid
, newsk
);
138 /* It is still raw copy of parent, so invalidate
139 * destructor and make plain sk_free() */
140 newsk
->sk_destruct
= NULL
;
146 * Step 3: Process LISTEN state
148 * Choose S.ISS (initial seqno) or set from Init Cookies
149 * Initialize S.GAR := S.ISS
150 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
153 /* See dccp_v4_conn_request */
154 newdmsk
->dccpms_sequence_window
= req
->rcv_wnd
;
156 newdp
->dccps_gar
= newdp
->dccps_isr
= dreq
->dreq_isr
;
157 dccp_update_gsr(newsk
, dreq
->dreq_isr
);
159 newdp
->dccps_iss
= dreq
->dreq_iss
;
160 dccp_update_gss(newsk
, dreq
->dreq_iss
);
163 * SWL and AWL are initially adjusted so that they are not less than
164 * the initial Sequence Numbers received and sent, respectively:
165 * SWL := max(GSR + 1 - floor(W/4), ISR),
166 * AWL := max(GSS - W' + 1, ISS).
167 * These adjustments MUST be applied only at the beginning of the
170 dccp_set_seqno(&newdp
->dccps_swl
,
171 max48(newdp
->dccps_swl
, newdp
->dccps_isr
));
172 dccp_set_seqno(&newdp
->dccps_awl
,
173 max48(newdp
->dccps_awl
, newdp
->dccps_iss
));
175 dccp_init_xmit_timers(newsk
);
177 DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS
);
182 EXPORT_SYMBOL_GPL(dccp_create_openreq_child
);
185 * Process an incoming packet for RESPOND sockets represented
186 * as an request_sock.
188 struct sock
*dccp_check_req(struct sock
*sk
, struct sk_buff
*skb
,
189 struct request_sock
*req
,
190 struct request_sock
**prev
)
192 struct sock
*child
= NULL
;
194 /* Check for retransmitted REQUEST */
195 if (dccp_hdr(skb
)->dccph_type
== DCCP_PKT_REQUEST
) {
196 struct dccp_request_sock
*dreq
= dccp_rsk(req
);
198 if (after48(DCCP_SKB_CB(skb
)->dccpd_seq
, dreq
->dreq_isr
)) {
199 dccp_pr_debug("Retransmitted REQUEST\n");
200 dreq
->dreq_isr
= DCCP_SKB_CB(skb
)->dccpd_seq
;
202 * Send another RESPONSE packet
203 * To protect against Request floods, increment retrans
204 * counter (backoff, monitored by dccp_response_timer).
207 req
->rsk_ops
->rtx_syn_ack(sk
, req
, NULL
);
209 /* Network Duplicate, discard packet */
213 DCCP_SKB_CB(skb
)->dccpd_reset_code
= DCCP_RESET_CODE_PACKET_ERROR
;
215 if (dccp_hdr(skb
)->dccph_type
!= DCCP_PKT_ACK
&&
216 dccp_hdr(skb
)->dccph_type
!= DCCP_PKT_DATAACK
)
220 if (DCCP_SKB_CB(skb
)->dccpd_ack_seq
!= dccp_rsk(req
)->dreq_iss
) {
221 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
224 DCCP_SKB_CB(skb
)->dccpd_ack_seq
,
226 dccp_rsk(req
)->dreq_iss
);
230 child
= inet_csk(sk
)->icsk_af_ops
->syn_recv_sock(sk
, skb
, req
, NULL
);
232 goto listen_overflow
;
234 /* FIXME: deal with options */
236 inet_csk_reqsk_queue_unlink(sk
, req
, prev
);
237 inet_csk_reqsk_queue_removed(sk
, req
);
238 inet_csk_reqsk_queue_add(sk
, req
, child
);
242 dccp_pr_debug("listen_overflow!\n");
243 DCCP_SKB_CB(skb
)->dccpd_reset_code
= DCCP_RESET_CODE_TOO_BUSY
;
245 if (dccp_hdr(skb
)->dccph_type
!= DCCP_PKT_RESET
)
246 req
->rsk_ops
->send_reset(sk
, skb
);
248 inet_csk_reqsk_queue_drop(sk
, req
, prev
);
252 EXPORT_SYMBOL_GPL(dccp_check_req
);
255 * Queue segment on the new socket if the new socket is active,
256 * otherwise we just shortcircuit this and continue with
259 int dccp_child_process(struct sock
*parent
, struct sock
*child
,
263 const int state
= child
->sk_state
;
265 if (!sock_owned_by_user(child
)) {
266 ret
= dccp_rcv_state_process(child
, skb
, dccp_hdr(skb
),
269 /* Wakeup parent, send SIGIO */
270 if (state
== DCCP_RESPOND
&& child
->sk_state
!= state
)
271 parent
->sk_data_ready(parent
, 0);
273 /* Alas, it is possible again, because we do lookup
274 * in main socket hash table and lock on listening
275 * socket does not protect us more.
277 sk_add_backlog(child
, skb
);
280 bh_unlock_sock(child
);
285 EXPORT_SYMBOL_GPL(dccp_child_process
);
287 void dccp_reqsk_send_ack(struct sk_buff
*skb
, struct request_sock
*rsk
)
289 DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state");
292 EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack
);
294 void dccp_reqsk_init(struct request_sock
*req
, struct sk_buff
*skb
)
296 inet_rsk(req
)->rmt_port
= dccp_hdr(skb
)->dccph_sport
;
297 inet_rsk(req
)->acked
= 0;
298 req
->rcv_wnd
= sysctl_dccp_feat_sequence_window
;
301 EXPORT_SYMBOL_GPL(dccp_reqsk_init
);