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/gfp.h>
15 #include <linux/kernel.h>
16 #include <linux/skbuff.h>
17 #include <linux/timer.h>
21 #include <net/inet_timewait_sock.h>
28 struct inet_timewait_death_row dccp_death_row
= {
29 .sysctl_max_tw_buckets
= NR_FILE
* 2,
30 .hashinfo
= &dccp_hashinfo
,
33 EXPORT_SYMBOL_GPL(dccp_death_row
);
35 void dccp_time_wait(struct sock
*sk
, int state
, int timeo
)
37 struct inet_timewait_sock
*tw
;
39 tw
= inet_twsk_alloc(sk
, &dccp_death_row
, state
);
42 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
43 const int rto
= (icsk
->icsk_rto
<< 2) - (icsk
->icsk_rto
>> 1);
44 #if IS_ENABLED(CONFIG_IPV6)
45 if (tw
->tw_family
== PF_INET6
) {
46 tw
->tw_v6_daddr
= sk
->sk_v6_daddr
;
47 tw
->tw_v6_rcv_saddr
= sk
->sk_v6_rcv_saddr
;
48 tw
->tw_ipv6only
= sk
->sk_ipv6only
;
52 /* Get the TIME_WAIT timeout firing. */
56 tw
->tw_timeout
= DCCP_TIMEWAIT_LEN
;
57 if (state
== DCCP_TIME_WAIT
)
58 timeo
= DCCP_TIMEWAIT_LEN
;
60 /* tw_timer is pinned, so we need to make sure BH are disabled
61 * in following section, otherwise timer handler could run before
62 * we complete the initialization.
65 inet_twsk_schedule(tw
, timeo
);
67 * Note that access to tw after this point is illegal.
69 inet_twsk_hashdance(tw
, sk
, &dccp_hashinfo
);
72 /* Sorry, if we're out of memory, just CLOSE this
73 * socket up. We've got bigger problems than
74 * non-graceful socket closings.
76 DCCP_WARN("time wait bucket table overflow\n");
82 struct sock
*dccp_create_openreq_child(const 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_lock(sk
, req
, GFP_ATOMIC
);
95 struct dccp_request_sock
*dreq
= dccp_rsk(req
);
96 struct inet_connection_sock
*newicsk
= inet_csk(newsk
);
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 newdp
->dccps_timestamp_echo
= dreq
->dreq_timestamp_echo
;
104 newdp
->dccps_timestamp_time
= dreq
->dreq_timestamp_time
;
105 newicsk
->icsk_rto
= DCCP_TIMEOUT_INIT
;
107 INIT_LIST_HEAD(&newdp
->dccps_featneg
);
109 * Step 3: Process LISTEN state
111 * Choose S.ISS (initial seqno) or set from Init Cookies
112 * Initialize S.GAR := S.ISS
113 * Set S.ISR, S.GSR from packet (or Init Cookies)
115 * Setting AWL/AWH and SWL/SWH happens as part of the feature
116 * activation below, as these windows all depend on the local
117 * and remote Sequence Window feature values (7.5.2).
119 newdp
->dccps_iss
= dreq
->dreq_iss
;
120 newdp
->dccps_gss
= dreq
->dreq_gss
;
121 newdp
->dccps_gar
= newdp
->dccps_iss
;
122 newdp
->dccps_isr
= dreq
->dreq_isr
;
123 newdp
->dccps_gsr
= dreq
->dreq_gsr
;
126 * Activate features: initialise CCIDs, sequence windows etc.
128 if (dccp_feat_activate_values(newsk
, &dreq
->dreq_featneg
)) {
129 sk_free_unlock_clone(newsk
);
132 dccp_init_xmit_timers(newsk
);
134 __DCCP_INC_STATS(DCCP_MIB_PASSIVEOPENS
);
139 EXPORT_SYMBOL_GPL(dccp_create_openreq_child
);
142 * Process an incoming packet for RESPOND sockets represented
143 * as an request_sock.
145 struct sock
*dccp_check_req(struct sock
*sk
, struct sk_buff
*skb
,
146 struct request_sock
*req
)
148 struct sock
*child
= NULL
;
149 struct dccp_request_sock
*dreq
= dccp_rsk(req
);
152 /* TCP/DCCP listeners became lockless.
153 * DCCP stores complex state in its request_sock, so we need
154 * a protection for them, now this code runs without being protected
155 * by the parent (listener) lock.
157 spin_lock_bh(&dreq
->dreq_lock
);
159 /* Check for retransmitted REQUEST */
160 if (dccp_hdr(skb
)->dccph_type
== DCCP_PKT_REQUEST
) {
162 if (after48(DCCP_SKB_CB(skb
)->dccpd_seq
, dreq
->dreq_gsr
)) {
163 dccp_pr_debug("Retransmitted REQUEST\n");
164 dreq
->dreq_gsr
= DCCP_SKB_CB(skb
)->dccpd_seq
;
166 * Send another RESPONSE packet
167 * To protect against Request floods, increment retrans
168 * counter (backoff, monitored by dccp_response_timer).
170 inet_rtx_syn_ack(sk
, req
);
172 /* Network Duplicate, discard packet */
176 DCCP_SKB_CB(skb
)->dccpd_reset_code
= DCCP_RESET_CODE_PACKET_ERROR
;
178 if (dccp_hdr(skb
)->dccph_type
!= DCCP_PKT_ACK
&&
179 dccp_hdr(skb
)->dccph_type
!= DCCP_PKT_DATAACK
)
183 if (!between48(DCCP_SKB_CB(skb
)->dccpd_ack_seq
,
184 dreq
->dreq_iss
, dreq
->dreq_gss
)) {
185 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
186 "dreq_iss=%llu, dreq_gss=%llu\n",
188 DCCP_SKB_CB(skb
)->dccpd_ack_seq
,
189 (unsigned long long) dreq
->dreq_iss
,
190 (unsigned long long) dreq
->dreq_gss
);
194 if (dccp_parse_options(sk
, dreq
, skb
))
197 child
= inet_csk(sk
)->icsk_af_ops
->syn_recv_sock(sk
, skb
, req
, NULL
,
200 child
= inet_csk_complete_hashdance(sk
, child
, req
, own_req
);
204 DCCP_SKB_CB(skb
)->dccpd_reset_code
= DCCP_RESET_CODE_TOO_BUSY
;
206 if (dccp_hdr(skb
)->dccph_type
!= DCCP_PKT_RESET
)
207 req
->rsk_ops
->send_reset(sk
, skb
);
209 inet_csk_reqsk_queue_drop(sk
, req
);
211 spin_unlock_bh(&dreq
->dreq_lock
);
215 EXPORT_SYMBOL_GPL(dccp_check_req
);
218 * Queue segment on the new socket if the new socket is active,
219 * otherwise we just shortcircuit this and continue with
222 int dccp_child_process(struct sock
*parent
, struct sock
*child
,
226 const int state
= child
->sk_state
;
228 if (!sock_owned_by_user(child
)) {
229 ret
= dccp_rcv_state_process(child
, skb
, dccp_hdr(skb
),
232 /* Wakeup parent, send SIGIO */
233 if (state
== DCCP_RESPOND
&& child
->sk_state
!= state
)
234 parent
->sk_data_ready(parent
);
236 /* Alas, it is possible again, because we do lookup
237 * in main socket hash table and lock on listening
238 * socket does not protect us more.
240 __sk_add_backlog(child
, skb
);
243 bh_unlock_sock(child
);
248 EXPORT_SYMBOL_GPL(dccp_child_process
);
250 void dccp_reqsk_send_ack(const struct sock
*sk
, struct sk_buff
*skb
,
251 struct request_sock
*rsk
)
253 DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state");
256 EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack
);
258 int dccp_reqsk_init(struct request_sock
*req
,
259 struct dccp_sock
const *dp
, struct sk_buff
const *skb
)
261 struct dccp_request_sock
*dreq
= dccp_rsk(req
);
263 spin_lock_init(&dreq
->dreq_lock
);
264 inet_rsk(req
)->ir_rmt_port
= dccp_hdr(skb
)->dccph_sport
;
265 inet_rsk(req
)->ir_num
= ntohs(dccp_hdr(skb
)->dccph_dport
);
266 inet_rsk(req
)->acked
= 0;
267 dreq
->dreq_timestamp_echo
= 0;
269 /* inherit feature negotiation options from listening socket */
270 return dccp_feat_clone_list(&dp
->dccps_featneg
, &dreq
->dreq_featneg
);
273 EXPORT_SYMBOL_GPL(dccp_reqsk_init
);