Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * net/dccp/minisocks.c
4 *
5 * An implementation of the DCCP protocol
6 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>
7 */
8
9#include <linux/dccp.h>
10#include <linux/gfp.h>
11#include <linux/kernel.h>
12#include <linux/skbuff.h>
13#include <linux/timer.h>
14
15#include <net/sock.h>
16#include <net/xfrm.h>
17#include <net/inet_timewait_sock.h>
18
19#include "ackvec.h"
20#include "ccid.h"
21#include "dccp.h"
22#include "feat.h"
23
24struct inet_timewait_death_row dccp_death_row = {
25 .sysctl_max_tw_buckets = NR_FILE * 2,
26 .hashinfo = &dccp_hashinfo,
27};
28
29EXPORT_SYMBOL_GPL(dccp_death_row);
30
31void dccp_time_wait(struct sock *sk, int state, int timeo)
32{
33 struct inet_timewait_sock *tw;
34
35 tw = inet_twsk_alloc(sk, &dccp_death_row, state);
36
37 if (tw != NULL) {
38 const struct inet_connection_sock *icsk = inet_csk(sk);
39 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
40#if IS_ENABLED(CONFIG_IPV6)
41 if (tw->tw_family == PF_INET6) {
42 tw->tw_v6_daddr = sk->sk_v6_daddr;
43 tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
44 tw->tw_ipv6only = sk->sk_ipv6only;
45 }
46#endif
47
48 /* Get the TIME_WAIT timeout firing. */
49 if (timeo < rto)
50 timeo = rto;
51
52 if (state == DCCP_TIME_WAIT)
53 timeo = DCCP_TIMEWAIT_LEN;
54
55 /* tw_timer is pinned, so we need to make sure BH are disabled
56 * in following section, otherwise timer handler could run before
57 * we complete the initialization.
58 */
59 local_bh_disable();
60 inet_twsk_schedule(tw, timeo);
61 /* Linkage updates.
62 * Note that access to tw after this point is illegal.
63 */
64 inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
65 local_bh_enable();
66 } else {
67 /* Sorry, if we're out of memory, just CLOSE this
68 * socket up. We've got bigger problems than
69 * non-graceful socket closings.
70 */
71 DCCP_WARN("time wait bucket table overflow\n");
72 }
73
74 dccp_done(sk);
75}
76
77struct sock *dccp_create_openreq_child(const struct sock *sk,
78 const struct request_sock *req,
79 const struct sk_buff *skb)
80{
81 /*
82 * Step 3: Process LISTEN state
83 *
84 * (* Generate a new socket and switch to that socket *)
85 * Set S := new socket for this port pair
86 */
87 struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
88
89 if (newsk != NULL) {
90 struct dccp_request_sock *dreq = dccp_rsk(req);
91 struct inet_connection_sock *newicsk = inet_csk(newsk);
92 struct dccp_sock *newdp = dccp_sk(newsk);
93
94 newdp->dccps_role = DCCP_ROLE_SERVER;
95 newdp->dccps_hc_rx_ackvec = NULL;
96 newdp->dccps_service_list = NULL;
97 newdp->dccps_hc_rx_ccid = NULL;
98 newdp->dccps_hc_tx_ccid = NULL;
99 newdp->dccps_service = dreq->dreq_service;
100 newdp->dccps_timestamp_echo = dreq->dreq_timestamp_echo;
101 newdp->dccps_timestamp_time = dreq->dreq_timestamp_time;
102 newicsk->icsk_rto = DCCP_TIMEOUT_INIT;
103
104 INIT_LIST_HEAD(&newdp->dccps_featneg);
105 /*
106 * Step 3: Process LISTEN state
107 *
108 * Choose S.ISS (initial seqno) or set from Init Cookies
109 * Initialize S.GAR := S.ISS
110 * Set S.ISR, S.GSR from packet (or Init Cookies)
111 *
112 * Setting AWL/AWH and SWL/SWH happens as part of the feature
113 * activation below, as these windows all depend on the local
114 * and remote Sequence Window feature values (7.5.2).
115 */
116 newdp->dccps_iss = dreq->dreq_iss;
117 newdp->dccps_gss = dreq->dreq_gss;
118 newdp->dccps_gar = newdp->dccps_iss;
119 newdp->dccps_isr = dreq->dreq_isr;
120 newdp->dccps_gsr = dreq->dreq_gsr;
121
122 /*
123 * Activate features: initialise CCIDs, sequence windows etc.
124 */
125 if (dccp_feat_activate_values(newsk, &dreq->dreq_featneg)) {
126 sk_free_unlock_clone(newsk);
127 return NULL;
128 }
129 dccp_init_xmit_timers(newsk);
130
131 __DCCP_INC_STATS(DCCP_MIB_PASSIVEOPENS);
132 }
133 return newsk;
134}
135
136EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
137
138/*
139 * Process an incoming packet for RESPOND sockets represented
140 * as an request_sock.
141 */
142struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
143 struct request_sock *req)
144{
145 struct sock *child = NULL;
146 struct dccp_request_sock *dreq = dccp_rsk(req);
147 bool own_req;
148
149 /* TCP/DCCP listeners became lockless.
150 * DCCP stores complex state in its request_sock, so we need
151 * a protection for them, now this code runs without being protected
152 * by the parent (listener) lock.
153 */
154 spin_lock_bh(&dreq->dreq_lock);
155
156 /* Check for retransmitted REQUEST */
157 if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
158
159 if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_gsr)) {
160 dccp_pr_debug("Retransmitted REQUEST\n");
161 dreq->dreq_gsr = DCCP_SKB_CB(skb)->dccpd_seq;
162 /*
163 * Send another RESPONSE packet
164 * To protect against Request floods, increment retrans
165 * counter (backoff, monitored by dccp_response_timer).
166 */
167 inet_rtx_syn_ack(sk, req);
168 }
169 /* Network Duplicate, discard packet */
170 goto out;
171 }
172
173 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
174
175 if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
176 dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
177 goto drop;
178
179 /* Invalid ACK */
180 if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
181 dreq->dreq_iss, dreq->dreq_gss)) {
182 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
183 "dreq_iss=%llu, dreq_gss=%llu\n",
184 (unsigned long long)
185 DCCP_SKB_CB(skb)->dccpd_ack_seq,
186 (unsigned long long) dreq->dreq_iss,
187 (unsigned long long) dreq->dreq_gss);
188 goto drop;
189 }
190
191 if (dccp_parse_options(sk, dreq, skb))
192 goto drop;
193
194 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
195 req, &own_req);
196 if (child) {
197 child = inet_csk_complete_hashdance(sk, child, req, own_req);
198 goto out;
199 }
200
201 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
202drop:
203 if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
204 req->rsk_ops->send_reset(sk, skb);
205
206 inet_csk_reqsk_queue_drop(sk, req);
207out:
208 spin_unlock_bh(&dreq->dreq_lock);
209 return child;
210}
211
212EXPORT_SYMBOL_GPL(dccp_check_req);
213
214/*
215 * Queue segment on the new socket if the new socket is active,
216 * otherwise we just shortcircuit this and continue with
217 * the new socket.
218 */
219int dccp_child_process(struct sock *parent, struct sock *child,
220 struct sk_buff *skb)
221 __releases(child)
222{
223 int ret = 0;
224 const int state = child->sk_state;
225
226 if (!sock_owned_by_user(child)) {
227 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
228 skb->len);
229
230 /* Wakeup parent, send SIGIO */
231 if (state == DCCP_RESPOND && child->sk_state != state)
232 parent->sk_data_ready(parent);
233 } else {
234 /* Alas, it is possible again, because we do lookup
235 * in main socket hash table and lock on listening
236 * socket does not protect us more.
237 */
238 __sk_add_backlog(child, skb);
239 }
240
241 bh_unlock_sock(child);
242 sock_put(child);
243 return ret;
244}
245
246EXPORT_SYMBOL_GPL(dccp_child_process);
247
248void dccp_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
249 struct request_sock *rsk)
250{
251 DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state");
252}
253
254EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack);
255
256int dccp_reqsk_init(struct request_sock *req,
257 struct dccp_sock const *dp, struct sk_buff const *skb)
258{
259 struct dccp_request_sock *dreq = dccp_rsk(req);
260
261 spin_lock_init(&dreq->dreq_lock);
262 inet_rsk(req)->ir_rmt_port = dccp_hdr(skb)->dccph_sport;
263 inet_rsk(req)->ir_num = ntohs(dccp_hdr(skb)->dccph_dport);
264 inet_rsk(req)->acked = 0;
265 dreq->dreq_timestamp_echo = 0;
266
267 /* inherit feature negotiation options from listening socket */
268 return dccp_feat_clone_list(&dp->dccps_featneg, &dreq->dreq_featneg);
269}
270
271EXPORT_SYMBOL_GPL(dccp_reqsk_init);
1/*
2 * net/dccp/minisocks.c
3 *
4 * An implementation of the DCCP protocol
5 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>
6 *
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.
11 */
12
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>
18
19#include <net/sock.h>
20#include <net/xfrm.h>
21#include <net/inet_timewait_sock.h>
22
23#include "ackvec.h"
24#include "ccid.h"
25#include "dccp.h"
26#include "feat.h"
27
28struct inet_timewait_death_row dccp_death_row = {
29 .sysctl_max_tw_buckets = NR_FILE * 2,
30 .hashinfo = &dccp_hashinfo,
31};
32
33EXPORT_SYMBOL_GPL(dccp_death_row);
34
35void dccp_time_wait(struct sock *sk, int state, int timeo)
36{
37 struct inet_timewait_sock *tw;
38
39 tw = inet_twsk_alloc(sk, &dccp_death_row, state);
40
41 if (tw != NULL) {
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;
49 }
50#endif
51
52 /* Get the TIME_WAIT timeout firing. */
53 if (timeo < rto)
54 timeo = rto;
55
56 tw->tw_timeout = DCCP_TIMEWAIT_LEN;
57 if (state == DCCP_TIME_WAIT)
58 timeo = DCCP_TIMEWAIT_LEN;
59
60 inet_twsk_schedule(tw, timeo);
61 /* Linkage updates. */
62 __inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
63 inet_twsk_put(tw);
64 } else {
65 /* Sorry, if we're out of memory, just CLOSE this
66 * socket up. We've got bigger problems than
67 * non-graceful socket closings.
68 */
69 DCCP_WARN("time wait bucket table overflow\n");
70 }
71
72 dccp_done(sk);
73}
74
75struct sock *dccp_create_openreq_child(const struct sock *sk,
76 const struct request_sock *req,
77 const struct sk_buff *skb)
78{
79 /*
80 * Step 3: Process LISTEN state
81 *
82 * (* Generate a new socket and switch to that socket *)
83 * Set S := new socket for this port pair
84 */
85 struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
86
87 if (newsk != NULL) {
88 struct dccp_request_sock *dreq = dccp_rsk(req);
89 struct inet_connection_sock *newicsk = inet_csk(newsk);
90 struct dccp_sock *newdp = dccp_sk(newsk);
91
92 newdp->dccps_role = DCCP_ROLE_SERVER;
93 newdp->dccps_hc_rx_ackvec = NULL;
94 newdp->dccps_service_list = NULL;
95 newdp->dccps_service = dreq->dreq_service;
96 newdp->dccps_timestamp_echo = dreq->dreq_timestamp_echo;
97 newdp->dccps_timestamp_time = dreq->dreq_timestamp_time;
98 newicsk->icsk_rto = DCCP_TIMEOUT_INIT;
99
100 INIT_LIST_HEAD(&newdp->dccps_featneg);
101 /*
102 * Step 3: Process LISTEN state
103 *
104 * Choose S.ISS (initial seqno) or set from Init Cookies
105 * Initialize S.GAR := S.ISS
106 * Set S.ISR, S.GSR from packet (or Init Cookies)
107 *
108 * Setting AWL/AWH and SWL/SWH happens as part of the feature
109 * activation below, as these windows all depend on the local
110 * and remote Sequence Window feature values (7.5.2).
111 */
112 newdp->dccps_iss = dreq->dreq_iss;
113 newdp->dccps_gss = dreq->dreq_gss;
114 newdp->dccps_gar = newdp->dccps_iss;
115 newdp->dccps_isr = dreq->dreq_isr;
116 newdp->dccps_gsr = dreq->dreq_gsr;
117
118 /*
119 * Activate features: initialise CCIDs, sequence windows etc.
120 */
121 if (dccp_feat_activate_values(newsk, &dreq->dreq_featneg)) {
122 /* It is still raw copy of parent, so invalidate
123 * destructor and make plain sk_free() */
124 newsk->sk_destruct = NULL;
125 bh_unlock_sock(newsk);
126 sk_free(newsk);
127 return NULL;
128 }
129 dccp_init_xmit_timers(newsk);
130
131 __DCCP_INC_STATS(DCCP_MIB_PASSIVEOPENS);
132 }
133 return newsk;
134}
135
136EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
137
138/*
139 * Process an incoming packet for RESPOND sockets represented
140 * as an request_sock.
141 */
142struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
143 struct request_sock *req)
144{
145 struct sock *child = NULL;
146 struct dccp_request_sock *dreq = dccp_rsk(req);
147 bool own_req;
148
149 /* TCP/DCCP listeners became lockless.
150 * DCCP stores complex state in its request_sock, so we need
151 * a protection for them, now this code runs without being protected
152 * by the parent (listener) lock.
153 */
154 spin_lock_bh(&dreq->dreq_lock);
155
156 /* Check for retransmitted REQUEST */
157 if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
158
159 if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_gsr)) {
160 dccp_pr_debug("Retransmitted REQUEST\n");
161 dreq->dreq_gsr = DCCP_SKB_CB(skb)->dccpd_seq;
162 /*
163 * Send another RESPONSE packet
164 * To protect against Request floods, increment retrans
165 * counter (backoff, monitored by dccp_response_timer).
166 */
167 inet_rtx_syn_ack(sk, req);
168 }
169 /* Network Duplicate, discard packet */
170 goto out;
171 }
172
173 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
174
175 if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
176 dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
177 goto drop;
178
179 /* Invalid ACK */
180 if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
181 dreq->dreq_iss, dreq->dreq_gss)) {
182 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
183 "dreq_iss=%llu, dreq_gss=%llu\n",
184 (unsigned long long)
185 DCCP_SKB_CB(skb)->dccpd_ack_seq,
186 (unsigned long long) dreq->dreq_iss,
187 (unsigned long long) dreq->dreq_gss);
188 goto drop;
189 }
190
191 if (dccp_parse_options(sk, dreq, skb))
192 goto drop;
193
194 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
195 req, &own_req);
196 if (child) {
197 child = inet_csk_complete_hashdance(sk, child, req, own_req);
198 goto out;
199 }
200
201 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
202drop:
203 if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
204 req->rsk_ops->send_reset(sk, skb);
205
206 inet_csk_reqsk_queue_drop(sk, req);
207out:
208 spin_unlock_bh(&dreq->dreq_lock);
209 return child;
210}
211
212EXPORT_SYMBOL_GPL(dccp_check_req);
213
214/*
215 * Queue segment on the new socket if the new socket is active,
216 * otherwise we just shortcircuit this and continue with
217 * the new socket.
218 */
219int dccp_child_process(struct sock *parent, struct sock *child,
220 struct sk_buff *skb)
221{
222 int ret = 0;
223 const int state = child->sk_state;
224
225 if (!sock_owned_by_user(child)) {
226 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
227 skb->len);
228
229 /* Wakeup parent, send SIGIO */
230 if (state == DCCP_RESPOND && child->sk_state != state)
231 parent->sk_data_ready(parent);
232 } else {
233 /* Alas, it is possible again, because we do lookup
234 * in main socket hash table and lock on listening
235 * socket does not protect us more.
236 */
237 __sk_add_backlog(child, skb);
238 }
239
240 bh_unlock_sock(child);
241 sock_put(child);
242 return ret;
243}
244
245EXPORT_SYMBOL_GPL(dccp_child_process);
246
247void dccp_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
248 struct request_sock *rsk)
249{
250 DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state");
251}
252
253EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack);
254
255int dccp_reqsk_init(struct request_sock *req,
256 struct dccp_sock const *dp, struct sk_buff const *skb)
257{
258 struct dccp_request_sock *dreq = dccp_rsk(req);
259
260 spin_lock_init(&dreq->dreq_lock);
261 inet_rsk(req)->ir_rmt_port = dccp_hdr(skb)->dccph_sport;
262 inet_rsk(req)->ir_num = ntohs(dccp_hdr(skb)->dccph_dport);
263 inet_rsk(req)->acked = 0;
264 dreq->dreq_timestamp_echo = 0;
265
266 /* inherit feature negotiation options from listening socket */
267 return dccp_feat_clone_list(&dp->dccps_featneg, &dreq->dreq_featneg);
268}
269
270EXPORT_SYMBOL_GPL(dccp_reqsk_init);