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_service = dreq->dreq_service;
98 newdp->dccps_timestamp_echo = dreq->dreq_timestamp_echo;
99 newdp->dccps_timestamp_time = dreq->dreq_timestamp_time;
100 newicsk->icsk_rto = DCCP_TIMEOUT_INIT;
101
102 INIT_LIST_HEAD(&newdp->dccps_featneg);
103 /*
104 * Step 3: Process LISTEN state
105 *
106 * Choose S.ISS (initial seqno) or set from Init Cookies
107 * Initialize S.GAR := S.ISS
108 * Set S.ISR, S.GSR from packet (or Init Cookies)
109 *
110 * Setting AWL/AWH and SWL/SWH happens as part of the feature
111 * activation below, as these windows all depend on the local
112 * and remote Sequence Window feature values (7.5.2).
113 */
114 newdp->dccps_iss = dreq->dreq_iss;
115 newdp->dccps_gss = dreq->dreq_gss;
116 newdp->dccps_gar = newdp->dccps_iss;
117 newdp->dccps_isr = dreq->dreq_isr;
118 newdp->dccps_gsr = dreq->dreq_gsr;
119
120 /*
121 * Activate features: initialise CCIDs, sequence windows etc.
122 */
123 if (dccp_feat_activate_values(newsk, &dreq->dreq_featneg)) {
124 sk_free_unlock_clone(newsk);
125 return NULL;
126 }
127 dccp_init_xmit_timers(newsk);
128
129 __DCCP_INC_STATS(DCCP_MIB_PASSIVEOPENS);
130 }
131 return newsk;
132}
133
134EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
135
136/*
137 * Process an incoming packet for RESPOND sockets represented
138 * as an request_sock.
139 */
140struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
141 struct request_sock *req)
142{
143 struct sock *child = NULL;
144 struct dccp_request_sock *dreq = dccp_rsk(req);
145 bool own_req;
146
147 /* TCP/DCCP listeners became lockless.
148 * DCCP stores complex state in its request_sock, so we need
149 * a protection for them, now this code runs without being protected
150 * by the parent (listener) lock.
151 */
152 spin_lock_bh(&dreq->dreq_lock);
153
154 /* Check for retransmitted REQUEST */
155 if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
156
157 if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_gsr)) {
158 dccp_pr_debug("Retransmitted REQUEST\n");
159 dreq->dreq_gsr = DCCP_SKB_CB(skb)->dccpd_seq;
160 /*
161 * Send another RESPONSE packet
162 * To protect against Request floods, increment retrans
163 * counter (backoff, monitored by dccp_response_timer).
164 */
165 inet_rtx_syn_ack(sk, req);
166 }
167 /* Network Duplicate, discard packet */
168 goto out;
169 }
170
171 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
172
173 if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
174 dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
175 goto drop;
176
177 /* Invalid ACK */
178 if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
179 dreq->dreq_iss, dreq->dreq_gss)) {
180 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
181 "dreq_iss=%llu, dreq_gss=%llu\n",
182 (unsigned long long)
183 DCCP_SKB_CB(skb)->dccpd_ack_seq,
184 (unsigned long long) dreq->dreq_iss,
185 (unsigned long long) dreq->dreq_gss);
186 goto drop;
187 }
188
189 if (dccp_parse_options(sk, dreq, skb))
190 goto drop;
191
192 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
193 req, &own_req);
194 if (child) {
195 child = inet_csk_complete_hashdance(sk, child, req, own_req);
196 goto out;
197 }
198
199 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
200drop:
201 if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
202 req->rsk_ops->send_reset(sk, skb);
203
204 inet_csk_reqsk_queue_drop(sk, req);
205out:
206 spin_unlock_bh(&dreq->dreq_lock);
207 return child;
208}
209
210EXPORT_SYMBOL_GPL(dccp_check_req);
211
212/*
213 * Queue segment on the new socket if the new socket is active,
214 * otherwise we just shortcircuit this and continue with
215 * the new socket.
216 */
217int dccp_child_process(struct sock *parent, struct sock *child,
218 struct sk_buff *skb)
219{
220 int ret = 0;
221 const int state = child->sk_state;
222
223 if (!sock_owned_by_user(child)) {
224 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
225 skb->len);
226
227 /* Wakeup parent, send SIGIO */
228 if (state == DCCP_RESPOND && child->sk_state != state)
229 parent->sk_data_ready(parent);
230 } else {
231 /* Alas, it is possible again, because we do lookup
232 * in main socket hash table and lock on listening
233 * socket does not protect us more.
234 */
235 __sk_add_backlog(child, skb);
236 }
237
238 bh_unlock_sock(child);
239 sock_put(child);
240 return ret;
241}
242
243EXPORT_SYMBOL_GPL(dccp_child_process);
244
245void dccp_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
246 struct request_sock *rsk)
247{
248 DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state");
249}
250
251EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack);
252
253int dccp_reqsk_init(struct request_sock *req,
254 struct dccp_sock const *dp, struct sk_buff const *skb)
255{
256 struct dccp_request_sock *dreq = dccp_rsk(req);
257
258 spin_lock_init(&dreq->dreq_lock);
259 inet_rsk(req)->ir_rmt_port = dccp_hdr(skb)->dccph_sport;
260 inet_rsk(req)->ir_num = ntohs(dccp_hdr(skb)->dccph_dport);
261 inet_rsk(req)->acked = 0;
262 dreq->dreq_timestamp_echo = 0;
263
264 /* inherit feature negotiation options from listening socket */
265 return dccp_feat_clone_list(&dp->dccps_featneg, &dreq->dreq_featneg);
266}
267
268EXPORT_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 .period = DCCP_TIMEWAIT_LEN / INET_TWDR_TWKILL_SLOTS,
31 .death_lock = __SPIN_LOCK_UNLOCKED(dccp_death_row.death_lock),
32 .hashinfo = &dccp_hashinfo,
33 .tw_timer = TIMER_INITIALIZER(inet_twdr_hangman, 0,
34 (unsigned long)&dccp_death_row),
35 .twkill_work = __WORK_INITIALIZER(dccp_death_row.twkill_work,
36 inet_twdr_twkill_work),
37/* Short-time timewait calendar */
38
39 .twcal_hand = -1,
40 .twcal_timer = TIMER_INITIALIZER(inet_twdr_twcal_tick, 0,
41 (unsigned long)&dccp_death_row),
42};
43
44EXPORT_SYMBOL_GPL(dccp_death_row);
45
46void dccp_time_wait(struct sock *sk, int state, int timeo)
47{
48 struct inet_timewait_sock *tw = NULL;
49
50 if (dccp_death_row.tw_count < dccp_death_row.sysctl_max_tw_buckets)
51 tw = inet_twsk_alloc(sk, state);
52
53 if (tw != NULL) {
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;
60
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;
66 }
67#endif
68 /* Linkage updates. */
69 __inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
70
71 /* Get the TIME_WAIT timeout firing. */
72 if (timeo < rto)
73 timeo = rto;
74
75 tw->tw_timeout = DCCP_TIMEWAIT_LEN;
76 if (state == DCCP_TIME_WAIT)
77 timeo = DCCP_TIMEWAIT_LEN;
78
79 inet_twsk_schedule(tw, &dccp_death_row, timeo,
80 DCCP_TIMEWAIT_LEN);
81 inet_twsk_put(tw);
82 } else {
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.
86 */
87 DCCP_WARN("time wait bucket table overflow\n");
88 }
89
90 dccp_done(sk);
91}
92
93struct sock *dccp_create_openreq_child(struct sock *sk,
94 const struct request_sock *req,
95 const struct sk_buff *skb)
96{
97 /*
98 * Step 3: Process LISTEN state
99 *
100 * (* Generate a new socket and switch to that socket *)
101 * Set S := new socket for this port pair
102 */
103 struct sock *newsk = inet_csk_clone(sk, req, GFP_ATOMIC);
104
105 if (newsk != NULL) {
106 struct dccp_request_sock *dreq = dccp_rsk(req);
107 struct inet_connection_sock *newicsk = inet_csk(newsk);
108 struct dccp_sock *newdp = dccp_sk(newsk);
109
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 newdp->dccps_timestamp_echo = dreq->dreq_timestamp_echo;
115 newdp->dccps_timestamp_time = dreq->dreq_timestamp_time;
116 newicsk->icsk_rto = DCCP_TIMEOUT_INIT;
117
118 INIT_LIST_HEAD(&newdp->dccps_featneg);
119 /*
120 * Step 3: Process LISTEN state
121 *
122 * Choose S.ISS (initial seqno) or set from Init Cookies
123 * Initialize S.GAR := S.ISS
124 * Set S.ISR, S.GSR from packet (or Init Cookies)
125 *
126 * Setting AWL/AWH and SWL/SWH happens as part of the feature
127 * activation below, as these windows all depend on the local
128 * and remote Sequence Window feature values (7.5.2).
129 */
130 newdp->dccps_gss = newdp->dccps_iss = dreq->dreq_iss;
131 newdp->dccps_gar = newdp->dccps_iss;
132 newdp->dccps_gsr = newdp->dccps_isr = dreq->dreq_isr;
133
134 /*
135 * Activate features: initialise CCIDs, sequence windows etc.
136 */
137 if (dccp_feat_activate_values(newsk, &dreq->dreq_featneg)) {
138 /* It is still raw copy of parent, so invalidate
139 * destructor and make plain sk_free() */
140 newsk->sk_destruct = NULL;
141 sk_free(newsk);
142 return NULL;
143 }
144 dccp_init_xmit_timers(newsk);
145
146 DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS);
147 }
148 return newsk;
149}
150
151EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
152
153/*
154 * Process an incoming packet for RESPOND sockets represented
155 * as an request_sock.
156 */
157struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
158 struct request_sock *req,
159 struct request_sock **prev)
160{
161 struct sock *child = NULL;
162 struct dccp_request_sock *dreq = dccp_rsk(req);
163
164 /* Check for retransmitted REQUEST */
165 if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
166
167 if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_isr)) {
168 dccp_pr_debug("Retransmitted REQUEST\n");
169 dreq->dreq_isr = DCCP_SKB_CB(skb)->dccpd_seq;
170 /*
171 * Send another RESPONSE packet
172 * To protect against Request floods, increment retrans
173 * counter (backoff, monitored by dccp_response_timer).
174 */
175 req->retrans++;
176 req->rsk_ops->rtx_syn_ack(sk, req, NULL);
177 }
178 /* Network Duplicate, discard packet */
179 return NULL;
180 }
181
182 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
183
184 if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
185 dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
186 goto drop;
187
188 /* Invalid ACK */
189 if (DCCP_SKB_CB(skb)->dccpd_ack_seq != dreq->dreq_iss) {
190 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
191 "dreq_iss=%llu\n",
192 (unsigned long long)
193 DCCP_SKB_CB(skb)->dccpd_ack_seq,
194 (unsigned long long) dreq->dreq_iss);
195 goto drop;
196 }
197
198 if (dccp_parse_options(sk, dreq, skb))
199 goto drop;
200
201 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
202 if (child == NULL)
203 goto listen_overflow;
204
205 inet_csk_reqsk_queue_unlink(sk, req, prev);
206 inet_csk_reqsk_queue_removed(sk, req);
207 inet_csk_reqsk_queue_add(sk, req, child);
208out:
209 return child;
210listen_overflow:
211 dccp_pr_debug("listen_overflow!\n");
212 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
213drop:
214 if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
215 req->rsk_ops->send_reset(sk, skb);
216
217 inet_csk_reqsk_queue_drop(sk, req, prev);
218 goto out;
219}
220
221EXPORT_SYMBOL_GPL(dccp_check_req);
222
223/*
224 * Queue segment on the new socket if the new socket is active,
225 * otherwise we just shortcircuit this and continue with
226 * the new socket.
227 */
228int dccp_child_process(struct sock *parent, struct sock *child,
229 struct sk_buff *skb)
230{
231 int ret = 0;
232 const int state = child->sk_state;
233
234 if (!sock_owned_by_user(child)) {
235 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
236 skb->len);
237
238 /* Wakeup parent, send SIGIO */
239 if (state == DCCP_RESPOND && child->sk_state != state)
240 parent->sk_data_ready(parent, 0);
241 } else {
242 /* Alas, it is possible again, because we do lookup
243 * in main socket hash table and lock on listening
244 * socket does not protect us more.
245 */
246 __sk_add_backlog(child, skb);
247 }
248
249 bh_unlock_sock(child);
250 sock_put(child);
251 return ret;
252}
253
254EXPORT_SYMBOL_GPL(dccp_child_process);
255
256void dccp_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
257 struct request_sock *rsk)
258{
259 DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state");
260}
261
262EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack);
263
264int dccp_reqsk_init(struct request_sock *req,
265 struct dccp_sock const *dp, struct sk_buff const *skb)
266{
267 struct dccp_request_sock *dreq = dccp_rsk(req);
268
269 inet_rsk(req)->rmt_port = dccp_hdr(skb)->dccph_sport;
270 inet_rsk(req)->loc_port = dccp_hdr(skb)->dccph_dport;
271 inet_rsk(req)->acked = 0;
272 dreq->dreq_timestamp_echo = 0;
273
274 /* inherit feature negotiation options from listening socket */
275 return dccp_feat_clone_list(&dp->dccps_featneg, &dreq->dreq_featneg);
276}
277
278EXPORT_SYMBOL_GPL(dccp_reqsk_init);