Loading...
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 IS_ENABLED(CONFIG_IPV6)
57 if (tw->tw_family == PF_INET6) {
58 const struct ipv6_pinfo *np = inet6_sk(sk);
59
60 tw->tw_v6_daddr = sk->sk_v6_daddr;
61 tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
62 tw->tw_ipv6only = np->ipv6only;
63 }
64#endif
65 /* Linkage updates. */
66 __inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
67
68 /* Get the TIME_WAIT timeout firing. */
69 if (timeo < rto)
70 timeo = rto;
71
72 tw->tw_timeout = DCCP_TIMEWAIT_LEN;
73 if (state == DCCP_TIME_WAIT)
74 timeo = DCCP_TIMEWAIT_LEN;
75
76 inet_twsk_schedule(tw, &dccp_death_row, timeo,
77 DCCP_TIMEWAIT_LEN);
78 inet_twsk_put(tw);
79 } else {
80 /* Sorry, if we're out of memory, just CLOSE this
81 * socket up. We've got bigger problems than
82 * non-graceful socket closings.
83 */
84 DCCP_WARN("time wait bucket table overflow\n");
85 }
86
87 dccp_done(sk);
88}
89
90struct sock *dccp_create_openreq_child(struct sock *sk,
91 const struct request_sock *req,
92 const struct sk_buff *skb)
93{
94 /*
95 * Step 3: Process LISTEN state
96 *
97 * (* Generate a new socket and switch to that socket *)
98 * Set S := new socket for this port pair
99 */
100 struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
101
102 if (newsk != NULL) {
103 struct dccp_request_sock *dreq = dccp_rsk(req);
104 struct inet_connection_sock *newicsk = inet_csk(newsk);
105 struct dccp_sock *newdp = dccp_sk(newsk);
106
107 newdp->dccps_role = DCCP_ROLE_SERVER;
108 newdp->dccps_hc_rx_ackvec = NULL;
109 newdp->dccps_service_list = NULL;
110 newdp->dccps_service = dreq->dreq_service;
111 newdp->dccps_timestamp_echo = dreq->dreq_timestamp_echo;
112 newdp->dccps_timestamp_time = dreq->dreq_timestamp_time;
113 newicsk->icsk_rto = DCCP_TIMEOUT_INIT;
114
115 INIT_LIST_HEAD(&newdp->dccps_featneg);
116 /*
117 * Step 3: Process LISTEN state
118 *
119 * Choose S.ISS (initial seqno) or set from Init Cookies
120 * Initialize S.GAR := S.ISS
121 * Set S.ISR, S.GSR from packet (or Init Cookies)
122 *
123 * Setting AWL/AWH and SWL/SWH happens as part of the feature
124 * activation below, as these windows all depend on the local
125 * and remote Sequence Window feature values (7.5.2).
126 */
127 newdp->dccps_iss = dreq->dreq_iss;
128 newdp->dccps_gss = dreq->dreq_gss;
129 newdp->dccps_gar = newdp->dccps_iss;
130 newdp->dccps_isr = dreq->dreq_isr;
131 newdp->dccps_gsr = dreq->dreq_gsr;
132
133 /*
134 * Activate features: initialise CCIDs, sequence windows etc.
135 */
136 if (dccp_feat_activate_values(newsk, &dreq->dreq_featneg)) {
137 /* It is still raw copy of parent, so invalidate
138 * destructor and make plain sk_free() */
139 newsk->sk_destruct = NULL;
140 sk_free(newsk);
141 return NULL;
142 }
143 dccp_init_xmit_timers(newsk);
144
145 DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS);
146 }
147 return newsk;
148}
149
150EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
151
152/*
153 * Process an incoming packet for RESPOND sockets represented
154 * as an request_sock.
155 */
156struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
157 struct request_sock *req,
158 struct request_sock **prev)
159{
160 struct sock *child = NULL;
161 struct dccp_request_sock *dreq = dccp_rsk(req);
162
163 /* Check for retransmitted REQUEST */
164 if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
165
166 if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_gsr)) {
167 dccp_pr_debug("Retransmitted REQUEST\n");
168 dreq->dreq_gsr = DCCP_SKB_CB(skb)->dccpd_seq;
169 /*
170 * Send another RESPONSE packet
171 * To protect against Request floods, increment retrans
172 * counter (backoff, monitored by dccp_response_timer).
173 */
174 inet_rtx_syn_ack(sk, req);
175 }
176 /* Network Duplicate, discard packet */
177 return NULL;
178 }
179
180 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
181
182 if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
183 dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
184 goto drop;
185
186 /* Invalid ACK */
187 if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
188 dreq->dreq_iss, dreq->dreq_gss)) {
189 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
190 "dreq_iss=%llu, dreq_gss=%llu\n",
191 (unsigned long long)
192 DCCP_SKB_CB(skb)->dccpd_ack_seq,
193 (unsigned long long) dreq->dreq_iss,
194 (unsigned long long) dreq->dreq_gss);
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);
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)->ir_rmt_port = dccp_hdr(skb)->dccph_sport;
270 inet_rsk(req)->ir_num = ntohs(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);
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 .tw_refcount = REFCOUNT_INIT(1),
26 .sysctl_max_tw_buckets = NR_FILE * 2,
27 .hashinfo = &dccp_hashinfo,
28};
29
30EXPORT_SYMBOL_GPL(dccp_death_row);
31
32void dccp_time_wait(struct sock *sk, int state, int timeo)
33{
34 struct inet_timewait_sock *tw;
35
36 tw = inet_twsk_alloc(sk, &dccp_death_row, state);
37
38 if (tw != NULL) {
39 const struct inet_connection_sock *icsk = inet_csk(sk);
40 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
41#if IS_ENABLED(CONFIG_IPV6)
42 if (tw->tw_family == PF_INET6) {
43 tw->tw_v6_daddr = sk->sk_v6_daddr;
44 tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
45 tw->tw_ipv6only = sk->sk_ipv6only;
46 }
47#endif
48
49 /* Get the TIME_WAIT timeout firing. */
50 if (timeo < rto)
51 timeo = rto;
52
53 if (state == DCCP_TIME_WAIT)
54 timeo = DCCP_TIMEWAIT_LEN;
55
56 /* tw_timer is pinned, so we need to make sure BH are disabled
57 * in following section, otherwise timer handler could run before
58 * we complete the initialization.
59 */
60 local_bh_disable();
61 inet_twsk_schedule(tw, timeo);
62 /* Linkage updates.
63 * Note that access to tw after this point is illegal.
64 */
65 inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
66 local_bh_enable();
67 } else {
68 /* Sorry, if we're out of memory, just CLOSE this
69 * socket up. We've got bigger problems than
70 * non-graceful socket closings.
71 */
72 DCCP_WARN("time wait bucket table overflow\n");
73 }
74
75 dccp_done(sk);
76}
77
78struct sock *dccp_create_openreq_child(const struct sock *sk,
79 const struct request_sock *req,
80 const struct sk_buff *skb)
81{
82 /*
83 * Step 3: Process LISTEN state
84 *
85 * (* Generate a new socket and switch to that socket *)
86 * Set S := new socket for this port pair
87 */
88 struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
89
90 if (newsk != NULL) {
91 struct dccp_request_sock *dreq = dccp_rsk(req);
92 struct inet_connection_sock *newicsk = inet_csk(newsk);
93 struct dccp_sock *newdp = dccp_sk(newsk);
94
95 newdp->dccps_role = DCCP_ROLE_SERVER;
96 newdp->dccps_hc_rx_ackvec = NULL;
97 newdp->dccps_service_list = NULL;
98 newdp->dccps_hc_rx_ccid = NULL;
99 newdp->dccps_hc_tx_ccid = NULL;
100 newdp->dccps_service = dreq->dreq_service;
101 newdp->dccps_timestamp_echo = dreq->dreq_timestamp_echo;
102 newdp->dccps_timestamp_time = dreq->dreq_timestamp_time;
103 newicsk->icsk_rto = DCCP_TIMEOUT_INIT;
104
105 INIT_LIST_HEAD(&newdp->dccps_featneg);
106 /*
107 * Step 3: Process LISTEN state
108 *
109 * Choose S.ISS (initial seqno) or set from Init Cookies
110 * Initialize S.GAR := S.ISS
111 * Set S.ISR, S.GSR from packet (or Init Cookies)
112 *
113 * Setting AWL/AWH and SWL/SWH happens as part of the feature
114 * activation below, as these windows all depend on the local
115 * and remote Sequence Window feature values (7.5.2).
116 */
117 newdp->dccps_iss = dreq->dreq_iss;
118 newdp->dccps_gss = dreq->dreq_gss;
119 newdp->dccps_gar = newdp->dccps_iss;
120 newdp->dccps_isr = dreq->dreq_isr;
121 newdp->dccps_gsr = dreq->dreq_gsr;
122
123 /*
124 * Activate features: initialise CCIDs, sequence windows etc.
125 */
126 if (dccp_feat_activate_values(newsk, &dreq->dreq_featneg)) {
127 sk_free_unlock_clone(newsk);
128 return NULL;
129 }
130 dccp_init_xmit_timers(newsk);
131
132 __DCCP_INC_STATS(DCCP_MIB_PASSIVEOPENS);
133 }
134 return newsk;
135}
136
137EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
138
139/*
140 * Process an incoming packet for RESPOND sockets represented
141 * as an request_sock.
142 */
143struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
144 struct request_sock *req)
145{
146 struct sock *child = NULL;
147 struct dccp_request_sock *dreq = dccp_rsk(req);
148 bool own_req;
149
150 /* TCP/DCCP listeners became lockless.
151 * DCCP stores complex state in its request_sock, so we need
152 * a protection for them, now this code runs without being protected
153 * by the parent (listener) lock.
154 */
155 spin_lock_bh(&dreq->dreq_lock);
156
157 /* Check for retransmitted REQUEST */
158 if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
159
160 if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_gsr)) {
161 dccp_pr_debug("Retransmitted REQUEST\n");
162 dreq->dreq_gsr = DCCP_SKB_CB(skb)->dccpd_seq;
163 /*
164 * Send another RESPONSE packet
165 * To protect against Request floods, increment retrans
166 * counter (backoff, monitored by dccp_response_timer).
167 */
168 inet_rtx_syn_ack(sk, req);
169 }
170 /* Network Duplicate, discard packet */
171 goto out;
172 }
173
174 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
175
176 if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
177 dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
178 goto drop;
179
180 /* Invalid ACK */
181 if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
182 dreq->dreq_iss, dreq->dreq_gss)) {
183 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
184 "dreq_iss=%llu, dreq_gss=%llu\n",
185 (unsigned long long)
186 DCCP_SKB_CB(skb)->dccpd_ack_seq,
187 (unsigned long long) dreq->dreq_iss,
188 (unsigned long long) dreq->dreq_gss);
189 goto drop;
190 }
191
192 if (dccp_parse_options(sk, dreq, skb))
193 goto drop;
194
195 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
196 req, &own_req);
197 if (child) {
198 child = inet_csk_complete_hashdance(sk, child, req, own_req);
199 goto out;
200 }
201
202 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
203drop:
204 if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
205 req->rsk_ops->send_reset(sk, skb);
206
207 inet_csk_reqsk_queue_drop(sk, req);
208out:
209 spin_unlock_bh(&dreq->dreq_lock);
210 return child;
211}
212
213EXPORT_SYMBOL_GPL(dccp_check_req);
214
215/*
216 * Queue segment on the new socket if the new socket is active,
217 * otherwise we just shortcircuit this and continue with
218 * the new socket.
219 */
220int dccp_child_process(struct sock *parent, struct sock *child,
221 struct sk_buff *skb)
222 __releases(child)
223{
224 int ret = 0;
225 const int state = child->sk_state;
226
227 if (!sock_owned_by_user(child)) {
228 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
229 skb->len);
230
231 /* Wakeup parent, send SIGIO */
232 if (state == DCCP_RESPOND && child->sk_state != state)
233 parent->sk_data_ready(parent);
234 } else {
235 /* Alas, it is possible again, because we do lookup
236 * in main socket hash table and lock on listening
237 * socket does not protect us more.
238 */
239 __sk_add_backlog(child, skb);
240 }
241
242 bh_unlock_sock(child);
243 sock_put(child);
244 return ret;
245}
246
247EXPORT_SYMBOL_GPL(dccp_child_process);
248
249void dccp_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
250 struct request_sock *rsk)
251{
252 DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state");
253}
254
255EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack);
256
257int dccp_reqsk_init(struct request_sock *req,
258 struct dccp_sock const *dp, struct sk_buff const *skb)
259{
260 struct dccp_request_sock *dreq = dccp_rsk(req);
261
262 spin_lock_init(&dreq->dreq_lock);
263 inet_rsk(req)->ir_rmt_port = dccp_hdr(skb)->dccph_sport;
264 inet_rsk(req)->ir_num = ntohs(dccp_hdr(skb)->dccph_dport);
265 inet_rsk(req)->acked = 0;
266 dreq->dreq_timestamp_echo = 0;
267
268 /* inherit feature negotiation options from listening socket */
269 return dccp_feat_clone_list(&dp->dccps_featneg, &dreq->dreq_featneg);
270}
271
272EXPORT_SYMBOL_GPL(dccp_reqsk_init);