1 |
3 |
xianfeng |
/*
|
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/kernel.h>
|
15 |
|
|
#include <linux/skbuff.h>
|
16 |
|
|
#include <linux/timer.h>
|
17 |
|
|
|
18 |
|
|
#include <net/sock.h>
|
19 |
|
|
#include <net/xfrm.h>
|
20 |
|
|
#include <net/inet_timewait_sock.h>
|
21 |
|
|
|
22 |
|
|
#include "ackvec.h"
|
23 |
|
|
#include "ccid.h"
|
24 |
|
|
#include "dccp.h"
|
25 |
|
|
#include "feat.h"
|
26 |
|
|
|
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 */
|
37 |
|
|
|
38 |
|
|
.twcal_hand = -1,
|
39 |
|
|
.twcal_timer = TIMER_INITIALIZER(inet_twdr_twcal_tick, 0,
|
40 |
|
|
(unsigned long)&dccp_death_row),
|
41 |
|
|
};
|
42 |
|
|
|
43 |
|
|
EXPORT_SYMBOL_GPL(dccp_death_row);
|
44 |
|
|
|
45 |
|
|
void dccp_minisock_init(struct dccp_minisock *dmsk)
|
46 |
|
|
{
|
47 |
|
|
dmsk->dccpms_sequence_window = sysctl_dccp_feat_sequence_window;
|
48 |
|
|
dmsk->dccpms_rx_ccid = sysctl_dccp_feat_rx_ccid;
|
49 |
|
|
dmsk->dccpms_tx_ccid = sysctl_dccp_feat_tx_ccid;
|
50 |
|
|
dmsk->dccpms_ack_ratio = sysctl_dccp_feat_ack_ratio;
|
51 |
|
|
dmsk->dccpms_send_ack_vector = sysctl_dccp_feat_send_ack_vector;
|
52 |
|
|
dmsk->dccpms_send_ndp_count = sysctl_dccp_feat_send_ndp_count;
|
53 |
|
|
}
|
54 |
|
|
|
55 |
|
|
void dccp_time_wait(struct sock *sk, int state, int timeo)
|
56 |
|
|
{
|
57 |
|
|
struct inet_timewait_sock *tw = NULL;
|
58 |
|
|
|
59 |
|
|
if (dccp_death_row.tw_count < dccp_death_row.sysctl_max_tw_buckets)
|
60 |
|
|
tw = inet_twsk_alloc(sk, state);
|
61 |
|
|
|
62 |
|
|
if (tw != NULL) {
|
63 |
|
|
const struct inet_connection_sock *icsk = inet_csk(sk);
|
64 |
|
|
const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
|
65 |
|
|
#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
|
66 |
|
|
if (tw->tw_family == PF_INET6) {
|
67 |
|
|
const struct ipv6_pinfo *np = inet6_sk(sk);
|
68 |
|
|
struct inet6_timewait_sock *tw6;
|
69 |
|
|
|
70 |
|
|
tw->tw_ipv6_offset = inet6_tw_offset(sk->sk_prot);
|
71 |
|
|
tw6 = inet6_twsk((struct sock *)tw);
|
72 |
|
|
ipv6_addr_copy(&tw6->tw_v6_daddr, &np->daddr);
|
73 |
|
|
ipv6_addr_copy(&tw6->tw_v6_rcv_saddr, &np->rcv_saddr);
|
74 |
|
|
tw->tw_ipv6only = np->ipv6only;
|
75 |
|
|
}
|
76 |
|
|
#endif
|
77 |
|
|
/* Linkage updates. */
|
78 |
|
|
__inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
|
79 |
|
|
|
80 |
|
|
/* Get the TIME_WAIT timeout firing. */
|
81 |
|
|
if (timeo < rto)
|
82 |
|
|
timeo = rto;
|
83 |
|
|
|
84 |
|
|
tw->tw_timeout = DCCP_TIMEWAIT_LEN;
|
85 |
|
|
if (state == DCCP_TIME_WAIT)
|
86 |
|
|
timeo = DCCP_TIMEWAIT_LEN;
|
87 |
|
|
|
88 |
|
|
inet_twsk_schedule(tw, &dccp_death_row, timeo,
|
89 |
|
|
DCCP_TIMEWAIT_LEN);
|
90 |
|
|
inet_twsk_put(tw);
|
91 |
|
|
} else {
|
92 |
|
|
/* Sorry, if we're out of memory, just CLOSE this
|
93 |
|
|
* socket up. We've got bigger problems than
|
94 |
|
|
* non-graceful socket closings.
|
95 |
|
|
*/
|
96 |
|
|
DCCP_WARN("time wait bucket table overflow\n");
|
97 |
|
|
}
|
98 |
|
|
|
99 |
|
|
dccp_done(sk);
|
100 |
|
|
}
|
101 |
|
|
|
102 |
|
|
struct sock *dccp_create_openreq_child(struct sock *sk,
|
103 |
|
|
const struct request_sock *req,
|
104 |
|
|
const struct sk_buff *skb)
|
105 |
|
|
{
|
106 |
|
|
/*
|
107 |
|
|
* Step 3: Process LISTEN state
|
108 |
|
|
*
|
109 |
|
|
* (* Generate a new socket and switch to that socket *)
|
110 |
|
|
* Set S := new socket for this port pair
|
111 |
|
|
*/
|
112 |
|
|
struct sock *newsk = inet_csk_clone(sk, req, GFP_ATOMIC);
|
113 |
|
|
|
114 |
|
|
if (newsk != NULL) {
|
115 |
|
|
const struct dccp_request_sock *dreq = dccp_rsk(req);
|
116 |
|
|
struct inet_connection_sock *newicsk = inet_csk(newsk);
|
117 |
|
|
struct dccp_sock *newdp = dccp_sk(newsk);
|
118 |
|
|
struct dccp_minisock *newdmsk = dccp_msk(newsk);
|
119 |
|
|
|
120 |
|
|
newdp->dccps_role = DCCP_ROLE_SERVER;
|
121 |
|
|
newdp->dccps_hc_rx_ackvec = NULL;
|
122 |
|
|
newdp->dccps_service_list = NULL;
|
123 |
|
|
newdp->dccps_service = dreq->dreq_service;
|
124 |
|
|
newicsk->icsk_rto = DCCP_TIMEOUT_INIT;
|
125 |
|
|
|
126 |
|
|
if (dccp_feat_clone(sk, newsk))
|
127 |
|
|
goto out_free;
|
128 |
|
|
|
129 |
|
|
if (newdmsk->dccpms_send_ack_vector) {
|
130 |
|
|
newdp->dccps_hc_rx_ackvec =
|
131 |
|
|
dccp_ackvec_alloc(GFP_ATOMIC);
|
132 |
|
|
if (unlikely(newdp->dccps_hc_rx_ackvec == NULL))
|
133 |
|
|
goto out_free;
|
134 |
|
|
}
|
135 |
|
|
|
136 |
|
|
newdp->dccps_hc_rx_ccid =
|
137 |
|
|
ccid_hc_rx_new(newdmsk->dccpms_rx_ccid,
|
138 |
|
|
newsk, GFP_ATOMIC);
|
139 |
|
|
newdp->dccps_hc_tx_ccid =
|
140 |
|
|
ccid_hc_tx_new(newdmsk->dccpms_tx_ccid,
|
141 |
|
|
newsk, GFP_ATOMIC);
|
142 |
|
|
if (unlikely(newdp->dccps_hc_rx_ccid == NULL ||
|
143 |
|
|
newdp->dccps_hc_tx_ccid == NULL)) {
|
144 |
|
|
dccp_ackvec_free(newdp->dccps_hc_rx_ackvec);
|
145 |
|
|
ccid_hc_rx_delete(newdp->dccps_hc_rx_ccid, newsk);
|
146 |
|
|
ccid_hc_tx_delete(newdp->dccps_hc_tx_ccid, newsk);
|
147 |
|
|
out_free:
|
148 |
|
|
/* It is still raw copy of parent, so invalidate
|
149 |
|
|
* destructor and make plain sk_free() */
|
150 |
|
|
newsk->sk_destruct = NULL;
|
151 |
|
|
sk_free(newsk);
|
152 |
|
|
return NULL;
|
153 |
|
|
}
|
154 |
|
|
|
155 |
|
|
/*
|
156 |
|
|
* Step 3: Process LISTEN state
|
157 |
|
|
*
|
158 |
|
|
* Choose S.ISS (initial seqno) or set from Init Cookies
|
159 |
|
|
* Initialize S.GAR := S.ISS
|
160 |
|
|
* Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
|
161 |
|
|
*/
|
162 |
|
|
|
163 |
|
|
/* See dccp_v4_conn_request */
|
164 |
|
|
newdmsk->dccpms_sequence_window = req->rcv_wnd;
|
165 |
|
|
|
166 |
|
|
newdp->dccps_gar = newdp->dccps_isr = dreq->dreq_isr;
|
167 |
|
|
dccp_update_gsr(newsk, dreq->dreq_isr);
|
168 |
|
|
|
169 |
|
|
newdp->dccps_iss = dreq->dreq_iss;
|
170 |
|
|
dccp_update_gss(newsk, dreq->dreq_iss);
|
171 |
|
|
|
172 |
|
|
/*
|
173 |
|
|
* SWL and AWL are initially adjusted so that they are not less than
|
174 |
|
|
* the initial Sequence Numbers received and sent, respectively:
|
175 |
|
|
* SWL := max(GSR + 1 - floor(W/4), ISR),
|
176 |
|
|
* AWL := max(GSS - W' + 1, ISS).
|
177 |
|
|
* These adjustments MUST be applied only at the beginning of the
|
178 |
|
|
* connection.
|
179 |
|
|
*/
|
180 |
|
|
dccp_set_seqno(&newdp->dccps_swl,
|
181 |
|
|
max48(newdp->dccps_swl, newdp->dccps_isr));
|
182 |
|
|
dccp_set_seqno(&newdp->dccps_awl,
|
183 |
|
|
max48(newdp->dccps_awl, newdp->dccps_iss));
|
184 |
|
|
|
185 |
|
|
dccp_init_xmit_timers(newsk);
|
186 |
|
|
|
187 |
|
|
DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS);
|
188 |
|
|
}
|
189 |
|
|
return newsk;
|
190 |
|
|
}
|
191 |
|
|
|
192 |
|
|
EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
|
193 |
|
|
|
194 |
|
|
/*
|
195 |
|
|
* Process an incoming packet for RESPOND sockets represented
|
196 |
|
|
* as an request_sock.
|
197 |
|
|
*/
|
198 |
|
|
struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
|
199 |
|
|
struct request_sock *req,
|
200 |
|
|
struct request_sock **prev)
|
201 |
|
|
{
|
202 |
|
|
struct sock *child = NULL;
|
203 |
|
|
|
204 |
|
|
/* Check for retransmitted REQUEST */
|
205 |
|
|
if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
|
206 |
|
|
struct dccp_request_sock *dreq = dccp_rsk(req);
|
207 |
|
|
|
208 |
|
|
if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_isr)) {
|
209 |
|
|
dccp_pr_debug("Retransmitted REQUEST\n");
|
210 |
|
|
dreq->dreq_isr = DCCP_SKB_CB(skb)->dccpd_seq;
|
211 |
|
|
/*
|
212 |
|
|
* Send another RESPONSE packet
|
213 |
|
|
* To protect against Request floods, increment retrans
|
214 |
|
|
* counter (backoff, monitored by dccp_response_timer).
|
215 |
|
|
*/
|
216 |
|
|
req->retrans++;
|
217 |
|
|
req->rsk_ops->rtx_syn_ack(sk, req, NULL);
|
218 |
|
|
}
|
219 |
|
|
/* Network Duplicate, discard packet */
|
220 |
|
|
return NULL;
|
221 |
|
|
}
|
222 |
|
|
|
223 |
|
|
DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
|
224 |
|
|
|
225 |
|
|
if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
|
226 |
|
|
dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
|
227 |
|
|
goto drop;
|
228 |
|
|
|
229 |
|
|
/* Invalid ACK */
|
230 |
|
|
if (DCCP_SKB_CB(skb)->dccpd_ack_seq != dccp_rsk(req)->dreq_iss) {
|
231 |
|
|
dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
|
232 |
|
|
"dreq_iss=%llu\n",
|
233 |
|
|
(unsigned long long)
|
234 |
|
|
DCCP_SKB_CB(skb)->dccpd_ack_seq,
|
235 |
|
|
(unsigned long long)
|
236 |
|
|
dccp_rsk(req)->dreq_iss);
|
237 |
|
|
goto drop;
|
238 |
|
|
}
|
239 |
|
|
|
240 |
|
|
child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
|
241 |
|
|
if (child == NULL)
|
242 |
|
|
goto listen_overflow;
|
243 |
|
|
|
244 |
|
|
/* FIXME: deal with options */
|
245 |
|
|
|
246 |
|
|
inet_csk_reqsk_queue_unlink(sk, req, prev);
|
247 |
|
|
inet_csk_reqsk_queue_removed(sk, req);
|
248 |
|
|
inet_csk_reqsk_queue_add(sk, req, child);
|
249 |
|
|
out:
|
250 |
|
|
return child;
|
251 |
|
|
listen_overflow:
|
252 |
|
|
dccp_pr_debug("listen_overflow!\n");
|
253 |
|
|
DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
|
254 |
|
|
drop:
|
255 |
|
|
if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
|
256 |
|
|
req->rsk_ops->send_reset(sk, skb);
|
257 |
|
|
|
258 |
|
|
inet_csk_reqsk_queue_drop(sk, req, prev);
|
259 |
|
|
goto out;
|
260 |
|
|
}
|
261 |
|
|
|
262 |
|
|
EXPORT_SYMBOL_GPL(dccp_check_req);
|
263 |
|
|
|
264 |
|
|
/*
|
265 |
|
|
* Queue segment on the new socket if the new socket is active,
|
266 |
|
|
* otherwise we just shortcircuit this and continue with
|
267 |
|
|
* the new socket.
|
268 |
|
|
*/
|
269 |
|
|
int dccp_child_process(struct sock *parent, struct sock *child,
|
270 |
|
|
struct sk_buff *skb)
|
271 |
|
|
{
|
272 |
|
|
int ret = 0;
|
273 |
|
|
const int state = child->sk_state;
|
274 |
|
|
|
275 |
|
|
if (!sock_owned_by_user(child)) {
|
276 |
|
|
ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
|
277 |
|
|
skb->len);
|
278 |
|
|
|
279 |
|
|
/* Wakeup parent, send SIGIO */
|
280 |
|
|
if (state == DCCP_RESPOND && child->sk_state != state)
|
281 |
|
|
parent->sk_data_ready(parent, 0);
|
282 |
|
|
} else {
|
283 |
|
|
/* Alas, it is possible again, because we do lookup
|
284 |
|
|
* in main socket hash table and lock on listening
|
285 |
|
|
* socket does not protect us more.
|
286 |
|
|
*/
|
287 |
|
|
sk_add_backlog(child, skb);
|
288 |
|
|
}
|
289 |
|
|
|
290 |
|
|
bh_unlock_sock(child);
|
291 |
|
|
sock_put(child);
|
292 |
|
|
return ret;
|
293 |
|
|
}
|
294 |
|
|
|
295 |
|
|
EXPORT_SYMBOL_GPL(dccp_child_process);
|
296 |
|
|
|
297 |
|
|
void dccp_reqsk_send_ack(struct sk_buff *skb, struct request_sock *rsk)
|
298 |
|
|
{
|
299 |
|
|
DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state");
|
300 |
|
|
}
|
301 |
|
|
|
302 |
|
|
EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack);
|
303 |
|
|
|
304 |
|
|
void dccp_reqsk_init(struct request_sock *req, struct sk_buff *skb)
|
305 |
|
|
{
|
306 |
|
|
inet_rsk(req)->rmt_port = dccp_hdr(skb)->dccph_sport;
|
307 |
|
|
inet_rsk(req)->acked = 0;
|
308 |
|
|
req->rcv_wnd = sysctl_dccp_feat_sequence_window;
|
309 |
|
|
}
|
310 |
|
|
|
311 |
|
|
EXPORT_SYMBOL_GPL(dccp_reqsk_init);
|