[PATCH] skge: version 1.5
[linux-2.6] / net / dccp / minisocks.c
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/config.h>
14 #include <linux/dccp.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,
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                                              &dccp_death_row),
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
44 EXPORT_SYMBOL_GPL(dccp_death_row);
45
46 void 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                 LIMIT_NETDEBUG(KERN_INFO "DCCP: time wait bucket "
88                                          "table overflow\n");
89         }
90
91         dccp_done(sk);
92 }
93
94 struct sock *dccp_create_openreq_child(struct sock *sk,
95                                        const struct request_sock *req,
96                                        const struct sk_buff *skb)
97 {
98         /*
99          * Step 3: Process LISTEN state
100          *
101          * // Generate a new socket and switch to that socket
102          * Set S := new socket for this port pair
103          */
104         struct sock *newsk = inet_csk_clone(sk, req, GFP_ATOMIC);
105
106         if (newsk != NULL) {
107                 const struct dccp_request_sock *dreq = dccp_rsk(req);
108                 struct inet_connection_sock *newicsk = inet_csk(sk);
109                 struct dccp_sock *newdp = dccp_sk(newsk);
110                 struct dccp_minisock *newdmsk = dccp_msk(newsk);
111
112                 newdp->dccps_role          = DCCP_ROLE_SERVER;
113                 newdp->dccps_hc_rx_ackvec  = NULL;
114                 newdp->dccps_service_list  = NULL;
115                 newdp->dccps_service       = dreq->dreq_service;
116                 newicsk->icsk_rto          = DCCP_TIMEOUT_INIT;
117                 do_gettimeofday(&newdp->dccps_epoch);
118
119                 if (dccp_feat_clone(sk, newsk))
120                         goto out_free;
121
122                 if (newdmsk->dccpms_send_ack_vector) {
123                         newdp->dccps_hc_rx_ackvec =
124                                                 dccp_ackvec_alloc(GFP_ATOMIC);
125                         if (unlikely(newdp->dccps_hc_rx_ackvec == NULL))
126                                 goto out_free;
127                 }
128
129                 newdp->dccps_hc_rx_ccid =
130                             ccid_hc_rx_new(newdmsk->dccpms_rx_ccid,
131                                            newsk, GFP_ATOMIC);
132                 newdp->dccps_hc_tx_ccid =
133                             ccid_hc_tx_new(newdmsk->dccpms_tx_ccid,
134                                            newsk, GFP_ATOMIC);
135                 if (unlikely(newdp->dccps_hc_rx_ccid == NULL ||
136                              newdp->dccps_hc_tx_ccid == NULL)) {
137                         dccp_ackvec_free(newdp->dccps_hc_rx_ackvec);
138                         ccid_hc_rx_delete(newdp->dccps_hc_rx_ccid, newsk);
139                         ccid_hc_tx_delete(newdp->dccps_hc_tx_ccid, newsk);
140 out_free:
141                         /* It is still raw copy of parent, so invalidate
142                          * destructor and make plain sk_free() */
143                         newsk->sk_destruct = NULL;
144                         sk_free(newsk);
145                         return NULL;
146                 }
147
148                 /*
149                  * Step 3: Process LISTEN state
150                  *
151                  *      Choose S.ISS (initial seqno) or set from Init Cookie
152                  *      Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init
153                  *      Cookie
154                  */
155
156                 /* See dccp_v4_conn_request */
157                 newdmsk->dccpms_sequence_window = req->rcv_wnd;
158
159                 newdp->dccps_gar = newdp->dccps_isr = dreq->dreq_isr;
160                 dccp_update_gsr(newsk, dreq->dreq_isr);
161
162                 newdp->dccps_iss = dreq->dreq_iss;
163                 dccp_update_gss(newsk, dreq->dreq_iss);
164
165                 /*
166                  * SWL and AWL are initially adjusted so that they are not less than
167                  * the initial Sequence Numbers received and sent, respectively:
168                  *      SWL := max(GSR + 1 - floor(W/4), ISR),
169                  *      AWL := max(GSS - W' + 1, ISS).
170                  * These adjustments MUST be applied only at the beginning of the
171                  * connection.
172                  */
173                 dccp_set_seqno(&newdp->dccps_swl,
174                                max48(newdp->dccps_swl, newdp->dccps_isr));
175                 dccp_set_seqno(&newdp->dccps_awl,
176                                max48(newdp->dccps_awl, newdp->dccps_iss));
177
178                 dccp_init_xmit_timers(newsk);
179
180                 DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS);
181         }
182         return newsk;
183 }
184
185 EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
186
187 /* 
188  * Process an incoming packet for RESPOND sockets represented
189  * as an request_sock.
190  */
191 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
192                             struct request_sock *req,
193                             struct request_sock **prev)
194 {
195         struct sock *child = NULL;
196
197         /* Check for retransmitted REQUEST */
198         if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
199                 if (after48(DCCP_SKB_CB(skb)->dccpd_seq,
200                             dccp_rsk(req)->dreq_isr)) {
201                         struct dccp_request_sock *dreq = dccp_rsk(req);
202
203                         dccp_pr_debug("Retransmitted REQUEST\n");
204                         /* Send another RESPONSE packet */
205                         dccp_set_seqno(&dreq->dreq_iss, dreq->dreq_iss + 1);
206                         dccp_set_seqno(&dreq->dreq_isr,
207                                        DCCP_SKB_CB(skb)->dccpd_seq);
208                         req->rsk_ops->rtx_syn_ack(sk, req, NULL);
209                 }
210                 /* Network Duplicate, discard packet */
211                 return NULL;
212         }
213
214         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
215
216         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
217             dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
218                 goto drop;
219
220         /* Invalid ACK */
221         if (DCCP_SKB_CB(skb)->dccpd_ack_seq != dccp_rsk(req)->dreq_iss) {
222                 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
223                               "dreq_iss=%llu\n",
224                               (unsigned long long)
225                               DCCP_SKB_CB(skb)->dccpd_ack_seq,
226                               (unsigned long long)
227                               dccp_rsk(req)->dreq_iss);
228                 goto drop;
229         }
230
231         child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
232         if (child == NULL)
233                 goto listen_overflow;
234
235         /* FIXME: deal with options */
236
237         inet_csk_reqsk_queue_unlink(sk, req, prev);
238         inet_csk_reqsk_queue_removed(sk, req);
239         inet_csk_reqsk_queue_add(sk, req, child);
240 out:
241         return child;
242 listen_overflow:
243         dccp_pr_debug("listen_overflow!\n");
244         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
245 drop:
246         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
247                 req->rsk_ops->send_reset(skb);
248
249         inet_csk_reqsk_queue_drop(sk, req, prev);
250         goto out;
251 }
252
253 EXPORT_SYMBOL_GPL(dccp_check_req);
254
255 /*
256  *  Queue segment on the new socket if the new socket is active,
257  *  otherwise we just shortcircuit this and continue with
258  *  the new socket.
259  */
260 int dccp_child_process(struct sock *parent, struct sock *child,
261                        struct sk_buff *skb)
262 {
263         int ret = 0;
264         const int state = child->sk_state;
265
266         if (!sock_owned_by_user(child)) {
267                 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
268                                              skb->len);
269
270                 /* Wakeup parent, send SIGIO */
271                 if (state == DCCP_RESPOND && child->sk_state != state)
272                         parent->sk_data_ready(parent, 0);
273         } else {
274                 /* Alas, it is possible again, because we do lookup
275                  * in main socket hash table and lock on listening
276                  * socket does not protect us more.
277                  */
278                 sk_add_backlog(child, skb);
279         }
280
281         bh_unlock_sock(child);
282         sock_put(child);
283         return ret;
284 }
285
286 EXPORT_SYMBOL_GPL(dccp_child_process);