Merge tag 'jg-20061012-00' of git://electric-eye.fr.zoreil.com/home/romieu/linux...
[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/dccp.h>
14 #include <linux/skbuff.h>
15 #include <linux/timer.h>
16
17 #include <net/sock.h>
18 #include <net/xfrm.h>
19 #include <net/inet_timewait_sock.h>
20
21 #include "ackvec.h"
22 #include "ccid.h"
23 #include "dccp.h"
24 #include "feat.h"
25
26 struct inet_timewait_death_row dccp_death_row = {
27         .sysctl_max_tw_buckets = NR_FILE * 2,
28         .period         = DCCP_TIMEWAIT_LEN / INET_TWDR_TWKILL_SLOTS,
29         .death_lock     = SPIN_LOCK_UNLOCKED,
30         .hashinfo       = &dccp_hashinfo,
31         .tw_timer       = TIMER_INITIALIZER(inet_twdr_hangman, 0,
32                                             (unsigned long)&dccp_death_row),
33         .twkill_work    = __WORK_INITIALIZER(dccp_death_row.twkill_work,
34                                              inet_twdr_twkill_work,
35                                              &dccp_death_row),
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_time_wait(struct sock *sk, int state, int timeo)
46 {
47         struct inet_timewait_sock *tw = NULL;
48
49         if (dccp_death_row.tw_count < dccp_death_row.sysctl_max_tw_buckets)
50                 tw = inet_twsk_alloc(sk, state);
51
52         if (tw != NULL) {
53                 const struct inet_connection_sock *icsk = inet_csk(sk);
54                 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
55 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
56                 if (tw->tw_family == PF_INET6) {
57                         const struct ipv6_pinfo *np = inet6_sk(sk);
58                         struct inet6_timewait_sock *tw6;
59
60                         tw->tw_ipv6_offset = inet6_tw_offset(sk->sk_prot);
61                         tw6 = inet6_twsk((struct sock *)tw);
62                         ipv6_addr_copy(&tw6->tw_v6_daddr, &np->daddr);
63                         ipv6_addr_copy(&tw6->tw_v6_rcv_saddr, &np->rcv_saddr);
64                         tw->tw_ipv6only = np->ipv6only;
65                 }
66 #endif
67                 /* Linkage updates. */
68                 __inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
69
70                 /* Get the TIME_WAIT timeout firing. */
71                 if (timeo < rto)
72                         timeo = rto;
73
74                 tw->tw_timeout = DCCP_TIMEWAIT_LEN;
75                 if (state == DCCP_TIME_WAIT)
76                         timeo = DCCP_TIMEWAIT_LEN;
77
78                 inet_twsk_schedule(tw, &dccp_death_row, timeo,
79                                    DCCP_TIMEWAIT_LEN);
80                 inet_twsk_put(tw);
81         } else {
82                 /* Sorry, if we're out of memory, just CLOSE this
83                  * socket up.  We've got bigger problems than
84                  * non-graceful socket closings.
85                  */
86                 LIMIT_NETDEBUG(KERN_INFO "DCCP: time wait bucket "
87                                          "table overflow\n");
88         }
89
90         dccp_done(sk);
91 }
92
93 struct 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                 const struct dccp_request_sock *dreq = dccp_rsk(req);
107                 struct inet_connection_sock *newicsk = inet_csk(sk);
108                 struct dccp_sock *newdp = dccp_sk(newsk);
109                 struct dccp_minisock *newdmsk = dccp_msk(newsk);
110
111                 newdp->dccps_role          = DCCP_ROLE_SERVER;
112                 newdp->dccps_hc_rx_ackvec  = NULL;
113                 newdp->dccps_service_list  = NULL;
114                 newdp->dccps_service       = dreq->dreq_service;
115                 newicsk->icsk_rto          = DCCP_TIMEOUT_INIT;
116                 do_gettimeofday(&newdp->dccps_epoch);
117
118                 if (dccp_feat_clone(sk, newsk))
119                         goto out_free;
120
121                 if (newdmsk->dccpms_send_ack_vector) {
122                         newdp->dccps_hc_rx_ackvec =
123                                                 dccp_ackvec_alloc(GFP_ATOMIC);
124                         if (unlikely(newdp->dccps_hc_rx_ackvec == NULL))
125                                 goto out_free;
126                 }
127
128                 newdp->dccps_hc_rx_ccid =
129                             ccid_hc_rx_new(newdmsk->dccpms_rx_ccid,
130                                            newsk, GFP_ATOMIC);
131                 newdp->dccps_hc_tx_ccid =
132                             ccid_hc_tx_new(newdmsk->dccpms_tx_ccid,
133                                            newsk, GFP_ATOMIC);
134                 if (unlikely(newdp->dccps_hc_rx_ccid == NULL ||
135                              newdp->dccps_hc_tx_ccid == NULL)) {
136                         dccp_ackvec_free(newdp->dccps_hc_rx_ackvec);
137                         ccid_hc_rx_delete(newdp->dccps_hc_rx_ccid, newsk);
138                         ccid_hc_tx_delete(newdp->dccps_hc_tx_ccid, newsk);
139 out_free:
140                         /* It is still raw copy of parent, so invalidate
141                          * destructor and make plain sk_free() */
142                         newsk->sk_destruct = NULL;
143                         sk_free(newsk);
144                         return NULL;
145                 }
146
147                 /*
148                  * Step 3: Process LISTEN state
149                  *
150                  *      Choose S.ISS (initial seqno) or set from Init Cookie
151                  *      Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init
152                  *      Cookie
153                  */
154
155                 /* See dccp_v4_conn_request */
156                 newdmsk->dccpms_sequence_window = req->rcv_wnd;
157
158                 newdp->dccps_gar = newdp->dccps_isr = dreq->dreq_isr;
159                 dccp_update_gsr(newsk, dreq->dreq_isr);
160
161                 newdp->dccps_iss = dreq->dreq_iss;
162                 dccp_update_gss(newsk, dreq->dreq_iss);
163
164                 /*
165                  * SWL and AWL are initially adjusted so that they are not less than
166                  * the initial Sequence Numbers received and sent, respectively:
167                  *      SWL := max(GSR + 1 - floor(W/4), ISR),
168                  *      AWL := max(GSS - W' + 1, ISS).
169                  * These adjustments MUST be applied only at the beginning of the
170                  * connection.
171                  */
172                 dccp_set_seqno(&newdp->dccps_swl,
173                                max48(newdp->dccps_swl, newdp->dccps_isr));
174                 dccp_set_seqno(&newdp->dccps_awl,
175                                max48(newdp->dccps_awl, newdp->dccps_iss));
176
177                 dccp_init_xmit_timers(newsk);
178
179                 DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS);
180         }
181         return newsk;
182 }
183
184 EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
185
186 /* 
187  * Process an incoming packet for RESPOND sockets represented
188  * as an request_sock.
189  */
190 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
191                             struct request_sock *req,
192                             struct request_sock **prev)
193 {
194         struct sock *child = NULL;
195
196         /* Check for retransmitted REQUEST */
197         if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
198                 if (after48(DCCP_SKB_CB(skb)->dccpd_seq,
199                             dccp_rsk(req)->dreq_isr)) {
200                         struct dccp_request_sock *dreq = dccp_rsk(req);
201
202                         dccp_pr_debug("Retransmitted REQUEST\n");
203                         /* Send another RESPONSE packet */
204                         dccp_set_seqno(&dreq->dreq_iss, dreq->dreq_iss + 1);
205                         dccp_set_seqno(&dreq->dreq_isr,
206                                        DCCP_SKB_CB(skb)->dccpd_seq);
207                         req->rsk_ops->rtx_syn_ack(sk, req, NULL);
208                 }
209                 /* Network Duplicate, discard packet */
210                 return NULL;
211         }
212
213         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
214
215         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
216             dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
217                 goto drop;
218
219         /* Invalid ACK */
220         if (DCCP_SKB_CB(skb)->dccpd_ack_seq != dccp_rsk(req)->dreq_iss) {
221                 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
222                               "dreq_iss=%llu\n",
223                               (unsigned long long)
224                               DCCP_SKB_CB(skb)->dccpd_ack_seq,
225                               (unsigned long long)
226                               dccp_rsk(req)->dreq_iss);
227                 goto drop;
228         }
229
230         child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
231         if (child == NULL)
232                 goto listen_overflow;
233
234         /* FIXME: deal with options */
235
236         inet_csk_reqsk_queue_unlink(sk, req, prev);
237         inet_csk_reqsk_queue_removed(sk, req);
238         inet_csk_reqsk_queue_add(sk, req, child);
239 out:
240         return child;
241 listen_overflow:
242         dccp_pr_debug("listen_overflow!\n");
243         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
244 drop:
245         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
246                 req->rsk_ops->send_reset(skb);
247
248         inet_csk_reqsk_queue_drop(sk, req, prev);
249         goto out;
250 }
251
252 EXPORT_SYMBOL_GPL(dccp_check_req);
253
254 /*
255  *  Queue segment on the new socket if the new socket is active,
256  *  otherwise we just shortcircuit this and continue with
257  *  the new socket.
258  */
259 int dccp_child_process(struct sock *parent, struct sock *child,
260                        struct sk_buff *skb)
261 {
262         int ret = 0;
263         const int state = child->sk_state;
264
265         if (!sock_owned_by_user(child)) {
266                 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
267                                              skb->len);
268
269                 /* Wakeup parent, send SIGIO */
270                 if (state == DCCP_RESPOND && child->sk_state != state)
271                         parent->sk_data_ready(parent, 0);
272         } else {
273                 /* Alas, it is possible again, because we do lookup
274                  * in main socket hash table and lock on listening
275                  * socket does not protect us more.
276                  */
277                 sk_add_backlog(child, skb);
278         }
279
280         bh_unlock_sock(child);
281         sock_put(child);
282         return ret;
283 }
284
285 EXPORT_SYMBOL_GPL(dccp_child_process);