Merge HEAD from ../scsi-misc-2.6-old
[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
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 void dccp_time_wait(struct sock *sk, int state, int timeo)
44 {
45         struct inet_timewait_sock *tw = NULL;
46
47         if (dccp_death_row.tw_count < dccp_death_row.sysctl_max_tw_buckets)
48                 tw = inet_twsk_alloc(sk, state);
49
50         if (tw != NULL) {
51                 const struct inet_connection_sock *icsk = inet_csk(sk);
52                 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
53
54                 /* Linkage updates. */
55                 __inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
56
57                 /* Get the TIME_WAIT timeout firing. */
58                 if (timeo < rto)
59                         timeo = rto;
60
61                 tw->tw_timeout = DCCP_TIMEWAIT_LEN;
62                 if (state == DCCP_TIME_WAIT)
63                         timeo = DCCP_TIMEWAIT_LEN;
64
65                 inet_twsk_schedule(tw, &dccp_death_row, timeo,
66                                    DCCP_TIMEWAIT_LEN);
67                 inet_twsk_put(tw);
68         } else {
69                 /* Sorry, if we're out of memory, just CLOSE this
70                  * socket up.  We've got bigger problems than
71                  * non-graceful socket closings.
72                  */
73                 LIMIT_NETDEBUG(KERN_INFO "DCCP: time wait bucket "
74                                          "table overflow\n");
75         }
76
77         dccp_done(sk);
78 }
79
80 struct sock *dccp_create_openreq_child(struct sock *sk,
81                                        const struct request_sock *req,
82                                        const struct sk_buff *skb)
83 {
84         /*
85          * Step 3: Process LISTEN state
86          *
87          * // Generate a new socket and switch to that socket
88          * Set S := new socket for this port pair
89          */
90         struct sock *newsk = inet_csk_clone(sk, req, GFP_ATOMIC);
91
92         if (newsk != NULL) {
93                 const struct dccp_request_sock *dreq = dccp_rsk(req);
94                 struct inet_connection_sock *newicsk = inet_csk(sk);
95                 struct dccp_sock *newdp = dccp_sk(newsk);
96
97                 newdp->dccps_role          = DCCP_ROLE_SERVER;
98                 newdp->dccps_hc_rx_ackvec  = NULL;
99                 newdp->dccps_service_list  = NULL;
100                 newdp->dccps_service       = dreq->dreq_service;
101                 newicsk->icsk_rto          = DCCP_TIMEOUT_INIT;
102                 do_gettimeofday(&newdp->dccps_epoch);
103
104                 if (newdp->dccps_options.dccpo_send_ack_vector) {
105                         newdp->dccps_hc_rx_ackvec =
106                                 dccp_ackvec_alloc(DCCP_MAX_ACKVEC_LEN,
107                                                   GFP_ATOMIC);
108                         /*
109                          * XXX: We're using the same CCIDs set on the parent,
110                          * i.e. sk_clone copied the master sock and left the
111                          * CCID pointers for this child, that is why we do the
112                          * __ccid_get calls.
113                          */
114                         if (unlikely(newdp->dccps_hc_rx_ackvec == NULL))
115                                 goto out_free;
116                 }
117
118                 if (unlikely(ccid_hc_rx_init(newdp->dccps_hc_rx_ccid,
119                                              newsk) != 0 ||
120                              ccid_hc_tx_init(newdp->dccps_hc_tx_ccid,
121                                              newsk) != 0)) {
122                         dccp_ackvec_free(newdp->dccps_hc_rx_ackvec);
123                         ccid_hc_rx_exit(newdp->dccps_hc_rx_ccid, newsk);
124                         ccid_hc_tx_exit(newdp->dccps_hc_tx_ccid, newsk);
125 out_free:
126                         /* It is still raw copy of parent, so invalidate
127                          * destructor and make plain sk_free() */
128                         newsk->sk_destruct = NULL;
129                         sk_free(newsk);
130                         return NULL;
131                 }
132
133                 __ccid_get(newdp->dccps_hc_rx_ccid);
134                 __ccid_get(newdp->dccps_hc_tx_ccid);
135
136                 /*
137                  * Step 3: Process LISTEN state
138                  *
139                  *      Choose S.ISS (initial seqno) or set from Init Cookie
140                  *      Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init
141                  *      Cookie
142                  */
143
144                 /* See dccp_v4_conn_request */
145                 newdp->dccps_options.dccpo_sequence_window = req->rcv_wnd;
146
147                 newdp->dccps_gar = newdp->dccps_isr = dreq->dreq_isr;
148                 dccp_update_gsr(newsk, dreq->dreq_isr);
149
150                 newdp->dccps_iss = dreq->dreq_iss;
151                 dccp_update_gss(newsk, dreq->dreq_iss);
152
153                 /*
154                  * SWL and AWL are initially adjusted so that they are not less than
155                  * the initial Sequence Numbers received and sent, respectively:
156                  *      SWL := max(GSR + 1 - floor(W/4), ISR),
157                  *      AWL := max(GSS - W' + 1, ISS).
158                  * These adjustments MUST be applied only at the beginning of the
159                  * connection.
160                  */
161                 dccp_set_seqno(&newdp->dccps_swl,
162                                max48(newdp->dccps_swl, newdp->dccps_isr));
163                 dccp_set_seqno(&newdp->dccps_awl,
164                                max48(newdp->dccps_awl, newdp->dccps_iss));
165
166                 dccp_init_xmit_timers(newsk);
167
168                 DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS);
169         }
170         return newsk;
171 }
172
173 /* 
174  * Process an incoming packet for RESPOND sockets represented
175  * as an request_sock.
176  */
177 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
178                             struct request_sock *req,
179                             struct request_sock **prev)
180 {
181         struct sock *child = NULL;
182
183         /* Check for retransmitted REQUEST */
184         if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
185                 if (after48(DCCP_SKB_CB(skb)->dccpd_seq,
186                             dccp_rsk(req)->dreq_isr)) {
187                         struct dccp_request_sock *dreq = dccp_rsk(req);
188
189                         dccp_pr_debug("Retransmitted REQUEST\n");
190                         /* Send another RESPONSE packet */
191                         dccp_set_seqno(&dreq->dreq_iss, dreq->dreq_iss + 1);
192                         dccp_set_seqno(&dreq->dreq_isr,
193                                        DCCP_SKB_CB(skb)->dccpd_seq);
194                         req->rsk_ops->rtx_syn_ack(sk, req, NULL);
195                 }
196                 /* Network Duplicate, discard packet */
197                 return NULL;
198         }
199
200         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
201
202         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
203             dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
204                 goto drop;
205
206         /* Invalid ACK */
207         if (DCCP_SKB_CB(skb)->dccpd_ack_seq != dccp_rsk(req)->dreq_iss) {
208                 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
209                               "dreq_iss=%llu\n",
210                               (unsigned long long)
211                               DCCP_SKB_CB(skb)->dccpd_ack_seq,
212                               (unsigned long long)
213                               dccp_rsk(req)->dreq_iss);
214                 goto drop;
215         }
216
217         child = dccp_v4_request_recv_sock(sk, skb, req, NULL);
218         if (child == NULL)
219                 goto listen_overflow;
220
221         /* FIXME: deal with options */
222
223         inet_csk_reqsk_queue_unlink(sk, req, prev);
224         inet_csk_reqsk_queue_removed(sk, req);
225         inet_csk_reqsk_queue_add(sk, req, child);
226 out:
227         return child;
228 listen_overflow:
229         dccp_pr_debug("listen_overflow!\n");
230         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
231 drop:
232         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
233                 req->rsk_ops->send_reset(skb);
234
235         inet_csk_reqsk_queue_drop(sk, req, prev);
236         goto out;
237 }
238
239 /*
240  *  Queue segment on the new socket if the new socket is active,
241  *  otherwise we just shortcircuit this and continue with
242  *  the new socket.
243  */
244 int dccp_child_process(struct sock *parent, struct sock *child,
245                        struct sk_buff *skb)
246 {
247         int ret = 0;
248         const int state = child->sk_state;
249
250         if (!sock_owned_by_user(child)) {
251                 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
252                                              skb->len);
253
254                 /* Wakeup parent, send SIGIO */
255                 if (state == DCCP_RESPOND && child->sk_state != state)
256                         parent->sk_data_ready(parent, 0);
257         } else {
258                 /* Alas, it is possible again, because we do lookup
259                  * in main socket hash table and lock on listening
260                  * socket does not protect us more.
261                  */
262                 sk_add_backlog(child, skb);
263         }
264
265         bh_unlock_sock(child);
266         sock_put(child);
267         return ret;
268 }