Merge branches 'release', 'cpuidle-2.6.25' and 'idle' into release
[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/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                 newdp->dccps_timestamp_echo = dreq->dreq_timestamp_echo;
125                 newdp->dccps_timestamp_time = dreq->dreq_timestamp_time;
126                 newicsk->icsk_rto           = DCCP_TIMEOUT_INIT;
127
128                 if (dccp_feat_clone(sk, newsk))
129                         goto out_free;
130
131                 if (newdmsk->dccpms_send_ack_vector) {
132                         newdp->dccps_hc_rx_ackvec =
133                                                 dccp_ackvec_alloc(GFP_ATOMIC);
134                         if (unlikely(newdp->dccps_hc_rx_ackvec == NULL))
135                                 goto out_free;
136                 }
137
138                 newdp->dccps_hc_rx_ccid =
139                             ccid_hc_rx_new(newdmsk->dccpms_rx_ccid,
140                                            newsk, GFP_ATOMIC);
141                 newdp->dccps_hc_tx_ccid =
142                             ccid_hc_tx_new(newdmsk->dccpms_tx_ccid,
143                                            newsk, GFP_ATOMIC);
144                 if (unlikely(newdp->dccps_hc_rx_ccid == NULL ||
145                              newdp->dccps_hc_tx_ccid == NULL)) {
146                         dccp_ackvec_free(newdp->dccps_hc_rx_ackvec);
147                         ccid_hc_rx_delete(newdp->dccps_hc_rx_ccid, newsk);
148                         ccid_hc_tx_delete(newdp->dccps_hc_tx_ccid, newsk);
149 out_free:
150                         /* It is still raw copy of parent, so invalidate
151                          * destructor and make plain sk_free() */
152                         newsk->sk_destruct = NULL;
153                         sk_free(newsk);
154                         return NULL;
155                 }
156
157                 /*
158                  * Step 3: Process LISTEN state
159                  *
160                  *    Choose S.ISS (initial seqno) or set from Init Cookies
161                  *    Initialize S.GAR := S.ISS
162                  *    Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
163                  */
164
165                 /* See dccp_v4_conn_request */
166                 newdmsk->dccpms_sequence_window = req->rcv_wnd;
167
168                 newdp->dccps_gar = newdp->dccps_isr = dreq->dreq_isr;
169                 dccp_update_gsr(newsk, dreq->dreq_isr);
170
171                 newdp->dccps_iss = dreq->dreq_iss;
172                 dccp_update_gss(newsk, dreq->dreq_iss);
173
174                 /*
175                  * SWL and AWL are initially adjusted so that they are not less than
176                  * the initial Sequence Numbers received and sent, respectively:
177                  *      SWL := max(GSR + 1 - floor(W/4), ISR),
178                  *      AWL := max(GSS - W' + 1, ISS).
179                  * These adjustments MUST be applied only at the beginning of the
180                  * connection.
181                  */
182                 dccp_set_seqno(&newdp->dccps_swl,
183                                max48(newdp->dccps_swl, newdp->dccps_isr));
184                 dccp_set_seqno(&newdp->dccps_awl,
185                                max48(newdp->dccps_awl, newdp->dccps_iss));
186
187                 dccp_init_xmit_timers(newsk);
188
189                 DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS);
190         }
191         return newsk;
192 }
193
194 EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
195
196 /*
197  * Process an incoming packet for RESPOND sockets represented
198  * as an request_sock.
199  */
200 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
201                             struct request_sock *req,
202                             struct request_sock **prev)
203 {
204         struct sock *child = NULL;
205         struct dccp_request_sock *dreq = dccp_rsk(req);
206
207         /* Check for retransmitted REQUEST */
208         if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
209
210                 if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_isr)) {
211                         dccp_pr_debug("Retransmitted REQUEST\n");
212                         dreq->dreq_isr = DCCP_SKB_CB(skb)->dccpd_seq;
213                         /*
214                          * Send another RESPONSE packet
215                          * To protect against Request floods, increment retrans
216                          * counter (backoff, monitored by dccp_response_timer).
217                          */
218                         req->retrans++;
219                         req->rsk_ops->rtx_syn_ack(sk, req, NULL);
220                 }
221                 /* Network Duplicate, discard packet */
222                 return NULL;
223         }
224
225         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
226
227         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
228             dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
229                 goto drop;
230
231         /* Invalid ACK */
232         if (DCCP_SKB_CB(skb)->dccpd_ack_seq != dreq->dreq_iss) {
233                 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
234                               "dreq_iss=%llu\n",
235                               (unsigned long long)
236                               DCCP_SKB_CB(skb)->dccpd_ack_seq,
237                               (unsigned long long) dreq->dreq_iss);
238                 goto drop;
239         }
240
241         if (dccp_parse_options(sk, dreq, skb))
242                  goto drop;
243
244         child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
245         if (child == NULL)
246                 goto listen_overflow;
247
248         inet_csk_reqsk_queue_unlink(sk, req, prev);
249         inet_csk_reqsk_queue_removed(sk, req);
250         inet_csk_reqsk_queue_add(sk, req, child);
251 out:
252         return child;
253 listen_overflow:
254         dccp_pr_debug("listen_overflow!\n");
255         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
256 drop:
257         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
258                 req->rsk_ops->send_reset(sk, skb);
259
260         inet_csk_reqsk_queue_drop(sk, req, prev);
261         goto out;
262 }
263
264 EXPORT_SYMBOL_GPL(dccp_check_req);
265
266 /*
267  *  Queue segment on the new socket if the new socket is active,
268  *  otherwise we just shortcircuit this and continue with
269  *  the new socket.
270  */
271 int dccp_child_process(struct sock *parent, struct sock *child,
272                        struct sk_buff *skb)
273 {
274         int ret = 0;
275         const int state = child->sk_state;
276
277         if (!sock_owned_by_user(child)) {
278                 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
279                                              skb->len);
280
281                 /* Wakeup parent, send SIGIO */
282                 if (state == DCCP_RESPOND && child->sk_state != state)
283                         parent->sk_data_ready(parent, 0);
284         } else {
285                 /* Alas, it is possible again, because we do lookup
286                  * in main socket hash table and lock on listening
287                  * socket does not protect us more.
288                  */
289                 sk_add_backlog(child, skb);
290         }
291
292         bh_unlock_sock(child);
293         sock_put(child);
294         return ret;
295 }
296
297 EXPORT_SYMBOL_GPL(dccp_child_process);
298
299 void dccp_reqsk_send_ack(struct sk_buff *skb, struct request_sock *rsk)
300 {
301         DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state");
302 }
303
304 EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack);
305
306 void dccp_reqsk_init(struct request_sock *req, struct sk_buff *skb)
307 {
308         struct dccp_request_sock *dreq = dccp_rsk(req);
309
310         inet_rsk(req)->rmt_port   = dccp_hdr(skb)->dccph_sport;
311         inet_rsk(req)->acked      = 0;
312         req->rcv_wnd              = sysctl_dccp_feat_sequence_window;
313         dreq->dreq_timestamp_echo = 0;
314 }
315
316 EXPORT_SYMBOL_GPL(dccp_reqsk_init);