Merge master.kernel.org:/pub/scm/linux/kernel/git/acme/net-2.6
[sfrench/cifs-2.6.git] / 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 "ccid.h"
23 #include "dccp.h"
24
25 struct inet_timewait_death_row dccp_death_row = {
26         .sysctl_max_tw_buckets = NR_FILE * 2,
27         .period         = DCCP_TIMEWAIT_LEN / INET_TWDR_TWKILL_SLOTS,
28         .death_lock     = SPIN_LOCK_UNLOCKED,
29         .hashinfo       = &dccp_hashinfo,
30         .tw_timer       = TIMER_INITIALIZER(inet_twdr_hangman, 0,
31                                             (unsigned long)&dccp_death_row),
32         .twkill_work    = __WORK_INITIALIZER(dccp_death_row.twkill_work,
33                                              inet_twdr_twkill_work,
34                                              &dccp_death_row),
35 /* Short-time timewait calendar */
36
37         .twcal_hand     = -1,
38         .twcal_timer    = TIMER_INITIALIZER(inet_twdr_twcal_tick, 0,
39                                             (unsigned long)&dccp_death_row),
40 };
41
42 void dccp_time_wait(struct sock *sk, int state, int timeo)
43 {
44         struct inet_timewait_sock *tw = NULL;
45
46         if (dccp_death_row.tw_count < dccp_death_row.sysctl_max_tw_buckets)
47                 tw = inet_twsk_alloc(sk, state);
48
49         if (tw != NULL) {
50                 const struct inet_connection_sock *icsk = inet_csk(sk);
51                 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
52
53                 /* Linkage updates. */
54                 __inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
55
56                 /* Get the TIME_WAIT timeout firing. */
57                 if (timeo < rto)
58                         timeo = rto;
59
60                 tw->tw_timeout = DCCP_TIMEWAIT_LEN;
61                 if (state == DCCP_TIME_WAIT)
62                         timeo = DCCP_TIMEWAIT_LEN;
63
64                 inet_twsk_schedule(tw, &dccp_death_row, timeo,
65                                    DCCP_TIMEWAIT_LEN);
66                 inet_twsk_put(tw);
67         } else {
68                 /* Sorry, if we're out of memory, just CLOSE this
69                  * socket up.  We've got bigger problems than
70                  * non-graceful socket closings.
71                  */
72                 LIMIT_NETDEBUG(KERN_INFO "DCCP: time wait bucket "
73                                          "table overflow\n");
74         }
75
76         dccp_done(sk);
77 }
78
79 struct sock *dccp_create_openreq_child(struct sock *sk,
80                                        const struct request_sock *req,
81                                        const struct sk_buff *skb)
82 {
83         /*
84          * Step 3: Process LISTEN state
85          *
86          * // Generate a new socket and switch to that socket
87          * Set S := new socket for this port pair
88          */
89         struct sock *newsk = inet_csk_clone(sk, req, GFP_ATOMIC);
90
91         if (newsk != NULL) {
92                 const struct dccp_request_sock *dreq = dccp_rsk(req);
93                 struct inet_connection_sock *newicsk = inet_csk(sk);
94                 struct dccp_sock *newdp = dccp_sk(newsk);
95
96                 newdp->dccps_hc_rx_ackpkts = NULL;
97                 newdp->dccps_role = DCCP_ROLE_SERVER;
98                 newicsk->icsk_rto = DCCP_TIMEOUT_INIT;
99                 do_gettimeofday(&newdp->dccps_epoch);
100
101                 if (newdp->dccps_options.dccpo_send_ack_vector) {
102                         newdp->dccps_hc_rx_ackpkts =
103                                 dccp_ackpkts_alloc(DCCP_MAX_ACK_VECTOR_LEN,
104                                                    GFP_ATOMIC);
105                         /*
106                          * XXX: We're using the same CCIDs set on the parent,
107                          * i.e. sk_clone copied the master sock and left the
108                          * CCID pointers for this child, that is why we do the
109                          * __ccid_get calls.
110                          */
111                         if (unlikely(newdp->dccps_hc_rx_ackpkts == NULL))
112                                 goto out_free;
113                 }
114
115                 if (unlikely(ccid_hc_rx_init(newdp->dccps_hc_rx_ccid,
116                                              newsk) != 0 ||
117                              ccid_hc_tx_init(newdp->dccps_hc_tx_ccid,
118                                              newsk) != 0)) {
119                         dccp_ackpkts_free(newdp->dccps_hc_rx_ackpkts);
120                         ccid_hc_rx_exit(newdp->dccps_hc_rx_ccid, newsk);
121                         ccid_hc_tx_exit(newdp->dccps_hc_tx_ccid, newsk);
122 out_free:
123                         /* It is still raw copy of parent, so invalidate
124                          * destructor and make plain sk_free() */
125                         newsk->sk_destruct = NULL;
126                         sk_free(newsk);
127                         return NULL;
128                 }
129
130                 __ccid_get(newdp->dccps_hc_rx_ccid);
131                 __ccid_get(newdp->dccps_hc_tx_ccid);
132
133                 /*
134                  * Step 3: Process LISTEN state
135                  *
136                  *      Choose S.ISS (initial seqno) or set from Init Cookie
137                  *      Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init
138                  *      Cookie
139                  */
140
141                 /* See dccp_v4_conn_request */
142                 newdp->dccps_options.dccpo_sequence_window = req->rcv_wnd;
143
144                 newdp->dccps_gar = newdp->dccps_isr = dreq->dreq_isr;
145                 dccp_update_gsr(newsk, dreq->dreq_isr);
146
147                 newdp->dccps_iss = dreq->dreq_iss;
148                 dccp_update_gss(newsk, dreq->dreq_iss);
149
150                 /*
151                  * SWL and AWL are initially adjusted so that they are not less than
152                  * the initial Sequence Numbers received and sent, respectively:
153                  *      SWL := max(GSR + 1 - floor(W/4), ISR),
154                  *      AWL := max(GSS - W' + 1, ISS).
155                  * These adjustments MUST be applied only at the beginning of the
156                  * connection.
157                  */
158                 dccp_set_seqno(&newdp->dccps_swl,
159                                max48(newdp->dccps_swl, newdp->dccps_isr));
160                 dccp_set_seqno(&newdp->dccps_awl,
161                                max48(newdp->dccps_awl, newdp->dccps_iss));
162
163                 dccp_init_xmit_timers(newsk);
164
165                 DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS);
166         }
167         return newsk;
168 }
169
170 /* 
171  * Process an incoming packet for RESPOND sockets represented
172  * as an request_sock.
173  */
174 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
175                             struct request_sock *req,
176                             struct request_sock **prev)
177 {
178         struct sock *child = NULL;
179
180         /* Check for retransmitted REQUEST */
181         if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
182                 if (after48(DCCP_SKB_CB(skb)->dccpd_seq,
183                             dccp_rsk(req)->dreq_isr)) {
184                         struct dccp_request_sock *dreq = dccp_rsk(req);
185
186                         dccp_pr_debug("Retransmitted REQUEST\n");
187                         /* Send another RESPONSE packet */
188                         dccp_set_seqno(&dreq->dreq_iss, dreq->dreq_iss + 1);
189                         dccp_set_seqno(&dreq->dreq_isr,
190                                        DCCP_SKB_CB(skb)->dccpd_seq);
191                         req->rsk_ops->rtx_syn_ack(sk, req, NULL);
192                 }
193                 /* Network Duplicate, discard packet */
194                 return NULL;
195         }
196
197         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
198
199         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
200             dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
201                 goto drop;
202
203         /* Invalid ACK */
204         if (DCCP_SKB_CB(skb)->dccpd_ack_seq != dccp_rsk(req)->dreq_iss) {
205                 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
206                               "dreq_iss=%llu\n",
207                               (unsigned long long)
208                               DCCP_SKB_CB(skb)->dccpd_ack_seq,
209                               (unsigned long long)
210                               dccp_rsk(req)->dreq_iss);
211                 goto drop;
212         }
213
214         child = dccp_v4_request_recv_sock(sk, skb, req, NULL);
215         if (child == NULL)
216                 goto listen_overflow;
217
218         /* FIXME: deal with options */
219
220         inet_csk_reqsk_queue_unlink(sk, req, prev);
221         inet_csk_reqsk_queue_removed(sk, req);
222         inet_csk_reqsk_queue_add(sk, req, child);
223 out:
224         return child;
225 listen_overflow:
226         dccp_pr_debug("listen_overflow!\n");
227         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
228 drop:
229         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
230                 req->rsk_ops->send_reset(skb);
231
232         inet_csk_reqsk_queue_drop(sk, req, prev);
233         goto out;
234 }
235
236 /*
237  *  Queue segment on the new socket if the new socket is active,
238  *  otherwise we just shortcircuit this and continue with
239  *  the new socket.
240  */
241 int dccp_child_process(struct sock *parent, struct sock *child,
242                        struct sk_buff *skb)
243 {
244         int ret = 0;
245         const int state = child->sk_state;
246
247         if (!sock_owned_by_user(child)) {
248                 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
249                                              skb->len);
250
251                 /* Wakeup parent, send SIGIO */
252                 if (state == DCCP_RESPOND && child->sk_state != state)
253                         parent->sk_data_ready(parent, 0);
254         } else {
255                 /* Alas, it is possible again, because we do lookup
256                  * in main socket hash table and lock on listening
257                  * socket does not protect us more.
258                  */
259                 sk_add_backlog(child, skb);
260         }
261
262         bh_unlock_sock(child);
263         sock_put(child);
264         return ret;
265 }