Pull cpumask into release branch
[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_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                 DCCP_WARN("time wait bucket table overflow\n");
87         }
88
89         dccp_done(sk);
90 }
91
92 struct sock *dccp_create_openreq_child(struct sock *sk,
93                                        const struct request_sock *req,
94                                        const struct sk_buff *skb)
95 {
96         /*
97          * Step 3: Process LISTEN state
98          *
99          *   (* Generate a new socket and switch to that socket *)
100          *   Set S := new socket for this port pair
101          */
102         struct sock *newsk = inet_csk_clone(sk, req, GFP_ATOMIC);
103
104         if (newsk != NULL) {
105                 struct dccp_request_sock *dreq = dccp_rsk(req);
106                 struct inet_connection_sock *newicsk = inet_csk(newsk);
107                 struct dccp_sock *newdp = dccp_sk(newsk);
108
109                 newdp->dccps_role           = DCCP_ROLE_SERVER;
110                 newdp->dccps_hc_rx_ackvec   = NULL;
111                 newdp->dccps_service_list   = NULL;
112                 newdp->dccps_service        = dreq->dreq_service;
113                 newdp->dccps_timestamp_echo = dreq->dreq_timestamp_echo;
114                 newdp->dccps_timestamp_time = dreq->dreq_timestamp_time;
115                 newicsk->icsk_rto           = DCCP_TIMEOUT_INIT;
116
117                 INIT_LIST_HEAD(&newdp->dccps_featneg);
118                 /*
119                  * Step 3: Process LISTEN state
120                  *
121                  *    Choose S.ISS (initial seqno) or set from Init Cookies
122                  *    Initialize S.GAR := S.ISS
123                  *    Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
124                  */
125                 newdp->dccps_gar = newdp->dccps_iss = dreq->dreq_iss;
126                 dccp_update_gss(newsk, dreq->dreq_iss);
127
128                 newdp->dccps_isr = dreq->dreq_isr;
129                 dccp_update_gsr(newsk, dreq->dreq_isr);
130
131                 /*
132                  * SWL and AWL are initially adjusted so that they are not less than
133                  * the initial Sequence Numbers received and sent, respectively:
134                  *      SWL := max(GSR + 1 - floor(W/4), ISR),
135                  *      AWL := max(GSS - W' + 1, ISS).
136                  * These adjustments MUST be applied only at the beginning of the
137                  * connection.
138                  */
139                 dccp_set_seqno(&newdp->dccps_swl,
140                                max48(newdp->dccps_swl, newdp->dccps_isr));
141                 dccp_set_seqno(&newdp->dccps_awl,
142                                max48(newdp->dccps_awl, newdp->dccps_iss));
143
144                 /*
145                  * Activate features after initialising the sequence numbers,
146                  * since CCID initialisation may depend on GSS, ISR, ISS etc.
147                  */
148                 if (dccp_feat_activate_values(newsk, &dreq->dreq_featneg)) {
149                         /* It is still raw copy of parent, so invalidate
150                          * destructor and make plain sk_free() */
151                         newsk->sk_destruct = NULL;
152                         sk_free(newsk);
153                         return NULL;
154                 }
155                 dccp_init_xmit_timers(newsk);
156
157                 DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS);
158         }
159         return newsk;
160 }
161
162 EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
163
164 /*
165  * Process an incoming packet for RESPOND sockets represented
166  * as an request_sock.
167  */
168 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
169                             struct request_sock *req,
170                             struct request_sock **prev)
171 {
172         struct sock *child = NULL;
173         struct dccp_request_sock *dreq = dccp_rsk(req);
174
175         /* Check for retransmitted REQUEST */
176         if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
177
178                 if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_isr)) {
179                         dccp_pr_debug("Retransmitted REQUEST\n");
180                         dreq->dreq_isr = DCCP_SKB_CB(skb)->dccpd_seq;
181                         /*
182                          * Send another RESPONSE packet
183                          * To protect against Request floods, increment retrans
184                          * counter (backoff, monitored by dccp_response_timer).
185                          */
186                         req->retrans++;
187                         req->rsk_ops->rtx_syn_ack(sk, req);
188                 }
189                 /* Network Duplicate, discard packet */
190                 return NULL;
191         }
192
193         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
194
195         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
196             dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
197                 goto drop;
198
199         /* Invalid ACK */
200         if (DCCP_SKB_CB(skb)->dccpd_ack_seq != dreq->dreq_iss) {
201                 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
202                               "dreq_iss=%llu\n",
203                               (unsigned long long)
204                               DCCP_SKB_CB(skb)->dccpd_ack_seq,
205                               (unsigned long long) dreq->dreq_iss);
206                 goto drop;
207         }
208
209         if (dccp_parse_options(sk, dreq, skb))
210                  goto drop;
211
212         child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
213         if (child == NULL)
214                 goto listen_overflow;
215
216         inet_csk_reqsk_queue_unlink(sk, req, prev);
217         inet_csk_reqsk_queue_removed(sk, req);
218         inet_csk_reqsk_queue_add(sk, req, child);
219 out:
220         return child;
221 listen_overflow:
222         dccp_pr_debug("listen_overflow!\n");
223         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
224 drop:
225         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
226                 req->rsk_ops->send_reset(sk, skb);
227
228         inet_csk_reqsk_queue_drop(sk, req, prev);
229         goto out;
230 }
231
232 EXPORT_SYMBOL_GPL(dccp_check_req);
233
234 /*
235  *  Queue segment on the new socket if the new socket is active,
236  *  otherwise we just shortcircuit this and continue with
237  *  the new socket.
238  */
239 int dccp_child_process(struct sock *parent, struct sock *child,
240                        struct sk_buff *skb)
241 {
242         int ret = 0;
243         const int state = child->sk_state;
244
245         if (!sock_owned_by_user(child)) {
246                 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
247                                              skb->len);
248
249                 /* Wakeup parent, send SIGIO */
250                 if (state == DCCP_RESPOND && child->sk_state != state)
251                         parent->sk_data_ready(parent, 0);
252         } else {
253                 /* Alas, it is possible again, because we do lookup
254                  * in main socket hash table and lock on listening
255                  * socket does not protect us more.
256                  */
257                 sk_add_backlog(child, skb);
258         }
259
260         bh_unlock_sock(child);
261         sock_put(child);
262         return ret;
263 }
264
265 EXPORT_SYMBOL_GPL(dccp_child_process);
266
267 void dccp_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
268                          struct request_sock *rsk)
269 {
270         DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state");
271 }
272
273 EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack);
274
275 int dccp_reqsk_init(struct request_sock *req,
276                     struct dccp_sock const *dp, struct sk_buff const *skb)
277 {
278         struct dccp_request_sock *dreq = dccp_rsk(req);
279
280         inet_rsk(req)->rmt_port   = dccp_hdr(skb)->dccph_sport;
281         inet_rsk(req)->loc_port   = dccp_hdr(skb)->dccph_dport;
282         inet_rsk(req)->acked      = 0;
283         dreq->dreq_timestamp_echo = 0;
284
285         /* inherit feature negotiation options from listening socket */
286         return dccp_feat_clone_list(&dp->dccps_featneg, &dreq->dreq_featneg);
287 }
288
289 EXPORT_SYMBOL_GPL(dccp_reqsk_init);