aio: don't include aio.h in sched.h
[linux-3.10.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/dccp.h>
14 #include <linux/gfp.h>
15 #include <linux/kernel.h>
16 #include <linux/skbuff.h>
17 #include <linux/timer.h>
18
19 #include <net/sock.h>
20 #include <net/xfrm.h>
21 #include <net/inet_timewait_sock.h>
22
23 #include "ackvec.h"
24 #include "ccid.h"
25 #include "dccp.h"
26 #include "feat.h"
27
28 struct inet_timewait_death_row dccp_death_row = {
29         .sysctl_max_tw_buckets = NR_FILE * 2,
30         .period         = DCCP_TIMEWAIT_LEN / INET_TWDR_TWKILL_SLOTS,
31         .death_lock     = __SPIN_LOCK_UNLOCKED(dccp_death_row.death_lock),
32         .hashinfo       = &dccp_hashinfo,
33         .tw_timer       = TIMER_INITIALIZER(inet_twdr_hangman, 0,
34                                             (unsigned long)&dccp_death_row),
35         .twkill_work    = __WORK_INITIALIZER(dccp_death_row.twkill_work,
36                                              inet_twdr_twkill_work),
37 /* Short-time timewait calendar */
38
39         .twcal_hand     = -1,
40         .twcal_timer    = TIMER_INITIALIZER(inet_twdr_twcal_tick, 0,
41                                             (unsigned long)&dccp_death_row),
42 };
43
44 EXPORT_SYMBOL_GPL(dccp_death_row);
45
46 void dccp_time_wait(struct sock *sk, int state, int timeo)
47 {
48         struct inet_timewait_sock *tw = NULL;
49
50         if (dccp_death_row.tw_count < dccp_death_row.sysctl_max_tw_buckets)
51                 tw = inet_twsk_alloc(sk, state);
52
53         if (tw != NULL) {
54                 const struct inet_connection_sock *icsk = inet_csk(sk);
55                 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
56 #if IS_ENABLED(CONFIG_IPV6)
57                 if (tw->tw_family == PF_INET6) {
58                         const struct ipv6_pinfo *np = inet6_sk(sk);
59                         struct inet6_timewait_sock *tw6;
60
61                         tw->tw_ipv6_offset = inet6_tw_offset(sk->sk_prot);
62                         tw6 = inet6_twsk((struct sock *)tw);
63                         tw6->tw_v6_daddr = np->daddr;
64                         tw6->tw_v6_rcv_saddr = np->rcv_saddr;
65                         tw->tw_ipv6only = np->ipv6only;
66                 }
67 #endif
68                 /* Linkage updates. */
69                 __inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
70
71                 /* Get the TIME_WAIT timeout firing. */
72                 if (timeo < rto)
73                         timeo = rto;
74
75                 tw->tw_timeout = DCCP_TIMEWAIT_LEN;
76                 if (state == DCCP_TIME_WAIT)
77                         timeo = DCCP_TIMEWAIT_LEN;
78
79                 inet_twsk_schedule(tw, &dccp_death_row, timeo,
80                                    DCCP_TIMEWAIT_LEN);
81                 inet_twsk_put(tw);
82         } else {
83                 /* Sorry, if we're out of memory, just CLOSE this
84                  * socket up.  We've got bigger problems than
85                  * non-graceful socket closings.
86                  */
87                 DCCP_WARN("time wait bucket 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_lock(sk, req, GFP_ATOMIC);
104
105         if (newsk != NULL) {
106                 struct dccp_request_sock *dreq = dccp_rsk(req);
107                 struct inet_connection_sock *newicsk = inet_csk(newsk);
108                 struct dccp_sock *newdp = dccp_sk(newsk);
109
110                 newdp->dccps_role           = DCCP_ROLE_SERVER;
111                 newdp->dccps_hc_rx_ackvec   = NULL;
112                 newdp->dccps_service_list   = NULL;
113                 newdp->dccps_service        = dreq->dreq_service;
114                 newdp->dccps_timestamp_echo = dreq->dreq_timestamp_echo;
115                 newdp->dccps_timestamp_time = dreq->dreq_timestamp_time;
116                 newicsk->icsk_rto           = DCCP_TIMEOUT_INIT;
117
118                 INIT_LIST_HEAD(&newdp->dccps_featneg);
119                 /*
120                  * Step 3: Process LISTEN state
121                  *
122                  *    Choose S.ISS (initial seqno) or set from Init Cookies
123                  *    Initialize S.GAR := S.ISS
124                  *    Set S.ISR, S.GSR from packet (or Init Cookies)
125                  *
126                  *    Setting AWL/AWH and SWL/SWH happens as part of the feature
127                  *    activation below, as these windows all depend on the local
128                  *    and remote Sequence Window feature values (7.5.2).
129                  */
130                 newdp->dccps_iss = dreq->dreq_iss;
131                 newdp->dccps_gss = dreq->dreq_gss;
132                 newdp->dccps_gar = newdp->dccps_iss;
133                 newdp->dccps_isr = dreq->dreq_isr;
134                 newdp->dccps_gsr = dreq->dreq_gsr;
135
136                 /*
137                  * Activate features: initialise CCIDs, sequence windows etc.
138                  */
139                 if (dccp_feat_activate_values(newsk, &dreq->dreq_featneg)) {
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                 dccp_init_xmit_timers(newsk);
147
148                 DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS);
149         }
150         return newsk;
151 }
152
153 EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
154
155 /*
156  * Process an incoming packet for RESPOND sockets represented
157  * as an request_sock.
158  */
159 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
160                             struct request_sock *req,
161                             struct request_sock **prev)
162 {
163         struct sock *child = NULL;
164         struct dccp_request_sock *dreq = dccp_rsk(req);
165
166         /* Check for retransmitted REQUEST */
167         if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
168
169                 if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_gsr)) {
170                         dccp_pr_debug("Retransmitted REQUEST\n");
171                         dreq->dreq_gsr = DCCP_SKB_CB(skb)->dccpd_seq;
172                         /*
173                          * Send another RESPONSE packet
174                          * To protect against Request floods, increment retrans
175                          * counter (backoff, monitored by dccp_response_timer).
176                          */
177                         inet_rtx_syn_ack(sk, req);
178                 }
179                 /* Network Duplicate, discard packet */
180                 return NULL;
181         }
182
183         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
184
185         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
186             dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
187                 goto drop;
188
189         /* Invalid ACK */
190         if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
191                                 dreq->dreq_iss, dreq->dreq_gss)) {
192                 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
193                               "dreq_iss=%llu, dreq_gss=%llu\n",
194                               (unsigned long long)
195                               DCCP_SKB_CB(skb)->dccpd_ack_seq,
196                               (unsigned long long) dreq->dreq_iss,
197                               (unsigned long long) dreq->dreq_gss);
198                 goto drop;
199         }
200
201         if (dccp_parse_options(sk, dreq, skb))
202                  goto drop;
203
204         child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
205         if (child == NULL)
206                 goto listen_overflow;
207
208         inet_csk_reqsk_queue_unlink(sk, req, prev);
209         inet_csk_reqsk_queue_removed(sk, req);
210         inet_csk_reqsk_queue_add(sk, req, child);
211 out:
212         return child;
213 listen_overflow:
214         dccp_pr_debug("listen_overflow!\n");
215         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
216 drop:
217         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
218                 req->rsk_ops->send_reset(sk, skb);
219
220         inet_csk_reqsk_queue_drop(sk, req, prev);
221         goto out;
222 }
223
224 EXPORT_SYMBOL_GPL(dccp_check_req);
225
226 /*
227  *  Queue segment on the new socket if the new socket is active,
228  *  otherwise we just shortcircuit this and continue with
229  *  the new socket.
230  */
231 int dccp_child_process(struct sock *parent, struct sock *child,
232                        struct sk_buff *skb)
233 {
234         int ret = 0;
235         const int state = child->sk_state;
236
237         if (!sock_owned_by_user(child)) {
238                 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
239                                              skb->len);
240
241                 /* Wakeup parent, send SIGIO */
242                 if (state == DCCP_RESPOND && child->sk_state != state)
243                         parent->sk_data_ready(parent, 0);
244         } else {
245                 /* Alas, it is possible again, because we do lookup
246                  * in main socket hash table and lock on listening
247                  * socket does not protect us more.
248                  */
249                 __sk_add_backlog(child, skb);
250         }
251
252         bh_unlock_sock(child);
253         sock_put(child);
254         return ret;
255 }
256
257 EXPORT_SYMBOL_GPL(dccp_child_process);
258
259 void dccp_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
260                          struct request_sock *rsk)
261 {
262         DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state");
263 }
264
265 EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack);
266
267 int dccp_reqsk_init(struct request_sock *req,
268                     struct dccp_sock const *dp, struct sk_buff const *skb)
269 {
270         struct dccp_request_sock *dreq = dccp_rsk(req);
271
272         inet_rsk(req)->rmt_port   = dccp_hdr(skb)->dccph_sport;
273         inet_rsk(req)->loc_port   = dccp_hdr(skb)->dccph_dport;
274         inet_rsk(req)->acked      = 0;
275         dreq->dreq_timestamp_echo = 0;
276
277         /* inherit feature negotiation options from listening socket */
278         return dccp_feat_clone_list(&dp->dccps_featneg, &dreq->dreq_featneg);
279 }
280
281 EXPORT_SYMBOL_GPL(dccp_reqsk_init);