dccp: Remove manual influence on NDP Count feature
[safe/jmp/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_send_ack_vector = sysctl_dccp_feat_send_ack_vector;
49 }
50
51 void dccp_time_wait(struct sock *sk, int state, int timeo)
52 {
53         struct inet_timewait_sock *tw = NULL;
54
55         if (dccp_death_row.tw_count < dccp_death_row.sysctl_max_tw_buckets)
56                 tw = inet_twsk_alloc(sk, state);
57
58         if (tw != NULL) {
59                 const struct inet_connection_sock *icsk = inet_csk(sk);
60                 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
61 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
62                 if (tw->tw_family == PF_INET6) {
63                         const struct ipv6_pinfo *np = inet6_sk(sk);
64                         struct inet6_timewait_sock *tw6;
65
66                         tw->tw_ipv6_offset = inet6_tw_offset(sk->sk_prot);
67                         tw6 = inet6_twsk((struct sock *)tw);
68                         ipv6_addr_copy(&tw6->tw_v6_daddr, &np->daddr);
69                         ipv6_addr_copy(&tw6->tw_v6_rcv_saddr, &np->rcv_saddr);
70                         tw->tw_ipv6only = np->ipv6only;
71                 }
72 #endif
73                 /* Linkage updates. */
74                 __inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
75
76                 /* Get the TIME_WAIT timeout firing. */
77                 if (timeo < rto)
78                         timeo = rto;
79
80                 tw->tw_timeout = DCCP_TIMEWAIT_LEN;
81                 if (state == DCCP_TIME_WAIT)
82                         timeo = DCCP_TIMEWAIT_LEN;
83
84                 inet_twsk_schedule(tw, &dccp_death_row, timeo,
85                                    DCCP_TIMEWAIT_LEN);
86                 inet_twsk_put(tw);
87         } else {
88                 /* Sorry, if we're out of memory, just CLOSE this
89                  * socket up.  We've got bigger problems than
90                  * non-graceful socket closings.
91                  */
92                 DCCP_WARN("time wait bucket table overflow\n");
93         }
94
95         dccp_done(sk);
96 }
97
98 struct sock *dccp_create_openreq_child(struct sock *sk,
99                                        const struct request_sock *req,
100                                        const struct sk_buff *skb)
101 {
102         /*
103          * Step 3: Process LISTEN state
104          *
105          *   (* Generate a new socket and switch to that socket *)
106          *   Set S := new socket for this port pair
107          */
108         struct sock *newsk = inet_csk_clone(sk, req, GFP_ATOMIC);
109
110         if (newsk != NULL) {
111                 struct dccp_request_sock *dreq = dccp_rsk(req);
112                 struct inet_connection_sock *newicsk = inet_csk(newsk);
113                 struct dccp_sock *newdp = dccp_sk(newsk);
114                 struct dccp_minisock *newdmsk = dccp_msk(newsk);
115
116                 newdp->dccps_role           = DCCP_ROLE_SERVER;
117                 newdp->dccps_hc_rx_ackvec   = NULL;
118                 newdp->dccps_service_list   = NULL;
119                 newdp->dccps_service        = dreq->dreq_service;
120                 newdp->dccps_timestamp_echo = dreq->dreq_timestamp_echo;
121                 newdp->dccps_timestamp_time = dreq->dreq_timestamp_time;
122                 newicsk->icsk_rto           = DCCP_TIMEOUT_INIT;
123
124                 INIT_LIST_HEAD(&newdp->dccps_featneg);
125                 /*
126                  * Step 3: Process LISTEN state
127                  *
128                  *    Choose S.ISS (initial seqno) or set from Init Cookies
129                  *    Initialize S.GAR := S.ISS
130                  *    Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
131                  */
132
133                 /* See dccp_v4_conn_request */
134                 newdmsk->dccpms_sequence_window = req->rcv_wnd;
135
136                 newdp->dccps_gar = newdp->dccps_iss = dreq->dreq_iss;
137                 dccp_update_gss(newsk, dreq->dreq_iss);
138
139                 newdp->dccps_isr = dreq->dreq_isr;
140                 dccp_update_gsr(newsk, dreq->dreq_isr);
141
142                 /*
143                  * SWL and AWL are initially adjusted so that they are not less than
144                  * the initial Sequence Numbers received and sent, respectively:
145                  *      SWL := max(GSR + 1 - floor(W/4), ISR),
146                  *      AWL := max(GSS - W' + 1, ISS).
147                  * These adjustments MUST be applied only at the beginning of the
148                  * connection.
149                  */
150                 dccp_set_seqno(&newdp->dccps_swl,
151                                max48(newdp->dccps_swl, newdp->dccps_isr));
152                 dccp_set_seqno(&newdp->dccps_awl,
153                                max48(newdp->dccps_awl, newdp->dccps_iss));
154
155                 /*
156                  * Activate features after initialising the sequence numbers,
157                  * since CCID initialisation may depend on GSS, ISR, ISS etc.
158                  */
159                 if (dccp_feat_activate_values(newsk, &dreq->dreq_featneg)) {
160                         /* It is still raw copy of parent, so invalidate
161                          * destructor and make plain sk_free() */
162                         newsk->sk_destruct = NULL;
163                         sk_free(newsk);
164                         return NULL;
165                 }
166                 dccp_init_xmit_timers(newsk);
167
168                 DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS);
169         }
170         return newsk;
171 }
172
173 EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
174
175 /*
176  * Process an incoming packet for RESPOND sockets represented
177  * as an request_sock.
178  */
179 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
180                             struct request_sock *req,
181                             struct request_sock **prev)
182 {
183         struct sock *child = NULL;
184         struct dccp_request_sock *dreq = dccp_rsk(req);
185
186         /* Check for retransmitted REQUEST */
187         if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
188
189                 if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_isr)) {
190                         dccp_pr_debug("Retransmitted REQUEST\n");
191                         dreq->dreq_isr = DCCP_SKB_CB(skb)->dccpd_seq;
192                         /*
193                          * Send another RESPONSE packet
194                          * To protect against Request floods, increment retrans
195                          * counter (backoff, monitored by dccp_response_timer).
196                          */
197                         req->retrans++;
198                         req->rsk_ops->rtx_syn_ack(sk, req);
199                 }
200                 /* Network Duplicate, discard packet */
201                 return NULL;
202         }
203
204         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
205
206         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
207             dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
208                 goto drop;
209
210         /* Invalid ACK */
211         if (DCCP_SKB_CB(skb)->dccpd_ack_seq != dreq->dreq_iss) {
212                 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
213                               "dreq_iss=%llu\n",
214                               (unsigned long long)
215                               DCCP_SKB_CB(skb)->dccpd_ack_seq,
216                               (unsigned long long) dreq->dreq_iss);
217                 goto drop;
218         }
219
220         if (dccp_parse_options(sk, dreq, skb))
221                  goto drop;
222
223         child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
224         if (child == NULL)
225                 goto listen_overflow;
226
227         inet_csk_reqsk_queue_unlink(sk, req, prev);
228         inet_csk_reqsk_queue_removed(sk, req);
229         inet_csk_reqsk_queue_add(sk, req, child);
230 out:
231         return child;
232 listen_overflow:
233         dccp_pr_debug("listen_overflow!\n");
234         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
235 drop:
236         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
237                 req->rsk_ops->send_reset(sk, skb);
238
239         inet_csk_reqsk_queue_drop(sk, req, prev);
240         goto out;
241 }
242
243 EXPORT_SYMBOL_GPL(dccp_check_req);
244
245 /*
246  *  Queue segment on the new socket if the new socket is active,
247  *  otherwise we just shortcircuit this and continue with
248  *  the new socket.
249  */
250 int dccp_child_process(struct sock *parent, struct sock *child,
251                        struct sk_buff *skb)
252 {
253         int ret = 0;
254         const int state = child->sk_state;
255
256         if (!sock_owned_by_user(child)) {
257                 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
258                                              skb->len);
259
260                 /* Wakeup parent, send SIGIO */
261                 if (state == DCCP_RESPOND && child->sk_state != state)
262                         parent->sk_data_ready(parent, 0);
263         } else {
264                 /* Alas, it is possible again, because we do lookup
265                  * in main socket hash table and lock on listening
266                  * socket does not protect us more.
267                  */
268                 sk_add_backlog(child, skb);
269         }
270
271         bh_unlock_sock(child);
272         sock_put(child);
273         return ret;
274 }
275
276 EXPORT_SYMBOL_GPL(dccp_child_process);
277
278 void dccp_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
279                          struct request_sock *rsk)
280 {
281         DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state");
282 }
283
284 EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack);
285
286 int dccp_reqsk_init(struct request_sock *req,
287                     struct dccp_sock const *dp, struct sk_buff const *skb)
288 {
289         struct dccp_request_sock *dreq = dccp_rsk(req);
290
291         inet_rsk(req)->rmt_port   = dccp_hdr(skb)->dccph_sport;
292         inet_rsk(req)->loc_port   = dccp_hdr(skb)->dccph_dport;
293         inet_rsk(req)->acked      = 0;
294         req->rcv_wnd              = sysctl_dccp_feat_sequence_window;
295         dreq->dreq_timestamp_echo = 0;
296
297         /* inherit feature negotiation options from listening socket */
298         return dccp_feat_clone_list(&dp->dccps_featneg, &dreq->dreq_featneg);
299 }
300
301 EXPORT_SYMBOL_GPL(dccp_reqsk_init);