]> git.karo-electronics.de Git - karo-tx-linux.git/blob - net/dccp/minisocks.c
ipv4: ip_tunnel: use net namespace from rtable not socket
[karo-tx-linux.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                         tw->tw_v6_daddr = sk->sk_v6_daddr;
59                         tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
60                         tw->tw_ipv6only = sk->sk_ipv6only;
61                 }
62 #endif
63                 /* Linkage updates. */
64                 __inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
65
66                 /* Get the TIME_WAIT timeout firing. */
67                 if (timeo < rto)
68                         timeo = rto;
69
70                 tw->tw_timeout = DCCP_TIMEWAIT_LEN;
71                 if (state == DCCP_TIME_WAIT)
72                         timeo = DCCP_TIMEWAIT_LEN;
73
74                 inet_twsk_schedule(tw, &dccp_death_row, timeo,
75                                    DCCP_TIMEWAIT_LEN);
76                 inet_twsk_put(tw);
77         } else {
78                 /* Sorry, if we're out of memory, just CLOSE this
79                  * socket up.  We've got bigger problems than
80                  * non-graceful socket closings.
81                  */
82                 DCCP_WARN("time wait bucket table overflow\n");
83         }
84
85         dccp_done(sk);
86 }
87
88 struct sock *dccp_create_openreq_child(struct sock *sk,
89                                        const struct request_sock *req,
90                                        const struct sk_buff *skb)
91 {
92         /*
93          * Step 3: Process LISTEN state
94          *
95          *   (* Generate a new socket and switch to that socket *)
96          *   Set S := new socket for this port pair
97          */
98         struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
99
100         if (newsk != NULL) {
101                 struct dccp_request_sock *dreq = dccp_rsk(req);
102                 struct inet_connection_sock *newicsk = inet_csk(newsk);
103                 struct dccp_sock *newdp = dccp_sk(newsk);
104
105                 newdp->dccps_role           = DCCP_ROLE_SERVER;
106                 newdp->dccps_hc_rx_ackvec   = NULL;
107                 newdp->dccps_service_list   = NULL;
108                 newdp->dccps_service        = dreq->dreq_service;
109                 newdp->dccps_timestamp_echo = dreq->dreq_timestamp_echo;
110                 newdp->dccps_timestamp_time = dreq->dreq_timestamp_time;
111                 newicsk->icsk_rto           = DCCP_TIMEOUT_INIT;
112
113                 INIT_LIST_HEAD(&newdp->dccps_featneg);
114                 /*
115                  * Step 3: Process LISTEN state
116                  *
117                  *    Choose S.ISS (initial seqno) or set from Init Cookies
118                  *    Initialize S.GAR := S.ISS
119                  *    Set S.ISR, S.GSR from packet (or Init Cookies)
120                  *
121                  *    Setting AWL/AWH and SWL/SWH happens as part of the feature
122                  *    activation below, as these windows all depend on the local
123                  *    and remote Sequence Window feature values (7.5.2).
124                  */
125                 newdp->dccps_iss = dreq->dreq_iss;
126                 newdp->dccps_gss = dreq->dreq_gss;
127                 newdp->dccps_gar = newdp->dccps_iss;
128                 newdp->dccps_isr = dreq->dreq_isr;
129                 newdp->dccps_gsr = dreq->dreq_gsr;
130
131                 /*
132                  * Activate features: initialise CCIDs, sequence windows etc.
133                  */
134                 if (dccp_feat_activate_values(newsk, &dreq->dreq_featneg)) {
135                         /* It is still raw copy of parent, so invalidate
136                          * destructor and make plain sk_free() */
137                         newsk->sk_destruct = NULL;
138                         sk_free(newsk);
139                         return NULL;
140                 }
141                 dccp_init_xmit_timers(newsk);
142
143                 DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS);
144         }
145         return newsk;
146 }
147
148 EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
149
150 /*
151  * Process an incoming packet for RESPOND sockets represented
152  * as an request_sock.
153  */
154 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
155                             struct request_sock *req)
156 {
157         struct sock *child = NULL;
158         struct dccp_request_sock *dreq = dccp_rsk(req);
159
160         /* Check for retransmitted REQUEST */
161         if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
162
163                 if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_gsr)) {
164                         dccp_pr_debug("Retransmitted REQUEST\n");
165                         dreq->dreq_gsr = DCCP_SKB_CB(skb)->dccpd_seq;
166                         /*
167                          * Send another RESPONSE packet
168                          * To protect against Request floods, increment retrans
169                          * counter (backoff, monitored by dccp_response_timer).
170                          */
171                         inet_rtx_syn_ack(sk, req);
172                 }
173                 /* Network Duplicate, discard packet */
174                 return NULL;
175         }
176
177         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
178
179         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
180             dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
181                 goto drop;
182
183         /* Invalid ACK */
184         if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
185                                 dreq->dreq_iss, dreq->dreq_gss)) {
186                 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
187                               "dreq_iss=%llu, dreq_gss=%llu\n",
188                               (unsigned long long)
189                               DCCP_SKB_CB(skb)->dccpd_ack_seq,
190                               (unsigned long long) dreq->dreq_iss,
191                               (unsigned long long) dreq->dreq_gss);
192                 goto drop;
193         }
194
195         if (dccp_parse_options(sk, dreq, skb))
196                  goto drop;
197
198         child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
199         if (child == NULL)
200                 goto listen_overflow;
201
202         inet_csk_reqsk_queue_unlink(sk, req);
203         inet_csk_reqsk_queue_removed(sk, req);
204         inet_csk_reqsk_queue_add(sk, req, child);
205 out:
206         return child;
207 listen_overflow:
208         dccp_pr_debug("listen_overflow!\n");
209         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
210 drop:
211         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
212                 req->rsk_ops->send_reset(sk, skb);
213
214         inet_csk_reqsk_queue_drop(sk, req);
215         goto out;
216 }
217
218 EXPORT_SYMBOL_GPL(dccp_check_req);
219
220 /*
221  *  Queue segment on the new socket if the new socket is active,
222  *  otherwise we just shortcircuit this and continue with
223  *  the new socket.
224  */
225 int dccp_child_process(struct sock *parent, struct sock *child,
226                        struct sk_buff *skb)
227 {
228         int ret = 0;
229         const int state = child->sk_state;
230
231         if (!sock_owned_by_user(child)) {
232                 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
233                                              skb->len);
234
235                 /* Wakeup parent, send SIGIO */
236                 if (state == DCCP_RESPOND && child->sk_state != state)
237                         parent->sk_data_ready(parent);
238         } else {
239                 /* Alas, it is possible again, because we do lookup
240                  * in main socket hash table and lock on listening
241                  * socket does not protect us more.
242                  */
243                 __sk_add_backlog(child, skb);
244         }
245
246         bh_unlock_sock(child);
247         sock_put(child);
248         return ret;
249 }
250
251 EXPORT_SYMBOL_GPL(dccp_child_process);
252
253 void dccp_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
254                          struct request_sock *rsk)
255 {
256         DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state");
257 }
258
259 EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack);
260
261 int dccp_reqsk_init(struct request_sock *req,
262                     struct dccp_sock const *dp, struct sk_buff const *skb)
263 {
264         struct dccp_request_sock *dreq = dccp_rsk(req);
265
266         inet_rsk(req)->ir_rmt_port = dccp_hdr(skb)->dccph_sport;
267         inet_rsk(req)->ir_num      = ntohs(dccp_hdr(skb)->dccph_dport);
268         inet_rsk(req)->acked       = 0;
269         dreq->dreq_timestamp_echo  = 0;
270
271         /* inherit feature negotiation options from listening socket */
272         return dccp_feat_clone_list(&dp->dccps_featneg, &dreq->dreq_featneg);
273 }
274
275 EXPORT_SYMBOL_GPL(dccp_reqsk_init);