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