SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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/config.h>
  14 #include <linux/dccp.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,
  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                                              &dccp_death_row),
  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 defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
  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                         ipv6_addr_copy(&tw6->tw_v6_daddr, &np->daddr);
  64                         ipv6_addr_copy(&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                 LIMIT_NETDEBUG(KERN_INFO "DCCP: time wait bucket "
  88                                          "table overflow\n");
  89         }
  90
  91         dccp_done(sk);
  92 }
  93
  94 struct sock *dccp_create_openreq_child(struct sock *sk,
  95                                        const struct request_sock *req,
  96                                        const struct sk_buff *skb)
  97 {
  98         /*
  99          * Step 3: Process LISTEN state
 100          *
 101          * // Generate a new socket and switch to that socket
 102          * Set S := new socket for this port pair
 103          */
 104         struct sock *newsk = inet_csk_clone(sk, req, GFP_ATOMIC);
 105
 106         if (newsk != NULL) {
 107                 const struct dccp_request_sock *dreq = dccp_rsk(req);
 108                 struct inet_connection_sock *newicsk = inet_csk(sk);
 109                 struct dccp_sock *newdp = dccp_sk(newsk);
 110
 111                 newdp->dccps_role          = DCCP_ROLE_SERVER;
 112                 newdp->dccps_hc_rx_ackvec  = NULL;
 113                 newdp->dccps_service_list  = NULL;
 114                 newdp->dccps_service       = dreq->dreq_service;
 115                 newicsk->icsk_rto          = DCCP_TIMEOUT_INIT;
 116                 do_gettimeofday(&newdp->dccps_epoch);
 117
 118                 if (dccp_feat_clone(sk, newsk))
 119                         goto out_free;
 120
 121                 if (newdp->dccps_options.dccpo_send_ack_vector) {
 122                         newdp->dccps_hc_rx_ackvec =
 123                                                 dccp_ackvec_alloc(GFP_ATOMIC);
 124                         /*
 125                          * XXX: We're using the same CCIDs set on the parent,
 126                          * i.e. sk_clone copied the master sock and left the
 127                          * CCID pointers for this child, that is why we do the
 128                          * __ccid_get calls.
 129                          */
 130                         if (unlikely(newdp->dccps_hc_rx_ackvec == NULL))
 131                                 goto out_free;
 132                 }
 133
 134                 if (unlikely(ccid_hc_rx_init(newdp->dccps_hc_rx_ccid,
 135                                              newsk) != 0 ||
 136                              ccid_hc_tx_init(newdp->dccps_hc_tx_ccid,
 137                                              newsk) != 0)) {
 138                         dccp_ackvec_free(newdp->dccps_hc_rx_ackvec);
 139                         ccid_hc_rx_exit(newdp->dccps_hc_rx_ccid, newsk);
 140                         ccid_hc_tx_exit(newdp->dccps_hc_tx_ccid, newsk);
 141 out_free:
 142                         /* It is still raw copy of parent, so invalidate
 143                          * destructor and make plain sk_free() */
 144                         newsk->sk_destruct = NULL;
 145                         sk_free(newsk);
 146                         return NULL;
 147                 }
 148
 149                 __ccid_get(newdp->dccps_hc_rx_ccid);
 150                 __ccid_get(newdp->dccps_hc_tx_ccid);
 151
 152                 /*
 153                  * Step 3: Process LISTEN state
 154                  *
 155                  *      Choose S.ISS (initial seqno) or set from Init Cookie
 156                  *      Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init
 157                  *      Cookie
 158                  */
 159
 160                 /* See dccp_v4_conn_request */
 161                 newdp->dccps_options.dccpo_sequence_window = req->rcv_wnd;
 162
 163                 newdp->dccps_gar = newdp->dccps_isr = dreq->dreq_isr;
 164                 dccp_update_gsr(newsk, dreq->dreq_isr);
 165
 166                 newdp->dccps_iss = dreq->dreq_iss;
 167                 dccp_update_gss(newsk, dreq->dreq_iss);
 168
 169                 /*
 170                  * SWL and AWL are initially adjusted so that they are not less than
 171                  * the initial Sequence Numbers received and sent, respectively:
 172                  *      SWL := max(GSR + 1 - floor(W/4), ISR),
 173                  *      AWL := max(GSS - W' + 1, ISS).
 174                  * These adjustments MUST be applied only at the beginning of the
 175                  * connection.
 176                  */
 177                 dccp_set_seqno(&newdp->dccps_swl,
 178                                max48(newdp->dccps_swl, newdp->dccps_isr));
 179                 dccp_set_seqno(&newdp->dccps_awl,
 180                                max48(newdp->dccps_awl, newdp->dccps_iss));
 181
 182                 dccp_init_xmit_timers(newsk);
 183
 184                 DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS);
 185         }
 186         return newsk;
 187 }
 188
 189 EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
 190
 191 /*
 192  * Process an incoming packet for RESPOND sockets represented
 193  * as an request_sock.
 194  */
 195 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
 196                             struct request_sock *req,
 197                             struct request_sock **prev)
 198 {
 199         struct sock *child = NULL;
 200
 201         /* Check for retransmitted REQUEST */
 202         if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
 203                 if (after48(DCCP_SKB_CB(skb)->dccpd_seq,
 204                             dccp_rsk(req)->dreq_isr)) {
 205                         struct dccp_request_sock *dreq = dccp_rsk(req);
 206
 207                         dccp_pr_debug("Retransmitted REQUEST\n");
 208                         /* Send another RESPONSE packet */
 209                         dccp_set_seqno(&dreq->dreq_iss, dreq->dreq_iss + 1);
 210                         dccp_set_seqno(&dreq->dreq_isr,
 211                                        DCCP_SKB_CB(skb)->dccpd_seq);
 212                         req->rsk_ops->rtx_syn_ack(sk, req, NULL);
 213                 }
 214                 /* Network Duplicate, discard packet */
 215                 return NULL;
 216         }
 217
 218         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
 219
 220         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
 221             dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
 222                 goto drop;
 223
 224         /* Invalid ACK */
 225         if (DCCP_SKB_CB(skb)->dccpd_ack_seq != dccp_rsk(req)->dreq_iss) {
 226                 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
 227                               "dreq_iss=%llu\n",
 228                               (unsigned long long)
 229                               DCCP_SKB_CB(skb)->dccpd_ack_seq,
 230                               (unsigned long long)
 231                               dccp_rsk(req)->dreq_iss);
 232                 goto drop;
 233         }
 234
 235         child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
 236         if (child == NULL)
 237                 goto listen_overflow;
 238
 239         /* FIXME: deal with options */
 240
 241         inet_csk_reqsk_queue_unlink(sk, req, prev);
 242         inet_csk_reqsk_queue_removed(sk, req);
 243         inet_csk_reqsk_queue_add(sk, req, child);
 244 out:
 245         return child;
 246 listen_overflow:
 247         dccp_pr_debug("listen_overflow!\n");
 248         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 249 drop:
 250         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
 251                 req->rsk_ops->send_reset(skb);
 252
 253         inet_csk_reqsk_queue_drop(sk, req, prev);
 254         goto out;
 255 }
 256
 257 EXPORT_SYMBOL_GPL(dccp_check_req);
 258
 259 /*
 260  *  Queue segment on the new socket if the new socket is active,
 261  *  otherwise we just shortcircuit this and continue with
 262  *  the new socket.
 263  */
 264 int dccp_child_process(struct sock *parent, struct sock *child,
 265                        struct sk_buff *skb)
 266 {
 267         int ret = 0;
 268         const int state = child->sk_state;
 269
 270         if (!sock_owned_by_user(child)) {
 271                 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
 272                                              skb->len);
 273
 274                 /* Wakeup parent, send SIGIO */
 275                 if (state == DCCP_RESPOND && child->sk_state != state)
 276                         parent->sk_data_ready(parent, 0);
 277         } else {
 278                 /* Alas, it is possible again, because we do lookup
 279                  * in main socket hash table and lock on listening
 280                  * socket does not protect us more.
 281                  */
 282                 sk_add_backlog(child, skb);
 283         }
 284
 285         bh_unlock_sock(child);
 286         sock_put(child);
 287         return ret;
 288 }
 289
 290 EXPORT_SYMBOL_GPL(dccp_child_process);