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 "ccid.h"
  23 #include "dccp.h"
  24
  25 struct inet_timewait_death_row dccp_death_row = {
  26         .sysctl_max_tw_buckets = NR_FILE * 2,
  27         .period         = DCCP_TIMEWAIT_LEN / INET_TWDR_TWKILL_SLOTS,
  28         .death_lock     = SPIN_LOCK_UNLOCKED,
  29         .hashinfo       = &dccp_hashinfo,
  30         .tw_timer       = TIMER_INITIALIZER(inet_twdr_hangman, 0,
  31                                             (unsigned long)&dccp_death_row),
  32         .twkill_work    = __WORK_INITIALIZER(dccp_death_row.twkill_work,
  33                                              inet_twdr_twkill_work,
  34                                              &dccp_death_row),
  35 /* Short-time timewait calendar */
  36
  37         .twcal_hand     = -1,
  38         .twcal_timer    = TIMER_INITIALIZER(inet_twdr_twcal_tick, 0,
  39                                             (unsigned long)&dccp_death_row),
  40 };
  41
  42 void dccp_time_wait(struct sock *sk, int state, int timeo)
  43 {
  44         struct inet_timewait_sock *tw = NULL;
  45
  46         if (dccp_death_row.tw_count < dccp_death_row.sysctl_max_tw_buckets)
  47                 tw = inet_twsk_alloc(sk, state);
  48
  49         if (tw != NULL) {
  50                 const struct inet_connection_sock *icsk = inet_csk(sk);
  51                 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
  52
  53                 /* Linkage updates. */
  54                 __inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
  55
  56                 /* Get the TIME_WAIT timeout firing. */
  57                 if (timeo < rto)
  58                         timeo = rto;
  59
  60                 tw->tw_timeout = DCCP_TIMEWAIT_LEN;
  61                 if (state == DCCP_TIME_WAIT)
  62                         timeo = DCCP_TIMEWAIT_LEN;
  63
  64                 inet_twsk_schedule(tw, &dccp_death_row, timeo,
  65                                    DCCP_TIMEWAIT_LEN);
  66                 inet_twsk_put(tw);
  67         } else {
  68                 /* Sorry, if we're out of memory, just CLOSE this
  69                  * socket up.  We've got bigger problems than
  70                  * non-graceful socket closings.
  71                  */
  72                 LIMIT_NETDEBUG(KERN_INFO "DCCP: time wait bucket "
  73                                          "table overflow\n");
  74         }
  75
  76         dccp_done(sk);
  77 }
  78
  79 struct sock *dccp_create_openreq_child(struct sock *sk,
  80                                        const struct request_sock *req,
  81                                        const struct sk_buff *skb)
  82 {
  83         /*
  84          * Step 3: Process LISTEN state
  85          *
  86          * // Generate a new socket and switch to that socket
  87          * Set S := new socket for this port pair
  88          */
  89         struct sock *newsk = inet_csk_clone(sk, req, GFP_ATOMIC);
  90
  91         if (newsk != NULL) {
  92                 const struct dccp_request_sock *dreq = dccp_rsk(req);
  93                 struct inet_connection_sock *newicsk = inet_csk(sk);
  94                 struct dccp_sock *newdp = dccp_sk(newsk);
  95
  96                 newdp->dccps_role          = DCCP_ROLE_SERVER;
  97                 newdp->dccps_hc_rx_ackpkts = NULL;
  98                 newdp->dccps_service_list  = NULL;
  99                 newdp->dccps_service       = dreq->dreq_service;
 100                 newicsk->icsk_rto          = DCCP_TIMEOUT_INIT;
 101                 do_gettimeofday(&newdp->dccps_epoch);
 102
 103                 if (newdp->dccps_options.dccpo_send_ack_vector) {
 104                         newdp->dccps_hc_rx_ackpkts =
 105                                 dccp_ackpkts_alloc(DCCP_MAX_ACK_VECTOR_LEN,
 106                                                    GFP_ATOMIC);
 107                         /*
 108                          * XXX: We're using the same CCIDs set on the parent,
 109                          * i.e. sk_clone copied the master sock and left the
 110                          * CCID pointers for this child, that is why we do the
 111                          * __ccid_get calls.
 112                          */
 113                         if (unlikely(newdp->dccps_hc_rx_ackpkts == NULL))
 114                                 goto out_free;
 115                 }
 116
 117                 if (unlikely(ccid_hc_rx_init(newdp->dccps_hc_rx_ccid,
 118                                              newsk) != 0 ||
 119                              ccid_hc_tx_init(newdp->dccps_hc_tx_ccid,
 120                                              newsk) != 0)) {
 121                         dccp_ackpkts_free(newdp->dccps_hc_rx_ackpkts);
 122                         ccid_hc_rx_exit(newdp->dccps_hc_rx_ccid, newsk);
 123                         ccid_hc_tx_exit(newdp->dccps_hc_tx_ccid, newsk);
 124 out_free:
 125                         /* It is still raw copy of parent, so invalidate
 126                          * destructor and make plain sk_free() */
 127                         newsk->sk_destruct = NULL;
 128                         sk_free(newsk);
 129                         return NULL;
 130                 }
 131
 132                 __ccid_get(newdp->dccps_hc_rx_ccid);
 133                 __ccid_get(newdp->dccps_hc_tx_ccid);
 134
 135                 /*
 136                  * Step 3: Process LISTEN state
 137                  *
 138                  *      Choose S.ISS (initial seqno) or set from Init Cookie
 139                  *      Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init
 140                  *      Cookie
 141                  */
 142
 143                 /* See dccp_v4_conn_request */
 144                 newdp->dccps_options.dccpo_sequence_window = req->rcv_wnd;
 145
 146                 newdp->dccps_gar = newdp->dccps_isr = dreq->dreq_isr;
 147                 dccp_update_gsr(newsk, dreq->dreq_isr);
 148
 149                 newdp->dccps_iss = dreq->dreq_iss;
 150                 dccp_update_gss(newsk, dreq->dreq_iss);
 151
 152                 /*
 153                  * SWL and AWL are initially adjusted so that they are not less than
 154                  * the initial Sequence Numbers received and sent, respectively:
 155                  *      SWL := max(GSR + 1 - floor(W/4), ISR),
 156                  *      AWL := max(GSS - W' + 1, ISS).
 157                  * These adjustments MUST be applied only at the beginning of the
 158                  * connection.
 159                  */
 160                 dccp_set_seqno(&newdp->dccps_swl,
 161                                max48(newdp->dccps_swl, newdp->dccps_isr));
 162                 dccp_set_seqno(&newdp->dccps_awl,
 163                                max48(newdp->dccps_awl, newdp->dccps_iss));
 164
 165                 dccp_init_xmit_timers(newsk);
 166
 167                 DCCP_INC_STATS_BH(DCCP_MIB_PASSIVEOPENS);
 168         }
 169         return newsk;
 170 }
 171
 172 /*
 173  * Process an incoming packet for RESPOND sockets represented
 174  * as an request_sock.
 175  */
 176 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
 177                             struct request_sock *req,
 178                             struct request_sock **prev)
 179 {
 180         struct sock *child = NULL;
 181
 182         /* Check for retransmitted REQUEST */
 183         if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
 184                 if (after48(DCCP_SKB_CB(skb)->dccpd_seq,
 185                             dccp_rsk(req)->dreq_isr)) {
 186                         struct dccp_request_sock *dreq = dccp_rsk(req);
 187
 188                         dccp_pr_debug("Retransmitted REQUEST\n");
 189                         /* Send another RESPONSE packet */
 190                         dccp_set_seqno(&dreq->dreq_iss, dreq->dreq_iss + 1);
 191                         dccp_set_seqno(&dreq->dreq_isr,
 192                                        DCCP_SKB_CB(skb)->dccpd_seq);
 193                         req->rsk_ops->rtx_syn_ack(sk, req, NULL);
 194                 }
 195                 /* Network Duplicate, discard packet */
 196                 return NULL;
 197         }
 198
 199         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
 200
 201         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
 202             dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
 203                 goto drop;
 204
 205         /* Invalid ACK */
 206         if (DCCP_SKB_CB(skb)->dccpd_ack_seq != dccp_rsk(req)->dreq_iss) {
 207                 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
 208                               "dreq_iss=%llu\n",
 209                               (unsigned long long)
 210                               DCCP_SKB_CB(skb)->dccpd_ack_seq,
 211                               (unsigned long long)
 212                               dccp_rsk(req)->dreq_iss);
 213                 goto drop;
 214         }
 215
 216         child = dccp_v4_request_recv_sock(sk, skb, req, NULL);
 217         if (child == NULL)
 218                 goto listen_overflow;
 219
 220         /* FIXME: deal with options */
 221
 222         inet_csk_reqsk_queue_unlink(sk, req, prev);
 223         inet_csk_reqsk_queue_removed(sk, req);
 224         inet_csk_reqsk_queue_add(sk, req, child);
 225 out:
 226         return child;
 227 listen_overflow:
 228         dccp_pr_debug("listen_overflow!\n");
 229         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 230 drop:
 231         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
 232                 req->rsk_ops->send_reset(skb);
 233
 234         inet_csk_reqsk_queue_drop(sk, req, prev);
 235         goto out;
 236 }
 237
 238 /*
 239  *  Queue segment on the new socket if the new socket is active,
 240  *  otherwise we just shortcircuit this and continue with
 241  *  the new socket.
 242  */
 243 int dccp_child_process(struct sock *parent, struct sock *child,
 244                        struct sk_buff *skb)
 245 {
 246         int ret = 0;
 247         const int state = child->sk_state;
 248
 249         if (!sock_owned_by_user(child)) {
 250                 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
 251                                              skb->len);
 252
 253                 /* Wakeup parent, send SIGIO */
 254                 if (state == DCCP_RESPOND && child->sk_state != state)
 255                         parent->sk_data_ready(parent, 0);
 256         } else {
 257                 /* Alas, it is possible again, because we do lookup
 258                  * in main socket hash table and lock on listening
 259                  * socket does not protect us more.
 260                  */
 261                 sk_add_backlog(child, skb);
 262         }
 263
 264         bh_unlock_sock(child);
 265         sock_put(child);
 266         return ret;
 267 }