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         .hashinfo       = &dccp_hashinfo,
  31 };
  32
  33 EXPORT_SYMBOL_GPL(dccp_death_row);
  34
  35 void dccp_time_wait(struct sock *sk, int state, int timeo)
  36 {
  37         struct inet_timewait_sock *tw;
  38
  39         tw = inet_twsk_alloc(sk, &dccp_death_row, state);
  40
  41         if (tw != NULL) {
  42                 const struct inet_connection_sock *icsk = inet_csk(sk);
  43                 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
  44 #if IS_ENABLED(CONFIG_IPV6)
  45                 if (tw->tw_family == PF_INET6) {
  46                         tw->tw_v6_daddr = sk->sk_v6_daddr;
  47                         tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
  48                         tw->tw_ipv6only = sk->sk_ipv6only;
  49                 }
  50 #endif
  51
  52                 /* Get the TIME_WAIT timeout firing. */
  53                 if (timeo < rto)
  54                         timeo = rto;
  55
  56                 tw->tw_timeout = DCCP_TIMEWAIT_LEN;
  57                 if (state == DCCP_TIME_WAIT)
  58                         timeo = DCCP_TIMEWAIT_LEN;
  59
  60                 /* tw_timer is pinned, so we need to make sure BH are disabled
  61                  * in following section, otherwise timer handler could run before
  62                  * we complete the initialization.
  63                  */
  64                 local_bh_disable();
  65                 inet_twsk_schedule(tw, timeo);
  66                 /* Linkage updates. */
  67                 __inet_twsk_hashdance(tw, sk, &dccp_hashinfo);
  68                 inet_twsk_put(tw);
  69                 local_bh_enable();
  70         } else {
  71                 /* Sorry, if we're out of memory, just CLOSE this
  72                  * socket up.  We've got bigger problems than
  73                  * non-graceful socket closings.
  74                  */
  75                 DCCP_WARN("time wait bucket table overflow\n");
  76         }
  77
  78         dccp_done(sk);
  79 }
  80
  81 struct sock *dccp_create_openreq_child(const struct sock *sk,
  82                                        const struct request_sock *req,
  83                                        const struct sk_buff *skb)
  84 {
  85         /*
  86          * Step 3: Process LISTEN state
  87          *
  88          *   (* Generate a new socket and switch to that socket *)
  89          *   Set S := new socket for this port pair
  90          */
  91         struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
  92
  93         if (newsk != NULL) {
  94                 struct dccp_request_sock *dreq = dccp_rsk(req);
  95                 struct inet_connection_sock *newicsk = inet_csk(newsk);
  96                 struct dccp_sock *newdp = dccp_sk(newsk);
  97
  98                 newdp->dccps_role           = DCCP_ROLE_SERVER;
  99                 newdp->dccps_hc_rx_ackvec   = NULL;
 100                 newdp->dccps_service_list   = NULL;
 101                 newdp->dccps_hc_rx_ccid     = NULL;
 102                 newdp->dccps_hc_tx_ccid     = NULL;
 103                 newdp->dccps_service        = dreq->dreq_service;
 104                 newdp->dccps_timestamp_echo = dreq->dreq_timestamp_echo;
 105                 newdp->dccps_timestamp_time = dreq->dreq_timestamp_time;
 106                 newicsk->icsk_rto           = DCCP_TIMEOUT_INIT;
 107
 108                 INIT_LIST_HEAD(&newdp->dccps_featneg);
 109                 /*
 110                  * Step 3: Process LISTEN state
 111                  *
 112                  *    Choose S.ISS (initial seqno) or set from Init Cookies
 113                  *    Initialize S.GAR := S.ISS
 114                  *    Set S.ISR, S.GSR from packet (or Init Cookies)
 115                  *
 116                  *    Setting AWL/AWH and SWL/SWH happens as part of the feature
 117                  *    activation below, as these windows all depend on the local
 118                  *    and remote Sequence Window feature values (7.5.2).
 119                  */
 120                 newdp->dccps_iss = dreq->dreq_iss;
 121                 newdp->dccps_gss = dreq->dreq_gss;
 122                 newdp->dccps_gar = newdp->dccps_iss;
 123                 newdp->dccps_isr = dreq->dreq_isr;
 124                 newdp->dccps_gsr = dreq->dreq_gsr;
 125
 126                 /*
 127                  * Activate features: initialise CCIDs, sequence windows etc.
 128                  */
 129                 if (dccp_feat_activate_values(newsk, &dreq->dreq_featneg)) {
 130                         /* It is still raw copy of parent, so invalidate
 131                          * destructor and make plain sk_free() */
 132                         newsk->sk_destruct = NULL;
 133                         bh_unlock_sock(newsk);
 134                         sk_free(newsk);
 135                         return NULL;
 136                 }
 137                 dccp_init_xmit_timers(newsk);
 138
 139                 __DCCP_INC_STATS(DCCP_MIB_PASSIVEOPENS);
 140         }
 141         return newsk;
 142 }
 143
 144 EXPORT_SYMBOL_GPL(dccp_create_openreq_child);
 145
 146 /*
 147  * Process an incoming packet for RESPOND sockets represented
 148  * as an request_sock.
 149  */
 150 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb,
 151                             struct request_sock *req)
 152 {
 153         struct sock *child = NULL;
 154         struct dccp_request_sock *dreq = dccp_rsk(req);
 155         bool own_req;
 156
 157         /* TCP/DCCP listeners became lockless.
 158          * DCCP stores complex state in its request_sock, so we need
 159          * a protection for them, now this code runs without being protected
 160          * by the parent (listener) lock.
 161          */
 162         spin_lock_bh(&dreq->dreq_lock);
 163
 164         /* Check for retransmitted REQUEST */
 165         if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
 166
 167                 if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_gsr)) {
 168                         dccp_pr_debug("Retransmitted REQUEST\n");
 169                         dreq->dreq_gsr = DCCP_SKB_CB(skb)->dccpd_seq;
 170                         /*
 171                          * Send another RESPONSE packet
 172                          * To protect against Request floods, increment retrans
 173                          * counter (backoff, monitored by dccp_response_timer).
 174                          */
 175                         inet_rtx_syn_ack(sk, req);
 176                 }
 177                 /* Network Duplicate, discard packet */
 178                 goto out;
 179         }
 180
 181         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
 182
 183         if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK &&
 184             dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK)
 185                 goto drop;
 186
 187         /* Invalid ACK */
 188         if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
 189                                 dreq->dreq_iss, dreq->dreq_gss)) {
 190                 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, "
 191                               "dreq_iss=%llu, dreq_gss=%llu\n",
 192                               (unsigned long long)
 193                               DCCP_SKB_CB(skb)->dccpd_ack_seq,
 194                               (unsigned long long) dreq->dreq_iss,
 195                               (unsigned long long) dreq->dreq_gss);
 196                 goto drop;
 197         }
 198
 199         if (dccp_parse_options(sk, dreq, skb))
 200                  goto drop;
 201
 202         child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
 203                                                          req, &own_req);
 204         if (child) {
 205                 child = inet_csk_complete_hashdance(sk, child, req, own_req);
 206                 goto out;
 207         }
 208
 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 out:
 216         spin_unlock_bh(&dreq->dreq_lock);
 217         return child;
 218 }
 219
 220 EXPORT_SYMBOL_GPL(dccp_check_req);
 221
 222 /*
 223  *  Queue segment on the new socket if the new socket is active,
 224  *  otherwise we just shortcircuit this and continue with
 225  *  the new socket.
 226  */
 227 int dccp_child_process(struct sock *parent, struct sock *child,
 228                        struct sk_buff *skb)
 229 {
 230         int ret = 0;
 231         const int state = child->sk_state;
 232
 233         if (!sock_owned_by_user(child)) {
 234                 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb),
 235                                              skb->len);
 236
 237                 /* Wakeup parent, send SIGIO */
 238                 if (state == DCCP_RESPOND && child->sk_state != state)
 239                         parent->sk_data_ready(parent);
 240         } else {
 241                 /* Alas, it is possible again, because we do lookup
 242                  * in main socket hash table and lock on listening
 243                  * socket does not protect us more.
 244                  */
 245                 __sk_add_backlog(child, skb);
 246         }
 247
 248         bh_unlock_sock(child);
 249         sock_put(child);
 250         return ret;
 251 }
 252
 253 EXPORT_SYMBOL_GPL(dccp_child_process);
 254
 255 void dccp_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
 256                          struct request_sock *rsk)
 257 {
 258         DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state");
 259 }
 260
 261 EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack);
 262
 263 int dccp_reqsk_init(struct request_sock *req,
 264                     struct dccp_sock const *dp, struct sk_buff const *skb)
 265 {
 266         struct dccp_request_sock *dreq = dccp_rsk(req);
 267
 268         spin_lock_init(&dreq->dreq_lock);
 269         inet_rsk(req)->ir_rmt_port = dccp_hdr(skb)->dccph_sport;
 270         inet_rsk(req)->ir_num      = ntohs(dccp_hdr(skb)->dccph_dport);
 271         inet_rsk(req)->acked       = 0;
 272         dreq->dreq_timestamp_echo  = 0;
 273
 274         /* inherit feature negotiation options from listening socket */
 275         return dccp_feat_clone_list(&dp->dccps_featneg, &dreq->dreq_featneg);
 276 }
 277
 278 EXPORT_SYMBOL_GPL(dccp_reqsk_init);