BSD 4_4_Lite1 release
[unix-history] / usr / src / sys / netinet / tcp_usrreq.c
index a618b6f..38a08d6 100644 (file)
-/* tcp_usrreq.c 1.44 81/12/20 */
-
-#include "../h/param.h"
-#include "../h/systm.h"
-#include "../h/mbuf.h"
-#include "../h/socket.h"
-#include "../h/socketvar.h"
-#include "../h/protosw.h"
-#include "../net/in.h"
-#include "../net/in_pcb.h"
-#include "../net/in_systm.h"
-#include "../net/if.h"
-#include "../net/ip.h"
-#include "../net/ip_var.h"
-#include "../net/tcp.h"
-#include "../net/tcp_fsm.h"
-#include "../net/tcp_seq.h"
-#include "../net/tcp_timer.h"
-#include "../net/tcp_var.h"
-#include "../net/tcpip.h"
-#include "../net/tcp_debug.h"
-#include "../errno.h"
-
-extern char *tcpstates[];
-struct tcpcb *tcp_newtcpcb();
 /*
 /*
- * Process a TCP user request for tcp tb.  If this is a send request
+ * Copyright (c) 1982, 1986, 1988, 1993
+ *     The Regents of the University of California.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *     This product includes software developed by the University of
+ *     California, Berkeley and its contributors.
+ * 4. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ *     @(#)tcp_usrreq.c        8.2 (Berkeley) 1/3/94
+ */
+
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/malloc.h>
+#include <sys/mbuf.h>
+#include <sys/socket.h>
+#include <sys/socketvar.h>
+#include <sys/protosw.h>
+#include <sys/errno.h>
+#include <sys/stat.h>
+
+#include <net/if.h>
+#include <net/route.h>
+
+#include <netinet/in.h>
+#include <netinet/in_systm.h>
+#include <netinet/ip.h>
+#include <netinet/in_pcb.h>
+#include <netinet/ip_var.h>
+#include <netinet/tcp.h>
+#include <netinet/tcp_fsm.h>
+#include <netinet/tcp_seq.h>
+#include <netinet/tcp_timer.h>
+#include <netinet/tcp_var.h>
+#include <netinet/tcpip.h>
+#include <netinet/tcp_debug.h>
+
+/*
+ * TCP protocol interface to socket abstraction.
+ */
+extern char *tcpstates[];
+
+/*
+ * Process a TCP user request for TCP tb.  If this is a send request
  * then m is the mbuf chain of send data.  If this is a timer expiration
  * (called from the software clock routine), then timertype tells which timer.
  */
  * then m is the mbuf chain of send data.  If this is a timer expiration
  * (called from the software clock routine), then timertype tells which timer.
  */
-tcp_usrreq(so, req, m, addr)
+/*ARGSUSED*/
+int
+tcp_usrreq(so, req, m, nam, control)
        struct socket *so;
        int req;
        struct socket *so;
        int req;
-       struct mbuf *m;
-       caddr_t addr;
+       struct mbuf *m, *nam, *control;
 {
 {
-       register struct inpcb *inp = sotoinpcb(so);
+       register struct inpcb *inp;
        register struct tcpcb *tp;
        register struct tcpcb *tp;
-       int s = splnet();
+       int s;
        int error = 0;
        int ostate;
        int error = 0;
        int ostate;
-COUNT(TCP_USRREQ);
 
 
+       if (req == PRU_CONTROL)
+               return (in_control(so, (int)m, (caddr_t)nam,
+                       (struct ifnet *)control));
+       if (control && control->m_len) {
+               m_freem(control);
+               if (m)
+                       m_freem(m);
+               return (EINVAL);
+       }
+
+       s = splnet();
+       inp = sotoinpcb(so);
        /*
        /*
-        * Make sure attached.  If not,
-        * only PRU_ATTACH is valid.
+        * When a TCP is attached to a socket, then there will be
+        * a (struct inpcb) pointed at by the socket, and this
+        * structure will point at a subsidary (struct tcpcb).
         */
        if (inp == 0 && req != PRU_ATTACH) {
                splx(s);
         */
        if (inp == 0 && req != PRU_ATTACH) {
                splx(s);
-               return (EINVAL);
+               return (EINVAL);                /* XXX */
        }
        if (inp) {
                tp = intotcpcb(inp);
        }
        if (inp) {
                tp = intotcpcb(inp);
+               /* WHAT IF TP IS 0? */
 #ifdef KPROF
                tcp_acounts[tp->t_state][req]++;
 #endif
                ostate = tp->t_state;
 #ifdef KPROF
                tcp_acounts[tp->t_state][req]++;
 #endif
                ostate = tp->t_state;
-       }
+       } else
+               ostate = 0;
        switch (req) {
 
        switch (req) {
 
+       /*
+        * TCP attaches to socket via PRU_ATTACH, reserving space,
+        * and an internet control block.
+        */
        case PRU_ATTACH:
                if (inp) {
                        error = EISCONN;
                        break;
                }
        case PRU_ATTACH:
                if (inp) {
                        error = EISCONN;
                        break;
                }
-               error = in_pcbattach(so, &tcb, 2048, 2048, (struct sockaddr_in *)addr);
+               error = tcp_attach(so);
                if (error)
                        break;
                if (error)
                        break;
-               inp = (struct inpcb *)so->so_pcb;
-               tp = tcp_newtcpcb(inp);
-               if (so->so_options & SO_ACCEPTCONN) {
-                       if (tp == 0) {
-                               in_pcbdetach(inp);
-                               error = ENOBUFS;
-                               break;
-                       }
-                       tp->t_state = TCPS_LISTEN;
-               } else
-                       tp->t_state = TCPS_CLOSED;
+               if ((so->so_options & SO_LINGER) && so->so_linger == 0)
+                       so->so_linger = TCP_LINGERTIME;
+               tp = sototcpcb(so);
                break;
 
                break;
 
+       /*
+        * PRU_DETACH detaches the TCP protocol from the socket.
+        * If the protocol state is non-embryonic, then can't
+        * do this directly: have to initiate a PRU_DISCONNECT,
+        * which may finish later; embryonic TCB's can just
+        * be discarded here.
+        */
        case PRU_DETACH:
        case PRU_DETACH:
-               if (tp)
-                       goto disconn;
-               in_pcbdetach(inp);
+               if (tp->t_state > TCPS_LISTEN)
+                       tp = tcp_disconnect(tp);
+               else
+                       tp = tcp_close(tp);
+               break;
+
+       /*
+        * Give the socket an address.
+        */
+       case PRU_BIND:
+               error = in_pcbbind(inp, nam);
+               if (error)
+                       break;
                break;
 
                break;
 
+       /*
+        * Prepare to accept connections.
+        */
+       case PRU_LISTEN:
+               if (inp->inp_lport == 0)
+                       error = in_pcbbind(inp, (struct mbuf *)0);
+               if (error == 0)
+                       tp->t_state = TCPS_LISTEN;
+               break;
+
+       /*
+        * Initiate connection to peer.
+        * Create a template for use in transmissions on this connection.
+        * Enter SYN_SENT state, and mark socket as connecting.
+        * Start keep-alive timer, and seed output sequence space.
+        * Send initial segment on connection.
+        */
        case PRU_CONNECT:
        case PRU_CONNECT:
-               error = in_pcbconnect(inp, (struct sockaddr_in *)addr);
+               if (inp->inp_lport == 0) {
+                       error = in_pcbbind(inp, (struct mbuf *)0);
+                       if (error)
+                               break;
+               }
+               error = in_pcbconnect(inp, nam);
                if (error)
                        break;
                tp->t_template = tcp_template(tp);
                if (error)
                        break;
                tp->t_template = tcp_template(tp);
-               if (tp->t_template == 0)
-                       goto badcon2;
-               tp->t_inpcb = inp;
-               inp->inp_ppcb = (caddr_t)tp;
+               if (tp->t_template == 0) {
+                       in_pcbdisconnect(inp);
+                       error = ENOBUFS;
+                       break;
+               }
+               /* Compute window scaling to request.  */
+               while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
+                   (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
+                       tp->request_r_scale++;
                soisconnecting(so);
                soisconnecting(so);
+               tcpstat.tcps_connattempt++;
                tp->t_state = TCPS_SYN_SENT;
                tp->t_state = TCPS_SYN_SENT;
-               tp->t_timer[TCPT_KEEP] = TCPTV_KEEP;
+               tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
                tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2;
                tcp_sendseqinit(tp);
                tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2;
                tcp_sendseqinit(tp);
-               (void) tcp_output(tp);
+               error = tcp_output(tp);
                break;
 
                break;
 
-badcon2:
-               (void) m_free(dtom(tp));
-badcon:
-               in_pcbdisconnect(inp);
-               error = ENOBUFS;
+       /*
+        * Create a TCP connection between two sockets.
+        */
+       case PRU_CONNECT2:
+               error = EOPNOTSUPP;
                break;
 
                break;
 
-       case PRU_ACCEPT:
-               *(struct sockaddr *)addr = so->so_addr;
+       /*
+        * Initiate disconnect from peer.
+        * If connection never passed embryonic stage, just drop;
+        * else if don't need to let data drain, then can just drop anyways,
+        * else have to begin TCP shutdown process: mark socket disconnecting,
+        * drain unread data, state switch to reflect user close, and
+        * send segment (e.g. FIN) to peer.  Socket will be really disconnected
+        * when peer sends FIN and acks ours.
+        *
+        * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
+        */
+       case PRU_DISCONNECT:
+               tp = tcp_disconnect(tp);
                break;
 
                break;
 
-       case PRU_DISCONNECT:
-disconn:
-               if (tp->t_state < TCPS_ESTABLISHED)
-                       tcp_close(tp);
-               else if ((so->so_state & SO_LETDATADRAIN) == 0)
-                       tcp_drop(tp, 0);
-               else {
-                       soisdisconnecting(so);
-                       sbflush(&so->so_rcv);
-                       tcp_usrclosed(tp);
-                       (void) tcp_output(tp);
-               }
+       /*
+        * Accept a connection.  Essentially all the work is
+        * done at higher levels; just return the address
+        * of the peer, storing through addr.
+        */
+       case PRU_ACCEPT:
+               in_setpeeraddr(inp, nam);
                break;
 
                break;
 
+       /*
+        * Mark the connection as being incapable of further output.
+        */
        case PRU_SHUTDOWN:
                socantsendmore(so);
        case PRU_SHUTDOWN:
                socantsendmore(so);
-               tcp_usrclosed(tp);
-               (void) tcp_output(tp);
+               tp = tcp_usrclosed(tp);
+               if (tp)
+                       error = tcp_output(tp);
                break;
 
                break;
 
+       /*
+        * After a receive, possibly send window update to peer.
+        */
        case PRU_RCVD:
                (void) tcp_output(tp);
                break;
 
        case PRU_RCVD:
                (void) tcp_output(tp);
                break;
 
+       /*
+        * Do a send by putting data in output queue and updating urgent
+        * marker if URG set.  Possibly send more data.
+        */
        case PRU_SEND:
                sbappend(&so->so_snd, m);
        case PRU_SEND:
                sbappend(&so->so_snd, m);
-/*
-               if (tp->t_flags & TF_PUSH)
-                       tp->snd_end = tp->snd_una + so->so_snd.sb_cc;
- */
-               if (tp->t_flags & TF_URG)
-                       tp->snd_up = tp->snd_una + so->so_snd.sb_cc + 1;
-               (void) tcp_output(tp);
+               error = tcp_output(tp);
                break;
 
                break;
 
+       /*
+        * Abort the TCP.
+        */
        case PRU_ABORT:
        case PRU_ABORT:
-               tcp_drop(tp, ECONNABORTED);
-               break;
-
-       case PRU_CONTROL:
-               error = EOPNOTSUPP;
+               tp = tcp_drop(tp, ECONNABORTED);
                break;
 
        case PRU_SENSE:
                break;
 
        case PRU_SENSE:
-               error = EOPNOTSUPP;
-               break;
+               ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat;
+               (void) splx(s);
+               return (0);
 
        case PRU_RCVOOB:
 
        case PRU_RCVOOB:
-               error = EOPNOTSUPP;
+               if ((so->so_oobmark == 0 &&
+                   (so->so_state & SS_RCVATMARK) == 0) ||
+                   so->so_options & SO_OOBINLINE ||
+                   tp->t_oobflags & TCPOOB_HADDATA) {
+                       error = EINVAL;
+                       break;
+               }
+               if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
+                       error = EWOULDBLOCK;
+                       break;
+               }
+               m->m_len = 1;
+               *mtod(m, caddr_t) = tp->t_iobc;
+               if (((int)nam & MSG_PEEK) == 0)
+                       tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
                break;
 
        case PRU_SENDOOB:
                break;
 
        case PRU_SENDOOB:
-               error = EOPNOTSUPP;
+               if (sbspace(&so->so_snd) < -512) {
+                       m_freem(m);
+                       error = ENOBUFS;
+                       break;
+               }
+               /*
+                * According to RFC961 (Assigned Protocols),
+                * the urgent pointer points to the last octet
+                * of urgent data.  We continue, however,
+                * to consider it to indicate the first octet
+                * of data past the urgent section.
+                * Otherwise, snd_up should be one lower.
+                */
+               sbappend(&so->so_snd, m);
+               tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
+               tp->t_force = 1;
+               error = tcp_output(tp);
+               tp->t_force = 0;
+               break;
+
+       case PRU_SOCKADDR:
+               in_setsockaddr(inp, nam);
+               break;
+
+       case PRU_PEERADDR:
+               in_setpeeraddr(inp, nam);
                break;
 
                break;
 
+       /*
+        * TCP slow timer went off; going through this
+        * routine for tracing's sake.
+        */
        case PRU_SLOWTIMO:
        case PRU_SLOWTIMO:
-               tcp_timers(tp, (int)addr);
-               req |= (int)addr << 8;          /* for debug's sake */
+               tp = tcp_timers(tp, (int)nam);
+               req |= (int)nam << 8;           /* for debug's sake */
                break;
 
        default:
                break;
 
        default:
@@ -182,16 +333,173 @@ disconn:
        return (error);
 }
 
        return (error);
 }
 
+int
+tcp_ctloutput(op, so, level, optname, mp)
+       int op;
+       struct socket *so;
+       int level, optname;
+       struct mbuf **mp;
+{
+       int error = 0, s;
+       struct inpcb *inp;
+       register struct tcpcb *tp;
+       register struct mbuf *m;
+       register int i;
+
+       s = splnet();
+       inp = sotoinpcb(so);
+       if (inp == NULL) {
+               splx(s);
+               if (op == PRCO_SETOPT && *mp)
+                       (void) m_free(*mp);
+               return (ECONNRESET);
+       }
+       if (level != IPPROTO_TCP) {
+               error = ip_ctloutput(op, so, level, optname, mp);
+               splx(s);
+               return (error);
+       }
+       tp = intotcpcb(inp);
+
+       switch (op) {
+
+       case PRCO_SETOPT:
+               m = *mp;
+               switch (optname) {
+
+               case TCP_NODELAY:
+                       if (m == NULL || m->m_len < sizeof (int))
+                               error = EINVAL;
+                       else if (*mtod(m, int *))
+                               tp->t_flags |= TF_NODELAY;
+                       else
+                               tp->t_flags &= ~TF_NODELAY;
+                       break;
+
+               case TCP_MAXSEG:
+                       if (m && (i = *mtod(m, int *)) > 0 && i <= tp->t_maxseg)
+                               tp->t_maxseg = i;
+                       else
+                               error = EINVAL;
+                       break;
+
+               default:
+                       error = ENOPROTOOPT;
+                       break;
+               }
+               if (m)
+                       (void) m_free(m);
+               break;
+
+       case PRCO_GETOPT:
+               *mp = m = m_get(M_WAIT, MT_SOOPTS);
+               m->m_len = sizeof(int);
+
+               switch (optname) {
+               case TCP_NODELAY:
+                       *mtod(m, int *) = tp->t_flags & TF_NODELAY;
+                       break;
+               case TCP_MAXSEG:
+                       *mtod(m, int *) = tp->t_maxseg;
+                       break;
+               default:
+                       error = ENOPROTOOPT;
+                       break;
+               }
+               break;
+       }
+       splx(s);
+       return (error);
+}
+
+u_long tcp_sendspace = 1024*8;
+u_long tcp_recvspace = 1024*8;
+
+/*
+ * Attach TCP protocol to socket, allocating
+ * internet protocol control block, tcp control block,
+ * bufer space, and entering LISTEN state if to accept connections.
+ */
+int
+tcp_attach(so)
+       struct socket *so;
+{
+       register struct tcpcb *tp;
+       struct inpcb *inp;
+       int error;
+
+       if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
+               error = soreserve(so, tcp_sendspace, tcp_recvspace);
+               if (error)
+                       return (error);
+       }
+       error = in_pcballoc(so, &tcb);
+       if (error)
+               return (error);
+       inp = sotoinpcb(so);
+       tp = tcp_newtcpcb(inp);
+       if (tp == 0) {
+               int nofd = so->so_state & SS_NOFDREF;   /* XXX */
+
+               so->so_state &= ~SS_NOFDREF;    /* don't free the socket yet */
+               in_pcbdetach(inp);
+               so->so_state |= nofd;
+               return (ENOBUFS);
+       }
+       tp->t_state = TCPS_CLOSED;
+       return (0);
+}
+
+/*
+ * Initiate (or continue) disconnect.
+ * If embryonic state, just send reset (once).
+ * If in ``let data drain'' option and linger null, just drop.
+ * Otherwise (hard), mark socket disconnecting and drop
+ * current input data; switch states based on user close, and
+ * send segment to peer (with FIN).
+ */
+struct tcpcb *
+tcp_disconnect(tp)
+       register struct tcpcb *tp;
+{
+       struct socket *so = tp->t_inpcb->inp_socket;
+
+       if (tp->t_state < TCPS_ESTABLISHED)
+               tp = tcp_close(tp);
+       else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
+               tp = tcp_drop(tp, 0);
+       else {
+               soisdisconnecting(so);
+               sbflush(&so->so_rcv);
+               tp = tcp_usrclosed(tp);
+               if (tp)
+                       (void) tcp_output(tp);
+       }
+       return (tp);
+}
+
+/*
+ * User issued close, and wish to trail through shutdown states:
+ * if never received SYN, just forget it.  If got a SYN from peer,
+ * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
+ * If already got a FIN from peer, then almost done; go to LAST_ACK
+ * state.  In all other cases, have already sent FIN to peer (e.g.
+ * after PRU_SHUTDOWN), and just have to play tedious game waiting
+ * for peer to send FIN or not respond to keep-alives, etc.
+ * We can let the user exit from the close as soon as the FIN is acked.
+ */
+struct tcpcb *
 tcp_usrclosed(tp)
 tcp_usrclosed(tp)
-       struct tcpcb *tp;
+       register struct tcpcb *tp;
 {
 
        switch (tp->t_state) {
 
 {
 
        switch (tp->t_state) {
 
+       case TCPS_CLOSED:
        case TCPS_LISTEN:
        case TCPS_SYN_SENT:
                tp->t_state = TCPS_CLOSED;
        case TCPS_LISTEN:
        case TCPS_SYN_SENT:
                tp->t_state = TCPS_CLOSED;
-               tcp_close(tp);
+               tp = tcp_close(tp);
                break;
 
        case TCPS_SYN_RECEIVED:
                break;
 
        case TCPS_SYN_RECEIVED:
@@ -203,4 +511,7 @@ tcp_usrclosed(tp)
                tp->t_state = TCPS_LAST_ACK;
                break;
        }
                tp->t_state = TCPS_LAST_ACK;
                break;
        }
+       if (tp && tp->t_state >= TCPS_FIN_WAIT_2)
+               soisdisconnected(tp->t_inpcb->inp_socket);
+       return (tp);
 }
 }