Home | History | Annotate | Line # | Download | only in kern
uipc_socket2.c revision 1.4
      1  1.1     cgd /*
      2  1.1     cgd  * Copyright (c) 1982, 1986, 1988, 1990 Regents of the University of California.
      3  1.1     cgd  * All rights reserved.
      4  1.1     cgd  *
      5  1.1     cgd  * Redistribution and use in source and binary forms, with or without
      6  1.1     cgd  * modification, are permitted provided that the following conditions
      7  1.1     cgd  * are met:
      8  1.1     cgd  * 1. Redistributions of source code must retain the above copyright
      9  1.1     cgd  *    notice, this list of conditions and the following disclaimer.
     10  1.1     cgd  * 2. Redistributions in binary form must reproduce the above copyright
     11  1.1     cgd  *    notice, this list of conditions and the following disclaimer in the
     12  1.1     cgd  *    documentation and/or other materials provided with the distribution.
     13  1.1     cgd  * 3. All advertising materials mentioning features or use of this software
     14  1.1     cgd  *    must display the following acknowledgement:
     15  1.1     cgd  *	This product includes software developed by the University of
     16  1.1     cgd  *	California, Berkeley and its contributors.
     17  1.1     cgd  * 4. Neither the name of the University nor the names of its contributors
     18  1.1     cgd  *    may be used to endorse or promote products derived from this software
     19  1.1     cgd  *    without specific prior written permission.
     20  1.1     cgd  *
     21  1.1     cgd  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     22  1.1     cgd  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     23  1.1     cgd  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     24  1.1     cgd  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     25  1.1     cgd  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     26  1.1     cgd  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     27  1.1     cgd  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     28  1.1     cgd  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     29  1.1     cgd  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     30  1.1     cgd  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     31  1.1     cgd  * SUCH DAMAGE.
     32  1.1     cgd  *
     33  1.3     cgd  *	from: @(#)uipc_socket2.c	7.17 (Berkeley) 5/4/91
     34  1.4  andrew  *	$Id: uipc_socket2.c,v 1.4 1993/06/27 06:02:00 andrew Exp $
     35  1.1     cgd  */
     36  1.1     cgd 
     37  1.1     cgd #include "param.h"
     38  1.1     cgd #include "systm.h"
     39  1.1     cgd #include "proc.h"
     40  1.1     cgd #include "file.h"
     41  1.1     cgd #include "buf.h"
     42  1.1     cgd #include "malloc.h"
     43  1.3     cgd #include "select.h"
     44  1.1     cgd #include "mbuf.h"
     45  1.1     cgd #include "protosw.h"
     46  1.1     cgd #include "socket.h"
     47  1.1     cgd #include "socketvar.h"
     48  1.1     cgd 
     49  1.1     cgd /*
     50  1.1     cgd  * Primitive routines for operating on sockets and socket buffers
     51  1.1     cgd  */
     52  1.1     cgd 
     53  1.1     cgd /* strings for sleep message: */
     54  1.1     cgd char	netio[] = "netio";
     55  1.1     cgd char	netcon[] = "netcon";
     56  1.1     cgd char	netcls[] = "netcls";
     57  1.1     cgd 
     58  1.1     cgd u_long	sb_max = SB_MAX;		/* patchable */
     59  1.1     cgd 
     60  1.1     cgd /*
     61  1.1     cgd  * Procedures to manipulate state flags of socket
     62  1.1     cgd  * and do appropriate wakeups.  Normal sequence from the
     63  1.1     cgd  * active (originating) side is that soisconnecting() is
     64  1.1     cgd  * called during processing of connect() call,
     65  1.1     cgd  * resulting in an eventual call to soisconnected() if/when the
     66  1.1     cgd  * connection is established.  When the connection is torn down
     67  1.1     cgd  * soisdisconnecting() is called during processing of disconnect() call,
     68  1.1     cgd  * and soisdisconnected() is called when the connection to the peer
     69  1.1     cgd  * is totally severed.  The semantics of these routines are such that
     70  1.1     cgd  * connectionless protocols can call soisconnected() and soisdisconnected()
     71  1.1     cgd  * only, bypassing the in-progress calls when setting up a ``connection''
     72  1.1     cgd  * takes no time.
     73  1.1     cgd  *
     74  1.1     cgd  * From the passive side, a socket is created with
     75  1.1     cgd  * two queues of sockets: so_q0 for connections in progress
     76  1.1     cgd  * and so_q for connections already made and awaiting user acceptance.
     77  1.1     cgd  * As a protocol is preparing incoming connections, it creates a socket
     78  1.1     cgd  * structure queued on so_q0 by calling sonewconn().  When the connection
     79  1.1     cgd  * is established, soisconnected() is called, and transfers the
     80  1.1     cgd  * socket structure to so_q, making it available to accept().
     81  1.1     cgd  *
     82  1.1     cgd  * If a socket is closed with sockets on either
     83  1.1     cgd  * so_q0 or so_q, these sockets are dropped.
     84  1.1     cgd  *
     85  1.1     cgd  * If higher level protocols are implemented in
     86  1.1     cgd  * the kernel, the wakeups done here will sometimes
     87  1.1     cgd  * cause software-interrupt process scheduling.
     88  1.1     cgd  */
     89  1.1     cgd 
     90  1.1     cgd soisconnecting(so)
     91  1.1     cgd 	register struct socket *so;
     92  1.1     cgd {
     93  1.1     cgd 
     94  1.1     cgd 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
     95  1.1     cgd 	so->so_state |= SS_ISCONNECTING;
     96  1.1     cgd }
     97  1.1     cgd 
     98  1.1     cgd soisconnected(so)
     99  1.1     cgd 	register struct socket *so;
    100  1.1     cgd {
    101  1.1     cgd 	register struct socket *head = so->so_head;
    102  1.1     cgd 
    103  1.1     cgd 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
    104  1.1     cgd 	so->so_state |= SS_ISCONNECTED;
    105  1.1     cgd 	if (head && soqremque(so, 0)) {
    106  1.1     cgd 		soqinsque(head, so, 1);
    107  1.1     cgd 		sorwakeup(head);
    108  1.1     cgd 		wakeup((caddr_t)&head->so_timeo);
    109  1.1     cgd 	} else {
    110  1.1     cgd 		wakeup((caddr_t)&so->so_timeo);
    111  1.1     cgd 		sorwakeup(so);
    112  1.1     cgd 		sowwakeup(so);
    113  1.1     cgd 	}
    114  1.1     cgd }
    115  1.1     cgd 
    116  1.1     cgd soisdisconnecting(so)
    117  1.1     cgd 	register struct socket *so;
    118  1.1     cgd {
    119  1.1     cgd 
    120  1.1     cgd 	so->so_state &= ~SS_ISCONNECTING;
    121  1.1     cgd 	so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
    122  1.1     cgd 	wakeup((caddr_t)&so->so_timeo);
    123  1.1     cgd 	sowwakeup(so);
    124  1.1     cgd 	sorwakeup(so);
    125  1.1     cgd }
    126  1.1     cgd 
    127  1.1     cgd soisdisconnected(so)
    128  1.1     cgd 	register struct socket *so;
    129  1.1     cgd {
    130  1.1     cgd 
    131  1.1     cgd 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
    132  1.1     cgd 	so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE);
    133  1.1     cgd 	wakeup((caddr_t)&so->so_timeo);
    134  1.1     cgd 	sowwakeup(so);
    135  1.1     cgd 	sorwakeup(so);
    136  1.1     cgd }
    137  1.1     cgd 
    138  1.1     cgd /*
    139  1.1     cgd  * When an attempt at a new connection is noted on a socket
    140  1.1     cgd  * which accepts connections, sonewconn is called.  If the
    141  1.1     cgd  * connection is possible (subject to space constraints, etc.)
    142  1.1     cgd  * then we allocate a new structure, propoerly linked into the
    143  1.1     cgd  * data structure of the original socket, and return this.
    144  1.1     cgd  * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
    145  1.1     cgd  *
    146  1.1     cgd  * Currently, sonewconn() is defined as sonewconn1() in socketvar.h
    147  1.1     cgd  * to catch calls that are missing the (new) second parameter.
    148  1.1     cgd  */
    149  1.1     cgd struct socket *
    150  1.1     cgd sonewconn1(head, connstatus)
    151  1.1     cgd 	register struct socket *head;
    152  1.1     cgd 	int connstatus;
    153  1.1     cgd {
    154  1.1     cgd 	register struct socket *so;
    155  1.1     cgd 	int soqueue = connstatus ? 1 : 0;
    156  1.1     cgd 
    157  1.1     cgd 	if (head->so_qlen + head->so_q0len > 3 * head->so_qlimit / 2)
    158  1.1     cgd 		return ((struct socket *)0);
    159  1.1     cgd 	MALLOC(so, struct socket *, sizeof(*so), M_SOCKET, M_DONTWAIT);
    160  1.1     cgd 	if (so == NULL)
    161  1.1     cgd 		return ((struct socket *)0);
    162  1.1     cgd 	bzero((caddr_t)so, sizeof(*so));
    163  1.1     cgd 	so->so_type = head->so_type;
    164  1.1     cgd 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
    165  1.1     cgd 	so->so_linger = head->so_linger;
    166  1.1     cgd 	so->so_state = head->so_state | SS_NOFDREF;
    167  1.1     cgd 	so->so_proto = head->so_proto;
    168  1.1     cgd 	so->so_timeo = head->so_timeo;
    169  1.1     cgd 	so->so_pgid = head->so_pgid;
    170  1.1     cgd 	(void) soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat);
    171  1.1     cgd 	soqinsque(head, so, soqueue);
    172  1.1     cgd 	if ((*so->so_proto->pr_usrreq)(so, PRU_ATTACH,
    173  1.1     cgd 	    (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0)) {
    174  1.1     cgd 		(void) soqremque(so, soqueue);
    175  1.1     cgd 		(void) free((caddr_t)so, M_SOCKET);
    176  1.1     cgd 		return ((struct socket *)0);
    177  1.1     cgd 	}
    178  1.1     cgd 	if (connstatus) {
    179  1.1     cgd 		sorwakeup(head);
    180  1.1     cgd 		wakeup((caddr_t)&head->so_timeo);
    181  1.1     cgd 		so->so_state |= connstatus;
    182  1.1     cgd 	}
    183  1.1     cgd 	return (so);
    184  1.1     cgd }
    185  1.1     cgd 
    186  1.1     cgd soqinsque(head, so, q)
    187  1.1     cgd 	register struct socket *head, *so;
    188  1.1     cgd 	int q;
    189  1.1     cgd {
    190  1.1     cgd 
    191  1.1     cgd 	register struct socket **prev;
    192  1.1     cgd 	so->so_head = head;
    193  1.1     cgd 	if (q == 0) {
    194  1.1     cgd 		head->so_q0len++;
    195  1.1     cgd 		so->so_q0 = 0;
    196  1.1     cgd 		for (prev = &(head->so_q0); *prev; )
    197  1.1     cgd 			prev = &((*prev)->so_q0);
    198  1.1     cgd 	} else {
    199  1.1     cgd 		head->so_qlen++;
    200  1.1     cgd 		so->so_q = 0;
    201  1.1     cgd 		for (prev = &(head->so_q); *prev; )
    202  1.1     cgd 			prev = &((*prev)->so_q);
    203  1.1     cgd 	}
    204  1.1     cgd 	*prev = so;
    205  1.1     cgd }
    206  1.1     cgd 
    207  1.1     cgd soqremque(so, q)
    208  1.1     cgd 	register struct socket *so;
    209  1.1     cgd 	int q;
    210  1.1     cgd {
    211  1.1     cgd 	register struct socket *head, *prev, *next;
    212  1.1     cgd 
    213  1.1     cgd 	head = so->so_head;
    214  1.1     cgd 	prev = head;
    215  1.1     cgd 	for (;;) {
    216  1.1     cgd 		next = q ? prev->so_q : prev->so_q0;
    217  1.1     cgd 		if (next == so)
    218  1.1     cgd 			break;
    219  1.1     cgd 		if (next == 0)
    220  1.1     cgd 			return (0);
    221  1.1     cgd 		prev = next;
    222  1.1     cgd 	}
    223  1.1     cgd 	if (q == 0) {
    224  1.1     cgd 		prev->so_q0 = next->so_q0;
    225  1.1     cgd 		head->so_q0len--;
    226  1.1     cgd 	} else {
    227  1.1     cgd 		prev->so_q = next->so_q;
    228  1.1     cgd 		head->so_qlen--;
    229  1.1     cgd 	}
    230  1.1     cgd 	next->so_q0 = next->so_q = 0;
    231  1.1     cgd 	next->so_head = 0;
    232  1.1     cgd 	return (1);
    233  1.1     cgd }
    234  1.1     cgd 
    235  1.1     cgd /*
    236  1.1     cgd  * Socantsendmore indicates that no more data will be sent on the
    237  1.1     cgd  * socket; it would normally be applied to a socket when the user
    238  1.1     cgd  * informs the system that no more data is to be sent, by the protocol
    239  1.1     cgd  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
    240  1.1     cgd  * will be received, and will normally be applied to the socket by a
    241  1.1     cgd  * protocol when it detects that the peer will send no more data.
    242  1.1     cgd  * Data queued for reading in the socket may yet be read.
    243  1.1     cgd  */
    244  1.1     cgd 
    245  1.4  andrew void
    246  1.1     cgd socantsendmore(so)
    247  1.1     cgd 	struct socket *so;
    248  1.1     cgd {
    249  1.1     cgd 
    250  1.1     cgd 	so->so_state |= SS_CANTSENDMORE;
    251  1.1     cgd 	sowwakeup(so);
    252  1.1     cgd }
    253  1.1     cgd 
    254  1.4  andrew void
    255  1.1     cgd socantrcvmore(so)
    256  1.1     cgd 	struct socket *so;
    257  1.1     cgd {
    258  1.1     cgd 
    259  1.1     cgd 	so->so_state |= SS_CANTRCVMORE;
    260  1.1     cgd 	sorwakeup(so);
    261  1.1     cgd }
    262  1.1     cgd 
    263  1.1     cgd /*
    264  1.1     cgd  * Socket select/wakeup routines.
    265  1.1     cgd  */
    266  1.1     cgd 
    267  1.1     cgd /*
    268  1.1     cgd  * Queue a process for a select on a socket buffer.
    269  1.1     cgd  */
    270  1.1     cgd sbselqueue(sb, cp)
    271  1.1     cgd 	struct sockbuf *sb;
    272  1.1     cgd 	struct proc *cp;
    273  1.1     cgd {
    274  1.3     cgd 	selrecord(cp, &sb->sb_sel);
    275  1.3     cgd 	sb->sb_flags |= SB_SEL;
    276  1.1     cgd }
    277  1.1     cgd 
    278  1.1     cgd /*
    279  1.1     cgd  * Wait for data to arrive at/drain from a socket buffer.
    280  1.1     cgd  */
    281  1.1     cgd sbwait(sb)
    282  1.1     cgd 	struct sockbuf *sb;
    283  1.1     cgd {
    284  1.1     cgd 
    285  1.1     cgd 	sb->sb_flags |= SB_WAIT;
    286  1.1     cgd 	return (tsleep((caddr_t)&sb->sb_cc,
    287  1.1     cgd 	    (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, netio,
    288  1.1     cgd 	    sb->sb_timeo));
    289  1.1     cgd }
    290  1.1     cgd 
    291  1.1     cgd /*
    292  1.1     cgd  * Lock a sockbuf already known to be locked;
    293  1.1     cgd  * return any error returned from sleep (EINTR).
    294  1.1     cgd  */
    295  1.1     cgd sb_lock(sb)
    296  1.1     cgd 	register struct sockbuf *sb;
    297  1.1     cgd {
    298  1.1     cgd 	int error;
    299  1.1     cgd 
    300  1.1     cgd 	while (sb->sb_flags & SB_LOCK) {
    301  1.1     cgd 		sb->sb_flags |= SB_WANT;
    302  1.1     cgd 		if (error = tsleep((caddr_t)&sb->sb_flags,
    303  1.1     cgd 		    (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
    304  1.1     cgd 		    netio, 0))
    305  1.1     cgd 			return (error);
    306  1.1     cgd 	}
    307  1.1     cgd 	sb->sb_flags |= SB_LOCK;
    308  1.1     cgd 	return (0);
    309  1.1     cgd }
    310  1.1     cgd 
    311  1.1     cgd /*
    312  1.1     cgd  * Wakeup processes waiting on a socket buffer.
    313  1.1     cgd  * Do asynchronous notification via SIGIO
    314  1.1     cgd  * if the socket has the SS_ASYNC flag set.
    315  1.1     cgd  */
    316  1.1     cgd sowakeup(so, sb)
    317  1.1     cgd 	register struct socket *so;
    318  1.1     cgd 	register struct sockbuf *sb;
    319  1.1     cgd {
    320  1.1     cgd 	struct proc *p;
    321  1.1     cgd 
    322  1.3     cgd 	selwakeup(&sb->sb_sel);
    323  1.3     cgd         sb->sb_flags &= ~SB_SEL;
    324  1.1     cgd 	if (sb->sb_flags & SB_WAIT) {
    325  1.1     cgd 		sb->sb_flags &= ~SB_WAIT;
    326  1.1     cgd 		wakeup((caddr_t)&sb->sb_cc);
    327  1.1     cgd 	}
    328  1.1     cgd 	if (so->so_state & SS_ASYNC) {
    329  1.1     cgd 		if (so->so_pgid < 0)
    330  1.1     cgd 			gsignal(-so->so_pgid, SIGIO);
    331  1.1     cgd 		else if (so->so_pgid > 0 && (p = pfind(so->so_pgid)) != 0)
    332  1.1     cgd 			psignal(p, SIGIO);
    333  1.1     cgd 	}
    334  1.1     cgd }
    335  1.1     cgd 
    336  1.1     cgd /*
    337  1.1     cgd  * Socket buffer (struct sockbuf) utility routines.
    338  1.1     cgd  *
    339  1.1     cgd  * Each socket contains two socket buffers: one for sending data and
    340  1.1     cgd  * one for receiving data.  Each buffer contains a queue of mbufs,
    341  1.1     cgd  * information about the number of mbufs and amount of data in the
    342  1.1     cgd  * queue, and other fields allowing select() statements and notification
    343  1.1     cgd  * on data availability to be implemented.
    344  1.1     cgd  *
    345  1.1     cgd  * Data stored in a socket buffer is maintained as a list of records.
    346  1.1     cgd  * Each record is a list of mbufs chained together with the m_next
    347  1.1     cgd  * field.  Records are chained together with the m_nextpkt field. The upper
    348  1.1     cgd  * level routine soreceive() expects the following conventions to be
    349  1.1     cgd  * observed when placing information in the receive buffer:
    350  1.1     cgd  *
    351  1.1     cgd  * 1. If the protocol requires each message be preceded by the sender's
    352  1.1     cgd  *    name, then a record containing that name must be present before
    353  1.1     cgd  *    any associated data (mbuf's must be of type MT_SONAME).
    354  1.1     cgd  * 2. If the protocol supports the exchange of ``access rights'' (really
    355  1.1     cgd  *    just additional data associated with the message), and there are
    356  1.1     cgd  *    ``rights'' to be received, then a record containing this data
    357  1.1     cgd  *    should be present (mbuf's must be of type MT_RIGHTS).
    358  1.1     cgd  * 3. If a name or rights record exists, then it must be followed by
    359  1.1     cgd  *    a data record, perhaps of zero length.
    360  1.1     cgd  *
    361  1.1     cgd  * Before using a new socket structure it is first necessary to reserve
    362  1.1     cgd  * buffer space to the socket, by calling sbreserve().  This should commit
    363  1.1     cgd  * some of the available buffer space in the system buffer pool for the
    364  1.1     cgd  * socket (currently, it does nothing but enforce limits).  The space
    365  1.1     cgd  * should be released by calling sbrelease() when the socket is destroyed.
    366  1.1     cgd  */
    367  1.1     cgd 
    368  1.1     cgd soreserve(so, sndcc, rcvcc)
    369  1.1     cgd 	register struct socket *so;
    370  1.1     cgd 	u_long sndcc, rcvcc;
    371  1.1     cgd {
    372  1.1     cgd 
    373  1.1     cgd 	if (sbreserve(&so->so_snd, sndcc) == 0)
    374  1.1     cgd 		goto bad;
    375  1.1     cgd 	if (sbreserve(&so->so_rcv, rcvcc) == 0)
    376  1.1     cgd 		goto bad2;
    377  1.1     cgd 	if (so->so_rcv.sb_lowat == 0)
    378  1.1     cgd 		so->so_rcv.sb_lowat = 1;
    379  1.1     cgd 	if (so->so_snd.sb_lowat == 0)
    380  1.1     cgd 		so->so_snd.sb_lowat = MCLBYTES;
    381  1.1     cgd 	if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
    382  1.1     cgd 		so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
    383  1.1     cgd 	return (0);
    384  1.1     cgd bad2:
    385  1.1     cgd 	sbrelease(&so->so_snd);
    386  1.1     cgd bad:
    387  1.1     cgd 	return (ENOBUFS);
    388  1.1     cgd }
    389  1.1     cgd 
    390  1.1     cgd /*
    391  1.1     cgd  * Allot mbufs to a sockbuf.
    392  1.1     cgd  * Attempt to scale mbmax so that mbcnt doesn't become limiting
    393  1.1     cgd  * if buffering efficiency is near the normal case.
    394  1.1     cgd  */
    395  1.1     cgd sbreserve(sb, cc)
    396  1.1     cgd 	struct sockbuf *sb;
    397  1.1     cgd 	u_long cc;
    398  1.1     cgd {
    399  1.1     cgd 
    400  1.1     cgd 	if (cc > sb_max * MCLBYTES / (MSIZE + MCLBYTES))
    401  1.1     cgd 		return (0);
    402  1.1     cgd 	sb->sb_hiwat = cc;
    403  1.1     cgd 	sb->sb_mbmax = min(cc * 2, sb_max);
    404  1.1     cgd 	if (sb->sb_lowat > sb->sb_hiwat)
    405  1.1     cgd 		sb->sb_lowat = sb->sb_hiwat;
    406  1.1     cgd 	return (1);
    407  1.1     cgd }
    408  1.1     cgd 
    409  1.1     cgd /*
    410  1.1     cgd  * Free mbufs held by a socket, and reserved mbuf space.
    411  1.1     cgd  */
    412  1.1     cgd sbrelease(sb)
    413  1.1     cgd 	struct sockbuf *sb;
    414  1.1     cgd {
    415  1.1     cgd 
    416  1.1     cgd 	sbflush(sb);
    417  1.1     cgd 	sb->sb_hiwat = sb->sb_mbmax = 0;
    418  1.1     cgd }
    419  1.1     cgd 
    420  1.1     cgd /*
    421  1.1     cgd  * Routines to add and remove
    422  1.1     cgd  * data from an mbuf queue.
    423  1.1     cgd  *
    424  1.1     cgd  * The routines sbappend() or sbappendrecord() are normally called to
    425  1.1     cgd  * append new mbufs to a socket buffer, after checking that adequate
    426  1.1     cgd  * space is available, comparing the function sbspace() with the amount
    427  1.1     cgd  * of data to be added.  sbappendrecord() differs from sbappend() in
    428  1.1     cgd  * that data supplied is treated as the beginning of a new record.
    429  1.1     cgd  * To place a sender's address, optional access rights, and data in a
    430  1.1     cgd  * socket receive buffer, sbappendaddr() should be used.  To place
    431  1.1     cgd  * access rights and data in a socket receive buffer, sbappendrights()
    432  1.1     cgd  * should be used.  In either case, the new data begins a new record.
    433  1.1     cgd  * Note that unlike sbappend() and sbappendrecord(), these routines check
    434  1.1     cgd  * for the caller that there will be enough space to store the data.
    435  1.1     cgd  * Each fails if there is not enough space, or if it cannot find mbufs
    436  1.1     cgd  * to store additional information in.
    437  1.1     cgd  *
    438  1.1     cgd  * Reliable protocols may use the socket send buffer to hold data
    439  1.1     cgd  * awaiting acknowledgement.  Data is normally copied from a socket
    440  1.1     cgd  * send buffer in a protocol with m_copy for output to a peer,
    441  1.1     cgd  * and then removing the data from the socket buffer with sbdrop()
    442  1.1     cgd  * or sbdroprecord() when the data is acknowledged by the peer.
    443  1.1     cgd  */
    444  1.1     cgd 
    445  1.1     cgd /*
    446  1.1     cgd  * Append mbuf chain m to the last record in the
    447  1.1     cgd  * socket buffer sb.  The additional space associated
    448  1.1     cgd  * the mbuf chain is recorded in sb.  Empty mbufs are
    449  1.1     cgd  * discarded and mbufs are compacted where possible.
    450  1.1     cgd  */
    451  1.1     cgd sbappend(sb, m)
    452  1.1     cgd 	struct sockbuf *sb;
    453  1.1     cgd 	struct mbuf *m;
    454  1.1     cgd {
    455  1.1     cgd 	register struct mbuf *n;
    456  1.1     cgd 
    457  1.1     cgd 	if (m == 0)
    458  1.1     cgd 		return;
    459  1.1     cgd 	if (n = sb->sb_mb) {
    460  1.1     cgd 		while (n->m_nextpkt)
    461  1.1     cgd 			n = n->m_nextpkt;
    462  1.1     cgd 		do {
    463  1.1     cgd 			if (n->m_flags & M_EOR) {
    464  1.1     cgd 				sbappendrecord(sb, m); /* XXXXXX!!!! */
    465  1.1     cgd 				return;
    466  1.1     cgd 			}
    467  1.1     cgd 		} while (n->m_next && (n = n->m_next));
    468  1.1     cgd 	}
    469  1.1     cgd 	sbcompress(sb, m, n);
    470  1.1     cgd }
    471  1.1     cgd 
    472  1.1     cgd #ifdef SOCKBUF_DEBUG
    473  1.1     cgd sbcheck(sb)
    474  1.1     cgd 	register struct sockbuf *sb;
    475  1.1     cgd {
    476  1.1     cgd 	register struct mbuf *m;
    477  1.1     cgd 	register int len = 0, mbcnt = 0;
    478  1.1     cgd 
    479  1.1     cgd 	for (m = sb->sb_mb; m; m = m->m_next) {
    480  1.1     cgd 		len += m->m_len;
    481  1.1     cgd 		mbcnt += MSIZE;
    482  1.1     cgd 		if (m->m_flags & M_EXT)
    483  1.1     cgd 			mbcnt += m->m_ext.ext_size;
    484  1.1     cgd 		if (m->m_nextpkt)
    485  1.1     cgd 			panic("sbcheck nextpkt");
    486  1.1     cgd 	}
    487  1.1     cgd 	if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
    488  1.1     cgd 		printf("cc %d != %d || mbcnt %d != %d\n", len, sb->sb_cc,
    489  1.1     cgd 		    mbcnt, sb->sb_mbcnt);
    490  1.1     cgd 		panic("sbcheck");
    491  1.1     cgd 	}
    492  1.1     cgd }
    493  1.1     cgd #endif
    494  1.1     cgd 
    495  1.1     cgd /*
    496  1.1     cgd  * As above, except the mbuf chain
    497  1.1     cgd  * begins a new record.
    498  1.1     cgd  */
    499  1.1     cgd sbappendrecord(sb, m0)
    500  1.1     cgd 	register struct sockbuf *sb;
    501  1.1     cgd 	register struct mbuf *m0;
    502  1.1     cgd {
    503  1.1     cgd 	register struct mbuf *m;
    504  1.1     cgd 
    505  1.1     cgd 	if (m0 == 0)
    506  1.1     cgd 		return;
    507  1.1     cgd 	if (m = sb->sb_mb)
    508  1.1     cgd 		while (m->m_nextpkt)
    509  1.1     cgd 			m = m->m_nextpkt;
    510  1.1     cgd 	/*
    511  1.1     cgd 	 * Put the first mbuf on the queue.
    512  1.1     cgd 	 * Note this permits zero length records.
    513  1.1     cgd 	 */
    514  1.1     cgd 	sballoc(sb, m0);
    515  1.1     cgd 	if (m)
    516  1.1     cgd 		m->m_nextpkt = m0;
    517  1.1     cgd 	else
    518  1.1     cgd 		sb->sb_mb = m0;
    519  1.1     cgd 	m = m0->m_next;
    520  1.1     cgd 	m0->m_next = 0;
    521  1.1     cgd 	if (m && (m0->m_flags & M_EOR)) {
    522  1.1     cgd 		m0->m_flags &= ~M_EOR;
    523  1.1     cgd 		m->m_flags |= M_EOR;
    524  1.1     cgd 	}
    525  1.1     cgd 	sbcompress(sb, m, m0);
    526  1.1     cgd }
    527  1.1     cgd 
    528  1.1     cgd /*
    529  1.1     cgd  * As above except that OOB data
    530  1.1     cgd  * is inserted at the beginning of the sockbuf,
    531  1.1     cgd  * but after any other OOB data.
    532  1.1     cgd  */
    533  1.1     cgd sbinsertoob(sb, m0)
    534  1.1     cgd 	register struct sockbuf *sb;
    535  1.1     cgd 	register struct mbuf *m0;
    536  1.1     cgd {
    537  1.1     cgd 	register struct mbuf *m;
    538  1.1     cgd 	register struct mbuf **mp;
    539  1.1     cgd 
    540  1.1     cgd 	if (m0 == 0)
    541  1.1     cgd 		return;
    542  1.1     cgd 	for (mp = &sb->sb_mb; m = *mp; mp = &((*mp)->m_nextpkt)) {
    543  1.1     cgd 	    again:
    544  1.1     cgd 		switch (m->m_type) {
    545  1.1     cgd 
    546  1.1     cgd 		case MT_OOBDATA:
    547  1.1     cgd 			continue;		/* WANT next train */
    548  1.1     cgd 
    549  1.1     cgd 		case MT_CONTROL:
    550  1.1     cgd 			if (m = m->m_next)
    551  1.1     cgd 				goto again;	/* inspect THIS train further */
    552  1.1     cgd 		}
    553  1.1     cgd 		break;
    554  1.1     cgd 	}
    555  1.1     cgd 	/*
    556  1.1     cgd 	 * Put the first mbuf on the queue.
    557  1.1     cgd 	 * Note this permits zero length records.
    558  1.1     cgd 	 */
    559  1.1     cgd 	sballoc(sb, m0);
    560  1.1     cgd 	m0->m_nextpkt = *mp;
    561  1.1     cgd 	*mp = m0;
    562  1.1     cgd 	m = m0->m_next;
    563  1.1     cgd 	m0->m_next = 0;
    564  1.1     cgd 	if (m && (m0->m_flags & M_EOR)) {
    565  1.1     cgd 		m0->m_flags &= ~M_EOR;
    566  1.1     cgd 		m->m_flags |= M_EOR;
    567  1.1     cgd 	}
    568  1.1     cgd 	sbcompress(sb, m, m0);
    569  1.1     cgd }
    570  1.1     cgd 
    571  1.1     cgd /*
    572  1.1     cgd  * Append address and data, and optionally, control (ancillary) data
    573  1.1     cgd  * to the receive queue of a socket.  If present,
    574  1.1     cgd  * m0 must include a packet header with total length.
    575  1.1     cgd  * Returns 0 if no space in sockbuf or insufficient mbufs.
    576  1.1     cgd  */
    577  1.1     cgd sbappendaddr(sb, asa, m0, control)
    578  1.1     cgd 	register struct sockbuf *sb;
    579  1.1     cgd 	struct sockaddr *asa;
    580  1.1     cgd 	struct mbuf *m0, *control;
    581  1.1     cgd {
    582  1.1     cgd 	register struct mbuf *m, *n;
    583  1.1     cgd 	int space = asa->sa_len;
    584  1.1     cgd 
    585  1.1     cgd if (m0 && (m0->m_flags & M_PKTHDR) == 0)
    586  1.1     cgd panic("sbappendaddr");
    587  1.1     cgd 	if (m0)
    588  1.1     cgd 		space += m0->m_pkthdr.len;
    589  1.1     cgd 	for (n = control; n; n = n->m_next) {
    590  1.1     cgd 		space += n->m_len;
    591  1.1     cgd 		if (n->m_next == 0)	/* keep pointer to last control buf */
    592  1.1     cgd 			break;
    593  1.1     cgd 	}
    594  1.1     cgd 	if (space > sbspace(sb))
    595  1.1     cgd 		return (0);
    596  1.1     cgd 	if (asa->sa_len > MLEN)
    597  1.1     cgd 		return (0);
    598  1.1     cgd 	MGET(m, M_DONTWAIT, MT_SONAME);
    599  1.1     cgd 	if (m == 0)
    600  1.1     cgd 		return (0);
    601  1.1     cgd 	m->m_len = asa->sa_len;
    602  1.1     cgd 	bcopy((caddr_t)asa, mtod(m, caddr_t), asa->sa_len);
    603  1.1     cgd 	if (n)
    604  1.1     cgd 		n->m_next = m0;		/* concatenate data to control */
    605  1.1     cgd 	else
    606  1.1     cgd 		control = m0;
    607  1.1     cgd 	m->m_next = control;
    608  1.1     cgd 	for (n = m; n; n = n->m_next)
    609  1.1     cgd 		sballoc(sb, n);
    610  1.1     cgd 	if (n = sb->sb_mb) {
    611  1.1     cgd 		while (n->m_nextpkt)
    612  1.1     cgd 			n = n->m_nextpkt;
    613  1.1     cgd 		n->m_nextpkt = m;
    614  1.1     cgd 	} else
    615  1.1     cgd 		sb->sb_mb = m;
    616  1.1     cgd 	return (1);
    617  1.1     cgd }
    618  1.1     cgd 
    619  1.1     cgd sbappendcontrol(sb, m0, control)
    620  1.1     cgd 	struct sockbuf *sb;
    621  1.1     cgd 	struct mbuf *control, *m0;
    622  1.1     cgd {
    623  1.1     cgd 	register struct mbuf *m, *n;
    624  1.1     cgd 	int space = 0;
    625  1.1     cgd 
    626  1.1     cgd 	if (control == 0)
    627  1.1     cgd 		panic("sbappendcontrol");
    628  1.1     cgd 	for (m = control; ; m = m->m_next) {
    629  1.1     cgd 		space += m->m_len;
    630  1.1     cgd 		if (m->m_next == 0)
    631  1.1     cgd 			break;
    632  1.1     cgd 	}
    633  1.1     cgd 	n = m;			/* save pointer to last control buffer */
    634  1.1     cgd 	for (m = m0; m; m = m->m_next)
    635  1.1     cgd 		space += m->m_len;
    636  1.1     cgd 	if (space > sbspace(sb))
    637  1.1     cgd 		return (0);
    638  1.1     cgd 	n->m_next = m0;			/* concatenate data to control */
    639  1.1     cgd 	for (m = control; m; m = m->m_next)
    640  1.1     cgd 		sballoc(sb, m);
    641  1.1     cgd 	if (n = sb->sb_mb) {
    642  1.1     cgd 		while (n->m_nextpkt)
    643  1.1     cgd 			n = n->m_nextpkt;
    644  1.1     cgd 		n->m_nextpkt = control;
    645  1.1     cgd 	} else
    646  1.1     cgd 		sb->sb_mb = control;
    647  1.1     cgd 	return (1);
    648  1.1     cgd }
    649  1.1     cgd 
    650  1.1     cgd /*
    651  1.1     cgd  * Compress mbuf chain m into the socket
    652  1.1     cgd  * buffer sb following mbuf n.  If n
    653  1.1     cgd  * is null, the buffer is presumed empty.
    654  1.1     cgd  */
    655  1.1     cgd sbcompress(sb, m, n)
    656  1.1     cgd 	register struct sockbuf *sb;
    657  1.1     cgd 	register struct mbuf *m, *n;
    658  1.1     cgd {
    659  1.1     cgd 	register int eor = 0;
    660  1.1     cgd 	register struct mbuf *o;
    661  1.1     cgd 
    662  1.1     cgd 	while (m) {
    663  1.1     cgd 		eor |= m->m_flags & M_EOR;
    664  1.1     cgd 		if (m->m_len == 0 &&
    665  1.1     cgd 		    (eor == 0 ||
    666  1.1     cgd 		     (((o = m->m_next) || (o = n)) &&
    667  1.1     cgd 		      o->m_type == m->m_type))) {
    668  1.1     cgd 			m = m_free(m);
    669  1.1     cgd 			continue;
    670  1.1     cgd 		}
    671  1.1     cgd 		if (n && (n->m_flags & (M_EXT | M_EOR)) == 0 &&
    672  1.1     cgd 		    (n->m_data + n->m_len + m->m_len) < &n->m_dat[MLEN] &&
    673  1.1     cgd 		    n->m_type == m->m_type) {
    674  1.1     cgd 			bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
    675  1.1     cgd 			    (unsigned)m->m_len);
    676  1.1     cgd 			n->m_len += m->m_len;
    677  1.1     cgd 			sb->sb_cc += m->m_len;
    678  1.1     cgd 			m = m_free(m);
    679  1.1     cgd 			continue;
    680  1.1     cgd 		}
    681  1.1     cgd 		if (n)
    682  1.1     cgd 			n->m_next = m;
    683  1.1     cgd 		else
    684  1.1     cgd 			sb->sb_mb = m;
    685  1.1     cgd 		sballoc(sb, m);
    686  1.1     cgd 		n = m;
    687  1.1     cgd 		m->m_flags &= ~M_EOR;
    688  1.1     cgd 		m = m->m_next;
    689  1.1     cgd 		n->m_next = 0;
    690  1.1     cgd 	}
    691  1.1     cgd 	if (eor) {
    692  1.1     cgd 		if (n)
    693  1.1     cgd 			n->m_flags |= eor;
    694  1.1     cgd 		else
    695  1.1     cgd 			printf("semi-panic: sbcompress\n");
    696  1.1     cgd 	}
    697  1.1     cgd }
    698  1.1     cgd 
    699  1.1     cgd /*
    700  1.1     cgd  * Free all mbufs in a sockbuf.
    701  1.1     cgd  * Check that all resources are reclaimed.
    702  1.1     cgd  */
    703  1.1     cgd sbflush(sb)
    704  1.1     cgd 	register struct sockbuf *sb;
    705  1.1     cgd {
    706  1.1     cgd 
    707  1.1     cgd 	if (sb->sb_flags & SB_LOCK)
    708  1.1     cgd 		panic("sbflush");
    709  1.1     cgd 	while (sb->sb_mbcnt)
    710  1.1     cgd 		sbdrop(sb, (int)sb->sb_cc);
    711  1.1     cgd 	if (sb->sb_cc || sb->sb_mb)
    712  1.1     cgd 		panic("sbflush 2");
    713  1.1     cgd }
    714  1.1     cgd 
    715  1.1     cgd /*
    716  1.1     cgd  * Drop data from (the front of) a sockbuf.
    717  1.1     cgd  */
    718  1.1     cgd sbdrop(sb, len)
    719  1.1     cgd 	register struct sockbuf *sb;
    720  1.1     cgd 	register int len;
    721  1.1     cgd {
    722  1.1     cgd 	register struct mbuf *m, *mn;
    723  1.1     cgd 	struct mbuf *next;
    724  1.1     cgd 
    725  1.1     cgd 	next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
    726  1.1     cgd 	while (len > 0) {
    727  1.1     cgd 		if (m == 0) {
    728  1.1     cgd 			if (next == 0)
    729  1.1     cgd 				panic("sbdrop");
    730  1.1     cgd 			m = next;
    731  1.1     cgd 			next = m->m_nextpkt;
    732  1.1     cgd 			continue;
    733  1.1     cgd 		}
    734  1.1     cgd 		if (m->m_len > len) {
    735  1.1     cgd 			m->m_len -= len;
    736  1.1     cgd 			m->m_data += len;
    737  1.1     cgd 			sb->sb_cc -= len;
    738  1.1     cgd 			break;
    739  1.1     cgd 		}
    740  1.1     cgd 		len -= m->m_len;
    741  1.1     cgd 		sbfree(sb, m);
    742  1.1     cgd 		MFREE(m, mn);
    743  1.1     cgd 		m = mn;
    744  1.1     cgd 	}
    745  1.1     cgd 	while (m && m->m_len == 0) {
    746  1.1     cgd 		sbfree(sb, m);
    747  1.1     cgd 		MFREE(m, mn);
    748  1.1     cgd 		m = mn;
    749  1.1     cgd 	}
    750  1.1     cgd 	if (m) {
    751  1.1     cgd 		sb->sb_mb = m;
    752  1.1     cgd 		m->m_nextpkt = next;
    753  1.1     cgd 	} else
    754  1.1     cgd 		sb->sb_mb = next;
    755  1.1     cgd }
    756  1.1     cgd 
    757  1.1     cgd /*
    758  1.1     cgd  * Drop a record off the front of a sockbuf
    759  1.1     cgd  * and move the next record to the front.
    760  1.1     cgd  */
    761  1.1     cgd sbdroprecord(sb)
    762  1.1     cgd 	register struct sockbuf *sb;
    763  1.1     cgd {
    764  1.1     cgd 	register struct mbuf *m, *mn;
    765  1.1     cgd 
    766  1.1     cgd 	m = sb->sb_mb;
    767  1.1     cgd 	if (m) {
    768  1.1     cgd 		sb->sb_mb = m->m_nextpkt;
    769  1.1     cgd 		do {
    770  1.1     cgd 			sbfree(sb, m);
    771  1.1     cgd 			MFREE(m, mn);
    772  1.1     cgd 		} while (m = mn);
    773  1.1     cgd 	}
    774  1.1     cgd }
    775