123431Smckusick /* 240706Skarels * Copyright (c) 1982, 1986, 1988, 1990 Regents of the University of California. 333187Sbostic * All rights reserved. 423431Smckusick * 533187Sbostic * Redistribution and use in source and binary forms are permitted 634858Sbostic * provided that the above copyright notice and this paragraph are 734858Sbostic * duplicated in all such forms and that any documentation, 834858Sbostic * advertising materials, and other materials related to such 934858Sbostic * distribution and use acknowledge that the software was developed 1034858Sbostic * by the University of California, Berkeley. The name of the 1134858Sbostic * University may not be used to endorse or promote products derived 1234858Sbostic * from this software without specific prior written permission. 1334858Sbostic * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR 1434858Sbostic * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED 1534858Sbostic * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. 1633187Sbostic * 17*44384Skarels * @(#)uipc_socket2.c 7.14 (Berkeley) 06/28/90 1823431Smckusick */ 194903Swnj 2017103Sbloom #include "param.h" 2117103Sbloom #include "systm.h" 2217103Sbloom #include "user.h" 2317103Sbloom #include "proc.h" 2417103Sbloom #include "file.h" 2517103Sbloom #include "buf.h" 2635385Skarels #include "malloc.h" 2717103Sbloom #include "mbuf.h" 2817103Sbloom #include "protosw.h" 2917103Sbloom #include "socket.h" 3017103Sbloom #include "socketvar.h" 314903Swnj 324903Swnj /* 334903Swnj * Primitive routines for operating on sockets and socket buffers 344903Swnj */ 354903Swnj 3640706Skarels /* strings for sleep message: */ 3740706Skarels char netio[] = "netio"; 3840706Skarels char netcon[] = "netcon"; 3940706Skarels char netcls[] = "netcls"; 4040706Skarels 4140706Skarels u_long sb_max = SB_MAX; /* patchable */ 4240706Skarels 434903Swnj /* 444903Swnj * Procedures to manipulate state flags of socket 457509Sroot * and do appropriate wakeups. Normal sequence from the 467509Sroot * active (originating) side is that soisconnecting() is 477509Sroot * called during processing of connect() call, 485169Swnj * resulting in an eventual call to soisconnected() if/when the 495169Swnj * connection is established. When the connection is torn down 505169Swnj * soisdisconnecting() is called during processing of disconnect() call, 515169Swnj * and soisdisconnected() is called when the connection to the peer 525169Swnj * is totally severed. The semantics of these routines are such that 535169Swnj * connectionless protocols can call soisconnected() and soisdisconnected() 545169Swnj * only, bypassing the in-progress calls when setting up a ``connection'' 555169Swnj * takes no time. 565169Swnj * 5712758Ssam * From the passive side, a socket is created with 5812758Ssam * two queues of sockets: so_q0 for connections in progress 597509Sroot * and so_q for connections already made and awaiting user acceptance. 607509Sroot * As a protocol is preparing incoming connections, it creates a socket 617509Sroot * structure queued on so_q0 by calling sonewconn(). When the connection 627509Sroot * is established, soisconnected() is called, and transfers the 637509Sroot * socket structure to so_q, making it available to accept(). 647509Sroot * 6512758Ssam * If a socket is closed with sockets on either 667509Sroot * so_q0 or so_q, these sockets are dropped. 677509Sroot * 6812758Ssam * If higher level protocols are implemented in 695169Swnj * the kernel, the wakeups done here will sometimes 7012758Ssam * cause software-interrupt process scheduling. 714903Swnj */ 725169Swnj 734903Swnj soisconnecting(so) 7412758Ssam register struct socket *so; 754903Swnj { 764903Swnj 774903Swnj so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 784903Swnj so->so_state |= SS_ISCONNECTING; 794903Swnj } 804903Swnj 814903Swnj soisconnected(so) 8212758Ssam register struct socket *so; 834903Swnj { 847509Sroot register struct socket *head = so->so_head; 854903Swnj 8640633Skarels so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 8740633Skarels so->so_state |= SS_ISCONNECTED; 8840633Skarels if (head && soqremque(so, 0)) { 897509Sroot soqinsque(head, so, 1); 9012758Ssam sorwakeup(head); 917509Sroot wakeup((caddr_t)&head->so_timeo); 9240633Skarels } else { 9340633Skarels wakeup((caddr_t)&so->so_timeo); 9440633Skarels sorwakeup(so); 9540633Skarels sowwakeup(so); 967509Sroot } 974903Swnj } 984903Swnj 994903Swnj soisdisconnecting(so) 10012758Ssam register struct socket *so; 1014903Swnj { 1024903Swnj 1035248Sroot so->so_state &= ~SS_ISCONNECTING; 1044903Swnj so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE); 1054903Swnj wakeup((caddr_t)&so->so_timeo); 1065170Swnj sowwakeup(so); 1075169Swnj sorwakeup(so); 1084903Swnj } 1094903Swnj 1104903Swnj soisdisconnected(so) 11112758Ssam register struct socket *so; 1124903Swnj { 1134903Swnj 1144903Swnj so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 1154903Swnj so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE); 1164903Swnj wakeup((caddr_t)&so->so_timeo); 1174903Swnj sowwakeup(so); 1184903Swnj sorwakeup(so); 1194903Swnj } 1204903Swnj 1215169Swnj /* 1227509Sroot * When an attempt at a new connection is noted on a socket 1237509Sroot * which accepts connections, sonewconn is called. If the 1247509Sroot * connection is possible (subject to space constraints, etc.) 1257509Sroot * then we allocate a new structure, propoerly linked into the 1267509Sroot * data structure of the original socket, and return this. 12740633Skarels * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED. 12840706Skarels * 12940706Skarels * Currently, sonewconn() is defined as sonewconn1() in socketvar.h 13040706Skarels * to catch calls that are missing the (new) second parameter. 1317509Sroot */ 1327509Sroot struct socket * 13340706Skarels sonewconn1(head, connstatus) 1347509Sroot register struct socket *head; 13540633Skarels int connstatus; 1367509Sroot { 1377509Sroot register struct socket *so; 13840633Skarels int soqueue = connstatus ? 1 : 0; 1397509Sroot 1407509Sroot if (head->so_qlen + head->so_q0len > 3 * head->so_qlimit / 2) 14137329Skarels return ((struct socket *)0); 14237329Skarels MALLOC(so, struct socket *, sizeof(*so), M_SOCKET, M_DONTWAIT); 14337329Skarels if (so == NULL) 14437329Skarels return ((struct socket *)0); 14537329Skarels bzero((caddr_t)so, sizeof(*so)); 1467509Sroot so->so_type = head->so_type; 1477509Sroot so->so_options = head->so_options &~ SO_ACCEPTCONN; 1487509Sroot so->so_linger = head->so_linger; 14910204Ssam so->so_state = head->so_state | SS_NOFDREF; 1507509Sroot so->so_proto = head->so_proto; 1517509Sroot so->so_timeo = head->so_timeo; 15235804Smarc so->so_pgid = head->so_pgid; 15335385Skarels (void) soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat); 15440633Skarels soqinsque(head, so, soqueue); 15512758Ssam if ((*so->so_proto->pr_usrreq)(so, PRU_ATTACH, 15612758Ssam (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0)) { 15740633Skarels (void) soqremque(so, soqueue); 15837329Skarels (void) free((caddr_t)so, M_SOCKET); 15937329Skarels return ((struct socket *)0); 1607509Sroot } 16140633Skarels if (connstatus) { 16240633Skarels sorwakeup(head); 16340633Skarels wakeup((caddr_t)&head->so_timeo); 16440633Skarels so->so_state |= connstatus; 16540633Skarels } 1667509Sroot return (so); 1677509Sroot } 1687509Sroot 1697509Sroot soqinsque(head, so, q) 1707509Sroot register struct socket *head, *so; 1717509Sroot int q; 1727509Sroot { 17340706Skarels 17440633Skarels register struct socket **prev; 1757509Sroot so->so_head = head; 1767509Sroot if (q == 0) { 1777509Sroot head->so_q0len++; 17840633Skarels so->so_q0 = 0; 17940633Skarels for (prev = &(head->so_q0); *prev; ) 18040633Skarels prev = &((*prev)->so_q0); 1817509Sroot } else { 1827509Sroot head->so_qlen++; 18340633Skarels so->so_q = 0; 18440633Skarels for (prev = &(head->so_q); *prev; ) 18540633Skarels prev = &((*prev)->so_q); 1867509Sroot } 18740633Skarels *prev = so; 1887509Sroot } 1897509Sroot 1907509Sroot soqremque(so, q) 1917509Sroot register struct socket *so; 1927509Sroot int q; 1937509Sroot { 1947509Sroot register struct socket *head, *prev, *next; 1957509Sroot 1967509Sroot head = so->so_head; 1977509Sroot prev = head; 1987509Sroot for (;;) { 1997509Sroot next = q ? prev->so_q : prev->so_q0; 2007509Sroot if (next == so) 2017509Sroot break; 20240633Skarels if (next == 0) 2037509Sroot return (0); 2047509Sroot prev = next; 2057509Sroot } 2067509Sroot if (q == 0) { 2077509Sroot prev->so_q0 = next->so_q0; 2087509Sroot head->so_q0len--; 2097509Sroot } else { 2107509Sroot prev->so_q = next->so_q; 2117509Sroot head->so_qlen--; 2127509Sroot } 2137509Sroot next->so_q0 = next->so_q = 0; 2147509Sroot next->so_head = 0; 2157509Sroot return (1); 2167509Sroot } 2177509Sroot 2187509Sroot /* 2195169Swnj * Socantsendmore indicates that no more data will be sent on the 2205169Swnj * socket; it would normally be applied to a socket when the user 2215169Swnj * informs the system that no more data is to be sent, by the protocol 2225169Swnj * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data 2235169Swnj * will be received, and will normally be applied to the socket by a 2245169Swnj * protocol when it detects that the peer will send no more data. 2255169Swnj * Data queued for reading in the socket may yet be read. 2265169Swnj */ 2275169Swnj 2284917Swnj socantsendmore(so) 2294917Swnj struct socket *so; 2304917Swnj { 2314917Swnj 2324917Swnj so->so_state |= SS_CANTSENDMORE; 2334917Swnj sowwakeup(so); 2344917Swnj } 2354917Swnj 2364917Swnj socantrcvmore(so) 2374917Swnj struct socket *so; 2384917Swnj { 2394917Swnj 2404917Swnj so->so_state |= SS_CANTRCVMORE; 2414917Swnj sorwakeup(so); 2424917Swnj } 2434917Swnj 2444903Swnj /* 2455169Swnj * Socket select/wakeup routines. 2464903Swnj */ 2475169Swnj 2485169Swnj /* 2494903Swnj * Queue a process for a select on a socket buffer. 2504903Swnj */ 2514903Swnj sbselqueue(sb) 2524903Swnj struct sockbuf *sb; 2534903Swnj { 25435385Skarels struct proc *p; 2554903Swnj 2564917Swnj if ((p = sb->sb_sel) && p->p_wchan == (caddr_t)&selwait) 2574903Swnj sb->sb_flags |= SB_COLL; 25844228Skarels else { 2594903Swnj sb->sb_sel = u.u_procp; 26044228Skarels sb->sb_flags |= SB_SEL; 26144228Skarels } 2624903Swnj } 2634903Swnj 2644903Swnj /* 2654917Swnj * Wait for data to arrive at/drain from a socket buffer. 2664917Swnj */ 2674917Swnj sbwait(sb) 2684917Swnj struct sockbuf *sb; 2694917Swnj { 2704917Swnj 2714917Swnj sb->sb_flags |= SB_WAIT; 27240706Skarels return (tsleep((caddr_t)&sb->sb_cc, 27340706Skarels (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, netio, 27440706Skarels sb->sb_timeo)); 2754917Swnj } 2764917Swnj 27740706Skarels /* 27840706Skarels * Lock a sockbuf already known to be locked; 27940706Skarels * return any error returned from sleep (EINTR). 28040706Skarels */ 28140706Skarels sb_lock(sb) 28240706Skarels register struct sockbuf *sb; 28340706Skarels { 28440706Skarels int error; 28540706Skarels 28640706Skarels while (sb->sb_flags & SB_LOCK) { 28740706Skarels sb->sb_flags |= SB_WANT; 28840706Skarels if (error = tsleep((caddr_t)&sb->sb_flags, 28940706Skarels (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH, 29040706Skarels netio, 0)) 29140706Skarels return (error); 29240706Skarels } 29340706Skarels sb->sb_flags |= SB_LOCK; 29440706Skarels return (0); 29540706Skarels } 29640706Skarels 2974917Swnj /* 2984903Swnj * Wakeup processes waiting on a socket buffer. 29935385Skarels * Do asynchronous notification via SIGIO 30035385Skarels * if the socket has the SS_ASYNC flag set. 3014903Swnj */ 30235385Skarels sowakeup(so, sb) 30335385Skarels register struct socket *so; 30412758Ssam register struct sockbuf *sb; 3054903Swnj { 30637329Skarels struct proc *p; 3074903Swnj 3084903Swnj if (sb->sb_sel) { 3094903Swnj selwakeup(sb->sb_sel, sb->sb_flags & SB_COLL); 3104903Swnj sb->sb_sel = 0; 31144228Skarels sb->sb_flags &= ~(SB_SEL|SB_COLL); 3124903Swnj } 3134903Swnj if (sb->sb_flags & SB_WAIT) { 3144903Swnj sb->sb_flags &= ~SB_WAIT; 3155013Swnj wakeup((caddr_t)&sb->sb_cc); 3164903Swnj } 31715829Scooper if (so->so_state & SS_ASYNC) { 31835804Smarc if (so->so_pgid < 0) 31935804Smarc gsignal(-so->so_pgid, SIGIO); 32035804Smarc else if (so->so_pgid > 0 && (p = pfind(so->so_pgid)) != 0) 32115829Scooper psignal(p, SIGIO); 32215829Scooper } 32315829Scooper } 32415829Scooper 32515829Scooper /* 3265169Swnj * Socket buffer (struct sockbuf) utility routines. 3275169Swnj * 3285169Swnj * Each socket contains two socket buffers: one for sending data and 3295169Swnj * one for receiving data. Each buffer contains a queue of mbufs, 3305169Swnj * information about the number of mbufs and amount of data in the 3315169Swnj * queue, and other fields allowing select() statements and notification 3325169Swnj * on data availability to be implemented. 3335169Swnj * 33416994Skarels * Data stored in a socket buffer is maintained as a list of records. 33516994Skarels * Each record is a list of mbufs chained together with the m_next 33635385Skarels * field. Records are chained together with the m_nextpkt field. The upper 33716994Skarels * level routine soreceive() expects the following conventions to be 33816994Skarels * observed when placing information in the receive buffer: 33916994Skarels * 34016994Skarels * 1. If the protocol requires each message be preceded by the sender's 34116994Skarels * name, then a record containing that name must be present before 34216994Skarels * any associated data (mbuf's must be of type MT_SONAME). 34316994Skarels * 2. If the protocol supports the exchange of ``access rights'' (really 34416994Skarels * just additional data associated with the message), and there are 34516994Skarels * ``rights'' to be received, then a record containing this data 34616994Skarels * should be present (mbuf's must be of type MT_RIGHTS). 34716994Skarels * 3. If a name or rights record exists, then it must be followed by 34816994Skarels * a data record, perhaps of zero length. 34916994Skarels * 3505169Swnj * Before using a new socket structure it is first necessary to reserve 35133406Skarels * buffer space to the socket, by calling sbreserve(). This should commit 3525169Swnj * some of the available buffer space in the system buffer pool for the 35333406Skarels * socket (currently, it does nothing but enforce limits). The space 35433406Skarels * should be released by calling sbrelease() when the socket is destroyed. 3555169Swnj */ 3565169Swnj 3579027Sroot soreserve(so, sndcc, rcvcc) 35812758Ssam register struct socket *so; 35933406Skarels u_long sndcc, rcvcc; 3609027Sroot { 3619027Sroot 3629027Sroot if (sbreserve(&so->so_snd, sndcc) == 0) 3639027Sroot goto bad; 3649027Sroot if (sbreserve(&so->so_rcv, rcvcc) == 0) 3659027Sroot goto bad2; 36640706Skarels if (so->so_rcv.sb_lowat == 0) 36740706Skarels so->so_rcv.sb_lowat = 1; 36840706Skarels if (so->so_snd.sb_lowat == 0) 36940706Skarels so->so_snd.sb_lowat = MCLBYTES; 37040706Skarels if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) 37140706Skarels so->so_snd.sb_lowat = so->so_snd.sb_hiwat; 3729027Sroot return (0); 3739027Sroot bad2: 3749027Sroot sbrelease(&so->so_snd); 3759027Sroot bad: 3769027Sroot return (ENOBUFS); 3779027Sroot } 3789027Sroot 3795169Swnj /* 3804903Swnj * Allot mbufs to a sockbuf. 38140706Skarels * Attempt to scale mbmax so that mbcnt doesn't become limiting 38226830Skarels * if buffering efficiency is near the normal case. 3834903Swnj */ 3844903Swnj sbreserve(sb, cc) 3854903Swnj struct sockbuf *sb; 38633406Skarels u_long cc; 3874903Swnj { 3884903Swnj 38940706Skarels if (cc > sb_max * MCLBYTES / (MSIZE + MCLBYTES)) 39017355Skarels return (0); 3914980Swnj sb->sb_hiwat = cc; 39240706Skarels sb->sb_mbmax = min(cc * 2, sb_max); 39340706Skarels if (sb->sb_lowat > sb->sb_hiwat) 39440706Skarels sb->sb_lowat = sb->sb_hiwat; 3954917Swnj return (1); 3964903Swnj } 3974903Swnj 3984903Swnj /* 3994903Swnj * Free mbufs held by a socket, and reserved mbuf space. 4004903Swnj */ 4014903Swnj sbrelease(sb) 4024903Swnj struct sockbuf *sb; 4034903Swnj { 4044903Swnj 4054903Swnj sbflush(sb); 4064980Swnj sb->sb_hiwat = sb->sb_mbmax = 0; 4074903Swnj } 4084903Swnj 4094903Swnj /* 41016994Skarels * Routines to add and remove 41116994Skarels * data from an mbuf queue. 41225630Skarels * 41325630Skarels * The routines sbappend() or sbappendrecord() are normally called to 41425630Skarels * append new mbufs to a socket buffer, after checking that adequate 41525630Skarels * space is available, comparing the function sbspace() with the amount 41625630Skarels * of data to be added. sbappendrecord() differs from sbappend() in 41725630Skarels * that data supplied is treated as the beginning of a new record. 41825630Skarels * To place a sender's address, optional access rights, and data in a 41925630Skarels * socket receive buffer, sbappendaddr() should be used. To place 42025630Skarels * access rights and data in a socket receive buffer, sbappendrights() 42125630Skarels * should be used. In either case, the new data begins a new record. 42225630Skarels * Note that unlike sbappend() and sbappendrecord(), these routines check 42325630Skarels * for the caller that there will be enough space to store the data. 42425630Skarels * Each fails if there is not enough space, or if it cannot find mbufs 42525630Skarels * to store additional information in. 42625630Skarels * 42725630Skarels * Reliable protocols may use the socket send buffer to hold data 42825630Skarels * awaiting acknowledgement. Data is normally copied from a socket 42925630Skarels * send buffer in a protocol with m_copy for output to a peer, 43025630Skarels * and then removing the data from the socket buffer with sbdrop() 43125630Skarels * or sbdroprecord() when the data is acknowledged by the peer. 4324903Swnj */ 4334903Swnj 4344903Swnj /* 43516994Skarels * Append mbuf chain m to the last record in the 43616994Skarels * socket buffer sb. The additional space associated 43716994Skarels * the mbuf chain is recorded in sb. Empty mbufs are 43816994Skarels * discarded and mbufs are compacted where possible. 4394903Swnj */ 4404903Swnj sbappend(sb, m) 44116994Skarels struct sockbuf *sb; 44216994Skarels struct mbuf *m; 4434903Swnj { 4446092Sroot register struct mbuf *n; 4454903Swnj 44616994Skarels if (m == 0) 44716994Skarels return; 44816994Skarels if (n = sb->sb_mb) { 44935385Skarels while (n->m_nextpkt) 45035385Skarels n = n->m_nextpkt; 4516092Sroot while (n->m_next) 45240633Skarels if (n->m_flags & M_EOR) { 45340633Skarels sbappendrecord(sb, m); /* XXXXXX!!!! */ 45440633Skarels return; 45540633Skarels } else 45640633Skarels n = n->m_next; 4574903Swnj } 45816994Skarels sbcompress(sb, m, n); 4594903Swnj } 4604903Swnj 461*44384Skarels #ifdef SOCKBUF_DEBUG 462*44384Skarels sbcheck(sb) 463*44384Skarels register struct sockbuf *sb; 464*44384Skarels { 465*44384Skarels register struct mbuf *m; 466*44384Skarels register int len = 0, mbcnt = 0; 467*44384Skarels 468*44384Skarels for (m = sb->sb_mb; m; m = m->m_next) { 469*44384Skarels len += m->m_len; 470*44384Skarels mbcnt += MSIZE; 471*44384Skarels if (m->m_flags & M_EXT) 472*44384Skarels mbcnt += m->m_ext.ext_size; 473*44384Skarels if (m->m_nextpkt) 474*44384Skarels panic("sbcheck nextpkt"); 475*44384Skarels } 476*44384Skarels if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) { 477*44384Skarels printf("cc %d != %d || mbcnt %d != %d\n", len, sb->sb_cc, 478*44384Skarels mbcnt, sb->sb_mbcnt); 479*44384Skarels panic("sbcheck"); 480*44384Skarels } 481*44384Skarels } 482*44384Skarels #endif 483*44384Skarels 4845169Swnj /* 48516994Skarels * As above, except the mbuf chain 48616994Skarels * begins a new record. 4875169Swnj */ 48816994Skarels sbappendrecord(sb, m0) 48916994Skarels register struct sockbuf *sb; 49016994Skarels register struct mbuf *m0; 4914928Swnj { 4924928Swnj register struct mbuf *m; 4934928Swnj 49416994Skarels if (m0 == 0) 49516994Skarels return; 49616994Skarels if (m = sb->sb_mb) 49735385Skarels while (m->m_nextpkt) 49835385Skarels m = m->m_nextpkt; 49916994Skarels /* 50016994Skarels * Put the first mbuf on the queue. 50116994Skarels * Note this permits zero length records. 50216994Skarels */ 50316994Skarels sballoc(sb, m0); 50416994Skarels if (m) 50535385Skarels m->m_nextpkt = m0; 50616994Skarels else 50716994Skarels sb->sb_mb = m0; 50816994Skarels m = m0->m_next; 50916994Skarels m0->m_next = 0; 51037329Skarels if (m && (m0->m_flags & M_EOR)) { 51137329Skarels m0->m_flags &= ~M_EOR; 51237329Skarels m->m_flags |= M_EOR; 51337329Skarels } 51416994Skarels sbcompress(sb, m, m0); 51516994Skarels } 51616994Skarels 51716994Skarels /* 51837329Skarels * As above except that OOB data 51937329Skarels * is inserted at the beginning of the sockbuf, 52037329Skarels * but after any other OOB data. 52137329Skarels */ 52237329Skarels sbinsertoob(sb, m0) 52337329Skarels register struct sockbuf *sb; 52437329Skarels register struct mbuf *m0; 52537329Skarels { 52637329Skarels register struct mbuf *m; 52737329Skarels register struct mbuf **mp; 52837329Skarels 52937329Skarels if (m0 == 0) 53037329Skarels return; 53137329Skarels for (mp = &sb->sb_mb; m = *mp; mp = &((*mp)->m_nextpkt)) { 53237329Skarels again: 53337329Skarels switch (m->m_type) { 53437329Skarels 53537329Skarels case MT_OOBDATA: 53637329Skarels continue; /* WANT next train */ 53737329Skarels 53837329Skarels case MT_CONTROL: 53937329Skarels if (m = m->m_next) 54037329Skarels goto again; /* inspect THIS train further */ 54137329Skarels } 54237329Skarels break; 54337329Skarels } 54437329Skarels /* 54537329Skarels * Put the first mbuf on the queue. 54637329Skarels * Note this permits zero length records. 54737329Skarels */ 54837329Skarels sballoc(sb, m0); 54937329Skarels m0->m_nextpkt = *mp; 55037329Skarels *mp = m0; 55137329Skarels m = m0->m_next; 55237329Skarels m0->m_next = 0; 55337329Skarels if (m && (m0->m_flags & M_EOR)) { 55437329Skarels m0->m_flags &= ~M_EOR; 55537329Skarels m->m_flags |= M_EOR; 55637329Skarels } 55737329Skarels sbcompress(sb, m, m0); 55837329Skarels } 55937329Skarels 56037329Skarels /* 56142259Skarels * Append address and data, and optionally, control (ancillary) data 56235385Skarels * to the receive queue of a socket. If present, 56342259Skarels * m0 must include a packet header with total length. 56442259Skarels * Returns 0 if no space in sockbuf or insufficient mbufs. 56516994Skarels */ 56642259Skarels sbappendaddr(sb, asa, m0, control) 56716994Skarels register struct sockbuf *sb; 56816994Skarels struct sockaddr *asa; 56942259Skarels struct mbuf *m0, *control; 57016994Skarels { 57116994Skarels register struct mbuf *m, *n; 57237329Skarels int space = asa->sa_len; 57316994Skarels 57435385Skarels if (m0 && (m0->m_flags & M_PKTHDR) == 0) 57535385Skarels panic("sbappendaddr"); 57635385Skarels if (m0) 57735385Skarels space += m0->m_pkthdr.len; 57842259Skarels for (n = control; n; n = n->m_next) { 57942259Skarels space += n->m_len; 58042259Skarels if (n->m_next == 0) /* keep pointer to last control buf */ 58142259Skarels break; 58242259Skarels } 58316994Skarels if (space > sbspace(sb)) 5844928Swnj return (0); 58542259Skarels if (asa->sa_len > MLEN) 58642259Skarels return (0); 58725630Skarels MGET(m, M_DONTWAIT, MT_SONAME); 58816994Skarels if (m == 0) 5894928Swnj return (0); 59037329Skarels m->m_len = asa->sa_len; 59137329Skarels bcopy((caddr_t)asa, mtod(m, caddr_t), asa->sa_len); 59242259Skarels if (n) 59342259Skarels n->m_next = m0; /* concatenate data to control */ 59442259Skarels else 59542259Skarels control = m0; 59642259Skarels m->m_next = control; 59742259Skarels for (n = m; n; n = n->m_next) 59842259Skarels sballoc(sb, n); 59916994Skarels if (n = sb->sb_mb) { 60035385Skarels while (n->m_nextpkt) 60135385Skarels n = n->m_nextpkt; 60235385Skarels n->m_nextpkt = m; 60316994Skarels } else 60416994Skarels sb->sb_mb = m; 60516994Skarels return (1); 60616994Skarels } 60716994Skarels 60842259Skarels sbappendcontrol(sb, m0, control) 60916994Skarels struct sockbuf *sb; 61042259Skarels struct mbuf *control, *m0; 61116994Skarels { 61216994Skarels register struct mbuf *m, *n; 61316994Skarels int space = 0; 61416994Skarels 61542259Skarels if (control == 0) 61642259Skarels panic("sbappendcontrol"); 61742259Skarels for (m = control; ; m = m->m_next) { 61842259Skarels space += m->m_len; 61942259Skarels if (m->m_next == 0) 62042259Skarels break; 62142259Skarels } 62242259Skarels n = m; /* save pointer to last control buffer */ 62325630Skarels for (m = m0; m; m = m->m_next) 62416994Skarels space += m->m_len; 62516994Skarels if (space > sbspace(sb)) 62612758Ssam return (0); 62742259Skarels n->m_next = m0; /* concatenate data to control */ 62842259Skarels for (m = control; m; m = m->m_next) 62942259Skarels sballoc(sb, m); 63016994Skarels if (n = sb->sb_mb) { 63135385Skarels while (n->m_nextpkt) 63235385Skarels n = n->m_nextpkt; 63342259Skarels n->m_nextpkt = control; 63416994Skarels } else 63542259Skarels sb->sb_mb = control; 6364928Swnj return (1); 6374928Swnj } 6384928Swnj 6394903Swnj /* 64016994Skarels * Compress mbuf chain m into the socket 64116994Skarels * buffer sb following mbuf n. If n 64216994Skarels * is null, the buffer is presumed empty. 6434903Swnj */ 64416994Skarels sbcompress(sb, m, n) 64516994Skarels register struct sockbuf *sb; 64616994Skarels register struct mbuf *m, *n; 64716994Skarels { 64842259Skarels register int eor = 0; 64916994Skarels 65016994Skarels while (m) { 65137329Skarels eor |= m->m_flags & M_EOR; 65216994Skarels if (m->m_len == 0) { 65316994Skarels m = m_free(m); 65416994Skarels continue; 65516994Skarels } 65637329Skarels if (n && (n->m_flags & (M_EXT | M_EOR)) == 0 && 65735385Skarels (n->m_data + n->m_len + m->m_len) < &n->m_dat[MLEN] && 65825630Skarels n->m_type == m->m_type) { 65916994Skarels bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len, 66016994Skarels (unsigned)m->m_len); 66116994Skarels n->m_len += m->m_len; 66216994Skarels sb->sb_cc += m->m_len; 66316994Skarels m = m_free(m); 66416994Skarels continue; 66516994Skarels } 66616994Skarels if (n) 66716994Skarels n->m_next = m; 66816994Skarels else 66916994Skarels sb->sb_mb = m; 67037329Skarels sballoc(sb, m); 67116994Skarels n = m; 67237329Skarels m->m_flags &= ~M_EOR; 67316994Skarels m = m->m_next; 67416994Skarels n->m_next = 0; 67516994Skarels } 67637329Skarels if (n) 67737329Skarels n->m_flags |= eor; 67816994Skarels } 67916994Skarels 68016994Skarels /* 68116994Skarels * Free all mbufs in a sockbuf. 68216994Skarels * Check that all resources are reclaimed. 68316994Skarels */ 6844903Swnj sbflush(sb) 68512758Ssam register struct sockbuf *sb; 6864903Swnj { 6874903Swnj 6884903Swnj if (sb->sb_flags & SB_LOCK) 6894903Swnj panic("sbflush"); 69026105Skarels while (sb->sb_mbcnt) 69126363Skarels sbdrop(sb, (int)sb->sb_cc); 69242259Skarels if (sb->sb_cc || sb->sb_mb) 6934903Swnj panic("sbflush 2"); 6944903Swnj } 6954903Swnj 6964903Swnj /* 69716994Skarels * Drop data from (the front of) a sockbuf. 6984903Swnj */ 6994903Swnj sbdrop(sb, len) 7004903Swnj register struct sockbuf *sb; 7014903Swnj register int len; 7024903Swnj { 70316994Skarels register struct mbuf *m, *mn; 70416994Skarels struct mbuf *next; 7054903Swnj 70635385Skarels next = (m = sb->sb_mb) ? m->m_nextpkt : 0; 7074903Swnj while (len > 0) { 70816994Skarels if (m == 0) { 70916994Skarels if (next == 0) 71016994Skarels panic("sbdrop"); 71116994Skarels m = next; 71235385Skarels next = m->m_nextpkt; 71316994Skarels continue; 71416994Skarels } 7155064Swnj if (m->m_len > len) { 7164903Swnj m->m_len -= len; 71735385Skarels m->m_data += len; 7184903Swnj sb->sb_cc -= len; 7194903Swnj break; 7204903Swnj } 7215064Swnj len -= m->m_len; 7225064Swnj sbfree(sb, m); 7235064Swnj MFREE(m, mn); 7245064Swnj m = mn; 7254903Swnj } 72617331Skarels while (m && m->m_len == 0) { 72717417Skarels sbfree(sb, m); 72817331Skarels MFREE(m, mn); 72917331Skarels m = mn; 73017331Skarels } 73116994Skarels if (m) { 73216994Skarels sb->sb_mb = m; 73335385Skarels m->m_nextpkt = next; 73416994Skarels } else 73516994Skarels sb->sb_mb = next; 7364903Swnj } 73716994Skarels 73816994Skarels /* 73916994Skarels * Drop a record off the front of a sockbuf 74016994Skarels * and move the next record to the front. 74116994Skarels */ 74216994Skarels sbdroprecord(sb) 74316994Skarels register struct sockbuf *sb; 74416994Skarels { 74516994Skarels register struct mbuf *m, *mn; 74616994Skarels 74716994Skarels m = sb->sb_mb; 74816994Skarels if (m) { 74935385Skarels sb->sb_mb = m->m_nextpkt; 75016994Skarels do { 75116994Skarels sbfree(sb, m); 75216994Skarels MFREE(m, mn); 75316994Skarels } while (m = mn); 75416994Skarels } 75516994Skarels } 756