4.3BSD - /usr/src/sys/sys/uipc_socket2.c

   1: /*
   2:  * Copyright (c) 1982, 1986 Regents of the University of California.
   3:  * All rights reserved.  The Berkeley software License Agreement
   4:  * specifies the terms and conditions for redistribution.
   5:  *
   6:  *	@(#)uipc_socket2.c	7.1 (Berkeley) 6/5/86
   7:  */
   8: 
   9: #include "param.h"
  10: #include "systm.h"
  11: #include "dir.h"
  12: #include "user.h"
  13: #include "proc.h"
  14: #include "file.h"
  15: #include "inode.h"
  16: #include "buf.h"
  17: #include "mbuf.h"
  18: #include "protosw.h"
  19: #include "socket.h"
  20: #include "socketvar.h"
  21: 
  22: /*
  23:  * Primitive routines for operating on sockets and socket buffers
  24:  */
  25: 
  26: /*
  27:  * Procedures to manipulate state flags of socket
  28:  * and do appropriate wakeups.  Normal sequence from the
  29:  * active (originating) side is that soisconnecting() is
  30:  * called during processing of connect() call,
  31:  * resulting in an eventual call to soisconnected() if/when the
  32:  * connection is established.  When the connection is torn down
  33:  * soisdisconnecting() is called during processing of disconnect() call,
  34:  * and soisdisconnected() is called when the connection to the peer
  35:  * is totally severed.  The semantics of these routines are such that
  36:  * connectionless protocols can call soisconnected() and soisdisconnected()
  37:  * only, bypassing the in-progress calls when setting up a ``connection''
  38:  * takes no time.
  39:  *
  40:  * From the passive side, a socket is created with
  41:  * two queues of sockets: so_q0 for connections in progress
  42:  * and so_q for connections already made and awaiting user acceptance.
  43:  * As a protocol is preparing incoming connections, it creates a socket
  44:  * structure queued on so_q0 by calling sonewconn().  When the connection
  45:  * is established, soisconnected() is called, and transfers the
  46:  * socket structure to so_q, making it available to accept().
  47:  *
  48:  * If a socket is closed with sockets on either
  49:  * so_q0 or so_q, these sockets are dropped.
  50:  *
  51:  * If higher level protocols are implemented in
  52:  * the kernel, the wakeups done here will sometimes
  53:  * cause software-interrupt process scheduling.
  54:  */
  55: 
  56: soisconnecting(so)
  57:     register struct socket *so;
  58: {
  59: 
  60:     so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
  61:     so->so_state |= SS_ISCONNECTING;
  62:     wakeup((caddr_t)&so->so_timeo);
  63: }
  64: 
  65: soisconnected(so)
  66:     register struct socket *so;
  67: {
  68:     register struct socket *head = so->so_head;
  69: 
  70:     if (head) {
  71:         if (soqremque(so, 0) == 0)
  72:             panic("soisconnected");
  73:         soqinsque(head, so, 1);
  74:         sorwakeup(head);
  75:         wakeup((caddr_t)&head->so_timeo);
  76:     }
  77:     so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING);
  78:     so->so_state |= SS_ISCONNECTED;
  79:     wakeup((caddr_t)&so->so_timeo);
  80:     sorwakeup(so);
  81:     sowwakeup(so);
  82: }
  83: 
  84: soisdisconnecting(so)
  85:     register struct socket *so;
  86: {
  87: 
  88:     so->so_state &= ~SS_ISCONNECTING;
  89:     so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
  90:     wakeup((caddr_t)&so->so_timeo);
  91:     sowwakeup(so);
  92:     sorwakeup(so);
  93: }
  94: 
  95: soisdisconnected(so)
  96:     register struct socket *so;
  97: {
  98: 
  99:     so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
 100:     so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE);
 101:     wakeup((caddr_t)&so->so_timeo);
 102:     sowwakeup(so);
 103:     sorwakeup(so);
 104: }
 105: 
 106: /*
 107:  * When an attempt at a new connection is noted on a socket
 108:  * which accepts connections, sonewconn is called.  If the
 109:  * connection is possible (subject to space constraints, etc.)
 110:  * then we allocate a new structure, propoerly linked into the
 111:  * data structure of the original socket, and return this.
 112:  */
 113: struct socket *
 114: sonewconn(head)
 115:     register struct socket *head;
 116: {
 117:     register struct socket *so;
 118:     register struct mbuf *m;
 119: 
 120:     if (head->so_qlen + head->so_q0len > 3 * head->so_qlimit / 2)
 121:         goto bad;
 122:     m = m_getclr(M_DONTWAIT, MT_SOCKET);
 123:     if (m == NULL)
 124:         goto bad;
 125:     so = mtod(m, struct socket *);
 126:     so->so_type = head->so_type;
 127:     so->so_options = head->so_options &~ SO_ACCEPTCONN;
 128:     so->so_linger = head->so_linger;
 129:     so->so_state = head->so_state | SS_NOFDREF;
 130:     so->so_proto = head->so_proto;
 131:     so->so_timeo = head->so_timeo;
 132:     so->so_pgrp = head->so_pgrp;
 133:     soqinsque(head, so, 0);
 134:     if ((*so->so_proto->pr_usrreq)(so, PRU_ATTACH,
 135:         (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0)) {
 136:         (void) soqremque(so, 0);
 137:         (void) m_free(m);
 138:         goto bad;
 139:     }
 140:     return (so);
 141: bad:
 142:     return ((struct socket *)0);
 143: }
 144: 
 145: soqinsque(head, so, q)
 146:     register struct socket *head, *so;
 147:     int q;
 148: {
 149: 
 150:     so->so_head = head;
 151:     if (q == 0) {
 152:         head->so_q0len++;
 153:         so->so_q0 = head->so_q0;
 154:         head->so_q0 = so;
 155:     } else {
 156:         head->so_qlen++;
 157:         so->so_q = head->so_q;
 158:         head->so_q = so;
 159:     }
 160: }
 161: 
 162: soqremque(so, q)
 163:     register struct socket *so;
 164:     int q;
 165: {
 166:     register struct socket *head, *prev, *next;
 167: 
 168:     head = so->so_head;
 169:     prev = head;
 170:     for (;;) {
 171:         next = q ? prev->so_q : prev->so_q0;
 172:         if (next == so)
 173:             break;
 174:         if (next == head)
 175:             return (0);
 176:         prev = next;
 177:     }
 178:     if (q == 0) {
 179:         prev->so_q0 = next->so_q0;
 180:         head->so_q0len--;
 181:     } else {
 182:         prev->so_q = next->so_q;
 183:         head->so_qlen--;
 184:     }
 185:     next->so_q0 = next->so_q = 0;
 186:     next->so_head = 0;
 187:     return (1);
 188: }
 189: 
 190: /*
 191:  * Socantsendmore indicates that no more data will be sent on the
 192:  * socket; it would normally be applied to a socket when the user
 193:  * informs the system that no more data is to be sent, by the protocol
 194:  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
 195:  * will be received, and will normally be applied to the socket by a
 196:  * protocol when it detects that the peer will send no more data.
 197:  * Data queued for reading in the socket may yet be read.
 198:  */
 199: 
 200: socantsendmore(so)
 201:     struct socket *so;
 202: {
 203: 
 204:     so->so_state |= SS_CANTSENDMORE;
 205:     sowwakeup(so);
 206: }
 207: 
 208: socantrcvmore(so)
 209:     struct socket *so;
 210: {
 211: 
 212:     so->so_state |= SS_CANTRCVMORE;
 213:     sorwakeup(so);
 214: }
 215: 
 216: /*
 217:  * Socket select/wakeup routines.
 218:  */
 219: 
 220: /*
 221:  * Queue a process for a select on a socket buffer.
 222:  */
 223: sbselqueue(sb)
 224:     struct sockbuf *sb;
 225: {
 226:     register struct proc *p;
 227: 
 228:     if ((p = sb->sb_sel) && p->p_wchan == (caddr_t)&selwait)
 229:         sb->sb_flags |= SB_COLL;
 230:     else
 231:         sb->sb_sel = u.u_procp;
 232: }
 233: 
 234: /*
 235:  * Wait for data to arrive at/drain from a socket buffer.
 236:  */
 237: sbwait(sb)
 238:     struct sockbuf *sb;
 239: {
 240: 
 241:     sb->sb_flags |= SB_WAIT;
 242:     sleep((caddr_t)&sb->sb_cc, PZERO+1);
 243: }
 244: 
 245: /*
 246:  * Wakeup processes waiting on a socket buffer.
 247:  */
 248: sbwakeup(sb)
 249:     register struct sockbuf *sb;
 250: {
 251: 
 252:     if (sb->sb_sel) {
 253:         selwakeup(sb->sb_sel, sb->sb_flags & SB_COLL);
 254:         sb->sb_sel = 0;
 255:         sb->sb_flags &= ~SB_COLL;
 256:     }
 257:     if (sb->sb_flags & SB_WAIT) {
 258:         sb->sb_flags &= ~SB_WAIT;
 259:         wakeup((caddr_t)&sb->sb_cc);
 260:     }
 261: }
 262: 
 263: /*
 264:  * Wakeup socket readers and writers.
 265:  * Do asynchronous notification via SIGIO
 266:  * if the socket has the SS_ASYNC flag set.
 267:  */
 268: sowakeup(so, sb)
 269:     register struct socket *so;
 270:     struct sockbuf *sb;
 271: {
 272:     register struct proc *p;
 273: 
 274:     sbwakeup(sb);
 275:     if (so->so_state & SS_ASYNC) {
 276:         if (so->so_pgrp < 0)
 277:             gsignal(-so->so_pgrp, SIGIO);
 278:         else if (so->so_pgrp > 0 && (p = pfind(so->so_pgrp)) != 0)
 279:             psignal(p, SIGIO);
 280:     }
 281: }
 282: 
 283: /*
 284:  * Socket buffer (struct sockbuf) utility routines.
 285:  *
 286:  * Each socket contains two socket buffers: one for sending data and
 287:  * one for receiving data.  Each buffer contains a queue of mbufs,
 288:  * information about the number of mbufs and amount of data in the
 289:  * queue, and other fields allowing select() statements and notification
 290:  * on data availability to be implemented.
 291:  *
 292:  * Data stored in a socket buffer is maintained as a list of records.
 293:  * Each record is a list of mbufs chained together with the m_next
 294:  * field.  Records are chained together with the m_act field. The upper
 295:  * level routine soreceive() expects the following conventions to be
 296:  * observed when placing information in the receive buffer:
 297:  *
 298:  * 1. If the protocol requires each message be preceded by the sender's
 299:  *    name, then a record containing that name must be present before
 300:  *    any associated data (mbuf's must be of type MT_SONAME).
 301:  * 2. If the protocol supports the exchange of ``access rights'' (really
 302:  *    just additional data associated with the message), and there are
 303:  *    ``rights'' to be received, then a record containing this data
 304:  *    should be present (mbuf's must be of type MT_RIGHTS).
 305:  * 3. If a name or rights record exists, then it must be followed by
 306:  *    a data record, perhaps of zero length.
 307:  *
 308:  * Before using a new socket structure it is first necessary to reserve
 309:  * buffer space to the socket, by calling sbreserve().  This commits
 310:  * some of the available buffer space in the system buffer pool for the
 311:  * socket.  The space should be released by calling sbrelease() when the
 312:  * socket is destroyed.
 313:  */
 314: 
 315: soreserve(so, sndcc, rcvcc)
 316:     register struct socket *so;
 317:     int sndcc, rcvcc;
 318: {
 319: 
 320:     if (sbreserve(&so->so_snd, sndcc) == 0)
 321:         goto bad;
 322:     if (sbreserve(&so->so_rcv, rcvcc) == 0)
 323:         goto bad2;
 324:     return (0);
 325: bad2:
 326:     sbrelease(&so->so_snd);
 327: bad:
 328:     return (ENOBUFS);
 329: }
 330: 
 331: /*
 332:  * Allot mbufs to a sockbuf.
 333:  * Attempt to scale cc so that mbcnt doesn't become limiting
 334:  * if buffering efficiency is near the normal case.
 335:  */
 336: sbreserve(sb, cc)
 337:     struct sockbuf *sb;
 338: {
 339: 
 340:     if ((unsigned) cc > (unsigned)SB_MAX * CLBYTES / (2 * MSIZE + CLBYTES))
 341:         return (0);
 342:     sb->sb_hiwat = cc;
 343:     sb->sb_mbmax = MIN(cc * 2, SB_MAX);
 344:     return (1);
 345: }
 346: 
 347: /*
 348:  * Free mbufs held by a socket, and reserved mbuf space.
 349:  */
 350: sbrelease(sb)
 351:     struct sockbuf *sb;
 352: {
 353: 
 354:     sbflush(sb);
 355:     sb->sb_hiwat = sb->sb_mbmax = 0;
 356: }
 357: 
 358: /*
 359:  * Routines to add and remove
 360:  * data from an mbuf queue.
 361:  *
 362:  * The routines sbappend() or sbappendrecord() are normally called to
 363:  * append new mbufs to a socket buffer, after checking that adequate
 364:  * space is available, comparing the function sbspace() with the amount
 365:  * of data to be added.  sbappendrecord() differs from sbappend() in
 366:  * that data supplied is treated as the beginning of a new record.
 367:  * To place a sender's address, optional access rights, and data in a
 368:  * socket receive buffer, sbappendaddr() should be used.  To place
 369:  * access rights and data in a socket receive buffer, sbappendrights()
 370:  * should be used.  In either case, the new data begins a new record.
 371:  * Note that unlike sbappend() and sbappendrecord(), these routines check
 372:  * for the caller that there will be enough space to store the data.
 373:  * Each fails if there is not enough space, or if it cannot find mbufs
 374:  * to store additional information in.
 375:  *
 376:  * Reliable protocols may use the socket send buffer to hold data
 377:  * awaiting acknowledgement.  Data is normally copied from a socket
 378:  * send buffer in a protocol with m_copy for output to a peer,
 379:  * and then removing the data from the socket buffer with sbdrop()
 380:  * or sbdroprecord() when the data is acknowledged by the peer.
 381:  */
 382: 
 383: /*
 384:  * Append mbuf chain m to the last record in the
 385:  * socket buffer sb.  The additional space associated
 386:  * the mbuf chain is recorded in sb.  Empty mbufs are
 387:  * discarded and mbufs are compacted where possible.
 388:  */
 389: sbappend(sb, m)
 390:     struct sockbuf *sb;
 391:     struct mbuf *m;
 392: {
 393:     register struct mbuf *n;
 394: 
 395:     if (m == 0)
 396:         return;
 397:     if (n = sb->sb_mb) {
 398:         while (n->m_act)
 399:             n = n->m_act;
 400:         while (n->m_next)
 401:             n = n->m_next;
 402:     }
 403:     sbcompress(sb, m, n);
 404: }
 405: 
 406: /*
 407:  * As above, except the mbuf chain
 408:  * begins a new record.
 409:  */
 410: sbappendrecord(sb, m0)
 411:     register struct sockbuf *sb;
 412:     register struct mbuf *m0;
 413: {
 414:     register struct mbuf *m;
 415: 
 416:     if (m0 == 0)
 417:         return;
 418:     if (m = sb->sb_mb)
 419:         while (m->m_act)
 420:             m = m->m_act;
 421:     /*
 422: 	 * Put the first mbuf on the queue.
 423: 	 * Note this permits zero length records.
 424: 	 */
 425:     sballoc(sb, m0);
 426:     if (m)
 427:         m->m_act = m0;
 428:     else
 429:         sb->sb_mb = m0;
 430:     m = m0->m_next;
 431:     m0->m_next = 0;
 432:     sbcompress(sb, m, m0);
 433: }
 434: 
 435: /*
 436:  * Append address and data, and optionally, rights
 437:  * to the receive queue of a socket.  Return 0 if
 438:  * no space in sockbuf or insufficient mbufs.
 439:  */
 440: sbappendaddr(sb, asa, m0, rights0)
 441:     register struct sockbuf *sb;
 442:     struct sockaddr *asa;
 443:     struct mbuf *rights0, *m0;
 444: {
 445:     register struct mbuf *m, *n;
 446:     int space = sizeof (*asa);
 447: 
 448:     for (m = m0; m; m = m->m_next)
 449:         space += m->m_len;
 450:     if (rights0)
 451:         space += rights0->m_len;
 452:     if (space > sbspace(sb))
 453:         return (0);
 454:     MGET(m, M_DONTWAIT, MT_SONAME);
 455:     if (m == 0)
 456:         return (0);
 457:     *mtod(m, struct sockaddr *) = *asa;
 458:     m->m_len = sizeof (*asa);
 459:     if (rights0 && rights0->m_len) {
 460:         m->m_next = m_copy(rights0, 0, rights0->m_len);
 461:         if (m->m_next == 0) {
 462:             m_freem(m);
 463:             return (0);
 464:         }
 465:         sballoc(sb, m->m_next);
 466:     }
 467:     sballoc(sb, m);
 468:     if (n = sb->sb_mb) {
 469:         while (n->m_act)
 470:             n = n->m_act;
 471:         n->m_act = m;
 472:     } else
 473:         sb->sb_mb = m;
 474:     if (m->m_next)
 475:         m = m->m_next;
 476:     if (m0)
 477:         sbcompress(sb, m0, m);
 478:     return (1);
 479: }
 480: 
 481: sbappendrights(sb, m0, rights)
 482:     struct sockbuf *sb;
 483:     struct mbuf *rights, *m0;
 484: {
 485:     register struct mbuf *m, *n;
 486:     int space = 0;
 487: 
 488:     if (rights == 0)
 489:         panic("sbappendrights");
 490:     for (m = m0; m; m = m->m_next)
 491:         space += m->m_len;
 492:     space += rights->m_len;
 493:     if (space > sbspace(sb))
 494:         return (0);
 495:     m = m_copy(rights, 0, rights->m_len);
 496:     if (m == 0)
 497:         return (0);
 498:     sballoc(sb, m);
 499:     if (n = sb->sb_mb) {
 500:         while (n->m_act)
 501:             n = n->m_act;
 502:         n->m_act = m;
 503:     } else
 504:         sb->sb_mb = m;
 505:     if (m0)
 506:         sbcompress(sb, m0, m);
 507:     return (1);
 508: }
 509: 
 510: /*
 511:  * Compress mbuf chain m into the socket
 512:  * buffer sb following mbuf n.  If n
 513:  * is null, the buffer is presumed empty.
 514:  */
 515: sbcompress(sb, m, n)
 516:     register struct sockbuf *sb;
 517:     register struct mbuf *m, *n;
 518: {
 519: 
 520:     while (m) {
 521:         if (m->m_len == 0) {
 522:             m = m_free(m);
 523:             continue;
 524:         }
 525:         if (n && n->m_off <= MMAXOFF && m->m_off <= MMAXOFF &&
 526:             (n->m_off + n->m_len + m->m_len) <= MMAXOFF &&
 527:             n->m_type == m->m_type) {
 528:             bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
 529:                 (unsigned)m->m_len);
 530:             n->m_len += m->m_len;
 531:             sb->sb_cc += m->m_len;
 532:             m = m_free(m);
 533:             continue;
 534:         }
 535:         sballoc(sb, m);
 536:         if (n)
 537:             n->m_next = m;
 538:         else
 539:             sb->sb_mb = m;
 540:         n = m;
 541:         m = m->m_next;
 542:         n->m_next = 0;
 543:     }
 544: }
 545: 
 546: /*
 547:  * Free all mbufs in a sockbuf.
 548:  * Check that all resources are reclaimed.
 549:  */
 550: sbflush(sb)
 551:     register struct sockbuf *sb;
 552: {
 553: 
 554:     if (sb->sb_flags & SB_LOCK)
 555:         panic("sbflush");
 556:     while (sb->sb_mbcnt)
 557:         sbdrop(sb, (int)sb->sb_cc);
 558:     if (sb->sb_cc || sb->sb_mbcnt || sb->sb_mb)
 559:         panic("sbflush 2");
 560: }
 561: 
 562: /*
 563:  * Drop data from (the front of) a sockbuf.
 564:  */
 565: sbdrop(sb, len)
 566:     register struct sockbuf *sb;
 567:     register int len;
 568: {
 569:     register struct mbuf *m, *mn;
 570:     struct mbuf *next;
 571: 
 572:     next = (m = sb->sb_mb) ? m->m_act : 0;
 573:     while (len > 0) {
 574:         if (m == 0) {
 575:             if (next == 0)
 576:                 panic("sbdrop");
 577:             m = next;
 578:             next = m->m_act;
 579:             continue;
 580:         }
 581:         if (m->m_len > len) {
 582:             m->m_len -= len;
 583:             m->m_off += len;
 584:             sb->sb_cc -= len;
 585:             break;
 586:         }
 587:         len -= m->m_len;
 588:         sbfree(sb, m);
 589:         MFREE(m, mn);
 590:         m = mn;
 591:     }
 592:     while (m && m->m_len == 0) {
 593:         sbfree(sb, m);
 594:         MFREE(m, mn);
 595:         m = mn;
 596:     }
 597:     if (m) {
 598:         sb->sb_mb = m;
 599:         m->m_act = next;
 600:     } else
 601:         sb->sb_mb = next;
 602: }
 603: 
 604: /*
 605:  * Drop a record off the front of a sockbuf
 606:  * and move the next record to the front.
 607:  */
 608: sbdroprecord(sb)
 609:     register struct sockbuf *sb;
 610: {
 611:     register struct mbuf *m, *mn;
 612: 
 613:     m = sb->sb_mb;
 614:     if (m) {
 615:         sb->sb_mb = m->m_act;
 616:         do {
 617:             sbfree(sb, m);
 618:             MFREE(m, mn);
 619:         } while (m = mn);
 620:     }
 621: }

Defined functions