sys/kern/uipc_usrreq.c

   1 /*
   2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. All advertising materials mentioning features or use of this software
  14  *    must display the following acknowledgement:
  15  *      This product includes software developed by the University of
  16  *      California, Berkeley and its contributors.
  17  * 4. Neither the name of the University nor the names of its contributors
  18  *    may be used to endorse or promote products derived from this software
  19  *    without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  *
  33  *      From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
  34  * $FreeBSD: src/sys/kern/uipc_usrreq.c,v 1.54.2.10 2003/03/04 17:28:09 nectar Exp $
  35  * $DragonFly: src/sys/kern/uipc_usrreq.c,v 1.44 2008/09/06 05:44:58 dillon Exp $
  36  */
  37
  38 #include <sys/param.h>
  39 #include <sys/systm.h>
  40 #include <sys/kernel.h>
  41 #include <sys/domain.h>
  42 #include <sys/fcntl.h>
  43 #include <sys/malloc.h>         /* XXX must be before <sys/file.h> */
  44 #include <sys/proc.h>
  45 #include <sys/file.h>
  46 #include <sys/filedesc.h>
  47 #include <sys/mbuf.h>
  48 #include <sys/nlookup.h>
  49 #include <sys/protosw.h>
  50 #include <sys/socket.h>
  51 #include <sys/socketvar.h>
  52 #include <sys/resourcevar.h>
  53 #include <sys/stat.h>
  54 #include <sys/mount.h>
  55 #include <sys/sysctl.h>
  56 #include <sys/un.h>
  57 #include <sys/unpcb.h>
  58 #include <sys/vnode.h>
  59 #include <sys/file2.h>
  60 #include <sys/spinlock2.h>
  61
  62
  63 static  MALLOC_DEFINE(M_UNPCB, "unpcb", "unpcb struct");
  64 static  unp_gen_t unp_gencnt;
  65 static  u_int unp_count;
  66
  67 static  struct unp_head unp_shead, unp_dhead;
  68
  69 /*
  70  * Unix communications domain.
  71  *
  72  * TODO:
  73  *      RDM
  74  *      rethink name space problems
  75  *      need a proper out-of-band
  76  *      lock pushdown
  77  */
  78 static struct   sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
  79 static ino_t    unp_ino = 1;            /* prototype for fake inode numbers */
  80 static struct spinlock unp_ino_spin = SPINLOCK_INITIALIZER(&unp_ino_spin);
  81
  82 static int     unp_attach (struct socket *, struct pru_attach_info *);
  83 static void    unp_detach (struct unpcb *);
  84 static int     unp_bind (struct unpcb *,struct sockaddr *, struct thread *);
  85 static int     unp_connect (struct socket *,struct sockaddr *,
  86                                 struct thread *);
  87 static void    unp_disconnect (struct unpcb *);
  88 static void    unp_shutdown (struct unpcb *);
  89 static void    unp_drop (struct unpcb *, int);
  90 static void    unp_gc (void);
  91 static int     unp_gc_clearmarks(struct file *, void *);
  92 static int     unp_gc_checkmarks(struct file *, void *);
  93 static int     unp_gc_checkrefs(struct file *, void *);
  94 static int     unp_revoke_gc_check(struct file *, void *);
  95 static void    unp_scan (struct mbuf *, void (*)(struct file *, void *),
  96                                 void *data);
  97 static void    unp_mark (struct file *, void *data);
  98 static void    unp_discard (struct file *, void *);
  99 static int     unp_internalize (struct mbuf *, struct thread *);
 100 static int     unp_listen (struct unpcb *, struct thread *);
 101 static void    unp_fp_externalize(struct proc *p, struct file *fp, int fd);
 102
 103 static int
 104 uipc_abort(struct socket *so)
 105 {
 106         struct unpcb *unp = so->so_pcb;
 107
 108         if (unp == NULL)
 109                 return EINVAL;
 110         unp_drop(unp, ECONNABORTED);
 111         unp_detach(unp);
 112         sofree(so);
 113         return 0;
 114 }
 115
 116 static int
 117 uipc_accept(struct socket *so, struct sockaddr **nam)
 118 {
 119         struct unpcb *unp = so->so_pcb;
 120
 121         if (unp == NULL)
 122                 return EINVAL;
 123
 124         /*
 125          * Pass back name of connected socket,
 126          * if it was bound and we are still connected
 127          * (our peer may have closed already!).
 128          */
 129         if (unp->unp_conn && unp->unp_conn->unp_addr) {
 130                 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr);
 131         } else {
 132                 *nam = dup_sockaddr((struct sockaddr *)&sun_noname);
 133         }
 134         return 0;
 135 }
 136
 137 static int
 138 uipc_attach(struct socket *so, int proto, struct pru_attach_info *ai)
 139 {
 140         struct unpcb *unp = so->so_pcb;
 141
 142         if (unp != NULL)
 143                 return EISCONN;
 144         return unp_attach(so, ai);
 145 }
 146
 147 static int
 148 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 149 {
 150         struct unpcb *unp = so->so_pcb;
 151
 152         if (unp == NULL)
 153                 return EINVAL;
 154         return unp_bind(unp, nam, td);
 155 }
 156
 157 static int
 158 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 159 {
 160         struct unpcb *unp = so->so_pcb;
 161
 162         if (unp == NULL)
 163                 return EINVAL;
 164         return unp_connect(so, nam, td);
 165 }
 166
 167 static int
 168 uipc_connect2(struct socket *so1, struct socket *so2)
 169 {
 170         struct unpcb *unp = so1->so_pcb;
 171
 172         if (unp == NULL)
 173                 return EINVAL;
 174
 175         return unp_connect2(so1, so2);
 176 }
 177
 178 /* control is EOPNOTSUPP */
 179
 180 static int
 181 uipc_detach(struct socket *so)
 182 {
 183         struct unpcb *unp = so->so_pcb;
 184
 185         if (unp == NULL)
 186                 return EINVAL;
 187
 188         unp_detach(unp);
 189         return 0;
 190 }
 191
 192 static int
 193 uipc_disconnect(struct socket *so)
 194 {
 195         struct unpcb *unp = so->so_pcb;
 196
 197         if (unp == NULL)
 198                 return EINVAL;
 199         unp_disconnect(unp);
 200         return 0;
 201 }
 202
 203 static int
 204 uipc_listen(struct socket *so, struct thread *td)
 205 {
 206         struct unpcb *unp = so->so_pcb;
 207
 208         if (unp == NULL || unp->unp_vnode == NULL)
 209                 return EINVAL;
 210         return unp_listen(unp, td);
 211 }
 212
 213 static int
 214 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
 215 {
 216         struct unpcb *unp = so->so_pcb;
 217
 218         if (unp == NULL)
 219                 return EINVAL;
 220         if (unp->unp_conn && unp->unp_conn->unp_addr)
 221                 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr);
 222         else {
 223                 /*
 224                  * XXX: It seems that this test always fails even when
 225                  * connection is established.  So, this else clause is
 226                  * added as workaround to return PF_LOCAL sockaddr.
 227                  */
 228                 *nam = dup_sockaddr((struct sockaddr *)&sun_noname);
 229         }
 230         return 0;
 231 }
 232
 233 static int
 234 uipc_rcvd(struct socket *so, int flags)
 235 {
 236         struct unpcb *unp = so->so_pcb;
 237         struct socket *so2;
 238
 239         if (unp == NULL)
 240                 return EINVAL;
 241         switch (so->so_type) {
 242         case SOCK_DGRAM:
 243                 panic("uipc_rcvd DGRAM?");
 244                 /*NOTREACHED*/
 245
 246         case SOCK_STREAM:
 247         case SOCK_SEQPACKET:
 248                 if (unp->unp_conn == NULL)
 249                         break;
 250                 /*
 251                  * Because we are transfering mbufs directly to the
 252                  * peer socket we have to use SSB_STOP on the sender
 253                  * to prevent it from building up infinite mbufs.
 254                  */
 255                 so2 = unp->unp_conn->unp_socket;
 256                 if (so->so_rcv.ssb_cc < so2->so_snd.ssb_hiwat &&
 257                     so->so_rcv.ssb_mbcnt < so2->so_snd.ssb_mbmax
 258                 ) {
 259                         so2->so_snd.ssb_flags &= ~SSB_STOP;
 260                         sowwakeup(so2);
 261                 }
 262                 break;
 263
 264         default:
 265                 panic("uipc_rcvd unknown socktype");
 266         }
 267         return 0;
 268 }
 269
 270 /* pru_rcvoob is EOPNOTSUPP */
 271
 272 static int
 273 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
 274           struct mbuf *control, struct thread *td)
 275 {
 276         int error = 0;
 277         struct unpcb *unp = so->so_pcb;
 278         struct socket *so2;
 279
 280         if (unp == NULL) {
 281                 error = EINVAL;
 282                 goto release;
 283         }
 284         if (flags & PRUS_OOB) {
 285                 error = EOPNOTSUPP;
 286                 goto release;
 287         }
 288
 289         if (control && (error = unp_internalize(control, td)))
 290                 goto release;
 291
 292         switch (so->so_type) {
 293         case SOCK_DGRAM:
 294         {
 295                 struct sockaddr *from;
 296
 297                 if (nam) {
 298                         if (unp->unp_conn) {
 299                                 error = EISCONN;
 300                                 break;
 301                         }
 302                         error = unp_connect(so, nam, td);
 303                         if (error)
 304                                 break;
 305                 } else {
 306                         if (unp->unp_conn == NULL) {
 307                                 error = ENOTCONN;
 308                                 break;
 309                         }
 310                 }
 311                 so2 = unp->unp_conn->unp_socket;
 312                 if (unp->unp_addr)
 313                         from = (struct sockaddr *)unp->unp_addr;
 314                 else
 315                         from = &sun_noname;
 316                 if (ssb_appendaddr(&so2->so_rcv, from, m, control)) {
 317                         sorwakeup(so2);
 318                         m = NULL;
 319                         control = NULL;
 320                 } else {
 321                         error = ENOBUFS;
 322                 }
 323                 if (nam)
 324                         unp_disconnect(unp);
 325                 break;
 326         }
 327
 328         case SOCK_STREAM:
 329         case SOCK_SEQPACKET:
 330                 /* Connect if not connected yet. */
 331                 /*
 332                  * Note: A better implementation would complain
 333                  * if not equal to the peer's address.
 334                  */
 335                 if (!(so->so_state & SS_ISCONNECTED)) {
 336                         if (nam) {
 337                                 error = unp_connect(so, nam, td);
 338                                 if (error)
 339                                         break;  /* XXX */
 340                         } else {
 341                                 error = ENOTCONN;
 342                                 break;
 343                         }
 344                 }
 345
 346                 if (so->so_state & SS_CANTSENDMORE) {
 347                         error = EPIPE;
 348                         break;
 349                 }
 350                 if (unp->unp_conn == NULL)
 351                         panic("uipc_send connected but no connection?");
 352                 so2 = unp->unp_conn->unp_socket;
 353                 /*
 354                  * Send to paired receive port, and then reduce
 355                  * send buffer hiwater marks to maintain backpressure.
 356                  * Wake up readers.
 357                  */
 358                 if (control) {
 359                         if (ssb_appendcontrol(&so2->so_rcv, m, control)) {
 360                                 control = NULL;
 361                                 m = NULL;
 362                         }
 363                 } else if (so->so_type == SOCK_SEQPACKET) {
 364                         sbappendrecord(&so2->so_rcv.sb, m);
 365                         m = NULL;
 366                 } else {
 367                         sbappend(&so2->so_rcv.sb, m);
 368                         m = NULL;
 369                 }
 370
 371                 /*
 372                  * Because we are transfering mbufs directly to the
 373                  * peer socket we have to use SSB_STOP on the sender
 374                  * to prevent it from building up infinite mbufs.
 375                  */
 376                 if (so2->so_rcv.ssb_cc >= so->so_snd.ssb_hiwat ||
 377                     so2->so_rcv.ssb_mbcnt >= so->so_snd.ssb_mbmax
 378                 ) {
 379                         so->so_snd.ssb_flags |= SSB_STOP;
 380                 }
 381                 sorwakeup(so2);
 382                 break;
 383
 384         default:
 385                 panic("uipc_send unknown socktype");
 386         }
 387
 388         /*
 389          * SEND_EOF is equivalent to a SEND followed by a SHUTDOWN.
 390          */
 391         if (flags & PRUS_EOF) {
 392                 socantsendmore(so);
 393                 unp_shutdown(unp);
 394         }
 395
 396         if (control && error != 0)
 397                 unp_dispose(control);
 398
 399 release:
 400         if (control)
 401                 m_freem(control);
 402         if (m)
 403                 m_freem(m);
 404         return error;
 405 }
 406
 407 /*
 408  * MPSAFE
 409  */
 410 static int
 411 uipc_sense(struct socket *so, struct stat *sb)
 412 {
 413         struct unpcb *unp = so->so_pcb;
 414
 415         if (unp == NULL)
 416                 return EINVAL;
 417         sb->st_blksize = so->so_snd.ssb_hiwat;
 418         sb->st_dev = NOUDEV;
 419         if (unp->unp_ino == 0) {        /* make up a non-zero inode number */
 420                 spin_lock_wr(&unp_ino_spin);
 421                 unp->unp_ino = unp_ino++;
 422                 spin_unlock_wr(&unp_ino_spin);
 423         }
 424         sb->st_ino = unp->unp_ino;
 425         return (0);
 426 }
 427
 428 static int
 429 uipc_shutdown(struct socket *so)
 430 {
 431         struct unpcb *unp = so->so_pcb;
 432
 433         if (unp == NULL)
 434                 return EINVAL;
 435         socantsendmore(so);
 436         unp_shutdown(unp);
 437         return 0;
 438 }
 439
 440 static int
 441 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
 442 {
 443         struct unpcb *unp = so->so_pcb;
 444
 445         if (unp == NULL)
 446                 return EINVAL;
 447         if (unp->unp_addr)
 448                 *nam = dup_sockaddr((struct sockaddr *)unp->unp_addr);
 449         return 0;
 450 }
 451
 452 struct pr_usrreqs uipc_usrreqs = {
 453         .pru_abort = uipc_abort,
 454         .pru_accept = uipc_accept,
 455         .pru_attach = uipc_attach,
 456         .pru_bind = uipc_bind,
 457         .pru_connect = uipc_connect,
 458         .pru_connect2 = uipc_connect2,
 459         .pru_control = pru_control_notsupp,
 460         .pru_detach = uipc_detach,
 461         .pru_disconnect = uipc_disconnect,
 462         .pru_listen = uipc_listen,
 463         .pru_peeraddr = uipc_peeraddr,
 464         .pru_rcvd = uipc_rcvd,
 465         .pru_rcvoob = pru_rcvoob_notsupp,
 466         .pru_send = uipc_send,
 467         .pru_sense = uipc_sense,
 468         .pru_shutdown = uipc_shutdown,
 469         .pru_sockaddr = uipc_sockaddr,
 470         .pru_sosend = sosend,
 471         .pru_soreceive = soreceive,
 472         .pru_sopoll = sopoll
 473 };
 474
 475 int
 476 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
 477 {
 478         struct unpcb *unp = so->so_pcb;
 479         int error = 0;
 480
 481         switch (sopt->sopt_dir) {
 482         case SOPT_GET:
 483                 switch (sopt->sopt_name) {
 484                 case LOCAL_PEERCRED:
 485                         if (unp->unp_flags & UNP_HAVEPC)
 486                                 soopt_from_kbuf(sopt, &unp->unp_peercred,
 487                                                 sizeof(unp->unp_peercred));
 488                         else {
 489                                 if (so->so_type == SOCK_STREAM)
 490                                         error = ENOTCONN;
 491                                 else if (so->so_type == SOCK_SEQPACKET)
 492                                         error = ENOTCONN;
 493                                 else
 494                                         error = EINVAL;
 495                         }
 496                         break;
 497                 default:
 498                         error = EOPNOTSUPP;
 499                         break;
 500                 }
 501                 break;
 502         case SOPT_SET:
 503         default:
 504                 error = EOPNOTSUPP;
 505                 break;
 506         }
 507         return (error);
 508 }
 509
 510 /*
 511  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
 512  * for stream sockets, although the total for sender and receiver is
 513  * actually only PIPSIZ.
 514  *
 515  * Datagram sockets really use the sendspace as the maximum datagram size,
 516  * and don't really want to reserve the sendspace.  Their recvspace should
 517  * be large enough for at least one max-size datagram plus address.
 518  *
 519  * We want the local send/recv space to be significant larger then lo0's
 520  * mtu of 16384.
 521  */
 522 #ifndef PIPSIZ
 523 #define PIPSIZ  57344
 524 #endif
 525 static u_long   unpst_sendspace = PIPSIZ;
 526 static u_long   unpst_recvspace = PIPSIZ;
 527 static u_long   unpdg_sendspace = 2*1024;       /* really max datagram size */
 528 static u_long   unpdg_recvspace = 4*1024;
 529
 530 static int      unp_rights;                     /* file descriptors in flight */
 531 static struct spinlock unp_spin = SPINLOCK_INITIALIZER(&unp_spin);
 532
 533 SYSCTL_DECL(_net_local_seqpacket);
 534 SYSCTL_DECL(_net_local_stream);
 535 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
 536            &unpst_sendspace, 0, "");
 537 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
 538            &unpst_recvspace, 0, "");
 539
 540 SYSCTL_DECL(_net_local_dgram);
 541 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
 542            &unpdg_sendspace, 0, "");
 543 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
 544            &unpdg_recvspace, 0, "");
 545
 546 SYSCTL_DECL(_net_local);
 547 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
 548
 549 static int
 550 unp_attach(struct socket *so, struct pru_attach_info *ai)
 551 {
 552         struct unpcb *unp;
 553         int error;
 554
 555         if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
 556                 switch (so->so_type) {
 557
 558                 case SOCK_STREAM:
 559                 case SOCK_SEQPACKET:
 560                         error = soreserve(so, unpst_sendspace, unpst_recvspace,
 561                                           ai->sb_rlimit);
 562                         break;
 563
 564                 case SOCK_DGRAM:
 565                         error = soreserve(so, unpdg_sendspace, unpdg_recvspace,
 566                                           ai->sb_rlimit);
 567                         break;
 568
 569                 default:
 570                         panic("unp_attach");
 571                 }
 572                 if (error)
 573                         return (error);
 574         }
 575         unp = kmalloc(sizeof(*unp), M_UNPCB, M_NOWAIT|M_ZERO);
 576         if (unp == NULL)
 577                 return (ENOBUFS);
 578         unp->unp_gencnt = ++unp_gencnt;
 579         unp_count++;
 580         LIST_INIT(&unp->unp_refs);
 581         unp->unp_socket = so;
 582         unp->unp_rvnode = ai->fd_rdir;          /* jail cruft XXX JH */
 583         LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
 584                          : &unp_shead, unp, unp_link);
 585         so->so_pcb = (caddr_t)unp;
 586         return (0);
 587 }
 588
 589 static void
 590 unp_detach(struct unpcb *unp)
 591 {
 592         LIST_REMOVE(unp, unp_link);
 593         unp->unp_gencnt = ++unp_gencnt;
 594         --unp_count;
 595         if (unp->unp_vnode) {
 596                 unp->unp_vnode->v_socket = NULL;
 597                 vrele(unp->unp_vnode);
 598                 unp->unp_vnode = NULL;
 599         }
 600         if (unp->unp_conn)
 601                 unp_disconnect(unp);
 602         while (!LIST_EMPTY(&unp->unp_refs))
 603                 unp_drop(LIST_FIRST(&unp->unp_refs), ECONNRESET);
 604         soisdisconnected(unp->unp_socket);
 605         unp->unp_socket->so_pcb = NULL;
 606         if (unp_rights) {
 607                 /*
 608                  * Normally the receive buffer is flushed later,
 609                  * in sofree, but if our receive buffer holds references
 610                  * to descriptors that are now garbage, we will dispose
 611                  * of those descriptor references after the garbage collector
 612                  * gets them (resulting in a "panic: closef: count < 0").
 613                  */
 614                 sorflush(unp->unp_socket);
 615                 unp_gc();
 616         }
 617         if (unp->unp_addr)
 618                 kfree(unp->unp_addr, M_SONAME);
 619         kfree(unp, M_UNPCB);
 620 }
 621
 622 static int
 623 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td)
 624 {
 625         struct proc *p = td->td_proc;
 626         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
 627         struct vnode *vp;
 628         struct vattr vattr;
 629         int error, namelen;
 630         struct nlookupdata nd;
 631         char buf[SOCK_MAXADDRLEN];
 632
 633         if (unp->unp_vnode != NULL)
 634                 return (EINVAL);
 635         namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
 636         if (namelen <= 0)
 637                 return (EINVAL);
 638         strncpy(buf, soun->sun_path, namelen);
 639         buf[namelen] = 0;       /* null-terminate the string */
 640         error = nlookup_init(&nd, buf, UIO_SYSSPACE,
 641                              NLC_LOCKVP | NLC_CREATE | NLC_REFDVP);
 642         if (error == 0)
 643                 error = nlookup(&nd);
 644         if (error == 0 && nd.nl_nch.ncp->nc_vp != NULL)
 645                 error = EADDRINUSE;
 646         if (error)
 647                 goto done;
 648
 649         VATTR_NULL(&vattr);
 650         vattr.va_type = VSOCK;
 651         vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask);
 652         error = VOP_NCREATE(&nd.nl_nch, nd.nl_dvp, &vp, nd.nl_cred, &vattr);
 653         if (error == 0) {
 654                 vp->v_socket = unp->unp_socket;
 655                 unp->unp_vnode = vp;
 656                 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam);
 657                 vn_unlock(vp);
 658         }
 659 done:
 660         nlookup_done(&nd);
 661         return (error);
 662 }
 663
 664 static int
 665 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 666 {
 667         struct proc *p = td->td_proc;
 668         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
 669         struct vnode *vp;
 670         struct socket *so2, *so3;
 671         struct unpcb *unp, *unp2, *unp3;
 672         int error, len;
 673         struct nlookupdata nd;
 674         char buf[SOCK_MAXADDRLEN];
 675
 676         KKASSERT(p);
 677
 678         len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
 679         if (len <= 0)
 680                 return EINVAL;
 681         strncpy(buf, soun->sun_path, len);
 682         buf[len] = 0;
 683
 684         vp = NULL;
 685         error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_FOLLOW);
 686         if (error == 0)
 687                 error = nlookup(&nd);
 688         if (error == 0)
 689                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
 690         nlookup_done(&nd);
 691         if (error)
 692                 return (error);
 693
 694         if (vp->v_type != VSOCK) {
 695                 error = ENOTSOCK;
 696                 goto bad;
 697         }
 698         error = VOP_ACCESS(vp, VWRITE, p->p_ucred);
 699         if (error)
 700                 goto bad;
 701         so2 = vp->v_socket;
 702         if (so2 == NULL) {
 703                 error = ECONNREFUSED;
 704                 goto bad;
 705         }
 706         if (so->so_type != so2->so_type) {
 707                 error = EPROTOTYPE;
 708                 goto bad;
 709         }
 710         if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
 711                 if (!(so2->so_options & SO_ACCEPTCONN) ||
 712                     (so3 = sonewconn(so2, 0)) == NULL) {
 713                         error = ECONNREFUSED;
 714                         goto bad;
 715                 }
 716                 unp = so->so_pcb;
 717                 unp2 = so2->so_pcb;
 718                 unp3 = so3->so_pcb;
 719                 if (unp2->unp_addr)
 720                         unp3->unp_addr = (struct sockaddr_un *)
 721                                 dup_sockaddr((struct sockaddr *)unp2->unp_addr);
 722
 723                 /*
 724                  * unp_peercred management:
 725                  *
 726                  * The connecter's (client's) credentials are copied
 727                  * from its process structure at the time of connect()
 728                  * (which is now).
 729                  */
 730                 cru2x(p->p_ucred, &unp3->unp_peercred);
 731                 unp3->unp_flags |= UNP_HAVEPC;
 732                 /*
 733                  * The receiver's (server's) credentials are copied
 734                  * from the unp_peercred member of socket on which the
 735                  * former called listen(); unp_listen() cached that
 736                  * process's credentials at that time so we can use
 737                  * them now.
 738                  */
 739                 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
 740                     ("unp_connect: listener without cached peercred"));
 741                 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
 742                     sizeof(unp->unp_peercred));
 743                 unp->unp_flags |= UNP_HAVEPC;
 744
 745                 so2 = so3;
 746         }
 747         error = unp_connect2(so, so2);
 748 bad:
 749         vput(vp);
 750         return (error);
 751 }
 752
 753 int
 754 unp_connect2(struct socket *so, struct socket *so2)
 755 {
 756         struct unpcb *unp = so->so_pcb;
 757         struct unpcb *unp2;
 758
 759         if (so2->so_type != so->so_type)
 760                 return (EPROTOTYPE);
 761         unp2 = so2->so_pcb;
 762         unp->unp_conn = unp2;
 763         switch (so->so_type) {
 764
 765         case SOCK_DGRAM:
 766                 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
 767                 soisconnected(so);
 768                 break;
 769
 770         case SOCK_STREAM:
 771         case SOCK_SEQPACKET:
 772                 unp2->unp_conn = unp;
 773                 soisconnected(so);
 774                 soisconnected(so2);
 775                 break;
 776
 777         default:
 778                 panic("unp_connect2");
 779         }
 780         return (0);
 781 }
 782
 783 static void
 784 unp_disconnect(struct unpcb *unp)
 785 {
 786         struct unpcb *unp2 = unp->unp_conn;
 787
 788         if (unp2 == NULL)
 789                 return;
 790
 791         unp->unp_conn = NULL;
 792
 793         switch (unp->unp_socket->so_type) {
 794         case SOCK_DGRAM:
 795                 LIST_REMOVE(unp, unp_reflink);
 796                 unp->unp_socket->so_state &= ~SS_ISCONNECTED;
 797                 break;
 798         case SOCK_STREAM:
 799         case SOCK_SEQPACKET:
 800                 soisdisconnected(unp->unp_socket);
 801                 unp2->unp_conn = NULL;
 802                 soisdisconnected(unp2->unp_socket);
 803                 break;
 804         }
 805 }
 806
 807 #ifdef notdef
 808 void
 809 unp_abort(struct unpcb *unp)
 810 {
 811
 812         unp_detach(unp);
 813 }
 814 #endif
 815
 816 static int
 817 prison_unpcb(struct thread *td, struct unpcb *unp)
 818 {
 819         struct proc *p;
 820
 821         if (td == NULL)
 822                 return (0);
 823         if ((p = td->td_proc) == NULL)
 824                 return (0);
 825         if (!p->p_ucred->cr_prison)
 826                 return (0);
 827         if (p->p_fd->fd_rdir == unp->unp_rvnode)
 828                 return (0);
 829         return (1);
 830 }
 831
 832 static int
 833 unp_pcblist(SYSCTL_HANDLER_ARGS)
 834 {
 835         int error, i, n;
 836         struct unpcb *unp, **unp_list;
 837         unp_gen_t gencnt;
 838         struct unp_head *head;
 839
 840         head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
 841
 842         KKASSERT(curproc != NULL);
 843
 844         /*
 845          * The process of preparing the PCB list is too time-consuming and
 846          * resource-intensive to repeat twice on every request.
 847          */
 848         if (req->oldptr == NULL) {
 849                 n = unp_count;
 850                 req->oldidx = (n + n/8) * sizeof(struct xunpcb);
 851                 return 0;
 852         }
 853
 854         if (req->newptr != NULL)
 855                 return EPERM;
 856
 857         /*
 858          * OK, now we're committed to doing something.
 859          */
 860         gencnt = unp_gencnt;
 861         n = unp_count;
 862
 863         unp_list = kmalloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
 864
 865         for (unp = LIST_FIRST(head), i = 0; unp && i < n;
 866              unp = LIST_NEXT(unp, unp_link)) {
 867                 if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->td, unp))
 868                         unp_list[i++] = unp;
 869         }
 870         n = i;                  /* in case we lost some during malloc */
 871
 872         error = 0;
 873         for (i = 0; i < n; i++) {
 874                 unp = unp_list[i];
 875                 if (unp->unp_gencnt <= gencnt) {
 876                         struct xunpcb xu;
 877                         xu.xu_len = sizeof xu;
 878                         xu.xu_unpp = unp;
 879                         /*
 880                          * XXX - need more locking here to protect against
 881                          * connect/disconnect races for SMP.
 882                          */
 883                         if (unp->unp_addr)
 884                                 bcopy(unp->unp_addr, &xu.xu_addr,
 885                                       unp->unp_addr->sun_len);
 886                         if (unp->unp_conn && unp->unp_conn->unp_addr)
 887                                 bcopy(unp->unp_conn->unp_addr,
 888                                       &xu.xu_caddr,
 889                                       unp->unp_conn->unp_addr->sun_len);
 890                         bcopy(unp, &xu.xu_unp, sizeof *unp);
 891                         sotoxsocket(unp->unp_socket, &xu.xu_socket);
 892                         error = SYSCTL_OUT(req, &xu, sizeof xu);
 893                 }
 894         }
 895         kfree(unp_list, M_TEMP);
 896         return error;
 897 }
 898
 899 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
 900             (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
 901             "List of active local datagram sockets");
 902 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
 903             (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
 904             "List of active local stream sockets");
 905 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, CTLFLAG_RD,
 906             (caddr_t)(long)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
 907             "List of active local seqpacket stream sockets");
 908
 909 static void
 910 unp_shutdown(struct unpcb *unp)
 911 {
 912         struct socket *so;
 913
 914         if ((unp->unp_socket->so_type == SOCK_STREAM ||
 915              unp->unp_socket->so_type == SOCK_SEQPACKET) &&
 916             unp->unp_conn != NULL && (so = unp->unp_conn->unp_socket)) {
 917                 socantrcvmore(so);
 918         }
 919 }
 920
 921 static void
 922 unp_drop(struct unpcb *unp, int err)
 923 {
 924         struct socket *so = unp->unp_socket;
 925
 926         so->so_error = err;
 927         unp_disconnect(unp);
 928 }
 929
 930 #ifdef notdef
 931 void
 932 unp_drain(void)
 933 {
 934
 935 }
 936 #endif
 937
 938 int
 939 unp_externalize(struct mbuf *rights)
 940 {
 941         struct proc *p = curproc;               /* XXX */
 942         int i;
 943         struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
 944         int *fdp;
 945         struct file **rp;
 946         struct file *fp;
 947         int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm))
 948                 / sizeof (struct file *);
 949         int f;
 950
 951         /*
 952          * if the new FD's will not fit, then we free them all
 953          */
 954         if (!fdavail(p, newfds)) {
 955                 rp = (struct file **)CMSG_DATA(cm);
 956                 for (i = 0; i < newfds; i++) {
 957                         fp = *rp;
 958                         /*
 959                          * zero the pointer before calling unp_discard,
 960                          * since it may end up in unp_gc()..
 961                          */
 962                         *rp++ = 0;
 963                         unp_discard(fp, NULL);
 964                 }
 965                 return (EMSGSIZE);
 966         }
 967
 968         /*
 969          * now change each pointer to an fd in the global table to
 970          * an integer that is the index to the local fd table entry
 971          * that we set up to point to the global one we are transferring.
 972          * If sizeof (struct file *) is bigger than or equal to sizeof int,
 973          * then do it in forward order. In that case, an integer will
 974          * always come in the same place or before its corresponding
 975          * struct file pointer.
 976          * If sizeof (struct file *) is smaller than sizeof int, then
 977          * do it in reverse order.
 978          */
 979         if (sizeof (struct file *) >= sizeof (int)) {
 980                 fdp = (int *)CMSG_DATA(cm);
 981                 rp = (struct file **)CMSG_DATA(cm);
 982                 for (i = 0; i < newfds; i++) {
 983                         if (fdalloc(p, 0, &f))
 984                                 panic("unp_externalize");
 985                         fp = *rp++;
 986                         unp_fp_externalize(p, fp, f);
 987                         *fdp++ = f;
 988                 }
 989         } else {
 990                 fdp = (int *)CMSG_DATA(cm) + newfds - 1;
 991                 rp = (struct file **)CMSG_DATA(cm) + newfds - 1;
 992                 for (i = 0; i < newfds; i++) {
 993                         if (fdalloc(p, 0, &f))
 994                                 panic("unp_externalize");
 995                         fp = *rp--;
 996                         unp_fp_externalize(p, fp, f);
 997                         *fdp-- = f;
 998                 }
 999         }
1000
1001         /*
1002          * Adjust length, in case sizeof(struct file *) and sizeof(int)
1003          * differs.
1004          */
1005         cm->cmsg_len = CMSG_LEN(newfds * sizeof(int));
1006         rights->m_len = cm->cmsg_len;
1007         return (0);
1008 }
1009
1010 static void
1011 unp_fp_externalize(struct proc *p, struct file *fp, int fd)
1012 {
1013         struct file *fx;
1014         int error;
1015
1016         if (p) {
1017                 KKASSERT(fd >= 0);
1018                 if (fp->f_flag & FREVOKED) {
1019                         kprintf("Warning: revoked fp exiting unix socket\n");
1020                         fx = NULL;
1021                         error = falloc(p, &fx, NULL);
1022                         if (error == 0)
1023                                 fsetfd(p, fx, fd);
1024                         else
1025                                 fsetfd(p, NULL, fd);
1026                         fdrop(fx);
1027                 } else {
1028                         fsetfd(p, fp, fd);
1029                 }
1030         }
1031         spin_lock_wr(&unp_spin);
1032         fp->f_msgcount--;
1033         unp_rights--;
1034         spin_unlock_wr(&unp_spin);
1035         fdrop(fp);
1036 }
1037
1038
1039 void
1040 unp_init(void)
1041 {
1042         LIST_INIT(&unp_dhead);
1043         LIST_INIT(&unp_shead);
1044         spin_init(&unp_spin);
1045 }
1046
1047 static int
1048 unp_internalize(struct mbuf *control, struct thread *td)
1049 {
1050         struct proc *p = td->td_proc;
1051         struct filedesc *fdescp;
1052         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1053         struct file **rp;
1054         struct file *fp;
1055         int i, fd, *fdp;
1056         struct cmsgcred *cmcred;
1057         int oldfds;
1058         u_int newlen;
1059
1060         KKASSERT(p);
1061         fdescp = p->p_fd;
1062         if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) ||
1063             cm->cmsg_level != SOL_SOCKET ||
1064             CMSG_ALIGN(cm->cmsg_len) != control->m_len) {
1065                 return (EINVAL);
1066         }
1067
1068         /*
1069          * Fill in credential information.
1070          */
1071         if (cm->cmsg_type == SCM_CREDS) {
1072                 cmcred = (struct cmsgcred *)CMSG_DATA(cm);
1073                 cmcred->cmcred_pid = p->p_pid;
1074                 cmcred->cmcred_uid = p->p_ucred->cr_ruid;
1075                 cmcred->cmcred_gid = p->p_ucred->cr_rgid;
1076                 cmcred->cmcred_euid = p->p_ucred->cr_uid;
1077                 cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups,
1078                                                         CMGROUP_MAX);
1079                 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1080                         cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i];
1081                 return(0);
1082         }
1083
1084         /*
1085          * cmsghdr may not be aligned, do not allow calculation(s) to
1086          * go negative.
1087          */
1088         if (cm->cmsg_len < CMSG_LEN(0))
1089                 return(EINVAL);
1090
1091         oldfds = (cm->cmsg_len - CMSG_LEN(0)) / sizeof (int);
1092
1093         /*
1094          * check that all the FDs passed in refer to legal OPEN files
1095          * If not, reject the entire operation.
1096          */
1097         fdp = (int *)CMSG_DATA(cm);
1098         for (i = 0; i < oldfds; i++) {
1099                 fd = *fdp++;
1100                 if ((unsigned)fd >= fdescp->fd_nfiles ||
1101                     fdescp->fd_files[fd].fp == NULL)
1102                         return (EBADF);
1103                 if (fdescp->fd_files[fd].fp->f_type == DTYPE_KQUEUE)
1104                         return (EOPNOTSUPP);
1105         }
1106         /*
1107          * Now replace the integer FDs with pointers to
1108          * the associated global file table entry..
1109          * Allocate a bigger buffer as necessary. But if an cluster is not
1110          * enough, return E2BIG.
1111          */
1112         newlen = CMSG_LEN(oldfds * sizeof(struct file *));
1113         if (newlen > MCLBYTES)
1114                 return (E2BIG);
1115         if (newlen - control->m_len > M_TRAILINGSPACE(control)) {
1116                 if (control->m_flags & M_EXT)
1117                         return (E2BIG);
1118                 MCLGET(control, MB_WAIT);
1119                 if (!(control->m_flags & M_EXT))
1120                         return (ENOBUFS);
1121
1122                 /* copy the data to the cluster */
1123                 memcpy(mtod(control, char *), cm, cm->cmsg_len);
1124                 cm = mtod(control, struct cmsghdr *);
1125         }
1126
1127         /*
1128          * Adjust length, in case sizeof(struct file *) and sizeof(int)
1129          * differs.
1130          */
1131         cm->cmsg_len = newlen;
1132         control->m_len = CMSG_ALIGN(newlen);
1133
1134         /*
1135          * Transform the file descriptors into struct file pointers.
1136          * If sizeof (struct file *) is bigger than or equal to sizeof int,
1137          * then do it in reverse order so that the int won't get until
1138          * we're done.
1139          * If sizeof (struct file *) is smaller than sizeof int, then
1140          * do it in forward order.
1141          */
1142         if (sizeof (struct file *) >= sizeof (int)) {
1143                 fdp = (int *)CMSG_DATA(cm) + oldfds - 1;
1144                 rp = (struct file **)CMSG_DATA(cm) + oldfds - 1;
1145                 for (i = 0; i < oldfds; i++) {
1146                         fp = fdescp->fd_files[*fdp--].fp;
1147                         *rp-- = fp;
1148                         fhold(fp);
1149                         spin_lock_wr(&unp_spin);
1150                         fp->f_msgcount++;
1151                         unp_rights++;
1152                         spin_unlock_wr(&unp_spin);
1153                 }
1154         } else {
1155                 fdp = (int *)CMSG_DATA(cm);
1156                 rp = (struct file **)CMSG_DATA(cm);
1157                 for (i = 0; i < oldfds; i++) {
1158                         fp = fdescp->fd_files[*fdp++].fp;
1159                         *rp++ = fp;
1160                         fhold(fp);
1161                         spin_lock_wr(&unp_spin);
1162                         fp->f_msgcount++;
1163                         unp_rights++;
1164                         spin_unlock_wr(&unp_spin);
1165                 }
1166         }
1167         return (0);
1168 }
1169
1170 /*
1171  * Garbage collect in-transit file descriptors that get lost due to
1172  * loops (i.e. when a socket is sent to another process over itself,
1173  * and more complex situations).
1174  *
1175  * NOT MPSAFE - TODO socket flush code and maybe closef.  Rest is MPSAFE.
1176  */
1177
1178 struct unp_gc_info {
1179         struct file **extra_ref;
1180         struct file *locked_fp;
1181         int defer;
1182         int index;
1183         int maxindex;
1184 };
1185
1186 static void
1187 unp_gc(void)
1188 {
1189         struct unp_gc_info info;
1190         static boolean_t unp_gcing;
1191         struct file **fpp;
1192         int i;
1193
1194         spin_lock_wr(&unp_spin);
1195         if (unp_gcing) {
1196                 spin_unlock_wr(&unp_spin);
1197                 return;
1198         }
1199         unp_gcing = TRUE;
1200         spin_unlock_wr(&unp_spin);
1201
1202         /*
1203          * before going through all this, set all FDs to
1204          * be NOT defered and NOT externally accessible
1205          */
1206         info.defer = 0;
1207         allfiles_scan_exclusive(unp_gc_clearmarks, NULL);
1208         do {
1209                 allfiles_scan_exclusive(unp_gc_checkmarks, &info);
1210         } while (info.defer);
1211
1212         /*
1213          * We grab an extra reference to each of the file table entries
1214          * that are not otherwise accessible and then free the rights
1215          * that are stored in messages on them.
1216          *
1217          * The bug in the orginal code is a little tricky, so I'll describe
1218          * what's wrong with it here.
1219          *
1220          * It is incorrect to simply unp_discard each entry for f_msgcount
1221          * times -- consider the case of sockets A and B that contain
1222          * references to each other.  On a last close of some other socket,
1223          * we trigger a gc since the number of outstanding rights (unp_rights)
1224          * is non-zero.  If during the sweep phase the gc code un_discards,
1225          * we end up doing a (full) closef on the descriptor.  A closef on A
1226          * results in the following chain.  Closef calls soo_close, which
1227          * calls soclose.   Soclose calls first (through the switch
1228          * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
1229          * returns because the previous instance had set unp_gcing, and
1230          * we return all the way back to soclose, which marks the socket
1231          * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
1232          * to free up the rights that are queued in messages on the socket A,
1233          * i.e., the reference on B.  The sorflush calls via the dom_dispose
1234          * switch unp_dispose, which unp_scans with unp_discard.  This second
1235          * instance of unp_discard just calls closef on B.
1236          *
1237          * Well, a similar chain occurs on B, resulting in a sorflush on B,
1238          * which results in another closef on A.  Unfortunately, A is already
1239          * being closed, and the descriptor has already been marked with
1240          * SS_NOFDREF, and soclose panics at this point.
1241          *
1242          * Here, we first take an extra reference to each inaccessible
1243          * descriptor.  Then, we call sorflush ourself, since we know
1244          * it is a Unix domain socket anyhow.  After we destroy all the
1245          * rights carried in messages, we do a last closef to get rid
1246          * of our extra reference.  This is the last close, and the
1247          * unp_detach etc will shut down the socket.
1248          *
1249          * 91/09/19, bsy@cs.cmu.edu
1250          */
1251         info.extra_ref = kmalloc(256 * sizeof(struct file *), M_FILE, M_WAITOK);
1252         info.maxindex = 256;
1253
1254         do {
1255                 /*
1256                  * Look for matches
1257                  */
1258                 info.index = 0;
1259                 allfiles_scan_exclusive(unp_gc_checkrefs, &info);
1260
1261                 /*
1262                  * For each FD on our hit list, do the following two things
1263                  */
1264                 for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp) {
1265                         struct file *tfp = *fpp;
1266                         if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL)
1267                                 sorflush((struct socket *)(tfp->f_data));
1268                 }
1269                 for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp)
1270                         closef(*fpp, NULL);
1271         } while (info.index == info.maxindex);
1272         kfree((caddr_t)info.extra_ref, M_FILE);
1273         unp_gcing = FALSE;
1274 }
1275
1276 /*
1277  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1278  */
1279 static int
1280 unp_gc_checkrefs(struct file *fp, void *data)
1281 {
1282         struct unp_gc_info *info = data;
1283
1284         if (fp->f_count == 0)
1285                 return(0);
1286         if (info->index == info->maxindex)
1287                 return(-1);
1288
1289         /*
1290          * If all refs are from msgs, and it's not marked accessible
1291          * then it must be referenced from some unreachable cycle
1292          * of (shut-down) FDs, so include it in our
1293          * list of FDs to remove
1294          */
1295         if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
1296                 info->extra_ref[info->index++] = fp;
1297                 fhold(fp);
1298         }
1299         return(0);
1300 }
1301
1302 /*
1303  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1304  */
1305 static int
1306 unp_gc_clearmarks(struct file *fp, void *data __unused)
1307 {
1308         atomic_clear_int(&fp->f_flag, FMARK | FDEFER);
1309         return(0);
1310 }
1311
1312 /*
1313  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1314  */
1315 static int
1316 unp_gc_checkmarks(struct file *fp, void *data)
1317 {
1318         struct unp_gc_info *info = data;
1319         struct socket *so;
1320
1321         /*
1322          * If the file is not open, skip it
1323          */
1324         if (fp->f_count == 0)
1325                 return(0);
1326         /*
1327          * If we already marked it as 'defer'  in a
1328          * previous pass, then try process it this time
1329          * and un-mark it
1330          */
1331         if (fp->f_flag & FDEFER) {
1332                 atomic_clear_int(&fp->f_flag, FDEFER);
1333                 --info->defer;
1334         } else {
1335                 /*
1336                  * if it's not defered, then check if it's
1337                  * already marked.. if so skip it
1338                  */
1339                 if (fp->f_flag & FMARK)
1340                         return(0);
1341                 /*
1342                  * If all references are from messages
1343                  * in transit, then skip it. it's not
1344                  * externally accessible.
1345                  */
1346                 if (fp->f_count == fp->f_msgcount)
1347                         return(0);
1348                 /*
1349                  * If it got this far then it must be
1350                  * externally accessible.
1351                  */
1352                 atomic_set_int(&fp->f_flag, FMARK);
1353         }
1354
1355         /*
1356          * either it was defered, or it is externally
1357          * accessible and not already marked so.
1358          * Now check if it is possibly one of OUR sockets.
1359          */
1360         if (fp->f_type != DTYPE_SOCKET ||
1361             (so = (struct socket *)fp->f_data) == NULL)
1362                 return(0);
1363         if (so->so_proto->pr_domain != &localdomain ||
1364             !(so->so_proto->pr_flags & PR_RIGHTS))
1365                 return(0);
1366 #ifdef notdef
1367         if (so->so_rcv.ssb_flags & SSB_LOCK) {
1368                 /*
1369                  * This is problematical; it's not clear
1370                  * we need to wait for the sockbuf to be
1371                  * unlocked (on a uniprocessor, at least),
1372                  * and it's also not clear what to do
1373                  * if sbwait returns an error due to receipt
1374                  * of a signal.  If sbwait does return
1375                  * an error, we'll go into an infinite
1376                  * loop.  Delete all of this for now.
1377                  */
1378                 sbwait(&so->so_rcv);
1379                 goto restart;
1380         }
1381 #endif
1382         /*
1383          * So, Ok, it's one of our sockets and it IS externally
1384          * accessible (or was defered). Now we look
1385          * to see if we hold any file descriptors in its
1386          * message buffers. Follow those links and mark them
1387          * as accessible too.
1388          */
1389         info->locked_fp = fp;
1390 /*      spin_lock_wr(&so->so_rcv.sb_spin); */
1391         unp_scan(so->so_rcv.ssb_mb, unp_mark, info);
1392 /*      spin_unlock_wr(&so->so_rcv.sb_spin);*/
1393         return (0);
1394 }
1395
1396 /*
1397  * Scan all unix domain sockets and replace any revoked file pointers
1398  * found with the dummy file pointer fx.  We don't worry about races
1399  * against file pointers being read out as those are handled in the
1400  * externalize code.
1401  */
1402
1403 #define REVOKE_GC_MAXFILES      32
1404
1405 struct unp_revoke_gc_info {
1406         struct file     *fx;
1407         struct file     *fary[REVOKE_GC_MAXFILES];
1408         int             fcount;
1409 };
1410
1411 void
1412 unp_revoke_gc(struct file *fx)
1413 {
1414         struct unp_revoke_gc_info info;
1415         int i;
1416
1417         info.fx = fx;
1418         do {
1419                 info.fcount = 0;
1420                 allfiles_scan_exclusive(unp_revoke_gc_check, &info);
1421                 for (i = 0; i < info.fcount; ++i)
1422                         unp_fp_externalize(NULL, info.fary[i], -1);
1423         } while (info.fcount == REVOKE_GC_MAXFILES);
1424 }
1425
1426 /*
1427  * Check for and replace revoked descriptors.
1428  *
1429  * WARNING:  This routine is not allowed to block.
1430  */
1431 static int
1432 unp_revoke_gc_check(struct file *fps, void *vinfo)
1433 {
1434         struct unp_revoke_gc_info *info = vinfo;
1435         struct file *fp;
1436         struct socket *so;
1437         struct mbuf *m0;
1438         struct mbuf *m;
1439         struct file **rp;
1440         struct cmsghdr *cm;
1441         int i;
1442         int qfds;
1443
1444         /*
1445          * Is this a unix domain socket with rights-passing abilities?
1446          */
1447         if (fps->f_type != DTYPE_SOCKET)
1448                 return (0);
1449         if ((so = (struct socket *)fps->f_data) == NULL)
1450                 return(0);
1451         if (so->so_proto->pr_domain != &localdomain)
1452                 return(0);
1453         if ((so->so_proto->pr_flags & PR_RIGHTS) == 0)
1454                 return(0);
1455
1456         /*
1457          * Scan the mbufs for control messages and replace any revoked
1458          * descriptors we find.
1459          */
1460         m0 = so->so_rcv.ssb_mb;
1461         while (m0) {
1462                 for (m = m0; m; m = m->m_next) {
1463                         if (m->m_type != MT_CONTROL)
1464                                 continue;
1465                         if (m->m_len < sizeof(*cm))
1466                                 continue;
1467                         cm = mtod(m, struct cmsghdr *);
1468                         if (cm->cmsg_level != SOL_SOCKET ||
1469                             cm->cmsg_type != SCM_RIGHTS) {
1470                                 continue;
1471                         }
1472                         qfds = (cm->cmsg_len - CMSG_LEN(0)) / sizeof(void *);
1473                         rp = (struct file **)CMSG_DATA(cm);
1474                         for (i = 0; i < qfds; i++) {
1475                                 fp = rp[i];
1476                                 if (fp->f_flag & FREVOKED) {
1477                                         kprintf("Warning: Removing revoked fp from unix domain socket queue\n");
1478                                         fhold(info->fx);
1479                                         info->fx->f_msgcount++;
1480                                         unp_rights++;
1481                                         rp[i] = info->fx;
1482                                         info->fary[info->fcount++] = fp;
1483                                 }
1484                                 if (info->fcount == REVOKE_GC_MAXFILES)
1485                                         break;
1486                         }
1487                         if (info->fcount == REVOKE_GC_MAXFILES)
1488                                 break;
1489                 }
1490                 m0 = m0->m_nextpkt;
1491                 if (info->fcount == REVOKE_GC_MAXFILES)
1492                         break;
1493         }
1494
1495         /*
1496          * Stop the scan if we filled up our array.
1497          */
1498         if (info->fcount == REVOKE_GC_MAXFILES)
1499                 return(-1);
1500         return(0);
1501 }
1502
1503 void
1504 unp_dispose(struct mbuf *m)
1505 {
1506         if (m)
1507                 unp_scan(m, unp_discard, NULL);
1508 }
1509
1510 static int
1511 unp_listen(struct unpcb *unp, struct thread *td)
1512 {
1513         struct proc *p = td->td_proc;
1514
1515         KKASSERT(p);
1516         cru2x(p->p_ucred, &unp->unp_peercred);
1517         unp->unp_flags |= UNP_HAVEPCCACHED;
1518         return (0);
1519 }
1520
1521 static void
1522 unp_scan(struct mbuf *m0, void (*op)(struct file *, void *), void *data)
1523 {
1524         struct mbuf *m;
1525         struct file **rp;
1526         struct cmsghdr *cm;
1527         int i;
1528         int qfds;
1529
1530         while (m0) {
1531                 for (m = m0; m; m = m->m_next) {
1532                         if (m->m_type == MT_CONTROL &&
1533                             m->m_len >= sizeof(*cm)) {
1534                                 cm = mtod(m, struct cmsghdr *);
1535                                 if (cm->cmsg_level != SOL_SOCKET ||
1536                                     cm->cmsg_type != SCM_RIGHTS)
1537                                         continue;
1538                                 qfds = (cm->cmsg_len - CMSG_LEN(0)) /
1539                                         sizeof(void *);
1540                                 rp = (struct file **)CMSG_DATA(cm);
1541                                 for (i = 0; i < qfds; i++)
1542                                         (*op)(*rp++, data);
1543                                 break;          /* XXX, but saves time */
1544                         }
1545                 }
1546                 m0 = m0->m_nextpkt;
1547         }
1548 }
1549
1550 static void
1551 unp_mark(struct file *fp, void *data)
1552 {
1553         struct unp_gc_info *info = data;
1554
1555         if ((fp->f_flag & FMARK) == 0) {
1556                 ++info->defer;
1557                 atomic_set_int(&fp->f_flag, FMARK | FDEFER);
1558         }
1559 }
1560
1561 static void
1562 unp_discard(struct file *fp, void *data __unused)
1563 {
1564         spin_lock_wr(&unp_spin);
1565         fp->f_msgcount--;
1566         unp_rights--;
1567         spin_unlock_wr(&unp_spin);
1568         closef(fp, NULL);
1569 }
1570