sys/kern/uipc_usrreq.c

   1 /*-
   2  * Copyright 2004-2005 Robert N. M. Watson
   3  * Copyright (c) 1982, 1986, 1989, 1991, 1993
   4  *      The Regents of the University of California.  All rights reserved.
   5  *
   6  * Redistribution and use in source and binary forms, with or without
   7  * modification, are permitted provided that the following conditions
   8  * are met:
   9  * 1. Redistributions of source code must retain the above copyright
  10  *    notice, this list of conditions and the following disclaimer.
  11  * 2. Redistributions in binary form must reproduce the above copyright
  12  *    notice, this list of conditions and the following disclaimer in the
  13  *    documentation and/or other materials provided with the distribution.
  14  * 4. Neither the name of the University nor the names of its contributors
  15  *    may be used to endorse or promote products derived from this software
  16  *    without specific prior written permission.
  17  *
  18  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  28  * SUCH DAMAGE.
  29  *
  30  *      From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
  31  */
  32
  33 #include <sys/cdefs.h>
  34 __FBSDID("$FreeBSD$");
  35
  36 #include "opt_mac.h"
  37
  38 #include <sys/param.h>
  39 #include <sys/domain.h>
  40 #include <sys/fcntl.h>
  41 #include <sys/malloc.h>         /* XXX must be before <sys/file.h> */
  42 #include <sys/file.h>
  43 #include <sys/filedesc.h>
  44 #include <sys/jail.h>
  45 #include <sys/kernel.h>
  46 #include <sys/lock.h>
  47 #include <sys/mac.h>
  48 #include <sys/mbuf.h>
  49 #include <sys/mutex.h>
  50 #include <sys/namei.h>
  51 #include <sys/proc.h>
  52 #include <sys/protosw.h>
  53 #include <sys/resourcevar.h>
  54 #include <sys/socket.h>
  55 #include <sys/socketvar.h>
  56 #include <sys/signalvar.h>
  57 #include <sys/stat.h>
  58 #include <sys/sx.h>
  59 #include <sys/sysctl.h>
  60 #include <sys/systm.h>
  61 #include <sys/un.h>
  62 #include <sys/unpcb.h>
  63 #include <sys/vnode.h>
  64
  65 #include <vm/uma.h>
  66
  67 static uma_zone_t unp_zone;
  68 static  unp_gen_t unp_gencnt;
  69 static  u_int unp_count;
  70
  71 static  struct unp_head unp_shead, unp_dhead;
  72
  73 /*
  74  * Unix communications domain.
  75  *
  76  * TODO:
  77  *      SEQPACKET, RDM
  78  *      rethink name space problems
  79  *      need a proper out-of-band
  80  *      lock pushdown
  81  */
  82 static const struct     sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
  83 static ino_t    unp_ino;                /* prototype for fake inode numbers */
  84 struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
  85
  86 /*
  87  * Currently, UNIX domain sockets are protected by a single subsystem lock,
  88  * which covers global data structures and variables, the contents of each
  89  * per-socket unpcb structure, and the so_pcb field in sockets attached to
  90  * the UNIX domain.  This provides for a moderate degree of paralellism, as
  91  * receive operations on UNIX domain sockets do not need to acquire the
  92  * subsystem lock.  Finer grained locking to permit send() without acquiring
  93  * a global lock would be a logical next step.
  94  *
  95  * The UNIX domain socket lock preceds all socket layer locks, including the
  96  * socket lock and socket buffer lock, permitting UNIX domain socket code to
  97  * call into socket support routines without releasing its locks.
  98  *
  99  * Some caution is required in areas where the UNIX domain socket code enters
 100  * VFS in order to create or find rendezvous points.  This results in
 101  * dropping of the UNIX domain socket subsystem lock, acquisition of the
 102  * Giant lock, and potential sleeping.  This increases the chances of races,
 103  * and exposes weaknesses in the socket->protocol API by offering poor
 104  * failure modes.
 105  */
 106 static struct mtx unp_mtx;
 107 #define UNP_LOCK_INIT() \
 108         mtx_init(&unp_mtx, "unp", NULL, MTX_DEF)
 109 #define UNP_LOCK()              mtx_lock(&unp_mtx)
 110 #define UNP_UNLOCK()            mtx_unlock(&unp_mtx)
 111 #define UNP_LOCK_ASSERT()       mtx_assert(&unp_mtx, MA_OWNED)
 112 #define UNP_UNLOCK_ASSERT()     mtx_assert(&unp_mtx, MA_NOTOWNED)
 113
 114 static int     unp_attach(struct socket *);
 115 static void    unp_detach(struct unpcb *);
 116 static int     unp_bind(struct unpcb *,struct sockaddr *, struct thread *);
 117 static int     unp_connect(struct socket *,struct sockaddr *, struct thread *);
 118 static int     unp_connect2(struct socket *so, struct socket *so2, int);
 119 static void    unp_disconnect(struct unpcb *);
 120 static void    unp_shutdown(struct unpcb *);
 121 static void    unp_drop(struct unpcb *, int);
 122 static void    unp_gc(void);
 123 static void    unp_scan(struct mbuf *, void (*)(struct file *));
 124 static void    unp_mark(struct file *);
 125 static void    unp_discard(struct file *);
 126 static void    unp_freerights(struct file **, int);
 127 static int     unp_internalize(struct mbuf **, struct thread *);
 128 static int     unp_listen(struct socket *, struct unpcb *, struct thread *);
 129
 130 static int
 131 uipc_abort(struct socket *so)
 132 {
 133         struct unpcb *unp;
 134
 135         UNP_LOCK();
 136         unp = sotounpcb(so);
 137         if (unp == NULL) {
 138                 UNP_UNLOCK();
 139                 return (EINVAL);
 140         }
 141         unp_drop(unp, ECONNABORTED);
 142         unp_detach(unp);
 143         UNP_UNLOCK_ASSERT();
 144         ACCEPT_LOCK();
 145         SOCK_LOCK(so);
 146         sotryfree(so);
 147         return (0);
 148 }
 149
 150 static int
 151 uipc_accept(struct socket *so, struct sockaddr **nam)
 152 {
 153         struct unpcb *unp;
 154         const struct sockaddr *sa;
 155
 156         /*
 157          * Pass back name of connected socket,
 158          * if it was bound and we are still connected
 159          * (our peer may have closed already!).
 160          */
 161         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 162         UNP_LOCK();
 163         unp = sotounpcb(so);
 164         if (unp == NULL) {
 165                 UNP_UNLOCK();
 166                 free(*nam, M_SONAME);
 167                 *nam = NULL;
 168                 return (EINVAL);
 169         }
 170         if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL)
 171                 sa = (struct sockaddr *) unp->unp_conn->unp_addr;
 172         else
 173                 sa = &sun_noname;
 174         bcopy(sa, *nam, sa->sa_len);
 175         UNP_UNLOCK();
 176         return (0);
 177 }
 178
 179 static int
 180 uipc_attach(struct socket *so, int proto, struct thread *td)
 181 {
 182         struct unpcb *unp = sotounpcb(so);
 183
 184         if (unp != NULL)
 185                 return (EISCONN);
 186         return (unp_attach(so));
 187 }
 188
 189 static int
 190 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 191 {
 192         struct unpcb *unp;
 193         int error;
 194
 195         UNP_LOCK();
 196         unp = sotounpcb(so);
 197         if (unp == NULL) {
 198                 UNP_UNLOCK();
 199                 return (EINVAL);
 200         }
 201         error = unp_bind(unp, nam, td);
 202         UNP_UNLOCK();
 203         return (error);
 204 }
 205
 206 static int
 207 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 208 {
 209         struct unpcb *unp;
 210         int error;
 211
 212         KASSERT(td == curthread, ("uipc_connect: td != curthread"));
 213
 214         UNP_LOCK();
 215         unp = sotounpcb(so);
 216         if (unp == NULL) {
 217                 UNP_UNLOCK();
 218                 return (EINVAL);
 219         }
 220         error = unp_connect(so, nam, td);
 221         UNP_UNLOCK();
 222         return (error);
 223 }
 224
 225 int
 226 uipc_connect2(struct socket *so1, struct socket *so2)
 227 {
 228         struct unpcb *unp;
 229         int error;
 230
 231         UNP_LOCK();
 232         unp = sotounpcb(so1);
 233         if (unp == NULL) {
 234                 UNP_UNLOCK();
 235                 return (EINVAL);
 236         }
 237         error = unp_connect2(so1, so2, PRU_CONNECT2);
 238         UNP_UNLOCK();
 239         return (error);
 240 }
 241
 242 /* control is EOPNOTSUPP */
 243
 244 static int
 245 uipc_detach(struct socket *so)
 246 {
 247         struct unpcb *unp;
 248
 249         UNP_LOCK();
 250         unp = sotounpcb(so);
 251         if (unp == NULL) {
 252                 UNP_UNLOCK();
 253                 return (EINVAL);
 254         }
 255         unp_detach(unp);
 256         UNP_UNLOCK_ASSERT();
 257         return (0);
 258 }
 259
 260 static int
 261 uipc_disconnect(struct socket *so)
 262 {
 263         struct unpcb *unp;
 264
 265         UNP_LOCK();
 266         unp = sotounpcb(so);
 267         if (unp == NULL) {
 268                 UNP_UNLOCK();
 269                 return (EINVAL);
 270         }
 271         unp_disconnect(unp);
 272         UNP_UNLOCK();
 273         return (0);
 274 }
 275
 276 static int
 277 uipc_listen(struct socket *so, struct thread *td)
 278 {
 279         struct unpcb *unp;
 280         int error;
 281
 282         UNP_LOCK();
 283         unp = sotounpcb(so);
 284         if (unp == NULL || unp->unp_vnode == NULL) {
 285                 UNP_UNLOCK();
 286                 return (EINVAL);
 287         }
 288         error = unp_listen(so, unp, td);
 289         UNP_UNLOCK();
 290         return (error);
 291 }
 292
 293 static int
 294 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
 295 {
 296         struct unpcb *unp;
 297         const struct sockaddr *sa;
 298
 299         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 300         UNP_LOCK();
 301         unp = sotounpcb(so);
 302         if (unp == NULL) {
 303                 UNP_UNLOCK();
 304                 free(*nam, M_SONAME);
 305                 *nam = NULL;
 306                 return (EINVAL);
 307         }
 308         if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL)
 309                 sa = (struct sockaddr *) unp->unp_conn->unp_addr;
 310         else {
 311                 /*
 312                  * XXX: It seems that this test always fails even when
 313                  * connection is established.  So, this else clause is
 314                  * added as workaround to return PF_LOCAL sockaddr.
 315                  */
 316                 sa = &sun_noname;
 317         }
 318         bcopy(sa, *nam, sa->sa_len);
 319         UNP_UNLOCK();
 320         return (0);
 321 }
 322
 323 static int
 324 uipc_rcvd(struct socket *so, int flags)
 325 {
 326         struct unpcb *unp;
 327         struct socket *so2;
 328         u_long newhiwat;
 329
 330         UNP_LOCK();
 331         unp = sotounpcb(so);
 332         if (unp == NULL) {
 333                 UNP_UNLOCK();
 334                 return (EINVAL);
 335         }
 336         switch (so->so_type) {
 337         case SOCK_DGRAM:
 338                 panic("uipc_rcvd DGRAM?");
 339                 /*NOTREACHED*/
 340
 341         case SOCK_STREAM:
 342                 if (unp->unp_conn == NULL)
 343                         break;
 344                 so2 = unp->unp_conn->unp_socket;
 345                 SOCKBUF_LOCK(&so2->so_snd);
 346                 SOCKBUF_LOCK(&so->so_rcv);
 347                 /*
 348                  * Adjust backpressure on sender
 349                  * and wakeup any waiting to write.
 350                  */
 351                 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt;
 352                 unp->unp_mbcnt = so->so_rcv.sb_mbcnt;
 353                 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc -
 354                     so->so_rcv.sb_cc;
 355                 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
 356                     newhiwat, RLIM_INFINITY);
 357                 unp->unp_cc = so->so_rcv.sb_cc;
 358                 SOCKBUF_UNLOCK(&so->so_rcv);
 359                 sowwakeup_locked(so2);
 360                 break;
 361
 362         default:
 363                 panic("uipc_rcvd unknown socktype");
 364         }
 365         UNP_UNLOCK();
 366         return (0);
 367 }
 368
 369 /* pru_rcvoob is EOPNOTSUPP */
 370
 371 static int
 372 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
 373     struct mbuf *control, struct thread *td)
 374 {
 375         int error = 0;
 376         struct unpcb *unp;
 377         struct socket *so2;
 378         u_long newhiwat;
 379
 380         unp = sotounpcb(so);
 381         if (unp == NULL) {
 382                 error = EINVAL;
 383                 goto release;
 384         }
 385         if (flags & PRUS_OOB) {
 386                 error = EOPNOTSUPP;
 387                 goto release;
 388         }
 389
 390         if (control != NULL && (error = unp_internalize(&control, td)))
 391                 goto release;
 392
 393         UNP_LOCK();
 394         unp = sotounpcb(so);
 395         if (unp == NULL) {
 396                 UNP_UNLOCK();
 397                 error = EINVAL;
 398                 goto dispose_release;
 399         }
 400
 401         switch (so->so_type) {
 402         case SOCK_DGRAM:
 403         {
 404                 const struct sockaddr *from;
 405
 406                 if (nam != NULL) {
 407                         if (unp->unp_conn != NULL) {
 408                                 error = EISCONN;
 409                                 break;
 410                         }
 411                         error = unp_connect(so, nam, td);
 412                         if (error)
 413                                 break;
 414                 } else {
 415                         if (unp->unp_conn == NULL) {
 416                                 error = ENOTCONN;
 417                                 break;
 418                         }
 419                 }
 420                 so2 = unp->unp_conn->unp_socket;
 421                 if (unp->unp_addr != NULL)
 422                         from = (struct sockaddr *)unp->unp_addr;
 423                 else
 424                         from = &sun_noname;
 425                 if (unp->unp_conn->unp_flags & UNP_WANTCRED)
 426                         control = unp_addsockcred(td, control);
 427                 SOCKBUF_LOCK(&so2->so_rcv);
 428                 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
 429                         sorwakeup_locked(so2);
 430                         m = NULL;
 431                         control = NULL;
 432                 } else {
 433                         SOCKBUF_UNLOCK(&so2->so_rcv);
 434                         error = ENOBUFS;
 435                 }
 436                 if (nam != NULL)
 437                         unp_disconnect(unp);
 438                 break;
 439         }
 440
 441         case SOCK_STREAM:
 442                 /* Connect if not connected yet. */
 443                 /*
 444                  * Note: A better implementation would complain
 445                  * if not equal to the peer's address.
 446                  */
 447                 if ((so->so_state & SS_ISCONNECTED) == 0) {
 448                         if (nam != NULL) {
 449                                 error = unp_connect(so, nam, td);
 450                                 if (error)
 451                                         break;  /* XXX */
 452                         } else {
 453                                 error = ENOTCONN;
 454                                 break;
 455                         }
 456                 }
 457
 458                 SOCKBUF_LOCK(&so->so_snd);
 459                 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
 460                         SOCKBUF_UNLOCK(&so->so_snd);
 461                         error = EPIPE;
 462                         break;
 463                 }
 464                 if (unp->unp_conn == NULL)
 465                         panic("uipc_send connected but no connection?");
 466                 so2 = unp->unp_conn->unp_socket;
 467                 SOCKBUF_LOCK(&so2->so_rcv);
 468                 if (unp->unp_conn->unp_flags & UNP_WANTCRED) {
 469                         /*
 470                          * Credentials are passed only once on
 471                          * SOCK_STREAM.
 472                          */
 473                         unp->unp_conn->unp_flags &= ~UNP_WANTCRED;
 474                         control = unp_addsockcred(td, control);
 475                 }
 476                 /*
 477                  * Send to paired receive port, and then reduce
 478                  * send buffer hiwater marks to maintain backpressure.
 479                  * Wake up readers.
 480                  */
 481                 if (control != NULL) {
 482                         if (sbappendcontrol_locked(&so2->so_rcv, m, control))
 483                                 control = NULL;
 484                 } else {
 485                         sbappend_locked(&so2->so_rcv, m);
 486                 }
 487                 so->so_snd.sb_mbmax -=
 488                         so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt;
 489                 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt;
 490                 newhiwat = so->so_snd.sb_hiwat -
 491                     (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc);
 492                 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
 493                     newhiwat, RLIM_INFINITY);
 494                 SOCKBUF_UNLOCK(&so->so_snd);
 495                 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc;
 496                 sorwakeup_locked(so2);
 497                 m = NULL;
 498                 break;
 499
 500         default:
 501                 panic("uipc_send unknown socktype");
 502         }
 503
 504         /*
 505          * SEND_EOF is equivalent to a SEND followed by
 506          * a SHUTDOWN.
 507          */
 508         if (flags & PRUS_EOF) {
 509                 socantsendmore(so);
 510                 unp_shutdown(unp);
 511         }
 512         UNP_UNLOCK();
 513
 514 dispose_release:
 515         if (control != NULL && error != 0)
 516                 unp_dispose(control);
 517
 518 release:
 519         if (control != NULL)
 520                 m_freem(control);
 521         if (m != NULL)
 522                 m_freem(m);
 523         return (error);
 524 }
 525
 526 static int
 527 uipc_sense(struct socket *so, struct stat *sb)
 528 {
 529         struct unpcb *unp;
 530         struct socket *so2;
 531
 532         UNP_LOCK();
 533         unp = sotounpcb(so);
 534         if (unp == NULL) {
 535                 UNP_UNLOCK();
 536                 return (EINVAL);
 537         }
 538         sb->st_blksize = so->so_snd.sb_hiwat;
 539         if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) {
 540                 so2 = unp->unp_conn->unp_socket;
 541                 sb->st_blksize += so2->so_rcv.sb_cc;
 542         }
 543         sb->st_dev = NODEV;
 544         if (unp->unp_ino == 0)
 545                 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
 546         sb->st_ino = unp->unp_ino;
 547         UNP_UNLOCK();
 548         return (0);
 549 }
 550
 551 static int
 552 uipc_shutdown(struct socket *so)
 553 {
 554         struct unpcb *unp;
 555
 556         UNP_LOCK();
 557         unp = sotounpcb(so);
 558         if (unp == NULL) {
 559                 UNP_UNLOCK();
 560                 return (EINVAL);
 561         }
 562         socantsendmore(so);
 563         unp_shutdown(unp);
 564         UNP_UNLOCK();
 565         return (0);
 566 }
 567
 568 static int
 569 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
 570 {
 571         struct unpcb *unp;
 572         const struct sockaddr *sa;
 573
 574         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 575         UNP_LOCK();
 576         unp = sotounpcb(so);
 577         if (unp == NULL) {
 578                 UNP_UNLOCK();
 579                 free(*nam, M_SONAME);
 580                 *nam = NULL;
 581                 return (EINVAL);
 582         }
 583         if (unp->unp_addr != NULL)
 584                 sa = (struct sockaddr *) unp->unp_addr;
 585         else
 586                 sa = &sun_noname;
 587         bcopy(sa, *nam, sa->sa_len);
 588         UNP_UNLOCK();
 589         return (0);
 590 }
 591
 592 struct pr_usrreqs uipc_usrreqs = {
 593         .pru_abort =            uipc_abort,
 594         .pru_accept =           uipc_accept,
 595         .pru_attach =           uipc_attach,
 596         .pru_bind =             uipc_bind,
 597         .pru_connect =          uipc_connect,
 598         .pru_connect2 =         uipc_connect2,
 599         .pru_detach =           uipc_detach,
 600         .pru_disconnect =       uipc_disconnect,
 601         .pru_listen =           uipc_listen,
 602         .pru_peeraddr =         uipc_peeraddr,
 603         .pru_rcvd =             uipc_rcvd,
 604         .pru_send =             uipc_send,
 605         .pru_sense =            uipc_sense,
 606         .pru_shutdown =         uipc_shutdown,
 607         .pru_sockaddr =         uipc_sockaddr,
 608         .pru_sosend =           sosend,
 609         .pru_soreceive =        soreceive,
 610         .pru_sopoll =           sopoll,
 611 };
 612
 613 int
 614 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
 615 {
 616         struct unpcb *unp;
 617         struct xucred xu;
 618         int error, optval;
 619
 620         if (sopt->sopt_level != 0)
 621                 return (EINVAL);
 622
 623         UNP_LOCK();
 624         unp = sotounpcb(so);
 625         if (unp == NULL) {
 626                 UNP_UNLOCK();
 627                 return (EINVAL);
 628         }
 629         error = 0;
 630
 631         switch (sopt->sopt_dir) {
 632         case SOPT_GET:
 633                 switch (sopt->sopt_name) {
 634                 case LOCAL_PEERCRED:
 635                         if (unp->unp_flags & UNP_HAVEPC)
 636                                 xu = unp->unp_peercred;
 637                         else {
 638                                 if (so->so_type == SOCK_STREAM)
 639                                         error = ENOTCONN;
 640                                 else
 641                                         error = EINVAL;
 642                         }
 643                         if (error == 0)
 644                                 error = sooptcopyout(sopt, &xu, sizeof(xu));
 645                         break;
 646                 case LOCAL_CREDS:
 647                         optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
 648                         error = sooptcopyout(sopt, &optval, sizeof(optval));
 649                         break;
 650                 case LOCAL_CONNWAIT:
 651                         optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
 652                         error = sooptcopyout(sopt, &optval, sizeof(optval));
 653                         break;
 654                 default:
 655                         error = EOPNOTSUPP;
 656                         break;
 657                 }
 658                 break;
 659         case SOPT_SET:
 660                 switch (sopt->sopt_name) {
 661                 case LOCAL_CREDS:
 662                 case LOCAL_CONNWAIT:
 663                         error = sooptcopyin(sopt, &optval, sizeof(optval),
 664                                             sizeof(optval));
 665                         if (error)
 666                                 break;
 667
 668 #define OPTSET(bit) \
 669         if (optval) \
 670                 unp->unp_flags |= bit; \
 671         else \
 672                 unp->unp_flags &= ~bit;
 673
 674                         switch (sopt->sopt_name) {
 675                         case LOCAL_CREDS:
 676                                 OPTSET(UNP_WANTCRED);
 677                                 break;
 678                         case LOCAL_CONNWAIT:
 679                                 OPTSET(UNP_CONNWAIT);
 680                                 break;
 681                         default:
 682                                 break;
 683                         }
 684                         break;
 685 #undef  OPTSET
 686                 default:
 687                         error = ENOPROTOOPT;
 688                         break;
 689                 }
 690                 break;
 691         default:
 692                 error = EOPNOTSUPP;
 693                 break;
 694         }
 695         UNP_UNLOCK();
 696         return (error);
 697 }
 698
 699 /*
 700  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
 701  * for stream sockets, although the total for sender and receiver is
 702  * actually only PIPSIZ.
 703  * Datagram sockets really use the sendspace as the maximum datagram size,
 704  * and don't really want to reserve the sendspace.  Their recvspace should
 705  * be large enough for at least one max-size datagram plus address.
 706  */
 707 #ifndef PIPSIZ
 708 #define PIPSIZ  8192
 709 #endif
 710 static u_long   unpst_sendspace = PIPSIZ;
 711 static u_long   unpst_recvspace = PIPSIZ;
 712 static u_long   unpdg_sendspace = 2*1024;       /* really max datagram size */
 713 static u_long   unpdg_recvspace = 4*1024;
 714
 715 static int      unp_rights;                     /* file descriptors in flight */
 716
 717 SYSCTL_DECL(_net_local_stream);
 718 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
 719            &unpst_sendspace, 0, "");
 720 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
 721            &unpst_recvspace, 0, "");
 722 SYSCTL_DECL(_net_local_dgram);
 723 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
 724            &unpdg_sendspace, 0, "");
 725 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
 726            &unpdg_recvspace, 0, "");
 727 SYSCTL_DECL(_net_local);
 728 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
 729
 730 static int
 731 unp_attach(struct socket *so)
 732 {
 733         struct unpcb *unp;
 734         int error;
 735
 736         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
 737                 switch (so->so_type) {
 738
 739                 case SOCK_STREAM:
 740                         error = soreserve(so, unpst_sendspace, unpst_recvspace);
 741                         break;
 742
 743                 case SOCK_DGRAM:
 744                         error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
 745                         break;
 746
 747                 default:
 748                         panic("unp_attach");
 749                 }
 750                 if (error)
 751                         return (error);
 752         }
 753         unp = uma_zalloc(unp_zone, M_WAITOK | M_ZERO);
 754         if (unp == NULL)
 755                 return (ENOBUFS);
 756         LIST_INIT(&unp->unp_refs);
 757         unp->unp_socket = so;
 758         so->so_pcb = unp;
 759
 760         UNP_LOCK();
 761         unp->unp_gencnt = ++unp_gencnt;
 762         unp_count++;
 763         LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
 764                          : &unp_shead, unp, unp_link);
 765         UNP_UNLOCK();
 766
 767         return (0);
 768 }
 769
 770 static void
 771 unp_detach(struct unpcb *unp)
 772 {
 773         struct vnode *vp;
 774
 775         UNP_LOCK_ASSERT();
 776
 777         LIST_REMOVE(unp, unp_link);
 778         unp->unp_gencnt = ++unp_gencnt;
 779         --unp_count;
 780         if ((vp = unp->unp_vnode) != NULL) {
 781                 /*
 782                  * XXXRW: should v_socket be frobbed only while holding
 783                  * Giant?
 784                  */
 785                 unp->unp_vnode->v_socket = NULL;
 786                 unp->unp_vnode = NULL;
 787         }
 788         if (unp->unp_conn != NULL)
 789                 unp_disconnect(unp);
 790         while (!LIST_EMPTY(&unp->unp_refs)) {
 791                 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
 792                 unp_drop(ref, ECONNRESET);
 793         }
 794         soisdisconnected(unp->unp_socket);
 795         unp->unp_socket->so_pcb = NULL;
 796         if (unp_rights) {
 797                 /*
 798                  * Normally the receive buffer is flushed later,
 799                  * in sofree, but if our receive buffer holds references
 800                  * to descriptors that are now garbage, we will dispose
 801                  * of those descriptor references after the garbage collector
 802                  * gets them (resulting in a "panic: closef: count < 0").
 803                  */
 804                 sorflush(unp->unp_socket);
 805                 unp_gc();       /* Will unlock UNP. */
 806         } else
 807                 UNP_UNLOCK();
 808         UNP_UNLOCK_ASSERT();
 809         if (unp->unp_addr != NULL)
 810                 FREE(unp->unp_addr, M_SONAME);
 811         uma_zfree(unp_zone, unp);
 812         if (vp) {
 813                 mtx_lock(&Giant);
 814                 vrele(vp);
 815                 mtx_unlock(&Giant);
 816         }
 817 }
 818
 819 static int
 820 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td)
 821 {
 822         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
 823         struct vnode *vp;
 824         struct mount *mp;
 825         struct vattr vattr;
 826         int error, namelen;
 827         struct nameidata nd;
 828         char *buf;
 829
 830         UNP_LOCK_ASSERT();
 831
 832         /*
 833          * XXXRW: This test-and-set of unp_vnode is non-atomic; the
 834          * unlocked read here is fine, but the value of unp_vnode needs
 835          * to be tested again after we do all the lookups to see if the
 836          * pcb is still unbound?
 837          */
 838         if (unp->unp_vnode != NULL)
 839                 return (EINVAL);
 840
 841         namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
 842         if (namelen <= 0)
 843                 return (EINVAL);
 844
 845         UNP_UNLOCK();
 846
 847         buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
 848         strlcpy(buf, soun->sun_path, namelen + 1);
 849
 850         mtx_lock(&Giant);
 851 restart:
 852         mtx_assert(&Giant, MA_OWNED);
 853         NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE,
 854             buf, td);
 855 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
 856         error = namei(&nd);
 857         if (error)
 858                 goto done;
 859         vp = nd.ni_vp;
 860         if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
 861                 NDFREE(&nd, NDF_ONLY_PNBUF);
 862                 if (nd.ni_dvp == vp)
 863                         vrele(nd.ni_dvp);
 864                 else
 865                         vput(nd.ni_dvp);
 866                 if (vp != NULL) {
 867                         vrele(vp);
 868                         error = EADDRINUSE;
 869                         goto done;
 870                 }
 871                 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
 872                 if (error)
 873                         goto done;
 874                 goto restart;
 875         }
 876         VATTR_NULL(&vattr);
 877         vattr.va_type = VSOCK;
 878         vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
 879 #ifdef MAC
 880         error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
 881             &vattr);
 882 #endif
 883         if (error == 0) {
 884                 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE);
 885                 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
 886         }
 887         NDFREE(&nd, NDF_ONLY_PNBUF);
 888         vput(nd.ni_dvp);
 889         if (error) {
 890                 vn_finished_write(mp);
 891                 goto done;
 892         }
 893         vp = nd.ni_vp;
 894         ASSERT_VOP_LOCKED(vp, "unp_bind");
 895         soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
 896         UNP_LOCK();
 897         vp->v_socket = unp->unp_socket;
 898         unp->unp_vnode = vp;
 899         unp->unp_addr = soun;
 900         UNP_UNLOCK();
 901         VOP_UNLOCK(vp, 0, td);
 902         vn_finished_write(mp);
 903 done:
 904         mtx_unlock(&Giant);
 905         free(buf, M_TEMP);
 906         UNP_LOCK();
 907         return (error);
 908 }
 909
 910 static int
 911 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 912 {
 913         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
 914         struct vnode *vp;
 915         struct socket *so2, *so3;
 916         struct unpcb *unp, *unp2, *unp3;
 917         int error, len;
 918         struct nameidata nd;
 919         char buf[SOCK_MAXADDRLEN];
 920         struct sockaddr *sa;
 921
 922         UNP_LOCK_ASSERT();
 923         unp = sotounpcb(so);
 924
 925         len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
 926         if (len <= 0)
 927                 return (EINVAL);
 928         strlcpy(buf, soun->sun_path, len + 1);
 929         UNP_UNLOCK();
 930         sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 931         mtx_lock(&Giant);
 932         NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td);
 933         error = namei(&nd);
 934         if (error)
 935                 vp = NULL;
 936         else
 937                 vp = nd.ni_vp;
 938         ASSERT_VOP_LOCKED(vp, "unp_connect");
 939         NDFREE(&nd, NDF_ONLY_PNBUF);
 940         if (error)
 941                 goto bad;
 942
 943         if (vp->v_type != VSOCK) {
 944                 error = ENOTSOCK;
 945                 goto bad;
 946         }
 947         error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
 948         if (error)
 949                 goto bad;
 950         mtx_unlock(&Giant);
 951         UNP_LOCK();
 952         unp = sotounpcb(so);
 953         if (unp == NULL) {
 954                 error = EINVAL;
 955                 goto bad2;
 956         }
 957         so2 = vp->v_socket;
 958         if (so2 == NULL) {
 959                 error = ECONNREFUSED;
 960                 goto bad2;
 961         }
 962         if (so->so_type != so2->so_type) {
 963                 error = EPROTOTYPE;
 964                 goto bad2;
 965         }
 966         if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
 967                 if (so2->so_options & SO_ACCEPTCONN) {
 968                         /*
 969                          * NB: drop locks here so unp_attach is entered
 970                          *     w/o locks; this avoids a recursive lock
 971                          *     of the head and holding sleep locks across
 972                          *     a (potentially) blocking malloc.
 973                          */
 974                         UNP_UNLOCK();
 975                         so3 = sonewconn(so2, 0);
 976                         UNP_LOCK();
 977                 } else
 978                         so3 = NULL;
 979                 if (so3 == NULL) {
 980                         error = ECONNREFUSED;
 981                         goto bad2;
 982                 }
 983                 unp = sotounpcb(so);
 984                 unp2 = sotounpcb(so2);
 985                 unp3 = sotounpcb(so3);
 986                 if (unp2->unp_addr != NULL) {
 987                         bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
 988                         unp3->unp_addr = (struct sockaddr_un *) sa;
 989                         sa = NULL;
 990                 }
 991                 /*
 992                  * unp_peercred management:
 993                  *
 994                  * The connecter's (client's) credentials are copied
 995                  * from its process structure at the time of connect()
 996                  * (which is now).
 997                  */
 998                 cru2x(td->td_ucred, &unp3->unp_peercred);
 999                 unp3->unp_flags |= UNP_HAVEPC;
1000                 /*
1001                  * The receiver's (server's) credentials are copied
1002                  * from the unp_peercred member of socket on which the
1003                  * former called listen(); unp_listen() cached that
1004                  * process's credentials at that time so we can use
1005                  * them now.
1006                  */
1007                 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1008                     ("unp_connect: listener without cached peercred"));
1009                 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1010                     sizeof(unp->unp_peercred));
1011                 unp->unp_flags |= UNP_HAVEPC;
1012 #ifdef MAC
1013                 SOCK_LOCK(so);
1014                 mac_set_socket_peer_from_socket(so, so3);
1015                 mac_set_socket_peer_from_socket(so3, so);
1016                 SOCK_UNLOCK(so);
1017 #endif
1018
1019                 so2 = so3;
1020         }
1021         error = unp_connect2(so, so2, PRU_CONNECT);
1022 bad2:
1023         UNP_UNLOCK();
1024         mtx_lock(&Giant);
1025 bad:
1026         mtx_assert(&Giant, MA_OWNED);
1027         if (vp != NULL)
1028                 vput(vp);
1029         mtx_unlock(&Giant);
1030         free(sa, M_SONAME);
1031         UNP_LOCK();
1032         return (error);
1033 }
1034
1035 static int
1036 unp_connect2(struct socket *so, struct socket *so2, int req)
1037 {
1038         struct unpcb *unp = sotounpcb(so);
1039         struct unpcb *unp2;
1040
1041         UNP_LOCK_ASSERT();
1042
1043         if (so2->so_type != so->so_type)
1044                 return (EPROTOTYPE);
1045         unp2 = sotounpcb(so2);
1046         unp->unp_conn = unp2;
1047         switch (so->so_type) {
1048
1049         case SOCK_DGRAM:
1050                 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1051                 soisconnected(so);
1052                 break;
1053
1054         case SOCK_STREAM:
1055                 unp2->unp_conn = unp;
1056                 if (req == PRU_CONNECT &&
1057                     ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1058                         soisconnecting(so);
1059                 else
1060                         soisconnected(so);
1061                 soisconnected(so2);
1062                 break;
1063
1064         default:
1065                 panic("unp_connect2");
1066         }
1067         return (0);
1068 }
1069
1070 static void
1071 unp_disconnect(struct unpcb *unp)
1072 {
1073         struct unpcb *unp2 = unp->unp_conn;
1074         struct socket *so;
1075
1076         UNP_LOCK_ASSERT();
1077
1078         if (unp2 == NULL)
1079                 return;
1080         unp->unp_conn = NULL;
1081         switch (unp->unp_socket->so_type) {
1082
1083         case SOCK_DGRAM:
1084                 LIST_REMOVE(unp, unp_reflink);
1085                 so = unp->unp_socket;
1086                 SOCK_LOCK(so);
1087                 so->so_state &= ~SS_ISCONNECTED;
1088                 SOCK_UNLOCK(so);
1089                 break;
1090
1091         case SOCK_STREAM:
1092                 soisdisconnected(unp->unp_socket);
1093                 unp2->unp_conn = NULL;
1094                 soisdisconnected(unp2->unp_socket);
1095                 break;
1096         }
1097 }
1098
1099 #ifdef notdef
1100 void
1101 unp_abort(struct unpcb *unp)
1102 {
1103
1104         unp_detach(unp);
1105         UNP_UNLOCK_ASSERT();
1106 }
1107 #endif
1108
1109 /*
1110  * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed
1111  * by the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers
1112  * are safe to reference.  It first scans the list of struct unpcb's to
1113  * generate a pointer list, then it rescans its list one entry at a time to
1114  * externalize and copyout.  It checks the generation number to see if a
1115  * struct unpcb has been reused, and will skip it if so.
1116  */
1117 static int
1118 unp_pcblist(SYSCTL_HANDLER_ARGS)
1119 {
1120         int error, i, n;
1121         struct unpcb *unp, **unp_list;
1122         unp_gen_t gencnt;
1123         struct xunpgen *xug;
1124         struct unp_head *head;
1125         struct xunpcb *xu;
1126
1127         head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
1128
1129         /*
1130          * The process of preparing the PCB list is too time-consuming and
1131          * resource-intensive to repeat twice on every request.
1132          */
1133         if (req->oldptr == NULL) {
1134                 n = unp_count;
1135                 req->oldidx = 2 * (sizeof *xug)
1136                         + (n + n/8) * sizeof(struct xunpcb);
1137                 return (0);
1138         }
1139
1140         if (req->newptr != NULL)
1141                 return (EPERM);
1142
1143         /*
1144          * OK, now we're committed to doing something.
1145          */
1146         xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1147         UNP_LOCK();
1148         gencnt = unp_gencnt;
1149         n = unp_count;
1150         UNP_UNLOCK();
1151
1152         xug->xug_len = sizeof *xug;
1153         xug->xug_count = n;
1154         xug->xug_gen = gencnt;
1155         xug->xug_sogen = so_gencnt;
1156         error = SYSCTL_OUT(req, xug, sizeof *xug);
1157         if (error) {
1158                 free(xug, M_TEMP);
1159                 return (error);
1160         }
1161
1162         unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1163
1164         UNP_LOCK();
1165         for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1166              unp = LIST_NEXT(unp, unp_link)) {
1167                 if (unp->unp_gencnt <= gencnt) {
1168                         if (cr_cansee(req->td->td_ucred,
1169                             unp->unp_socket->so_cred))
1170                                 continue;
1171                         unp_list[i++] = unp;
1172                 }
1173         }
1174         UNP_UNLOCK();
1175         n = i;                  /* in case we lost some during malloc */
1176
1177         error = 0;
1178         xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1179         for (i = 0; i < n; i++) {
1180                 unp = unp_list[i];
1181                 if (unp->unp_gencnt <= gencnt) {
1182                         xu->xu_len = sizeof *xu;
1183                         xu->xu_unpp = unp;
1184                         /*
1185                          * XXX - need more locking here to protect against
1186                          * connect/disconnect races for SMP.
1187                          */
1188                         if (unp->unp_addr != NULL)
1189                                 bcopy(unp->unp_addr, &xu->xu_addr,
1190                                       unp->unp_addr->sun_len);
1191                         if (unp->unp_conn != NULL &&
1192                             unp->unp_conn->unp_addr != NULL)
1193                                 bcopy(unp->unp_conn->unp_addr,
1194                                       &xu->xu_caddr,
1195                                       unp->unp_conn->unp_addr->sun_len);
1196                         bcopy(unp, &xu->xu_unp, sizeof *unp);
1197                         sotoxsocket(unp->unp_socket, &xu->xu_socket);
1198                         error = SYSCTL_OUT(req, xu, sizeof *xu);
1199                 }
1200         }
1201         free(xu, M_TEMP);
1202         if (!error) {
1203                 /*
1204                  * Give the user an updated idea of our state.
1205                  * If the generation differs from what we told
1206                  * her before, she knows that something happened
1207                  * while we were processing this request, and it
1208                  * might be necessary to retry.
1209                  */
1210                 xug->xug_gen = unp_gencnt;
1211                 xug->xug_sogen = so_gencnt;
1212                 xug->xug_count = unp_count;
1213                 error = SYSCTL_OUT(req, xug, sizeof *xug);
1214         }
1215         free(unp_list, M_TEMP);
1216         free(xug, M_TEMP);
1217         return (error);
1218 }
1219
1220 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1221             (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1222             "List of active local datagram sockets");
1223 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1224             (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1225             "List of active local stream sockets");
1226
1227 static void
1228 unp_shutdown(struct unpcb *unp)
1229 {
1230         struct socket *so;
1231
1232         UNP_LOCK_ASSERT();
1233
1234         if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
1235             (so = unp->unp_conn->unp_socket))
1236                 socantrcvmore(so);
1237 }
1238
1239 static void
1240 unp_drop(struct unpcb *unp, int errno)
1241 {
1242         struct socket *so = unp->unp_socket;
1243
1244         UNP_LOCK_ASSERT();
1245
1246         so->so_error = errno;
1247         unp_disconnect(unp);
1248 }
1249
1250 #ifdef notdef
1251 void
1252 unp_drain(void)
1253 {
1254
1255 }
1256 #endif
1257
1258 static void
1259 unp_freerights(struct file **rp, int fdcount)
1260 {
1261         int i;
1262         struct file *fp;
1263
1264         for (i = 0; i < fdcount; i++) {
1265                 fp = *rp;
1266                 /*
1267                  * zero the pointer before calling
1268                  * unp_discard since it may end up
1269                  * in unp_gc()..
1270                  */
1271                 *rp++ = 0;
1272                 unp_discard(fp);
1273         }
1274 }
1275
1276 int
1277 unp_externalize(struct mbuf *control, struct mbuf **controlp)
1278 {
1279         struct thread *td = curthread;          /* XXX */
1280         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1281         int i;
1282         int *fdp;
1283         struct file **rp;
1284         struct file *fp;
1285         void *data;
1286         socklen_t clen = control->m_len, datalen;
1287         int error, newfds;
1288         int f;
1289         u_int newlen;
1290
1291         UNP_UNLOCK_ASSERT();
1292
1293         error = 0;
1294         if (controlp != NULL) /* controlp == NULL => free control messages */
1295                 *controlp = NULL;
1296
1297         while (cm != NULL) {
1298                 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1299                         error = EINVAL;
1300                         break;
1301                 }
1302
1303                 data = CMSG_DATA(cm);
1304                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1305
1306                 if (cm->cmsg_level == SOL_SOCKET
1307                     && cm->cmsg_type == SCM_RIGHTS) {
1308                         newfds = datalen / sizeof(struct file *);
1309                         rp = data;
1310
1311                         /* If we're not outputting the descriptors free them. */
1312                         if (error || controlp == NULL) {
1313                                 unp_freerights(rp, newfds);
1314                                 goto next;
1315                         }
1316                         FILEDESC_LOCK(td->td_proc->p_fd);
1317                         /* if the new FD's will not fit free them.  */
1318                         if (!fdavail(td, newfds)) {
1319                                 FILEDESC_UNLOCK(td->td_proc->p_fd);
1320                                 error = EMSGSIZE;
1321                                 unp_freerights(rp, newfds);
1322                                 goto next;
1323                         }
1324                         /*
1325                          * now change each pointer to an fd in the global
1326                          * table to an integer that is the index to the
1327                          * local fd table entry that we set up to point
1328                          * to the global one we are transferring.
1329                          */
1330                         newlen = newfds * sizeof(int);
1331                         *controlp = sbcreatecontrol(NULL, newlen,
1332                             SCM_RIGHTS, SOL_SOCKET);
1333                         if (*controlp == NULL) {
1334                                 FILEDESC_UNLOCK(td->td_proc->p_fd);
1335                                 error = E2BIG;
1336                                 unp_freerights(rp, newfds);
1337                                 goto next;
1338                         }
1339
1340                         fdp = (int *)
1341                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1342                         for (i = 0; i < newfds; i++) {
1343                                 if (fdalloc(td, 0, &f))
1344                                         panic("unp_externalize fdalloc failed");
1345                                 fp = *rp++;
1346                                 td->td_proc->p_fd->fd_ofiles[f] = fp;
1347                                 FILE_LOCK(fp);
1348                                 fp->f_msgcount--;
1349                                 FILE_UNLOCK(fp);
1350                                 unp_rights--;
1351                                 *fdp++ = f;
1352                         }
1353                         FILEDESC_UNLOCK(td->td_proc->p_fd);
1354                 } else { /* We can just copy anything else across */
1355                         if (error || controlp == NULL)
1356                                 goto next;
1357                         *controlp = sbcreatecontrol(NULL, datalen,
1358                             cm->cmsg_type, cm->cmsg_level);
1359                         if (*controlp == NULL) {
1360                                 error = ENOBUFS;
1361                                 goto next;
1362                         }
1363                         bcopy(data,
1364                             CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1365                             datalen);
1366                 }
1367
1368                 controlp = &(*controlp)->m_next;
1369
1370 next:
1371                 if (CMSG_SPACE(datalen) < clen) {
1372                         clen -= CMSG_SPACE(datalen);
1373                         cm = (struct cmsghdr *)
1374                             ((caddr_t)cm + CMSG_SPACE(datalen));
1375                 } else {
1376                         clen = 0;
1377                         cm = NULL;
1378                 }
1379         }
1380
1381         m_freem(control);
1382
1383         return (error);
1384 }
1385
1386 void
1387 unp_init(void)
1388 {
1389         unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1390             NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1391         if (unp_zone == NULL)
1392                 panic("unp_init");
1393         uma_zone_set_max(unp_zone, nmbclusters);
1394         LIST_INIT(&unp_dhead);
1395         LIST_INIT(&unp_shead);
1396
1397         UNP_LOCK_INIT();
1398 }
1399
1400 static int
1401 unp_internalize(struct mbuf **controlp, struct thread *td)
1402 {
1403         struct mbuf *control = *controlp;
1404         struct proc *p = td->td_proc;
1405         struct filedesc *fdescp = p->p_fd;
1406         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1407         struct cmsgcred *cmcred;
1408         struct file **rp;
1409         struct file *fp;
1410         struct timeval *tv;
1411         int i, fd, *fdp;
1412         void *data;
1413         socklen_t clen = control->m_len, datalen;
1414         int error, oldfds;
1415         u_int newlen;
1416
1417         UNP_UNLOCK_ASSERT();
1418
1419         error = 0;
1420         *controlp = NULL;
1421
1422         while (cm != NULL) {
1423                 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1424                     || cm->cmsg_len > clen) {
1425                         error = EINVAL;
1426                         goto out;
1427                 }
1428
1429                 data = CMSG_DATA(cm);
1430                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1431
1432                 switch (cm->cmsg_type) {
1433                 /*
1434                  * Fill in credential information.
1435                  */
1436                 case SCM_CREDS:
1437                         *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1438                             SCM_CREDS, SOL_SOCKET);
1439                         if (*controlp == NULL) {
1440                                 error = ENOBUFS;
1441                                 goto out;
1442                         }
1443
1444                         cmcred = (struct cmsgcred *)
1445                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1446                         cmcred->cmcred_pid = p->p_pid;
1447                         cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1448                         cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1449                         cmcred->cmcred_euid = td->td_ucred->cr_uid;
1450                         cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1451                                                         CMGROUP_MAX);
1452                         for (i = 0; i < cmcred->cmcred_ngroups; i++)
1453                                 cmcred->cmcred_groups[i] =
1454                                     td->td_ucred->cr_groups[i];
1455                         break;
1456
1457                 case SCM_RIGHTS:
1458                         oldfds = datalen / sizeof (int);
1459                         /*
1460                          * check that all the FDs passed in refer to legal files
1461                          * If not, reject the entire operation.
1462                          */
1463                         fdp = data;
1464                         FILEDESC_LOCK(fdescp);
1465                         for (i = 0; i < oldfds; i++) {
1466                                 fd = *fdp++;
1467                                 if ((unsigned)fd >= fdescp->fd_nfiles ||
1468                                     fdescp->fd_ofiles[fd] == NULL) {
1469                                         FILEDESC_UNLOCK(fdescp);
1470                                         error = EBADF;
1471                                         goto out;
1472                                 }
1473                                 fp = fdescp->fd_ofiles[fd];
1474                                 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1475                                         FILEDESC_UNLOCK(fdescp);
1476                                         error = EOPNOTSUPP;
1477                                         goto out;
1478                                 }
1479
1480                         }
1481                         /*
1482                          * Now replace the integer FDs with pointers to
1483                          * the associated global file table entry..
1484                          */
1485                         newlen = oldfds * sizeof(struct file *);
1486                         *controlp = sbcreatecontrol(NULL, newlen,
1487                             SCM_RIGHTS, SOL_SOCKET);
1488                         if (*controlp == NULL) {
1489                                 FILEDESC_UNLOCK(fdescp);
1490                                 error = E2BIG;
1491                                 goto out;
1492                         }
1493
1494                         fdp = data;
1495                         rp = (struct file **)
1496                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1497                         for (i = 0; i < oldfds; i++) {
1498                                 fp = fdescp->fd_ofiles[*fdp++];
1499                                 *rp++ = fp;
1500                                 FILE_LOCK(fp);
1501                                 fp->f_count++;
1502                                 fp->f_msgcount++;
1503                                 FILE_UNLOCK(fp);
1504                                 unp_rights++;
1505                         }
1506                         FILEDESC_UNLOCK(fdescp);
1507                         break;
1508
1509                 case SCM_TIMESTAMP:
1510                         *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1511                             SCM_TIMESTAMP, SOL_SOCKET);
1512                         if (*controlp == NULL) {
1513                                 error = ENOBUFS;
1514                                 goto out;
1515                         }
1516                         tv = (struct timeval *)
1517                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1518                         microtime(tv);
1519                         break;
1520
1521                 default:
1522                         error = EINVAL;
1523                         goto out;
1524                 }
1525
1526                 controlp = &(*controlp)->m_next;
1527
1528                 if (CMSG_SPACE(datalen) < clen) {
1529                         clen -= CMSG_SPACE(datalen);
1530                         cm = (struct cmsghdr *)
1531                             ((caddr_t)cm + CMSG_SPACE(datalen));
1532                 } else {
1533                         clen = 0;
1534                         cm = NULL;
1535                 }
1536         }
1537
1538 out:
1539         m_freem(control);
1540
1541         return (error);
1542 }
1543
1544 struct mbuf *
1545 unp_addsockcred(struct thread *td, struct mbuf *control)
1546 {
1547         struct mbuf *m, *n;
1548         struct sockcred *sc;
1549         int ngroups;
1550         int i;
1551
1552         ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
1553
1554         m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
1555         if (m == NULL)
1556                 return (control);
1557         m->m_next = NULL;
1558
1559         sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
1560         sc->sc_uid = td->td_ucred->cr_ruid;
1561         sc->sc_euid = td->td_ucred->cr_uid;
1562         sc->sc_gid = td->td_ucred->cr_rgid;
1563         sc->sc_egid = td->td_ucred->cr_gid;
1564         sc->sc_ngroups = ngroups;
1565         for (i = 0; i < sc->sc_ngroups; i++)
1566                 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
1567
1568         /*
1569          * If a control message already exists, append us to the end.
1570          */
1571         if (control != NULL) {
1572                 for (n = control; n->m_next != NULL; n = n->m_next)
1573                         ;
1574                 n->m_next = m;
1575         } else
1576                 control = m;
1577
1578         return (control);
1579 }
1580
1581 /*
1582  * unp_defer is thread-local during garbage collection, and does not require
1583  * explicit synchronization.  unp_gcing prevents other threads from entering
1584  * garbage collection, and perhaps should be an sx lock instead.
1585  */
1586 static int      unp_defer, unp_gcing;
1587
1588 static void
1589 unp_gc(void)
1590 {
1591         struct file *fp, *nextfp;
1592         struct socket *so;
1593         struct file **extra_ref, **fpp;
1594         int nunref, i;
1595         int nfiles_snap;
1596         int nfiles_slack = 20;
1597
1598         UNP_LOCK_ASSERT();
1599
1600         if (unp_gcing) {
1601                 UNP_UNLOCK();
1602                 return;
1603         }
1604         unp_gcing = 1;
1605         unp_defer = 0;
1606         UNP_UNLOCK();
1607         /*
1608          * before going through all this, set all FDs to
1609          * be NOT defered and NOT externally accessible
1610          */
1611         sx_slock(&filelist_lock);
1612         LIST_FOREACH(fp, &filehead, f_list)
1613                 fp->f_gcflag &= ~(FMARK|FDEFER);
1614         do {
1615                 LIST_FOREACH(fp, &filehead, f_list) {
1616                         FILE_LOCK(fp);
1617                         /*
1618                          * If the file is not open, skip it
1619                          */
1620                         if (fp->f_count == 0) {
1621                                 FILE_UNLOCK(fp);
1622                                 continue;
1623                         }
1624                         /*
1625                          * If we already marked it as 'defer'  in a
1626                          * previous pass, then try process it this time
1627                          * and un-mark it
1628                          */
1629                         if (fp->f_gcflag & FDEFER) {
1630                                 fp->f_gcflag &= ~FDEFER;
1631                                 unp_defer--;
1632                         } else {
1633                                 /*
1634                                  * if it's not defered, then check if it's
1635                                  * already marked.. if so skip it
1636                                  */
1637                                 if (fp->f_gcflag & FMARK) {
1638                                         FILE_UNLOCK(fp);
1639                                         continue;
1640                                 }
1641                                 /*
1642                                  * If all references are from messages
1643                                  * in transit, then skip it. it's not
1644                                  * externally accessible.
1645                                  */
1646                                 if (fp->f_count == fp->f_msgcount) {
1647                                         FILE_UNLOCK(fp);
1648                                         continue;
1649                                 }
1650                                 /*
1651                                  * If it got this far then it must be
1652                                  * externally accessible.
1653                                  */
1654                                 fp->f_gcflag |= FMARK;
1655                         }
1656                         /*
1657                          * either it was defered, or it is externally
1658                          * accessible and not already marked so.
1659                          * Now check if it is possibly one of OUR sockets.
1660                          */
1661                         if (fp->f_type != DTYPE_SOCKET ||
1662                             (so = fp->f_data) == NULL) {
1663                                 FILE_UNLOCK(fp);
1664                                 continue;
1665                         }
1666                         FILE_UNLOCK(fp);
1667                         if (so->so_proto->pr_domain != &localdomain ||
1668                             (so->so_proto->pr_flags&PR_RIGHTS) == 0)
1669                                 continue;
1670 #ifdef notdef
1671                         if (so->so_rcv.sb_flags & SB_LOCK) {
1672                                 /*
1673                                  * This is problematical; it's not clear
1674                                  * we need to wait for the sockbuf to be
1675                                  * unlocked (on a uniprocessor, at least),
1676                                  * and it's also not clear what to do
1677                                  * if sbwait returns an error due to receipt
1678                                  * of a signal.  If sbwait does return
1679                                  * an error, we'll go into an infinite
1680                                  * loop.  Delete all of this for now.
1681                                  */
1682                                 (void) sbwait(&so->so_rcv);
1683                                 goto restart;
1684                         }
1685 #endif
1686                         /*
1687                          * So, Ok, it's one of our sockets and it IS externally
1688                          * accessible (or was defered). Now we look
1689                          * to see if we hold any file descriptors in its
1690                          * message buffers. Follow those links and mark them
1691                          * as accessible too.
1692                          */
1693                         SOCKBUF_LOCK(&so->so_rcv);
1694                         unp_scan(so->so_rcv.sb_mb, unp_mark);
1695                         SOCKBUF_UNLOCK(&so->so_rcv);
1696                 }
1697         } while (unp_defer);
1698         sx_sunlock(&filelist_lock);
1699         /*
1700          * We grab an extra reference to each of the file table entries
1701          * that are not otherwise accessible and then free the rights
1702          * that are stored in messages on them.
1703          *
1704          * The bug in the orginal code is a little tricky, so I'll describe
1705          * what's wrong with it here.
1706          *
1707          * It is incorrect to simply unp_discard each entry for f_msgcount
1708          * times -- consider the case of sockets A and B that contain
1709          * references to each other.  On a last close of some other socket,
1710          * we trigger a gc since the number of outstanding rights (unp_rights)
1711          * is non-zero.  If during the sweep phase the gc code un_discards,
1712          * we end up doing a (full) closef on the descriptor.  A closef on A
1713          * results in the following chain.  Closef calls soo_close, which
1714          * calls soclose.   Soclose calls first (through the switch
1715          * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
1716          * returns because the previous instance had set unp_gcing, and
1717          * we return all the way back to soclose, which marks the socket
1718          * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
1719          * to free up the rights that are queued in messages on the socket A,
1720          * i.e., the reference on B.  The sorflush calls via the dom_dispose
1721          * switch unp_dispose, which unp_scans with unp_discard.  This second
1722          * instance of unp_discard just calls closef on B.
1723          *
1724          * Well, a similar chain occurs on B, resulting in a sorflush on B,
1725          * which results in another closef on A.  Unfortunately, A is already
1726          * being closed, and the descriptor has already been marked with
1727          * SS_NOFDREF, and soclose panics at this point.
1728          *
1729          * Here, we first take an extra reference to each inaccessible
1730          * descriptor.  Then, we call sorflush ourself, since we know
1731          * it is a Unix domain socket anyhow.  After we destroy all the
1732          * rights carried in messages, we do a last closef to get rid
1733          * of our extra reference.  This is the last close, and the
1734          * unp_detach etc will shut down the socket.
1735          *
1736          * 91/09/19, bsy@cs.cmu.edu
1737          */
1738 again:
1739         nfiles_snap = openfiles + nfiles_slack; /* some slack */
1740         extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP,
1741             M_WAITOK);
1742         sx_slock(&filelist_lock);
1743         if (nfiles_snap < openfiles) {
1744                 sx_sunlock(&filelist_lock);
1745                 free(extra_ref, M_TEMP);
1746                 nfiles_slack += 20;
1747                 goto again;
1748         }
1749         for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref;
1750             fp != NULL; fp = nextfp) {
1751                 nextfp = LIST_NEXT(fp, f_list);
1752                 FILE_LOCK(fp);
1753                 /*
1754                  * If it's not open, skip it
1755                  */
1756                 if (fp->f_count == 0) {
1757                         FILE_UNLOCK(fp);
1758                         continue;
1759                 }
1760                 /*
1761                  * If all refs are from msgs, and it's not marked accessible
1762                  * then it must be referenced from some unreachable cycle
1763                  * of (shut-down) FDs, so include it in our
1764                  * list of FDs to remove
1765                  */
1766                 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) {
1767                         *fpp++ = fp;
1768                         nunref++;
1769                         fp->f_count++;
1770                 }
1771                 FILE_UNLOCK(fp);
1772         }
1773         sx_sunlock(&filelist_lock);
1774         /*
1775          * for each FD on our hit list, do the following two things
1776          */
1777         for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1778                 struct file *tfp = *fpp;
1779                 FILE_LOCK(tfp);
1780                 if (tfp->f_type == DTYPE_SOCKET &&
1781                     tfp->f_data != NULL) {
1782                         FILE_UNLOCK(tfp);
1783                         sorflush(tfp->f_data);
1784                 } else {
1785                         FILE_UNLOCK(tfp);
1786                 }
1787         }
1788         for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
1789                 closef(*fpp, (struct thread *) NULL);
1790         free(extra_ref, M_TEMP);
1791         unp_gcing = 0;
1792
1793         UNP_UNLOCK_ASSERT();
1794 }
1795
1796 void
1797 unp_dispose(struct mbuf *m)
1798 {
1799
1800         if (m)
1801                 unp_scan(m, unp_discard);
1802 }
1803
1804 static int
1805 unp_listen(struct socket *so, struct unpcb *unp, struct thread *td)
1806 {
1807         int error;
1808
1809         UNP_LOCK_ASSERT();
1810
1811         SOCK_LOCK(so);
1812         error = solisten_proto_check(so);
1813         if (error == 0) {
1814                 cru2x(td->td_ucred, &unp->unp_peercred);
1815                 unp->unp_flags |= UNP_HAVEPCCACHED;
1816                 solisten_proto(so);
1817         }
1818         SOCK_UNLOCK(so);
1819         return (error);
1820 }
1821
1822 static void
1823 unp_scan(struct mbuf *m0, void (*op)(struct file *))
1824 {
1825         struct mbuf *m;
1826         struct file **rp;
1827         struct cmsghdr *cm;
1828         void *data;
1829         int i;
1830         socklen_t clen, datalen;
1831         int qfds;
1832
1833         while (m0 != NULL) {
1834                 for (m = m0; m; m = m->m_next) {
1835                         if (m->m_type != MT_CONTROL)
1836                                 continue;
1837
1838                         cm = mtod(m, struct cmsghdr *);
1839                         clen = m->m_len;
1840
1841                         while (cm != NULL) {
1842                                 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
1843                                         break;
1844
1845                                 data = CMSG_DATA(cm);
1846                                 datalen = (caddr_t)cm + cm->cmsg_len
1847                                     - (caddr_t)data;
1848
1849                                 if (cm->cmsg_level == SOL_SOCKET &&
1850                                     cm->cmsg_type == SCM_RIGHTS) {
1851                                         qfds = datalen / sizeof (struct file *);
1852                                         rp = data;
1853                                         for (i = 0; i < qfds; i++)
1854                                                 (*op)(*rp++);
1855                                 }
1856
1857                                 if (CMSG_SPACE(datalen) < clen) {
1858                                         clen -= CMSG_SPACE(datalen);
1859                                         cm = (struct cmsghdr *)
1860                                             ((caddr_t)cm + CMSG_SPACE(datalen));
1861                                 } else {
1862                                         clen = 0;
1863                                         cm = NULL;
1864                                 }
1865                         }
1866                 }
1867                 m0 = m0->m_act;
1868         }
1869 }
1870
1871 static void
1872 unp_mark(struct file *fp)
1873 {
1874         if (fp->f_gcflag & FMARK)
1875                 return;
1876         unp_defer++;
1877         fp->f_gcflag |= (FMARK|FDEFER);
1878 }
1879
1880 static void
1881 unp_discard(struct file *fp)
1882 {
1883         FILE_LOCK(fp);
1884         fp->f_msgcount--;
1885         unp_rights--;
1886         FILE_UNLOCK(fp);
1887         (void) closef(fp, (struct thread *)NULL);
1888 }