sys/kern/uipc_usrreq.c

   1 /*-
   2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
   3  *      The Regents of the University of California.
   4  * Copyright (c) 2004-2007 Robert N. M. Watson
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer.
  12  * 2. Redistributions in binary form must reproduce the above copyright
  13  *    notice, this list of conditions and the following disclaimer in the
  14  *    documentation and/or other materials provided with the distribution.
  15  * 4. Neither the name of the University nor the names of its contributors
  16  *    may be used to endorse or promote products derived from this software
  17  *    without specific prior written permission.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  29  * SUCH DAMAGE.
  30  *
  31  *      From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
  32  */
  33
  34 /*
  35  * UNIX Domain (Local) Sockets
  36  *
  37  * This is an implementation of UNIX (local) domain sockets.  Each socket has
  38  * an associated struct unpcb (UNIX protocol control block).  Stream sockets
  39  * may be connected to 0 or 1 other socket.  Datagram sockets may be
  40  * connected to 0, 1, or many other sockets.  Sockets may be created and
  41  * connected in pairs (socketpair(2)), or bound/connected to using the file
  42  * system name space.  For most purposes, only the receive socket buffer is
  43  * used, as sending on one socket delivers directly to the receive socket
  44  * buffer of a second socket.
  45  *
  46  * The implementation is substantially complicated by the fact that
  47  * "ancillary data", such as file descriptors or credentials, may be passed
  48  * across UNIX domain sockets.  The potential for passing UNIX domain sockets
  49  * over other UNIX domain sockets requires the implementation of a simple
  50  * garbage collector to find and tear down cycles of disconnected sockets.
  51  *
  52  * TODO:
  53  *      SEQPACKET, RDM
  54  *      rethink name space problems
  55  *      need a proper out-of-band
  56  *      lock pushdown
  57  */
  58
  59 #include <sys/cdefs.h>
  60 __FBSDID("$FreeBSD$");
  61
  62 #include "opt_ddb.h"
  63 #include "opt_mac.h"
  64
  65 #include <sys/param.h>
  66 #include <sys/domain.h>
  67 #include <sys/fcntl.h>
  68 #include <sys/malloc.h>         /* XXX must be before <sys/file.h> */
  69 #include <sys/eventhandler.h>
  70 #include <sys/file.h>
  71 #include <sys/filedesc.h>
  72 #include <sys/jail.h>
  73 #include <sys/kernel.h>
  74 #include <sys/lock.h>
  75 #include <sys/mbuf.h>
  76 #include <sys/mount.h>
  77 #include <sys/mutex.h>
  78 #include <sys/namei.h>
  79 #include <sys/proc.h>
  80 #include <sys/protosw.h>
  81 #include <sys/resourcevar.h>
  82 #include <sys/rwlock.h>
  83 #include <sys/socket.h>
  84 #include <sys/socketvar.h>
  85 #include <sys/signalvar.h>
  86 #include <sys/stat.h>
  87 #include <sys/sx.h>
  88 #include <sys/sysctl.h>
  89 #include <sys/systm.h>
  90 #include <sys/taskqueue.h>
  91 #include <sys/un.h>
  92 #include <sys/unpcb.h>
  93 #include <sys/vnode.h>
  94
  95 #ifdef DDB
  96 #include <ddb/ddb.h>
  97 #endif
  98
  99 #include <security/mac/mac_framework.h>
 100
 101 #include <vm/uma.h>
 102
 103 static uma_zone_t       unp_zone;
 104 static unp_gen_t        unp_gencnt;
 105 static u_int            unp_count;      /* Count of local sockets. */
 106 static ino_t            unp_ino;        /* Prototype for fake inode numbers. */
 107 static int              unp_rights;     /* File descriptors in flight. */
 108 static struct unp_head  unp_shead;      /* List of local stream sockets. */
 109 static struct unp_head  unp_dhead;      /* List of local datagram sockets. */
 110
 111 static const struct sockaddr    sun_noname = { sizeof(sun_noname), AF_LOCAL };
 112
 113 /*
 114  * Garbage collection of cyclic file descriptor/socket references occurs
 115  * asynchronously in a taskqueue context in order to avoid recursion and
 116  * reentrance in the UNIX domain socket, file descriptor, and socket layer
 117  * code.  See unp_gc() for a full description.
 118  */
 119 static struct task      unp_gc_task;
 120
 121 /*
 122  * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
 123  * stream sockets, although the total for sender and receiver is actually
 124  * only PIPSIZ.
 125  *
 126  * Datagram sockets really use the sendspace as the maximum datagram size,
 127  * and don't really want to reserve the sendspace.  Their recvspace should be
 128  * large enough for at least one max-size datagram plus address.
 129  */
 130 #ifndef PIPSIZ
 131 #define PIPSIZ  8192
 132 #endif
 133 static u_long   unpst_sendspace = PIPSIZ;
 134 static u_long   unpst_recvspace = PIPSIZ;
 135 static u_long   unpdg_sendspace = 2*1024;       /* really max datagram size */
 136 static u_long   unpdg_recvspace = 4*1024;
 137
 138 SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
 139 SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, "SOCK_STREAM");
 140 SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
 141
 142 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
 143            &unpst_sendspace, 0, "");
 144 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
 145            &unpst_recvspace, 0, "");
 146 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
 147            &unpdg_sendspace, 0, "");
 148 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
 149            &unpdg_recvspace, 0, "");
 150 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
 151
 152 /*-
 153  * Locking and synchronization:
 154  *
 155  * The global UNIX domain socket rwlock (unp_global_rwlock) protects all
 156  * global variables, including the linked lists tracking the set of allocated
 157  * UNIX domain sockets.  The global rwlock also serves to prevent deadlock
 158  * when more than one PCB lock is acquired at a time (i.e., during
 159  * connect()).  Finally, the global rwlock protects uncounted references from
 160  * vnodes to sockets bound to those vnodes: to safely dereference the
 161  * v_socket pointer, the global rwlock must be held while a full reference is
 162  * acquired.
 163  *
 164  * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
 165  * allocated in pru_attach() and freed in pru_detach().  The validity of that
 166  * pointer is an invariant, so no lock is required to dereference the so_pcb
 167  * pointer if a valid socket reference is held by the caller.  In practice,
 168  * this is always true during operations performed on a socket.  Each unpcb
 169  * has a back-pointer to its socket, unp_socket, which will be stable under
 170  * the same circumstances.
 171  *
 172  * This pointer may only be safely dereferenced as long as a valid reference
 173  * to the unpcb is held.  Typically, this reference will be from the socket,
 174  * or from another unpcb when the referring unpcb's lock is held (in order
 175  * that the reference not be invalidated during use).  For example, to follow
 176  * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
 177  * as unp_socket remains valid as long as the reference to unp_conn is valid.
 178  *
 179  * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx.  Individual
 180  * atomic reads without the lock may be performed "lockless", but more
 181  * complex reads and read-modify-writes require the mutex to be held.  No
 182  * lock order is defined between unpcb locks -- multiple unpcb locks may be
 183  * acquired at the same time only when holding the global UNIX domain socket
 184  * rwlock exclusively, which prevents deadlocks.
 185  *
 186  * Blocking with UNIX domain sockets is a tricky issue: unlike most network
 187  * protocols, bind() is a non-atomic operation, and connect() requires
 188  * potential sleeping in the protocol, due to potentially waiting on local or
 189  * distributed file systems.  We try to separate "lookup" operations, which
 190  * may sleep, and the IPC operations themselves, which typically can occur
 191  * with relative atomicity as locks can be held over the entire operation.
 192  *
 193  * Another tricky issue is simultaneous multi-threaded or multi-process
 194  * access to a single UNIX domain socket.  These are handled by the flags
 195  * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
 196  * binding, both of which involve dropping UNIX domain socket locks in order
 197  * to perform namei() and other file system operations.
 198  */
 199 static struct rwlock    unp_global_rwlock;
 200
 201 #define UNP_GLOBAL_LOCK_INIT()          rw_init(&unp_global_rwlock,     \
 202                                             "unp_global_rwlock")
 203
 204 #define UNP_GLOBAL_LOCK_ASSERT()        rw_assert(&unp_global_rwlock,   \
 205                                             RA_LOCKED)
 206 #define UNP_GLOBAL_UNLOCK_ASSERT()      rw_assert(&unp_global_rwlock,   \
 207                                             RA_UNLOCKED)
 208
 209 #define UNP_GLOBAL_WLOCK()              rw_wlock(&unp_global_rwlock)
 210 #define UNP_GLOBAL_WUNLOCK()            rw_wunlock(&unp_global_rwlock)
 211 #define UNP_GLOBAL_WLOCK_ASSERT()       rw_assert(&unp_global_rwlock,   \
 212                                             RA_WLOCKED)
 213 #define UNP_GLOBAL_WOWNED()             rw_wowned(&unp_global_rwlock)
 214
 215 #define UNP_GLOBAL_RLOCK()              rw_rlock(&unp_global_rwlock)
 216 #define UNP_GLOBAL_RUNLOCK()            rw_runlock(&unp_global_rwlock)
 217 #define UNP_GLOBAL_RLOCK_ASSERT()       rw_assert(&unp_global_rwlock,   \
 218                                             RA_RLOCKED)
 219
 220 #define UNP_PCB_LOCK_INIT(unp)          mtx_init(&(unp)->unp_mtx,       \
 221                                             "unp_mtx", "unp_mtx",       \
 222                                             MTX_DUPOK|MTX_DEF|MTX_RECURSE)
 223 #define UNP_PCB_LOCK_DESTROY(unp)       mtx_destroy(&(unp)->unp_mtx)
 224 #define UNP_PCB_LOCK(unp)               mtx_lock(&(unp)->unp_mtx)
 225 #define UNP_PCB_UNLOCK(unp)             mtx_unlock(&(unp)->unp_mtx)
 226 #define UNP_PCB_LOCK_ASSERT(unp)        mtx_assert(&(unp)->unp_mtx, MA_OWNED)
 227
 228 static int      unp_connect(struct socket *, struct sockaddr *,
 229                     struct thread *);
 230 static int      unp_connect2(struct socket *so, struct socket *so2, int);
 231 static void     unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
 232 static void     unp_shutdown(struct unpcb *);
 233 static void     unp_drop(struct unpcb *, int);
 234 static void     unp_gc(__unused void *, int);
 235 static void     unp_scan(struct mbuf *, void (*)(struct file *));
 236 static void     unp_mark(struct file *);
 237 static void     unp_discard(struct file *);
 238 static void     unp_freerights(struct file **, int);
 239 static int      unp_internalize(struct mbuf **, struct thread *);
 240 static struct mbuf      *unp_addsockcred(struct thread *, struct mbuf *);
 241
 242 /*
 243  * Definitions of protocols supported in the LOCAL domain.
 244  */
 245 static struct domain localdomain;
 246 static struct protosw localsw[] = {
 247 {
 248         .pr_type =              SOCK_STREAM,
 249         .pr_domain =            &localdomain,
 250         .pr_flags =             PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
 251         .pr_ctloutput =         &uipc_ctloutput,
 252         .pr_usrreqs =           &uipc_usrreqs
 253 },
 254 {
 255         .pr_type =              SOCK_DGRAM,
 256         .pr_domain =            &localdomain,
 257         .pr_flags =             PR_ATOMIC|PR_ADDR|PR_RIGHTS,
 258         .pr_usrreqs =           &uipc_usrreqs
 259 },
 260 };
 261
 262 static struct domain localdomain = {
 263         .dom_family =           AF_LOCAL,
 264         .dom_name =             "local",
 265         .dom_init =             unp_init,
 266         .dom_externalize =      unp_externalize,
 267         .dom_dispose =          unp_dispose,
 268         .dom_protosw =          localsw,
 269         .dom_protoswNPROTOSW =  &localsw[sizeof(localsw)/sizeof(localsw[0])]
 270 };
 271 DOMAIN_SET(local);
 272
 273 static void
 274 uipc_abort(struct socket *so)
 275 {
 276         struct unpcb *unp, *unp2;
 277
 278         unp = sotounpcb(so);
 279         KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
 280
 281         UNP_GLOBAL_WLOCK();
 282         UNP_PCB_LOCK(unp);
 283         unp2 = unp->unp_conn;
 284         if (unp2 != NULL) {
 285                 UNP_PCB_LOCK(unp2);
 286                 unp_drop(unp2, ECONNABORTED);
 287                 UNP_PCB_UNLOCK(unp2);
 288         }
 289         UNP_PCB_UNLOCK(unp);
 290         UNP_GLOBAL_WUNLOCK();
 291 }
 292
 293 static int
 294 uipc_accept(struct socket *so, struct sockaddr **nam)
 295 {
 296         struct unpcb *unp, *unp2;
 297         const struct sockaddr *sa;
 298
 299         /*
 300          * Pass back name of connected socket, if it was bound and we are
 301          * still connected (our peer may have closed already!).
 302          */
 303         unp = sotounpcb(so);
 304         KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
 305
 306         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 307         UNP_GLOBAL_RLOCK();
 308         unp2 = unp->unp_conn;
 309         if (unp2 != NULL && unp2->unp_addr != NULL) {
 310                 UNP_PCB_LOCK(unp2);
 311                 sa = (struct sockaddr *) unp2->unp_addr;
 312                 bcopy(sa, *nam, sa->sa_len);
 313                 UNP_PCB_UNLOCK(unp2);
 314         } else {
 315                 sa = &sun_noname;
 316                 bcopy(sa, *nam, sa->sa_len);
 317         }
 318         UNP_GLOBAL_RUNLOCK();
 319         return (0);
 320 }
 321
 322 static int
 323 uipc_attach(struct socket *so, int proto, struct thread *td)
 324 {
 325         u_long sendspace, recvspace;
 326         struct unpcb *unp;
 327         int error, locked;
 328
 329         KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
 330         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
 331                 switch (so->so_type) {
 332                 case SOCK_STREAM:
 333                         sendspace = unpst_sendspace;
 334                         recvspace = unpst_recvspace;
 335                         break;
 336
 337                 case SOCK_DGRAM:
 338                         sendspace = unpdg_sendspace;
 339                         recvspace = unpdg_recvspace;
 340                         break;
 341
 342                 default:
 343                         panic("uipc_attach");
 344                 }
 345                 error = soreserve(so, sendspace, recvspace);
 346                 if (error)
 347                         return (error);
 348         }
 349         unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
 350         if (unp == NULL)
 351                 return (ENOBUFS);
 352         LIST_INIT(&unp->unp_refs);
 353         UNP_PCB_LOCK_INIT(unp);
 354         unp->unp_socket = so;
 355         so->so_pcb = unp;
 356         unp->unp_refcount = 1;
 357
 358         /*
 359          * uipc_attach() may be called indirectly from within the UNIX domain
 360          * socket code via sonewconn() in unp_connect().  Since rwlocks can
 361          * not be recursed, we do the closest thing.
 362          */
 363         locked = 0;
 364         if (!UNP_GLOBAL_WOWNED()) {
 365                 UNP_GLOBAL_WLOCK();
 366                 locked = 1;
 367         }
 368         unp->unp_gencnt = ++unp_gencnt;
 369         unp_count++;
 370         LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead : &unp_shead,
 371             unp, unp_link);
 372         if (locked)
 373                 UNP_GLOBAL_WUNLOCK();
 374
 375         return (0);
 376 }
 377
 378 static int
 379 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 380 {
 381         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
 382         struct vattr vattr;
 383         int error, namelen, vfslocked;
 384         struct nameidata nd;
 385         struct unpcb *unp;
 386         struct vnode *vp;
 387         struct mount *mp;
 388         char *buf;
 389
 390         unp = sotounpcb(so);
 391         KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
 392
 393         namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
 394         if (namelen <= 0)
 395                 return (EINVAL);
 396
 397         /*
 398          * We don't allow simultaneous bind() calls on a single UNIX domain
 399          * socket, so flag in-progress operations, and return an error if an
 400          * operation is already in progress.
 401          *
 402          * Historically, we have not allowed a socket to be rebound, so this
 403          * also returns an error.  Not allowing re-binding simplifies the
 404          * implementation and avoids a great many possible failure modes.
 405          */
 406         UNP_PCB_LOCK(unp);
 407         if (unp->unp_vnode != NULL) {
 408                 UNP_PCB_UNLOCK(unp);
 409                 return (EINVAL);
 410         }
 411         if (unp->unp_flags & UNP_BINDING) {
 412                 UNP_PCB_UNLOCK(unp);
 413                 return (EALREADY);
 414         }
 415         unp->unp_flags |= UNP_BINDING;
 416         UNP_PCB_UNLOCK(unp);
 417
 418         buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
 419         strlcpy(buf, soun->sun_path, namelen + 1);
 420
 421 restart:
 422         vfslocked = 0;
 423         NDINIT(&nd, CREATE, MPSAFE | NOFOLLOW | LOCKPARENT | SAVENAME,
 424             UIO_SYSSPACE, buf, td);
 425 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
 426         error = namei(&nd);
 427         if (error)
 428                 goto error;
 429         vp = nd.ni_vp;
 430         vfslocked = NDHASGIANT(&nd);
 431         if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
 432                 NDFREE(&nd, NDF_ONLY_PNBUF);
 433                 if (nd.ni_dvp == vp)
 434                         vrele(nd.ni_dvp);
 435                 else
 436                         vput(nd.ni_dvp);
 437                 if (vp != NULL) {
 438                         vrele(vp);
 439                         error = EADDRINUSE;
 440                         goto error;
 441                 }
 442                 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
 443                 if (error)
 444                         goto error;
 445                 VFS_UNLOCK_GIANT(vfslocked);
 446                 goto restart;
 447         }
 448         VATTR_NULL(&vattr);
 449         vattr.va_type = VSOCK;
 450         vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
 451 #ifdef MAC
 452         error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
 453             &vattr);
 454 #endif
 455         if (error == 0) {
 456                 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE);
 457                 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
 458         }
 459         NDFREE(&nd, NDF_ONLY_PNBUF);
 460         vput(nd.ni_dvp);
 461         if (error) {
 462                 vn_finished_write(mp);
 463                 goto error;
 464         }
 465         vp = nd.ni_vp;
 466         ASSERT_VOP_ELOCKED(vp, "uipc_bind");
 467         soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
 468
 469         UNP_GLOBAL_WLOCK();
 470         UNP_PCB_LOCK(unp);
 471         vp->v_socket = unp->unp_socket;
 472         unp->unp_vnode = vp;
 473         unp->unp_addr = soun;
 474         unp->unp_flags &= ~UNP_BINDING;
 475         UNP_PCB_UNLOCK(unp);
 476         UNP_GLOBAL_WUNLOCK();
 477         VOP_UNLOCK(vp, 0, td);
 478         vn_finished_write(mp);
 479         VFS_UNLOCK_GIANT(vfslocked);
 480         free(buf, M_TEMP);
 481         return (0);
 482
 483 error:
 484         VFS_UNLOCK_GIANT(vfslocked);
 485         UNP_PCB_LOCK(unp);
 486         unp->unp_flags &= ~UNP_BINDING;
 487         UNP_PCB_UNLOCK(unp);
 488         free(buf, M_TEMP);
 489         return (error);
 490 }
 491
 492 static int
 493 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 494 {
 495         int error;
 496
 497         KASSERT(td == curthread, ("uipc_connect: td != curthread"));
 498         UNP_GLOBAL_WLOCK();
 499         error = unp_connect(so, nam, td);
 500         UNP_GLOBAL_WUNLOCK();
 501         return (error);
 502 }
 503
 504 static void
 505 uipc_close(struct socket *so)
 506 {
 507         struct unpcb *unp, *unp2;
 508
 509         unp = sotounpcb(so);
 510         KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
 511
 512         UNP_GLOBAL_WLOCK();
 513         UNP_PCB_LOCK(unp);
 514         unp2 = unp->unp_conn;
 515         if (unp2 != NULL) {
 516                 UNP_PCB_LOCK(unp2);
 517                 unp_disconnect(unp, unp2);
 518                 UNP_PCB_UNLOCK(unp2);
 519         }
 520         UNP_PCB_UNLOCK(unp);
 521         UNP_GLOBAL_WUNLOCK();
 522 }
 523
 524 int
 525 uipc_connect2(struct socket *so1, struct socket *so2)
 526 {
 527         struct unpcb *unp, *unp2;
 528         int error;
 529
 530         UNP_GLOBAL_WLOCK();
 531         unp = so1->so_pcb;
 532         KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
 533         UNP_PCB_LOCK(unp);
 534         unp2 = so2->so_pcb;
 535         KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
 536         UNP_PCB_LOCK(unp2);
 537         error = unp_connect2(so1, so2, PRU_CONNECT2);
 538         UNP_PCB_UNLOCK(unp2);
 539         UNP_PCB_UNLOCK(unp);
 540         UNP_GLOBAL_WUNLOCK();
 541         return (error);
 542 }
 543
 544 /* control is EOPNOTSUPP */
 545
 546 static void
 547 uipc_detach(struct socket *so)
 548 {
 549         struct unpcb *unp, *unp2;
 550         struct sockaddr_un *saved_unp_addr;
 551         struct vnode *vp;
 552         int freeunp, local_unp_rights;
 553
 554         unp = sotounpcb(so);
 555         KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
 556
 557         UNP_GLOBAL_WLOCK();
 558         UNP_PCB_LOCK(unp);
 559
 560         LIST_REMOVE(unp, unp_link);
 561         unp->unp_gencnt = ++unp_gencnt;
 562         --unp_count;
 563
 564         /*
 565          * XXXRW: Should assert vp->v_socket == so.
 566          */
 567         if ((vp = unp->unp_vnode) != NULL) {
 568                 unp->unp_vnode->v_socket = NULL;
 569                 unp->unp_vnode = NULL;
 570         }
 571         unp2 = unp->unp_conn;
 572         if (unp2 != NULL) {
 573                 UNP_PCB_LOCK(unp2);
 574                 unp_disconnect(unp, unp2);
 575                 UNP_PCB_UNLOCK(unp2);
 576         }
 577
 578         /*
 579          * We hold the global lock, so it's OK to acquire multiple pcb locks
 580          * at a time.
 581          */
 582         while (!LIST_EMPTY(&unp->unp_refs)) {
 583                 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
 584
 585                 UNP_PCB_LOCK(ref);
 586                 unp_drop(ref, ECONNRESET);
 587                 UNP_PCB_UNLOCK(ref);
 588         }
 589         UNP_GLOBAL_WUNLOCK();
 590         unp->unp_socket->so_pcb = NULL;
 591         local_unp_rights = unp_rights;
 592         saved_unp_addr = unp->unp_addr;
 593         unp->unp_addr = NULL;
 594         unp->unp_refcount--;
 595         freeunp = (unp->unp_refcount == 0);
 596         if (saved_unp_addr != NULL)
 597                 FREE(saved_unp_addr, M_SONAME);
 598         if (freeunp) {
 599                 UNP_PCB_LOCK_DESTROY(unp);
 600                 uma_zfree(unp_zone, unp);
 601         } else
 602                 UNP_PCB_UNLOCK(unp);
 603         if (vp) {
 604                 int vfslocked;
 605
 606                 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 607                 vrele(vp);
 608                 VFS_UNLOCK_GIANT(vfslocked);
 609         }
 610         if (local_unp_rights)
 611                 taskqueue_enqueue(taskqueue_thread, &unp_gc_task);
 612 }
 613
 614 static int
 615 uipc_disconnect(struct socket *so)
 616 {
 617         struct unpcb *unp, *unp2;
 618
 619         unp = sotounpcb(so);
 620         KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
 621
 622         UNP_GLOBAL_WLOCK();
 623         UNP_PCB_LOCK(unp);
 624         unp2 = unp->unp_conn;
 625         if (unp2 != NULL) {
 626                 UNP_PCB_LOCK(unp2);
 627                 unp_disconnect(unp, unp2);
 628                 UNP_PCB_UNLOCK(unp2);
 629         }
 630         UNP_PCB_UNLOCK(unp);
 631         UNP_GLOBAL_WUNLOCK();
 632         return (0);
 633 }
 634
 635 static int
 636 uipc_listen(struct socket *so, int backlog, struct thread *td)
 637 {
 638         struct unpcb *unp;
 639         int error;
 640
 641         unp = sotounpcb(so);
 642         KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
 643
 644         UNP_PCB_LOCK(unp);
 645         if (unp->unp_vnode == NULL) {
 646                 UNP_PCB_UNLOCK(unp);
 647                 return (EINVAL);
 648         }
 649
 650         SOCK_LOCK(so);
 651         error = solisten_proto_check(so);
 652         if (error == 0) {
 653                 cru2x(td->td_ucred, &unp->unp_peercred);
 654                 unp->unp_flags |= UNP_HAVEPCCACHED;
 655                 solisten_proto(so, backlog);
 656         }
 657         SOCK_UNLOCK(so);
 658         UNP_PCB_UNLOCK(unp);
 659         return (error);
 660 }
 661
 662 static int
 663 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
 664 {
 665         struct unpcb *unp, *unp2;
 666         const struct sockaddr *sa;
 667
 668         unp = sotounpcb(so);
 669         KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
 670
 671         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 672         UNP_PCB_LOCK(unp);
 673         /*
 674          * XXX: It seems that this test always fails even when connection is
 675          * established.  So, this else clause is added as workaround to
 676          * return PF_LOCAL sockaddr.
 677          */
 678         unp2 = unp->unp_conn;
 679         if (unp2 != NULL) {
 680                 UNP_PCB_LOCK(unp2);
 681                 if (unp2->unp_addr != NULL)
 682                         sa = (struct sockaddr *) unp->unp_conn->unp_addr;
 683                 else
 684                         sa = &sun_noname;
 685                 bcopy(sa, *nam, sa->sa_len);
 686                 UNP_PCB_UNLOCK(unp2);
 687         } else {
 688                 sa = &sun_noname;
 689                 bcopy(sa, *nam, sa->sa_len);
 690         }
 691         UNP_PCB_UNLOCK(unp);
 692         return (0);
 693 }
 694
 695 static int
 696 uipc_rcvd(struct socket *so, int flags)
 697 {
 698         struct unpcb *unp, *unp2;
 699         struct socket *so2;
 700         u_int mbcnt, sbcc;
 701         u_long newhiwat;
 702
 703         unp = sotounpcb(so);
 704         KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
 705
 706         if (so->so_type == SOCK_DGRAM)
 707                 panic("uipc_rcvd DGRAM?");
 708
 709         if (so->so_type != SOCK_STREAM)
 710                 panic("uipc_rcvd unknown socktype");
 711
 712         /*
 713          * Adjust backpressure on sender and wakeup any waiting to write.
 714          *
 715          * The unp lock is acquired to maintain the validity of the unp_conn
 716          * pointer; no lock on unp2 is required as unp2->unp_socket will be
 717          * static as long as we don't permit unp2 to disconnect from unp,
 718          * which is prevented by the lock on unp.  We cache values from
 719          * so_rcv to avoid holding the so_rcv lock over the entire
 720          * transaction on the remote so_snd.
 721          */
 722         SOCKBUF_LOCK(&so->so_rcv);
 723         mbcnt = so->so_rcv.sb_mbcnt;
 724         sbcc = so->so_rcv.sb_cc;
 725         SOCKBUF_UNLOCK(&so->so_rcv);
 726         UNP_PCB_LOCK(unp);
 727         unp2 = unp->unp_conn;
 728         if (unp2 == NULL) {
 729                 UNP_PCB_UNLOCK(unp);
 730                 return (0);
 731         }
 732         so2 = unp2->unp_socket;
 733         SOCKBUF_LOCK(&so2->so_snd);
 734         so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt;
 735         newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc;
 736         (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
 737             newhiwat, RLIM_INFINITY);
 738         sowwakeup_locked(so2);
 739         unp->unp_mbcnt = mbcnt;
 740         unp->unp_cc = sbcc;
 741         UNP_PCB_UNLOCK(unp);
 742         return (0);
 743 }
 744
 745 /* pru_rcvoob is EOPNOTSUPP */
 746
 747 static int
 748 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
 749     struct mbuf *control, struct thread *td)
 750 {
 751         struct unpcb *unp, *unp2;
 752         struct socket *so2;
 753         u_int mbcnt, sbcc;
 754         u_long newhiwat;
 755         int error = 0;
 756
 757         unp = sotounpcb(so);
 758         KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
 759
 760         if (flags & PRUS_OOB) {
 761                 error = EOPNOTSUPP;
 762                 goto release;
 763         }
 764
 765         if (control != NULL && (error = unp_internalize(&control, td)))
 766                 goto release;
 767
 768         if ((nam != NULL) || (flags & PRUS_EOF))
 769                 UNP_GLOBAL_WLOCK();
 770         else
 771                 UNP_GLOBAL_RLOCK();
 772
 773         switch (so->so_type) {
 774         case SOCK_DGRAM:
 775         {
 776                 const struct sockaddr *from;
 777
 778                 unp2 = unp->unp_conn;
 779                 if (nam != NULL) {
 780                         UNP_GLOBAL_WLOCK_ASSERT();
 781                         if (unp2 != NULL) {
 782                                 error = EISCONN;
 783                                 break;
 784                         }
 785                         error = unp_connect(so, nam, td);
 786                         if (error)
 787                                 break;
 788                         unp2 = unp->unp_conn;
 789                 }
 790                 /*
 791                  * Because connect() and send() are non-atomic in a sendto()
 792                  * with a target address, it's possible that the socket will
 793                  * have disconnected before the send() can run.  In that case
 794                  * return the slightly counter-intuitive but otherwise
 795                  * correct error that the socket is not connected.
 796                  */
 797                 if (unp2 == NULL) {
 798                         error = ENOTCONN;
 799                         break;
 800                 }
 801                 /* Lockless read. */
 802                 if (unp2->unp_flags & UNP_WANTCRED)
 803                         control = unp_addsockcred(td, control);
 804                 UNP_PCB_LOCK(unp);
 805                 if (unp->unp_addr != NULL)
 806                         from = (struct sockaddr *)unp->unp_addr;
 807                 else
 808                         from = &sun_noname;
 809                 so2 = unp2->unp_socket;
 810                 SOCKBUF_LOCK(&so2->so_rcv);
 811                 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
 812                         sorwakeup_locked(so2);
 813                         m = NULL;
 814                         control = NULL;
 815                 } else {
 816                         SOCKBUF_UNLOCK(&so2->so_rcv);
 817                         error = ENOBUFS;
 818                 }
 819                 if (nam != NULL) {
 820                         UNP_GLOBAL_WLOCK_ASSERT();
 821                         UNP_PCB_LOCK(unp2);
 822                         unp_disconnect(unp, unp2);
 823                         UNP_PCB_UNLOCK(unp2);
 824                 }
 825                 UNP_PCB_UNLOCK(unp);
 826                 break;
 827         }
 828
 829         case SOCK_STREAM:
 830                 /*
 831                  * Connect if not connected yet.
 832                  *
 833                  * Note: A better implementation would complain if not equal
 834                  * to the peer's address.
 835                  */
 836                 if ((so->so_state & SS_ISCONNECTED) == 0) {
 837                         if (nam != NULL) {
 838                                 UNP_GLOBAL_WLOCK_ASSERT();
 839                                 error = unp_connect(so, nam, td);
 840                                 if (error)
 841                                         break;  /* XXX */
 842                         } else {
 843                                 error = ENOTCONN;
 844                                 break;
 845                         }
 846                 }
 847
 848                 /* Lockless read. */
 849                 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
 850                         error = EPIPE;
 851                         break;
 852                 }
 853                 /*
 854                  * Because connect() and send() are non-atomic in a sendto()
 855                  * with a target address, it's possible that the socket will
 856                  * have disconnected before the send() can run.  In that case
 857                  * return the slightly counter-intuitive but otherwise
 858                  * correct error that the socket is not connected.
 859                  *
 860                  * Locking here must be done carefully: the global lock
 861                  * prevents interconnections between unpcbs from changing, so
 862                  * we can traverse from unp to unp2 without acquiring unp's
 863                  * lock.  Socket buffer locks follow unpcb locks, so we can
 864                  * acquire both remote and lock socket buffer locks.
 865                  */
 866                 unp2 = unp->unp_conn;
 867                 if (unp2 == NULL) {
 868                         error = ENOTCONN;
 869                         break;
 870                 }
 871                 so2 = unp2->unp_socket;
 872                 UNP_PCB_LOCK(unp2);
 873                 SOCKBUF_LOCK(&so2->so_rcv);
 874                 if (unp2->unp_flags & UNP_WANTCRED) {
 875                         /*
 876                          * Credentials are passed only once on SOCK_STREAM.
 877                          */
 878                         unp2->unp_flags &= ~UNP_WANTCRED;
 879                         control = unp_addsockcred(td, control);
 880                 }
 881                 /*
 882                  * Send to paired receive port, and then reduce send buffer
 883                  * hiwater marks to maintain backpressure.  Wake up readers.
 884                  */
 885                 if (control != NULL) {
 886                         if (sbappendcontrol_locked(&so2->so_rcv, m, control))
 887                                 control = NULL;
 888                 } else
 889                         sbappend_locked(&so2->so_rcv, m);
 890                 mbcnt = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt;
 891                 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt;
 892                 sbcc = so2->so_rcv.sb_cc;
 893                 sorwakeup_locked(so2);
 894
 895                 SOCKBUF_LOCK(&so->so_snd);
 896                 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc);
 897                 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
 898                     newhiwat, RLIM_INFINITY);
 899                 so->so_snd.sb_mbmax -= mbcnt;
 900                 SOCKBUF_UNLOCK(&so->so_snd);
 901                 unp2->unp_cc = sbcc;
 902                 UNP_PCB_UNLOCK(unp2);
 903                 m = NULL;
 904                 break;
 905
 906         default:
 907                 panic("uipc_send unknown socktype");
 908         }
 909
 910         /*
 911          * SEND_EOF is equivalent to a SEND followed by a SHUTDOWN.
 912          */
 913         if (flags & PRUS_EOF) {
 914                 UNP_PCB_LOCK(unp);
 915                 socantsendmore(so);
 916                 unp_shutdown(unp);
 917                 UNP_PCB_UNLOCK(unp);
 918         }
 919
 920         if ((nam != NULL) || (flags & PRUS_EOF))
 921                 UNP_GLOBAL_WUNLOCK();
 922         else
 923                 UNP_GLOBAL_RUNLOCK();
 924
 925         if (control != NULL && error != 0)
 926                 unp_dispose(control);
 927
 928 release:
 929         if (control != NULL)
 930                 m_freem(control);
 931         if (m != NULL)
 932                 m_freem(m);
 933         return (error);
 934 }
 935
 936 static int
 937 uipc_sense(struct socket *so, struct stat *sb)
 938 {
 939         struct unpcb *unp, *unp2;
 940         struct socket *so2;
 941
 942         unp = sotounpcb(so);
 943         KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
 944
 945         sb->st_blksize = so->so_snd.sb_hiwat;
 946         UNP_GLOBAL_RLOCK();
 947         UNP_PCB_LOCK(unp);
 948         unp2 = unp->unp_conn;
 949         if (so->so_type == SOCK_STREAM && unp2 != NULL) {
 950                 so2 = unp2->unp_socket;
 951                 sb->st_blksize += so2->so_rcv.sb_cc;
 952         }
 953         sb->st_dev = NODEV;
 954         if (unp->unp_ino == 0)
 955                 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
 956         sb->st_ino = unp->unp_ino;
 957         UNP_PCB_UNLOCK(unp);
 958         UNP_GLOBAL_RUNLOCK();
 959         return (0);
 960 }
 961
 962 static int
 963 uipc_shutdown(struct socket *so)
 964 {
 965         struct unpcb *unp;
 966
 967         unp = sotounpcb(so);
 968         KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
 969
 970         UNP_GLOBAL_WLOCK();
 971         UNP_PCB_LOCK(unp);
 972         socantsendmore(so);
 973         unp_shutdown(unp);
 974         UNP_PCB_UNLOCK(unp);
 975         UNP_GLOBAL_WUNLOCK();
 976         return (0);
 977 }
 978
 979 static int
 980 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
 981 {
 982         struct unpcb *unp;
 983         const struct sockaddr *sa;
 984
 985         unp = sotounpcb(so);
 986         KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
 987
 988         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 989         UNP_PCB_LOCK(unp);
 990         if (unp->unp_addr != NULL)
 991                 sa = (struct sockaddr *) unp->unp_addr;
 992         else
 993                 sa = &sun_noname;
 994         bcopy(sa, *nam, sa->sa_len);
 995         UNP_PCB_UNLOCK(unp);
 996         return (0);
 997 }
 998
 999 struct pr_usrreqs uipc_usrreqs = {
1000         .pru_abort =            uipc_abort,
1001         .pru_accept =           uipc_accept,
1002         .pru_attach =           uipc_attach,
1003         .pru_bind =             uipc_bind,
1004         .pru_connect =          uipc_connect,
1005         .pru_connect2 =         uipc_connect2,
1006         .pru_detach =           uipc_detach,
1007         .pru_disconnect =       uipc_disconnect,
1008         .pru_listen =           uipc_listen,
1009         .pru_peeraddr =         uipc_peeraddr,
1010         .pru_rcvd =             uipc_rcvd,
1011         .pru_send =             uipc_send,
1012         .pru_sense =            uipc_sense,
1013         .pru_shutdown =         uipc_shutdown,
1014         .pru_sockaddr =         uipc_sockaddr,
1015         .pru_close =            uipc_close,
1016 };
1017
1018 int
1019 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1020 {
1021         struct unpcb *unp;
1022         struct xucred xu;
1023         int error, optval;
1024
1025         if (sopt->sopt_level != 0)
1026                 return (EINVAL);
1027
1028         unp = sotounpcb(so);
1029         KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1030         error = 0;
1031         switch (sopt->sopt_dir) {
1032         case SOPT_GET:
1033                 switch (sopt->sopt_name) {
1034                 case LOCAL_PEERCRED:
1035                         UNP_PCB_LOCK(unp);
1036                         if (unp->unp_flags & UNP_HAVEPC)
1037                                 xu = unp->unp_peercred;
1038                         else {
1039                                 if (so->so_type == SOCK_STREAM)
1040                                         error = ENOTCONN;
1041                                 else
1042                                         error = EINVAL;
1043                         }
1044                         UNP_PCB_UNLOCK(unp);
1045                         if (error == 0)
1046                                 error = sooptcopyout(sopt, &xu, sizeof(xu));
1047                         break;
1048
1049                 case LOCAL_CREDS:
1050                         /* Unocked read. */
1051                         optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1052                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1053                         break;
1054
1055                 case LOCAL_CONNWAIT:
1056                         /* Unocked read. */
1057                         optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1058                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1059                         break;
1060
1061                 default:
1062                         error = EOPNOTSUPP;
1063                         break;
1064                 }
1065                 break;
1066
1067         case SOPT_SET:
1068                 switch (sopt->sopt_name) {
1069                 case LOCAL_CREDS:
1070                 case LOCAL_CONNWAIT:
1071                         error = sooptcopyin(sopt, &optval, sizeof(optval),
1072                                             sizeof(optval));
1073                         if (error)
1074                                 break;
1075
1076 #define OPTSET(bit) do {                                                \
1077         UNP_PCB_LOCK(unp);                                              \
1078         if (optval)                                                     \
1079                 unp->unp_flags |= bit;                                  \
1080         else                                                            \
1081                 unp->unp_flags &= ~bit;                                 \
1082         UNP_PCB_UNLOCK(unp);                                            \
1083 } while (0)
1084
1085                         switch (sopt->sopt_name) {
1086                         case LOCAL_CREDS:
1087                                 OPTSET(UNP_WANTCRED);
1088                                 break;
1089
1090                         case LOCAL_CONNWAIT:
1091                                 OPTSET(UNP_CONNWAIT);
1092                                 break;
1093
1094                         default:
1095                                 break;
1096                         }
1097                         break;
1098 #undef  OPTSET
1099                 default:
1100                         error = ENOPROTOOPT;
1101                         break;
1102                 }
1103                 break;
1104
1105         default:
1106                 error = EOPNOTSUPP;
1107                 break;
1108         }
1109         return (error);
1110 }
1111
1112 static int
1113 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1114 {
1115         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1116         struct vnode *vp;
1117         struct socket *so2, *so3;
1118         struct unpcb *unp, *unp2, *unp3;
1119         int error, len, vfslocked;
1120         struct nameidata nd;
1121         char buf[SOCK_MAXADDRLEN];
1122         struct sockaddr *sa;
1123
1124         UNP_GLOBAL_WLOCK_ASSERT();
1125         UNP_GLOBAL_WUNLOCK();
1126
1127         unp = sotounpcb(so);
1128         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1129
1130         len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1131         if (len <= 0)
1132                 return (EINVAL);
1133         strlcpy(buf, soun->sun_path, len + 1);
1134
1135         UNP_PCB_LOCK(unp);
1136         if (unp->unp_flags & UNP_CONNECTING) {
1137                 UNP_PCB_UNLOCK(unp);
1138                 return (EALREADY);
1139         }
1140         unp->unp_flags |= UNP_CONNECTING;
1141         UNP_PCB_UNLOCK(unp);
1142
1143         sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1144         NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf,
1145             td);
1146         error = namei(&nd);
1147         if (error)
1148                 vp = NULL;
1149         else
1150                 vp = nd.ni_vp;
1151         ASSERT_VOP_LOCKED(vp, "unp_connect");
1152         vfslocked = NDHASGIANT(&nd);
1153         NDFREE(&nd, NDF_ONLY_PNBUF);
1154         if (error)
1155                 goto bad;
1156
1157         if (vp->v_type != VSOCK) {
1158                 error = ENOTSOCK;
1159                 goto bad;
1160         }
1161 #ifdef MAC
1162         error = mac_check_vnode_open(td->td_ucred, vp, VWRITE | VREAD);
1163         if (error)
1164                 goto bad;
1165 #endif
1166         error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1167         if (error)
1168                 goto bad;
1169         VFS_UNLOCK_GIANT(vfslocked);
1170
1171         unp = sotounpcb(so);
1172         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1173
1174         /*
1175          * Lock global lock for two reasons: make sure v_socket is stable,
1176          * and to protect simultaneous locking of multiple pcbs.
1177          */
1178         UNP_GLOBAL_WLOCK();
1179         so2 = vp->v_socket;
1180         if (so2 == NULL) {
1181                 error = ECONNREFUSED;
1182                 goto bad2;
1183         }
1184         if (so->so_type != so2->so_type) {
1185                 error = EPROTOTYPE;
1186                 goto bad2;
1187         }
1188         if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1189                 if (so2->so_options & SO_ACCEPTCONN) {
1190                         /*
1191                          * We can't drop the global lock here or 'so2' may
1192                          * become invalid.  As a result, we need to handle
1193                          * possibly lock recursion in uipc_attach.
1194                          */
1195                         so3 = sonewconn(so2, 0);
1196                 } else
1197                         so3 = NULL;
1198                 if (so3 == NULL) {
1199                         error = ECONNREFUSED;
1200                         goto bad2;
1201                 }
1202                 unp = sotounpcb(so);
1203                 unp2 = sotounpcb(so2);
1204                 unp3 = sotounpcb(so3);
1205                 UNP_PCB_LOCK(unp);
1206                 UNP_PCB_LOCK(unp2);
1207                 UNP_PCB_LOCK(unp3);
1208                 if (unp2->unp_addr != NULL) {
1209                         bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1210                         unp3->unp_addr = (struct sockaddr_un *) sa;
1211                         sa = NULL;
1212                 }
1213                 /*
1214                  * unp_peercred management:
1215                  *
1216                  * The connecter's (client's) credentials are copied from its
1217                  * process structure at the time of connect() (which is now).
1218                  */
1219                 cru2x(td->td_ucred, &unp3->unp_peercred);
1220                 unp3->unp_flags |= UNP_HAVEPC;
1221                 /*
1222                  * The receiver's (server's) credentials are copied from the
1223                  * unp_peercred member of socket on which the former called
1224                  * listen(); uipc_listen() cached that process's credentials
1225                  * at that time so we can use them now.
1226                  */
1227                 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1228                     ("unp_connect: listener without cached peercred"));
1229                 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1230                     sizeof(unp->unp_peercred));
1231                 unp->unp_flags |= UNP_HAVEPC;
1232                 if (unp2->unp_flags & UNP_WANTCRED)
1233                         unp3->unp_flags |= UNP_WANTCRED;
1234                 UNP_PCB_UNLOCK(unp3);
1235                 UNP_PCB_UNLOCK(unp2);
1236                 UNP_PCB_UNLOCK(unp);
1237 #ifdef MAC
1238                 SOCK_LOCK(so);
1239                 mac_set_socket_peer_from_socket(so, so3);
1240                 mac_set_socket_peer_from_socket(so3, so);
1241                 SOCK_UNLOCK(so);
1242 #endif
1243
1244                 so2 = so3;
1245         }
1246         unp = sotounpcb(so);
1247         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1248         unp2 = sotounpcb(so2);
1249         KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1250         UNP_PCB_LOCK(unp);
1251         UNP_PCB_LOCK(unp2);
1252         error = unp_connect2(so, so2, PRU_CONNECT);
1253         UNP_PCB_UNLOCK(unp2);
1254         UNP_PCB_UNLOCK(unp);
1255 bad2:
1256         UNP_GLOBAL_WUNLOCK();
1257         if (vfslocked)
1258                 /*
1259                  * Giant has been previously acquired. This means filesystem
1260                  * isn't MPSAFE. Do it once again.
1261                  */
1262                 mtx_lock(&Giant);
1263 bad:
1264         if (vp != NULL)
1265                 vput(vp);
1266         VFS_UNLOCK_GIANT(vfslocked);
1267         free(sa, M_SONAME);
1268         UNP_GLOBAL_WLOCK();
1269         UNP_PCB_LOCK(unp);
1270         unp->unp_flags &= ~UNP_CONNECTING;
1271         UNP_PCB_UNLOCK(unp);
1272         return (error);
1273 }
1274
1275 static int
1276 unp_connect2(struct socket *so, struct socket *so2, int req)
1277 {
1278         struct unpcb *unp;
1279         struct unpcb *unp2;
1280
1281         unp = sotounpcb(so);
1282         KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1283         unp2 = sotounpcb(so2);
1284         KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1285
1286         UNP_GLOBAL_WLOCK_ASSERT();
1287         UNP_PCB_LOCK_ASSERT(unp);
1288         UNP_PCB_LOCK_ASSERT(unp2);
1289
1290         if (so2->so_type != so->so_type)
1291                 return (EPROTOTYPE);
1292         unp->unp_conn = unp2;
1293
1294         switch (so->so_type) {
1295         case SOCK_DGRAM:
1296                 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1297                 soisconnected(so);
1298                 break;
1299
1300         case SOCK_STREAM:
1301                 unp2->unp_conn = unp;
1302                 if (req == PRU_CONNECT &&
1303                     ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1304                         soisconnecting(so);
1305                 else
1306                         soisconnected(so);
1307                 soisconnected(so2);
1308                 break;
1309
1310         default:
1311                 panic("unp_connect2");
1312         }
1313         return (0);
1314 }
1315
1316 static void
1317 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1318 {
1319         struct socket *so;
1320
1321         KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1322
1323         UNP_GLOBAL_WLOCK_ASSERT();
1324         UNP_PCB_LOCK_ASSERT(unp);
1325         UNP_PCB_LOCK_ASSERT(unp2);
1326
1327         unp->unp_conn = NULL;
1328         switch (unp->unp_socket->so_type) {
1329         case SOCK_DGRAM:
1330                 LIST_REMOVE(unp, unp_reflink);
1331                 so = unp->unp_socket;
1332                 SOCK_LOCK(so);
1333                 so->so_state &= ~SS_ISCONNECTED;
1334                 SOCK_UNLOCK(so);
1335                 break;
1336
1337         case SOCK_STREAM:
1338                 soisdisconnected(unp->unp_socket);
1339                 unp2->unp_conn = NULL;
1340                 soisdisconnected(unp2->unp_socket);
1341                 break;
1342         }
1343 }
1344
1345 /*
1346  * unp_pcblist() walks the global list of struct unpcb's to generate a
1347  * pointer list, bumping the refcount on each unpcb.  It then copies them out
1348  * sequentially, validating the generation number on each to see if it has
1349  * been detached.  All of this is necessary because copyout() may sleep on
1350  * disk I/O.
1351  */
1352 static int
1353 unp_pcblist(SYSCTL_HANDLER_ARGS)
1354 {
1355         int error, i, n;
1356         int freeunp;
1357         struct unpcb *unp, **unp_list;
1358         unp_gen_t gencnt;
1359         struct xunpgen *xug;
1360         struct unp_head *head;
1361         struct xunpcb *xu;
1362
1363         head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
1364
1365         /*
1366          * The process of preparing the PCB list is too time-consuming and
1367          * resource-intensive to repeat twice on every request.
1368          */
1369         if (req->oldptr == NULL) {
1370                 n = unp_count;
1371                 req->oldidx = 2 * (sizeof *xug)
1372                         + (n + n/8) * sizeof(struct xunpcb);
1373                 return (0);
1374         }
1375
1376         if (req->newptr != NULL)
1377                 return (EPERM);
1378
1379         /*
1380          * OK, now we're committed to doing something.
1381          */
1382         xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1383         UNP_GLOBAL_RLOCK();
1384         gencnt = unp_gencnt;
1385         n = unp_count;
1386         UNP_GLOBAL_RUNLOCK();
1387
1388         xug->xug_len = sizeof *xug;
1389         xug->xug_count = n;
1390         xug->xug_gen = gencnt;
1391         xug->xug_sogen = so_gencnt;
1392         error = SYSCTL_OUT(req, xug, sizeof *xug);
1393         if (error) {
1394                 free(xug, M_TEMP);
1395                 return (error);
1396         }
1397
1398         unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1399
1400         UNP_GLOBAL_RLOCK();
1401         for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1402              unp = LIST_NEXT(unp, unp_link)) {
1403                 UNP_PCB_LOCK(unp);
1404                 if (unp->unp_gencnt <= gencnt) {
1405                         if (cr_cansee(req->td->td_ucred,
1406                             unp->unp_socket->so_cred)) {
1407                                 UNP_PCB_UNLOCK(unp);
1408                                 continue;
1409                         }
1410                         unp_list[i++] = unp;
1411                         unp->unp_refcount++;
1412                 }
1413                 UNP_PCB_UNLOCK(unp);
1414         }
1415         UNP_GLOBAL_RUNLOCK();
1416         n = i;                  /* In case we lost some during malloc. */
1417
1418         error = 0;
1419         xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1420         for (i = 0; i < n; i++) {
1421                 unp = unp_list[i];
1422                 UNP_PCB_LOCK(unp);
1423                 unp->unp_refcount--;
1424                 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1425                         xu->xu_len = sizeof *xu;
1426                         xu->xu_unpp = unp;
1427                         /*
1428                          * XXX - need more locking here to protect against
1429                          * connect/disconnect races for SMP.
1430                          */
1431                         if (unp->unp_addr != NULL)
1432                                 bcopy(unp->unp_addr, &xu->xu_addr,
1433                                       unp->unp_addr->sun_len);
1434                         if (unp->unp_conn != NULL &&
1435                             unp->unp_conn->unp_addr != NULL)
1436                                 bcopy(unp->unp_conn->unp_addr,
1437                                       &xu->xu_caddr,
1438                                       unp->unp_conn->unp_addr->sun_len);
1439                         bcopy(unp, &xu->xu_unp, sizeof *unp);
1440                         sotoxsocket(unp->unp_socket, &xu->xu_socket);
1441                         UNP_PCB_UNLOCK(unp);
1442                         error = SYSCTL_OUT(req, xu, sizeof *xu);
1443                 } else {
1444                         freeunp = (unp->unp_refcount == 0);
1445                         UNP_PCB_UNLOCK(unp);
1446                         if (freeunp) {
1447                                 UNP_PCB_LOCK_DESTROY(unp);
1448                                 uma_zfree(unp_zone, unp);
1449                         }
1450                 }
1451         }
1452         free(xu, M_TEMP);
1453         if (!error) {
1454                 /*
1455                  * Give the user an updated idea of our state.  If the
1456                  * generation differs from what we told her before, she knows
1457                  * that something happened while we were processing this
1458                  * request, and it might be necessary to retry.
1459                  */
1460                 xug->xug_gen = unp_gencnt;
1461                 xug->xug_sogen = so_gencnt;
1462                 xug->xug_count = unp_count;
1463                 error = SYSCTL_OUT(req, xug, sizeof *xug);
1464         }
1465         free(unp_list, M_TEMP);
1466         free(xug, M_TEMP);
1467         return (error);
1468 }
1469
1470 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1471             (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1472             "List of active local datagram sockets");
1473 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1474             (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1475             "List of active local stream sockets");
1476
1477 static void
1478 unp_shutdown(struct unpcb *unp)
1479 {
1480         struct unpcb *unp2;
1481         struct socket *so;
1482
1483         UNP_GLOBAL_WLOCK_ASSERT();
1484         UNP_PCB_LOCK_ASSERT(unp);
1485
1486         unp2 = unp->unp_conn;
1487         if (unp->unp_socket->so_type == SOCK_STREAM && unp2 != NULL) {
1488                 so = unp2->unp_socket;
1489                 if (so != NULL)
1490                         socantrcvmore(so);
1491         }
1492 }
1493
1494 static void
1495 unp_drop(struct unpcb *unp, int errno)
1496 {
1497         struct socket *so = unp->unp_socket;
1498         struct unpcb *unp2;
1499
1500         UNP_GLOBAL_WLOCK_ASSERT();
1501         UNP_PCB_LOCK_ASSERT(unp);
1502
1503         so->so_error = errno;
1504         unp2 = unp->unp_conn;
1505         if (unp2 == NULL)
1506                 return;
1507
1508         UNP_PCB_LOCK(unp2);
1509         unp_disconnect(unp, unp2);
1510         UNP_PCB_UNLOCK(unp2);
1511 }
1512
1513 static void
1514 unp_freerights(struct file **rp, int fdcount)
1515 {
1516         int i;
1517         struct file *fp;
1518
1519         for (i = 0; i < fdcount; i++) {
1520                 /*
1521                  * Zero the pointer before calling unp_discard since it may
1522                  * end up in unp_gc()..
1523                  *
1524                  * XXXRW: This is less true than it used to be.
1525                  */
1526                 fp = *rp;
1527                 *rp++ = NULL;
1528                 unp_discard(fp);
1529         }
1530 }
1531
1532 int
1533 unp_externalize(struct mbuf *control, struct mbuf **controlp)
1534 {
1535         struct thread *td = curthread;          /* XXX */
1536         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1537         int i;
1538         int *fdp;
1539         struct file **rp;
1540         struct file *fp;
1541         void *data;
1542         socklen_t clen = control->m_len, datalen;
1543         int error, newfds;
1544         int f;
1545         u_int newlen;
1546
1547         UNP_GLOBAL_UNLOCK_ASSERT();
1548
1549         error = 0;
1550         if (controlp != NULL) /* controlp == NULL => free control messages */
1551                 *controlp = NULL;
1552
1553         while (cm != NULL) {
1554                 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1555                         error = EINVAL;
1556                         break;
1557                 }
1558
1559                 data = CMSG_DATA(cm);
1560                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1561
1562                 if (cm->cmsg_level == SOL_SOCKET
1563                     && cm->cmsg_type == SCM_RIGHTS) {
1564                         newfds = datalen / sizeof(struct file *);
1565                         rp = data;
1566
1567                         /* If we're not outputting the descriptors free them. */
1568                         if (error || controlp == NULL) {
1569                                 unp_freerights(rp, newfds);
1570                                 goto next;
1571                         }
1572                         FILEDESC_XLOCK(td->td_proc->p_fd);
1573                         /* if the new FD's will not fit free them.  */
1574                         if (!fdavail(td, newfds)) {
1575                                 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1576                                 error = EMSGSIZE;
1577                                 unp_freerights(rp, newfds);
1578                                 goto next;
1579                         }
1580                         /*
1581                          * Now change each pointer to an fd in the global
1582                          * table to an integer that is the index to the local
1583                          * fd table entry that we set up to point to the
1584                          * global one we are transferring.
1585                          */
1586                         newlen = newfds * sizeof(int);
1587                         *controlp = sbcreatecontrol(NULL, newlen,
1588                             SCM_RIGHTS, SOL_SOCKET);
1589                         if (*controlp == NULL) {
1590                                 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1591                                 error = E2BIG;
1592                                 unp_freerights(rp, newfds);
1593                                 goto next;
1594                         }
1595
1596                         fdp = (int *)
1597                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1598                         for (i = 0; i < newfds; i++) {
1599                                 if (fdalloc(td, 0, &f))
1600                                         panic("unp_externalize fdalloc failed");
1601                                 fp = *rp++;
1602                                 td->td_proc->p_fd->fd_ofiles[f] = fp;
1603                                 FILE_LOCK(fp);
1604                                 fp->f_msgcount--;
1605                                 FILE_UNLOCK(fp);
1606                                 unp_rights--;
1607                                 *fdp++ = f;
1608                         }
1609                         FILEDESC_XUNLOCK(td->td_proc->p_fd);
1610                 } else {
1611                         /* We can just copy anything else across. */
1612                         if (error || controlp == NULL)
1613                                 goto next;
1614                         *controlp = sbcreatecontrol(NULL, datalen,
1615                             cm->cmsg_type, cm->cmsg_level);
1616                         if (*controlp == NULL) {
1617                                 error = ENOBUFS;
1618                                 goto next;
1619                         }
1620                         bcopy(data,
1621                             CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1622                             datalen);
1623                 }
1624
1625                 controlp = &(*controlp)->m_next;
1626
1627 next:
1628                 if (CMSG_SPACE(datalen) < clen) {
1629                         clen -= CMSG_SPACE(datalen);
1630                         cm = (struct cmsghdr *)
1631                             ((caddr_t)cm + CMSG_SPACE(datalen));
1632                 } else {
1633                         clen = 0;
1634                         cm = NULL;
1635                 }
1636         }
1637
1638         m_freem(control);
1639
1640         return (error);
1641 }
1642
1643 static void
1644 unp_zone_change(void *tag)
1645 {
1646
1647         uma_zone_set_max(unp_zone, maxsockets);
1648 }
1649
1650 void
1651 unp_init(void)
1652 {
1653
1654         unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1655             NULL, NULL, UMA_ALIGN_PTR, 0);
1656         if (unp_zone == NULL)
1657                 panic("unp_init");
1658         uma_zone_set_max(unp_zone, maxsockets);
1659         EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1660             NULL, EVENTHANDLER_PRI_ANY);
1661         LIST_INIT(&unp_dhead);
1662         LIST_INIT(&unp_shead);
1663         TASK_INIT(&unp_gc_task, 0, unp_gc, NULL);
1664         UNP_GLOBAL_LOCK_INIT();
1665 }
1666
1667 static int
1668 unp_internalize(struct mbuf **controlp, struct thread *td)
1669 {
1670         struct mbuf *control = *controlp;
1671         struct proc *p = td->td_proc;
1672         struct filedesc *fdescp = p->p_fd;
1673         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1674         struct cmsgcred *cmcred;
1675         struct file **rp;
1676         struct file *fp;
1677         struct timeval *tv;
1678         int i, fd, *fdp;
1679         void *data;
1680         socklen_t clen = control->m_len, datalen;
1681         int error, oldfds;
1682         u_int newlen;
1683
1684         UNP_GLOBAL_UNLOCK_ASSERT();
1685
1686         error = 0;
1687         *controlp = NULL;
1688
1689         while (cm != NULL) {
1690                 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1691                     || cm->cmsg_len > clen) {
1692                         error = EINVAL;
1693                         goto out;
1694                 }
1695
1696                 data = CMSG_DATA(cm);
1697                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1698
1699                 switch (cm->cmsg_type) {
1700                 /*
1701                  * Fill in credential information.
1702                  */
1703                 case SCM_CREDS:
1704                         *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1705                             SCM_CREDS, SOL_SOCKET);
1706                         if (*controlp == NULL) {
1707                                 error = ENOBUFS;
1708                                 goto out;
1709                         }
1710
1711                         cmcred = (struct cmsgcred *)
1712                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1713                         cmcred->cmcred_pid = p->p_pid;
1714                         cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1715                         cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1716                         cmcred->cmcred_euid = td->td_ucred->cr_uid;
1717                         cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1718                                                         CMGROUP_MAX);
1719                         for (i = 0; i < cmcred->cmcred_ngroups; i++)
1720                                 cmcred->cmcred_groups[i] =
1721                                     td->td_ucred->cr_groups[i];
1722                         break;
1723
1724                 case SCM_RIGHTS:
1725                         oldfds = datalen / sizeof (int);
1726                         /*
1727                          * Check that all the FDs passed in refer to legal
1728                          * files.  If not, reject the entire operation.
1729                          */
1730                         fdp = data;
1731                         FILEDESC_SLOCK(fdescp);
1732                         for (i = 0; i < oldfds; i++) {
1733                                 fd = *fdp++;
1734                                 if ((unsigned)fd >= fdescp->fd_nfiles ||
1735                                     fdescp->fd_ofiles[fd] == NULL) {
1736                                         FILEDESC_SUNLOCK(fdescp);
1737                                         error = EBADF;
1738                                         goto out;
1739                                 }
1740                                 fp = fdescp->fd_ofiles[fd];
1741                                 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1742                                         FILEDESC_SUNLOCK(fdescp);
1743                                         error = EOPNOTSUPP;
1744                                         goto out;
1745                                 }
1746
1747                         }
1748
1749                         /*
1750                          * Now replace the integer FDs with pointers to
1751                          * the associated global file table entry..
1752                          */
1753                         newlen = oldfds * sizeof(struct file *);
1754                         *controlp = sbcreatecontrol(NULL, newlen,
1755                             SCM_RIGHTS, SOL_SOCKET);
1756                         if (*controlp == NULL) {
1757                                 FILEDESC_SUNLOCK(fdescp);
1758                                 error = E2BIG;
1759                                 goto out;
1760                         }
1761
1762                         fdp = data;
1763                         rp = (struct file **)
1764                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1765                         for (i = 0; i < oldfds; i++) {
1766                                 fp = fdescp->fd_ofiles[*fdp++];
1767                                 *rp++ = fp;
1768                                 FILE_LOCK(fp);
1769                                 fp->f_count++;
1770                                 fp->f_msgcount++;
1771                                 FILE_UNLOCK(fp);
1772                                 unp_rights++;
1773                         }
1774                         FILEDESC_SUNLOCK(fdescp);
1775                         break;
1776
1777                 case SCM_TIMESTAMP:
1778                         *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1779                             SCM_TIMESTAMP, SOL_SOCKET);
1780                         if (*controlp == NULL) {
1781                                 error = ENOBUFS;
1782                                 goto out;
1783                         }
1784                         tv = (struct timeval *)
1785                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1786                         microtime(tv);
1787                         break;
1788
1789                 default:
1790                         error = EINVAL;
1791                         goto out;
1792                 }
1793
1794                 controlp = &(*controlp)->m_next;
1795
1796                 if (CMSG_SPACE(datalen) < clen) {
1797                         clen -= CMSG_SPACE(datalen);
1798                         cm = (struct cmsghdr *)
1799                             ((caddr_t)cm + CMSG_SPACE(datalen));
1800                 } else {
1801                         clen = 0;
1802                         cm = NULL;
1803                 }
1804         }
1805
1806 out:
1807         m_freem(control);
1808
1809         return (error);
1810 }
1811
1812 static struct mbuf *
1813 unp_addsockcred(struct thread *td, struct mbuf *control)
1814 {
1815         struct mbuf *m, *n, *n_prev;
1816         struct sockcred *sc;
1817         const struct cmsghdr *cm;
1818         int ngroups;
1819         int i;
1820
1821         ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
1822
1823         m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
1824         if (m == NULL)
1825                 return (control);
1826
1827         sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
1828         sc->sc_uid = td->td_ucred->cr_ruid;
1829         sc->sc_euid = td->td_ucred->cr_uid;
1830         sc->sc_gid = td->td_ucred->cr_rgid;
1831         sc->sc_egid = td->td_ucred->cr_gid;
1832         sc->sc_ngroups = ngroups;
1833         for (i = 0; i < sc->sc_ngroups; i++)
1834                 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
1835
1836         /*
1837          * Unlink SCM_CREDS control messages (struct cmsgcred), since just
1838          * created SCM_CREDS control message (struct sockcred) has another
1839          * format.
1840          */
1841         if (control != NULL)
1842                 for (n = control, n_prev = NULL; n != NULL;) {
1843                         cm = mtod(n, struct cmsghdr *);
1844                         if (cm->cmsg_level == SOL_SOCKET &&
1845                             cm->cmsg_type == SCM_CREDS) {
1846                                 if (n_prev == NULL)
1847                                         control = n->m_next;
1848                                 else
1849                                         n_prev->m_next = n->m_next;
1850                                 n = m_free(n);
1851                         } else {
1852                                 n_prev = n;
1853                                 n = n->m_next;
1854                         }
1855                 }
1856
1857         /* Prepend it to the head. */
1858         m->m_next = control;
1859
1860         return (m);
1861 }
1862
1863 /*
1864  * unp_defer indicates whether additional work has been defered for a future
1865  * pass through unp_gc().  It is thread local and does not require explicit
1866  * synchronization.
1867  */
1868 static int      unp_defer;
1869
1870 static int unp_taskcount;
1871 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, "");
1872
1873 static int unp_recycled;
1874 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, "");
1875
1876 static void
1877 unp_gc(__unused void *arg, int pending)
1878 {
1879         struct file *fp, *nextfp;
1880         struct socket *so;
1881         struct file **extra_ref, **fpp;
1882         int nunref, i;
1883         int nfiles_snap;
1884         int nfiles_slack = 20;
1885
1886         unp_taskcount++;
1887         unp_defer = 0;
1888         /*
1889          * Before going through all this, set all FDs to be NOT deferred and
1890          * NOT externally accessible.
1891          */
1892         sx_slock(&filelist_lock);
1893         LIST_FOREACH(fp, &filehead, f_list)
1894                 fp->f_gcflag &= ~(FMARK|FDEFER);
1895         do {
1896                 KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer));
1897                 LIST_FOREACH(fp, &filehead, f_list) {
1898                         FILE_LOCK(fp);
1899                         /*
1900                          * If the file is not open, skip it -- could be a
1901                          * file in the process of being opened, or in the
1902                          * process of being closed.  If the file is
1903                          * "closing", it may have been marked for deferred
1904                          * consideration.  Clear the flag now if so.
1905                          */
1906                         if (fp->f_count == 0) {
1907                                 if (fp->f_gcflag & FDEFER)
1908                                         unp_defer--;
1909                                 fp->f_gcflag &= ~(FMARK|FDEFER);
1910                                 FILE_UNLOCK(fp);
1911                                 continue;
1912                         }
1913                         /*
1914                          * If we already marked it as 'defer' in a
1915                          * previous pass, then try to process it this
1916                          * time and un-mark it.
1917                          */
1918                         if (fp->f_gcflag & FDEFER) {
1919                                 fp->f_gcflag &= ~FDEFER;
1920                                 unp_defer--;
1921                         } else {
1922                                 /*
1923                                  * If it's not deferred, then check if it's
1924                                  * already marked.. if so skip it
1925                                  */
1926                                 if (fp->f_gcflag & FMARK) {
1927                                         FILE_UNLOCK(fp);
1928                                         continue;
1929                                 }
1930                                 /*
1931                                  * If all references are from messages in
1932                                  * transit, then skip it. it's not externally
1933                                  * accessible.
1934                                  */
1935                                 if (fp->f_count == fp->f_msgcount) {
1936                                         FILE_UNLOCK(fp);
1937                                         continue;
1938                                 }
1939                                 /*
1940                                  * If it got this far then it must be
1941                                  * externally accessible.
1942                                  */
1943                                 fp->f_gcflag |= FMARK;
1944                         }
1945                         /*
1946                          * Either it was deferred, or it is externally
1947                          * accessible and not already marked so.  Now check
1948                          * if it is possibly one of OUR sockets.
1949                          */
1950                         if (fp->f_type != DTYPE_SOCKET ||
1951                             (so = fp->f_data) == NULL) {
1952                                 FILE_UNLOCK(fp);
1953                                 continue;
1954                         }
1955                         if (so->so_proto->pr_domain != &localdomain ||
1956                             (so->so_proto->pr_flags & PR_RIGHTS) == 0) {
1957                                 FILE_UNLOCK(fp);
1958                                 continue;
1959                         }
1960
1961                         /*
1962                          * Tell any other threads that do a subsequent
1963                          * fdrop() that we are scanning the message
1964                          * buffers.
1965                          */
1966                         fp->f_gcflag |= FWAIT;
1967                         FILE_UNLOCK(fp);
1968
1969                         /*
1970                          * So, Ok, it's one of our sockets and it IS
1971                          * externally accessible (or was deferred).  Now we
1972                          * look to see if we hold any file descriptors in its
1973                          * message buffers. Follow those links and mark them
1974                          * as accessible too.
1975                          */
1976                         SOCKBUF_LOCK(&so->so_rcv);
1977                         unp_scan(so->so_rcv.sb_mb, unp_mark);
1978                         SOCKBUF_UNLOCK(&so->so_rcv);
1979
1980                         /*
1981                          * Wake up any threads waiting in fdrop().
1982                          */
1983                         FILE_LOCK(fp);
1984                         fp->f_gcflag &= ~FWAIT;
1985                         wakeup(&fp->f_gcflag);
1986                         FILE_UNLOCK(fp);
1987                 }
1988         } while (unp_defer);
1989         sx_sunlock(&filelist_lock);
1990         /*
1991          * XXXRW: The following comments need updating for a post-SMPng and
1992          * deferred unp_gc() world, but are still generally accurate.
1993          *
1994          * We grab an extra reference to each of the file table entries that
1995          * are not otherwise accessible and then free the rights that are
1996          * stored in messages on them.
1997          *
1998          * The bug in the orginal code is a little tricky, so I'll describe
1999          * what's wrong with it here.
2000          *
2001          * It is incorrect to simply unp_discard each entry for f_msgcount
2002          * times -- consider the case of sockets A and B that contain
2003          * references to each other.  On a last close of some other socket,
2004          * we trigger a gc since the number of outstanding rights (unp_rights)
2005          * is non-zero.  If during the sweep phase the gc code unp_discards,
2006          * we end up doing a (full) closef on the descriptor.  A closef on A
2007          * results in the following chain.  Closef calls soo_close, which
2008          * calls soclose.   Soclose calls first (through the switch
2009          * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
2010          * returns because the previous instance had set unp_gcing, and we
2011          * return all the way back to soclose, which marks the socket with
2012          * SS_NOFDREF, and then calls sofree.  Sofree calls sorflush to free
2013          * up the rights that are queued in messages on the socket A, i.e.,
2014          * the reference on B.  The sorflush calls via the dom_dispose switch
2015          * unp_dispose, which unp_scans with unp_discard.  This second
2016          * instance of unp_discard just calls closef on B.
2017          *
2018          * Well, a similar chain occurs on B, resulting in a sorflush on B,
2019          * which results in another closef on A.  Unfortunately, A is already
2020          * being closed, and the descriptor has already been marked with
2021          * SS_NOFDREF, and soclose panics at this point.
2022          *
2023          * Here, we first take an extra reference to each inaccessible
2024          * descriptor.  Then, we call sorflush ourself, since we know it is a
2025          * Unix domain socket anyhow.  After we destroy all the rights
2026          * carried in messages, we do a last closef to get rid of our extra
2027          * reference.  This is the last close, and the unp_detach etc will
2028          * shut down the socket.
2029          *
2030          * 91/09/19, bsy@cs.cmu.edu
2031          */
2032 again:
2033         nfiles_snap = openfiles + nfiles_slack; /* some slack */
2034         extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP,
2035             M_WAITOK);
2036         sx_slock(&filelist_lock);
2037         if (nfiles_snap < openfiles) {
2038                 sx_sunlock(&filelist_lock);
2039                 free(extra_ref, M_TEMP);
2040                 nfiles_slack += 20;
2041                 goto again;
2042         }
2043         for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref;
2044             fp != NULL; fp = nextfp) {
2045                 nextfp = LIST_NEXT(fp, f_list);
2046                 FILE_LOCK(fp);
2047                 /*
2048                  * If it's not open, skip it
2049                  */
2050                 if (fp->f_count == 0) {
2051                         FILE_UNLOCK(fp);
2052                         continue;
2053                 }
2054                 /*
2055                  * If all refs are from msgs, and it's not marked accessible
2056                  * then it must be referenced from some unreachable cycle of
2057                  * (shut-down) FDs, so include it in our list of FDs to
2058                  * remove.
2059                  */
2060                 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) {
2061                         *fpp++ = fp;
2062                         nunref++;
2063                         fp->f_count++;
2064                 }
2065                 FILE_UNLOCK(fp);
2066         }
2067         sx_sunlock(&filelist_lock);
2068         /*
2069          * For each FD on our hit list, do the following two things:
2070          */
2071         for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
2072                 struct file *tfp = *fpp;
2073                 FILE_LOCK(tfp);
2074                 if (tfp->f_type == DTYPE_SOCKET &&
2075                     tfp->f_data != NULL) {
2076                         FILE_UNLOCK(tfp);
2077                         sorflush(tfp->f_data);
2078                 } else {
2079                         FILE_UNLOCK(tfp);
2080                 }
2081         }
2082         for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
2083                 closef(*fpp, (struct thread *) NULL);
2084                 unp_recycled++;
2085         }
2086         free(extra_ref, M_TEMP);
2087 }
2088
2089 void
2090 unp_dispose(struct mbuf *m)
2091 {
2092
2093         if (m)
2094                 unp_scan(m, unp_discard);
2095 }
2096
2097 static void
2098 unp_scan(struct mbuf *m0, void (*op)(struct file *))
2099 {
2100         struct mbuf *m;
2101         struct file **rp;
2102         struct cmsghdr *cm;
2103         void *data;
2104         int i;
2105         socklen_t clen, datalen;
2106         int qfds;
2107
2108         while (m0 != NULL) {
2109                 for (m = m0; m; m = m->m_next) {
2110                         if (m->m_type != MT_CONTROL)
2111                                 continue;
2112
2113                         cm = mtod(m, struct cmsghdr *);
2114                         clen = m->m_len;
2115
2116                         while (cm != NULL) {
2117                                 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2118                                         break;
2119
2120                                 data = CMSG_DATA(cm);
2121                                 datalen = (caddr_t)cm + cm->cmsg_len
2122                                     - (caddr_t)data;
2123
2124                                 if (cm->cmsg_level == SOL_SOCKET &&
2125                                     cm->cmsg_type == SCM_RIGHTS) {
2126                                         qfds = datalen / sizeof (struct file *);
2127                                         rp = data;
2128                                         for (i = 0; i < qfds; i++)
2129                                                 (*op)(*rp++);
2130                                 }
2131
2132                                 if (CMSG_SPACE(datalen) < clen) {
2133                                         clen -= CMSG_SPACE(datalen);
2134                                         cm = (struct cmsghdr *)
2135                                             ((caddr_t)cm + CMSG_SPACE(datalen));
2136                                 } else {
2137                                         clen = 0;
2138                                         cm = NULL;
2139                                 }
2140                         }
2141                 }
2142                 m0 = m0->m_act;
2143         }
2144 }
2145
2146 static void
2147 unp_mark(struct file *fp)
2148 {
2149
2150         /* XXXRW: Should probably assert file list lock here. */
2151
2152         if (fp->f_gcflag & FMARK)
2153                 return;
2154         unp_defer++;
2155         fp->f_gcflag |= (FMARK|FDEFER);
2156 }
2157
2158 static void
2159 unp_discard(struct file *fp)
2160 {
2161
2162         UNP_GLOBAL_WLOCK();
2163         FILE_LOCK(fp);
2164         fp->f_msgcount--;
2165         unp_rights--;
2166         FILE_UNLOCK(fp);
2167         UNP_GLOBAL_WUNLOCK();
2168         (void) closef(fp, (struct thread *)NULL);
2169 }
2170
2171 #ifdef DDB
2172 static void
2173 db_print_indent(int indent)
2174 {
2175         int i;
2176
2177         for (i = 0; i < indent; i++)
2178                 db_printf(" ");
2179 }
2180
2181 static void
2182 db_print_unpflags(int unp_flags)
2183 {
2184         int comma;
2185
2186         comma = 0;
2187         if (unp_flags & UNP_HAVEPC) {
2188                 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2189                 comma = 1;
2190         }
2191         if (unp_flags & UNP_HAVEPCCACHED) {
2192                 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2193                 comma = 1;
2194         }
2195         if (unp_flags & UNP_WANTCRED) {
2196                 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2197                 comma = 1;
2198         }
2199         if (unp_flags & UNP_CONNWAIT) {
2200                 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2201                 comma = 1;
2202         }
2203         if (unp_flags & UNP_CONNECTING) {
2204                 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2205                 comma = 1;
2206         }
2207         if (unp_flags & UNP_BINDING) {
2208                 db_printf("%sUNP_BINDING", comma ? ", " : "");
2209                 comma = 1;
2210         }
2211 }
2212
2213 static void
2214 db_print_xucred(int indent, struct xucred *xu)
2215 {
2216         int comma, i;
2217
2218         db_print_indent(indent);
2219         db_printf("cr_version: %u   cr_uid: %u   cr_ngroups: %d\n",
2220             xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2221         db_print_indent(indent);
2222         db_printf("cr_groups: ");
2223         comma = 0;
2224         for (i = 0; i < xu->cr_ngroups; i++) {
2225                 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2226                 comma = 1;
2227         }
2228         db_printf("\n");
2229 }
2230
2231 static void
2232 db_print_unprefs(int indent, struct unp_head *uh)
2233 {
2234         struct unpcb *unp;
2235         int counter;
2236
2237         counter = 0;
2238         LIST_FOREACH(unp, uh, unp_reflink) {
2239                 if (counter % 4 == 0)
2240                         db_print_indent(indent);
2241                 db_printf("%p  ", unp);
2242                 if (counter % 4 == 3)
2243                         db_printf("\n");
2244                 counter++;
2245         }
2246         if (counter != 0 && counter % 4 != 0)
2247                 db_printf("\n");
2248 }
2249
2250 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2251 {
2252         struct unpcb *unp;
2253
2254         if (!have_addr) {
2255                 db_printf("usage: show unpcb <addr>\n");
2256                 return;
2257         }
2258         unp = (struct unpcb *)addr;
2259
2260         db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
2261             unp->unp_vnode);
2262
2263         db_printf("unp_ino: %d   unp_conn: %p\n", unp->unp_ino,
2264             unp->unp_conn);
2265
2266         db_printf("unp_refs:\n");
2267         db_print_unprefs(2, &unp->unp_refs);
2268
2269         /* XXXRW: Would be nice to print the full address, if any. */
2270         db_printf("unp_addr: %p\n", unp->unp_addr);
2271
2272         db_printf("unp_cc: %d   unp_mbcnt: %d   unp_gencnt: %llu\n",
2273             unp->unp_cc, unp->unp_mbcnt,
2274             (unsigned long long)unp->unp_gencnt);
2275
2276         db_printf("unp_flags: %x (", unp->unp_flags);
2277         db_print_unpflags(unp->unp_flags);
2278         db_printf(")\n");
2279
2280         db_printf("unp_peercred:\n");
2281         db_print_xucred(2, &unp->unp_peercred);
2282
2283         db_printf("unp_refcount: %u\n", unp->unp_refcount);
2284 }
2285 #endif