sys/kern/uipc_usrreq.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-3-Clause
   3  *
   4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
   5  *      The Regents of the University of California.
   6  * Copyright (c) 2004-2009 Robert N. M. Watson
   7  * All rights reserved.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in the
  16  *    documentation and/or other materials provided with the distribution.
  17  * 3. Neither the name of the University nor the names of its contributors
  18  *    may be used to endorse or promote products derived from this software
  19  *    without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  *
  33  *      From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
  34  */
  35
  36 /*
  37  * UNIX Domain (Local) Sockets
  38  *
  39  * This is an implementation of UNIX (local) domain sockets.  Each socket has
  40  * an associated struct unpcb (UNIX protocol control block).  Stream sockets
  41  * may be connected to 0 or 1 other socket.  Datagram sockets may be
  42  * connected to 0, 1, or many other sockets.  Sockets may be created and
  43  * connected in pairs (socketpair(2)), or bound/connected to using the file
  44  * system name space.  For most purposes, only the receive socket buffer is
  45  * used, as sending on one socket delivers directly to the receive socket
  46  * buffer of a second socket.
  47  *
  48  * The implementation is substantially complicated by the fact that
  49  * "ancillary data", such as file descriptors or credentials, may be passed
  50  * across UNIX domain sockets.  The potential for passing UNIX domain sockets
  51  * over other UNIX domain sockets requires the implementation of a simple
  52  * garbage collector to find and tear down cycles of disconnected sockets.
  53  *
  54  * TODO:
  55  *      RDM
  56  *      rethink name space problems
  57  *      need a proper out-of-band
  58  */
  59
  60 #include <sys/cdefs.h>
  61 __FBSDID("$FreeBSD$");
  62
  63 #include "opt_ddb.h"
  64
  65 #include <sys/param.h>
  66 #include <sys/capsicum.h>
  67 #include <sys/domain.h>
  68 #include <sys/fcntl.h>
  69 #include <sys/malloc.h>         /* XXX must be before <sys/file.h> */
  70 #include <sys/eventhandler.h>
  71 #include <sys/file.h>
  72 #include <sys/filedesc.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/queue.h>
  82 #include <sys/resourcevar.h>
  83 #include <sys/rwlock.h>
  84 #include <sys/socket.h>
  85 #include <sys/socketvar.h>
  86 #include <sys/signalvar.h>
  87 #include <sys/stat.h>
  88 #include <sys/sx.h>
  89 #include <sys/sysctl.h>
  90 #include <sys/systm.h>
  91 #include <sys/taskqueue.h>
  92 #include <sys/un.h>
  93 #include <sys/unpcb.h>
  94 #include <sys/vnode.h>
  95
  96 #include <net/vnet.h>
  97
  98 #ifdef DDB
  99 #include <ddb/ddb.h>
 100 #endif
 101
 102 #include <security/mac/mac_framework.h>
 103
 104 #include <vm/uma.h>
 105
 106 MALLOC_DECLARE(M_FILECAPS);
 107
 108 /*
 109  * Locking key:
 110  * (l)  Locked using list lock
 111  * (g)  Locked using linkage lock
 112  */
 113
 114 static uma_zone_t       unp_zone;
 115 static unp_gen_t        unp_gencnt;     /* (l) */
 116 static u_int            unp_count;      /* (l) Count of local sockets. */
 117 static ino_t            unp_ino;        /* Prototype for fake inode numbers. */
 118 static int              unp_rights;     /* (g) File descriptors in flight. */
 119 static struct unp_head  unp_shead;      /* (l) List of stream sockets. */
 120 static struct unp_head  unp_dhead;      /* (l) List of datagram sockets. */
 121 static struct unp_head  unp_sphead;     /* (l) List of seqpacket sockets. */
 122
 123 struct unp_defer {
 124         SLIST_ENTRY(unp_defer) ud_link;
 125         struct file *ud_fp;
 126 };
 127 static SLIST_HEAD(, unp_defer) unp_defers;
 128 static int unp_defers_count;
 129
 130 static const struct sockaddr    sun_noname = { sizeof(sun_noname), AF_LOCAL };
 131
 132 /*
 133  * Garbage collection of cyclic file descriptor/socket references occurs
 134  * asynchronously in a taskqueue context in order to avoid recursion and
 135  * reentrance in the UNIX domain socket, file descriptor, and socket layer
 136  * code.  See unp_gc() for a full description.
 137  */
 138 static struct timeout_task unp_gc_task;
 139
 140 /*
 141  * The close of unix domain sockets attached as SCM_RIGHTS is
 142  * postponed to the taskqueue, to avoid arbitrary recursion depth.
 143  * The attached sockets might have another sockets attached.
 144  */
 145 static struct task      unp_defer_task;
 146
 147 /*
 148  * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
 149  * stream sockets, although the total for sender and receiver is actually
 150  * only PIPSIZ.
 151  *
 152  * Datagram sockets really use the sendspace as the maximum datagram size,
 153  * and don't really want to reserve the sendspace.  Their recvspace should be
 154  * large enough for at least one max-size datagram plus address.
 155  */
 156 #ifndef PIPSIZ
 157 #define PIPSIZ  8192
 158 #endif
 159 static u_long   unpst_sendspace = PIPSIZ;
 160 static u_long   unpst_recvspace = PIPSIZ;
 161 static u_long   unpdg_sendspace = 2*1024;       /* really max datagram size */
 162 static u_long   unpdg_recvspace = 4*1024;
 163 static u_long   unpsp_sendspace = PIPSIZ;       /* really max datagram size */
 164 static u_long   unpsp_recvspace = PIPSIZ;
 165
 166 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
 167 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0,
 168     "SOCK_STREAM");
 169 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
 170 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0,
 171     "SOCK_SEQPACKET");
 172
 173 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
 174            &unpst_sendspace, 0, "Default stream send space.");
 175 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
 176            &unpst_recvspace, 0, "Default stream receive space.");
 177 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
 178            &unpdg_sendspace, 0, "Default datagram send space.");
 179 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
 180            &unpdg_recvspace, 0, "Default datagram receive space.");
 181 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
 182            &unpsp_sendspace, 0, "Default seqpacket send space.");
 183 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
 184            &unpsp_recvspace, 0, "Default seqpacket receive space.");
 185 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
 186     "File descriptors in flight.");
 187 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
 188     &unp_defers_count, 0,
 189     "File descriptors deferred to taskqueue for close.");
 190
 191 /*
 192  * Locking and synchronization:
 193  *
 194  * Two types of locks exist in the local domain socket implementation: a
 195  * a global linkage rwlock and per-unpcb mutexes.  The linkage lock protects
 196  * the socket count, global generation number, stream/datagram global lists and
 197  * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be
 198  * held exclusively over the acquisition of multiple unpcb locks to prevent
 199  * deadlock.
 200  *
 201  * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
 202  * allocated in pru_attach() and freed in pru_detach().  The validity of that
 203  * pointer is an invariant, so no lock is required to dereference the so_pcb
 204  * pointer if a valid socket reference is held by the caller.  In practice,
 205  * this is always true during operations performed on a socket.  Each unpcb
 206  * has a back-pointer to its socket, unp_socket, which will be stable under
 207  * the same circumstances.
 208  *
 209  * This pointer may only be safely dereferenced as long as a valid reference
 210  * to the unpcb is held.  Typically, this reference will be from the socket,
 211  * or from another unpcb when the referring unpcb's lock is held (in order
 212  * that the reference not be invalidated during use).  For example, to follow
 213  * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
 214  * as unp_socket remains valid as long as the reference to unp_conn is valid.
 215  *
 216  * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx.  Individual
 217  * atomic reads without the lock may be performed "lockless", but more
 218  * complex reads and read-modify-writes require the mutex to be held.  No
 219  * lock order is defined between unpcb locks -- multiple unpcb locks may be
 220  * acquired at the same time only when holding the linkage rwlock
 221  * exclusively, which prevents deadlocks.
 222  *
 223  * Blocking with UNIX domain sockets is a tricky issue: unlike most network
 224  * protocols, bind() is a non-atomic operation, and connect() requires
 225  * potential sleeping in the protocol, due to potentially waiting on local or
 226  * distributed file systems.  We try to separate "lookup" operations, which
 227  * may sleep, and the IPC operations themselves, which typically can occur
 228  * with relative atomicity as locks can be held over the entire operation.
 229  *
 230  * Another tricky issue is simultaneous multi-threaded or multi-process
 231  * access to a single UNIX domain socket.  These are handled by the flags
 232  * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
 233  * binding, both of which involve dropping UNIX domain socket locks in order
 234  * to perform namei() and other file system operations.
 235  */
 236 static struct rwlock    unp_link_rwlock;
 237 static struct mtx       unp_defers_lock;
 238
 239 #define UNP_LINK_LOCK_INIT()            rw_init(&unp_link_rwlock,       \
 240                                             "unp_link_rwlock")
 241
 242 #define UNP_LINK_LOCK_ASSERT()  rw_assert(&unp_link_rwlock,     \
 243                                             RA_LOCKED)
 244 #define UNP_LINK_UNLOCK_ASSERT()        rw_assert(&unp_link_rwlock,     \
 245                                             RA_UNLOCKED)
 246
 247 #define UNP_LINK_RLOCK()                rw_rlock(&unp_link_rwlock)
 248 #define UNP_LINK_RUNLOCK()              rw_runlock(&unp_link_rwlock)
 249 #define UNP_LINK_WLOCK()                rw_wlock(&unp_link_rwlock)
 250 #define UNP_LINK_WUNLOCK()              rw_wunlock(&unp_link_rwlock)
 251 #define UNP_LINK_WLOCK_ASSERT()         rw_assert(&unp_link_rwlock,     \
 252                                             RA_WLOCKED)
 253 #define UNP_LINK_WOWNED()               rw_wowned(&unp_link_rwlock)
 254
 255 #define UNP_DEFERRED_LOCK_INIT()        mtx_init(&unp_defers_lock, \
 256                                             "unp_defer", NULL, MTX_DEF)
 257 #define UNP_DEFERRED_LOCK()             mtx_lock(&unp_defers_lock)
 258 #define UNP_DEFERRED_UNLOCK()           mtx_unlock(&unp_defers_lock)
 259
 260 #define UNP_PCB_LOCK_INIT(unp)          mtx_init(&(unp)->unp_mtx,       \
 261                                             "unp_mtx", "unp_mtx",       \
 262                                             MTX_DUPOK|MTX_DEF|MTX_RECURSE)
 263 #define UNP_PCB_LOCK_DESTROY(unp)       mtx_destroy(&(unp)->unp_mtx)
 264 #define UNP_PCB_LOCK(unp)               mtx_lock(&(unp)->unp_mtx)
 265 #define UNP_PCB_UNLOCK(unp)             mtx_unlock(&(unp)->unp_mtx)
 266 #define UNP_PCB_LOCK_ASSERT(unp)        mtx_assert(&(unp)->unp_mtx, MA_OWNED)
 267
 268 static int      uipc_connect2(struct socket *, struct socket *);
 269 static int      uipc_ctloutput(struct socket *, struct sockopt *);
 270 static int      unp_connect(struct socket *, struct sockaddr *,
 271                     struct thread *);
 272 static int      unp_connectat(int, struct socket *, struct sockaddr *,
 273                     struct thread *);
 274 static int      unp_connect2(struct socket *so, struct socket *so2, int);
 275 static void     unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
 276 static void     unp_dispose(struct socket *so);
 277 static void     unp_dispose_mbuf(struct mbuf *);
 278 static void     unp_shutdown(struct unpcb *);
 279 static void     unp_drop(struct unpcb *);
 280 static void     unp_gc(__unused void *, int);
 281 static void     unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
 282 static void     unp_discard(struct file *);
 283 static void     unp_freerights(struct filedescent **, int);
 284 static void     unp_init(void);
 285 static int      unp_internalize(struct mbuf **, struct thread *);
 286 static void     unp_internalize_fp(struct file *);
 287 static int      unp_externalize(struct mbuf *, struct mbuf **, int);
 288 static int      unp_externalize_fp(struct file *);
 289 static struct mbuf      *unp_addsockcred(struct thread *, struct mbuf *);
 290 static void     unp_process_defers(void * __unused, int);
 291
 292 /*
 293  * Definitions of protocols supported in the LOCAL domain.
 294  */
 295 static struct domain localdomain;
 296 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
 297 static struct pr_usrreqs uipc_usrreqs_seqpacket;
 298 static struct protosw localsw[] = {
 299 {
 300         .pr_type =              SOCK_STREAM,
 301         .pr_domain =            &localdomain,
 302         .pr_flags =             PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
 303         .pr_ctloutput =         &uipc_ctloutput,
 304         .pr_usrreqs =           &uipc_usrreqs_stream
 305 },
 306 {
 307         .pr_type =              SOCK_DGRAM,
 308         .pr_domain =            &localdomain,
 309         .pr_flags =             PR_ATOMIC|PR_ADDR|PR_RIGHTS,
 310         .pr_ctloutput =         &uipc_ctloutput,
 311         .pr_usrreqs =           &uipc_usrreqs_dgram
 312 },
 313 {
 314         .pr_type =              SOCK_SEQPACKET,
 315         .pr_domain =            &localdomain,
 316
 317         /*
 318          * XXXRW: For now, PR_ADDR because soreceive will bump into them
 319          * due to our use of sbappendaddr.  A new sbappend variants is needed
 320          * that supports both atomic record writes and control data.
 321          */
 322         .pr_flags =             PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
 323                                     PR_RIGHTS,
 324         .pr_ctloutput =         &uipc_ctloutput,
 325         .pr_usrreqs =           &uipc_usrreqs_seqpacket,
 326 },
 327 };
 328
 329 static struct domain localdomain = {
 330         .dom_family =           AF_LOCAL,
 331         .dom_name =             "local",
 332         .dom_init =             unp_init,
 333         .dom_externalize =      unp_externalize,
 334         .dom_dispose =          unp_dispose,
 335         .dom_protosw =          localsw,
 336         .dom_protoswNPROTOSW =  &localsw[nitems(localsw)]
 337 };
 338 DOMAIN_SET(local);
 339
 340 static void
 341 uipc_abort(struct socket *so)
 342 {
 343         struct unpcb *unp, *unp2;
 344
 345         unp = sotounpcb(so);
 346         KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
 347
 348         UNP_LINK_WLOCK();
 349         UNP_PCB_LOCK(unp);
 350         unp2 = unp->unp_conn;
 351         if (unp2 != NULL) {
 352                 UNP_PCB_LOCK(unp2);
 353                 unp_drop(unp2);
 354                 UNP_PCB_UNLOCK(unp2);
 355         }
 356         UNP_PCB_UNLOCK(unp);
 357         UNP_LINK_WUNLOCK();
 358 }
 359
 360 static int
 361 uipc_accept(struct socket *so, struct sockaddr **nam)
 362 {
 363         struct unpcb *unp, *unp2;
 364         const struct sockaddr *sa;
 365
 366         /*
 367          * Pass back name of connected socket, if it was bound and we are
 368          * still connected (our peer may have closed already!).
 369          */
 370         unp = sotounpcb(so);
 371         KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
 372
 373         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 374         UNP_LINK_RLOCK();
 375         unp2 = unp->unp_conn;
 376         if (unp2 != NULL && unp2->unp_addr != NULL) {
 377                 UNP_PCB_LOCK(unp2);
 378                 sa = (struct sockaddr *) unp2->unp_addr;
 379                 bcopy(sa, *nam, sa->sa_len);
 380                 UNP_PCB_UNLOCK(unp2);
 381         } else {
 382                 sa = &sun_noname;
 383                 bcopy(sa, *nam, sa->sa_len);
 384         }
 385         UNP_LINK_RUNLOCK();
 386         return (0);
 387 }
 388
 389 static int
 390 uipc_attach(struct socket *so, int proto, struct thread *td)
 391 {
 392         u_long sendspace, recvspace;
 393         struct unpcb *unp;
 394         int error;
 395         bool locked;
 396
 397         KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
 398         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
 399                 switch (so->so_type) {
 400                 case SOCK_STREAM:
 401                         sendspace = unpst_sendspace;
 402                         recvspace = unpst_recvspace;
 403                         break;
 404
 405                 case SOCK_DGRAM:
 406                         sendspace = unpdg_sendspace;
 407                         recvspace = unpdg_recvspace;
 408                         break;
 409
 410                 case SOCK_SEQPACKET:
 411                         sendspace = unpsp_sendspace;
 412                         recvspace = unpsp_recvspace;
 413                         break;
 414
 415                 default:
 416                         panic("uipc_attach");
 417                 }
 418                 error = soreserve(so, sendspace, recvspace);
 419                 if (error)
 420                         return (error);
 421         }
 422         unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
 423         if (unp == NULL)
 424                 return (ENOBUFS);
 425         LIST_INIT(&unp->unp_refs);
 426         UNP_PCB_LOCK_INIT(unp);
 427         unp->unp_socket = so;
 428         so->so_pcb = unp;
 429         unp->unp_refcount = 1;
 430         if (so->so_listen != NULL)
 431                 unp->unp_flags |= UNP_NASCENT;
 432
 433         if ((locked = UNP_LINK_WOWNED()) == false)
 434                 UNP_LINK_WLOCK();
 435
 436         unp->unp_gencnt = ++unp_gencnt;
 437         unp_count++;
 438         switch (so->so_type) {
 439         case SOCK_STREAM:
 440                 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
 441                 break;
 442
 443         case SOCK_DGRAM:
 444                 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
 445                 break;
 446
 447         case SOCK_SEQPACKET:
 448                 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
 449                 break;
 450
 451         default:
 452                 panic("uipc_attach");
 453         }
 454
 455         if (locked == false)
 456                 UNP_LINK_WUNLOCK();
 457
 458         return (0);
 459 }
 460
 461 static int
 462 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
 463 {
 464         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
 465         struct vattr vattr;
 466         int error, namelen;
 467         struct nameidata nd;
 468         struct unpcb *unp;
 469         struct vnode *vp;
 470         struct mount *mp;
 471         cap_rights_t rights;
 472         char *buf;
 473
 474         if (nam->sa_family != AF_UNIX)
 475                 return (EAFNOSUPPORT);
 476
 477         unp = sotounpcb(so);
 478         KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
 479
 480         if (soun->sun_len > sizeof(struct sockaddr_un))
 481                 return (EINVAL);
 482         namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
 483         if (namelen <= 0)
 484                 return (EINVAL);
 485
 486         /*
 487          * We don't allow simultaneous bind() calls on a single UNIX domain
 488          * socket, so flag in-progress operations, and return an error if an
 489          * operation is already in progress.
 490          *
 491          * Historically, we have not allowed a socket to be rebound, so this
 492          * also returns an error.  Not allowing re-binding simplifies the
 493          * implementation and avoids a great many possible failure modes.
 494          */
 495         UNP_PCB_LOCK(unp);
 496         if (unp->unp_vnode != NULL) {
 497                 UNP_PCB_UNLOCK(unp);
 498                 return (EINVAL);
 499         }
 500         if (unp->unp_flags & UNP_BINDING) {
 501                 UNP_PCB_UNLOCK(unp);
 502                 return (EALREADY);
 503         }
 504         unp->unp_flags |= UNP_BINDING;
 505         UNP_PCB_UNLOCK(unp);
 506
 507         buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
 508         bcopy(soun->sun_path, buf, namelen);
 509         buf[namelen] = 0;
 510
 511 restart:
 512         NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE,
 513             UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td);
 514 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
 515         error = namei(&nd);
 516         if (error)
 517                 goto error;
 518         vp = nd.ni_vp;
 519         if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
 520                 NDFREE(&nd, NDF_ONLY_PNBUF);
 521                 if (nd.ni_dvp == vp)
 522                         vrele(nd.ni_dvp);
 523                 else
 524                         vput(nd.ni_dvp);
 525                 if (vp != NULL) {
 526                         vrele(vp);
 527                         error = EADDRINUSE;
 528                         goto error;
 529                 }
 530                 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
 531                 if (error)
 532                         goto error;
 533                 goto restart;
 534         }
 535         VATTR_NULL(&vattr);
 536         vattr.va_type = VSOCK;
 537         vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
 538 #ifdef MAC
 539         error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
 540             &vattr);
 541 #endif
 542         if (error == 0)
 543                 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
 544         NDFREE(&nd, NDF_ONLY_PNBUF);
 545         vput(nd.ni_dvp);
 546         if (error) {
 547                 vn_finished_write(mp);
 548                 goto error;
 549         }
 550         vp = nd.ni_vp;
 551         ASSERT_VOP_ELOCKED(vp, "uipc_bind");
 552         soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
 553
 554         UNP_LINK_WLOCK();
 555         UNP_PCB_LOCK(unp);
 556         VOP_UNP_BIND(vp, unp);
 557         unp->unp_vnode = vp;
 558         unp->unp_addr = soun;
 559         unp->unp_flags &= ~UNP_BINDING;
 560         UNP_PCB_UNLOCK(unp);
 561         UNP_LINK_WUNLOCK();
 562         VOP_UNLOCK(vp, 0);
 563         vn_finished_write(mp);
 564         free(buf, M_TEMP);
 565         return (0);
 566
 567 error:
 568         UNP_PCB_LOCK(unp);
 569         unp->unp_flags &= ~UNP_BINDING;
 570         UNP_PCB_UNLOCK(unp);
 571         free(buf, M_TEMP);
 572         return (error);
 573 }
 574
 575 static int
 576 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 577 {
 578
 579         return (uipc_bindat(AT_FDCWD, so, nam, td));
 580 }
 581
 582 static int
 583 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 584 {
 585         int error;
 586
 587         KASSERT(td == curthread, ("uipc_connect: td != curthread"));
 588         UNP_LINK_WLOCK();
 589         error = unp_connect(so, nam, td);
 590         UNP_LINK_WUNLOCK();
 591         return (error);
 592 }
 593
 594 static int
 595 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
 596     struct thread *td)
 597 {
 598         int error;
 599
 600         KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
 601         UNP_LINK_WLOCK();
 602         error = unp_connectat(fd, so, nam, td);
 603         UNP_LINK_WUNLOCK();
 604         return (error);
 605 }
 606
 607 static void
 608 uipc_close(struct socket *so)
 609 {
 610         struct unpcb *unp, *unp2;
 611         struct vnode *vp = NULL;
 612
 613         unp = sotounpcb(so);
 614         KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
 615
 616         UNP_LINK_WLOCK();
 617         UNP_PCB_LOCK(unp);
 618         unp2 = unp->unp_conn;
 619         if (unp2 != NULL) {
 620                 UNP_PCB_LOCK(unp2);
 621                 unp_disconnect(unp, unp2);
 622                 UNP_PCB_UNLOCK(unp2);
 623         }
 624         if (SOLISTENING(so) && ((vp = unp->unp_vnode) != NULL)) {
 625                 VOP_UNP_DETACH(vp);
 626                 unp->unp_vnode = NULL;
 627         }
 628         UNP_PCB_UNLOCK(unp);
 629         UNP_LINK_WUNLOCK();
 630         if (vp)
 631                 vrele(vp);
 632 }
 633
 634 static int
 635 uipc_connect2(struct socket *so1, struct socket *so2)
 636 {
 637         struct unpcb *unp, *unp2;
 638         int error;
 639
 640         UNP_LINK_WLOCK();
 641         unp = so1->so_pcb;
 642         KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
 643         UNP_PCB_LOCK(unp);
 644         unp2 = so2->so_pcb;
 645         KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
 646         UNP_PCB_LOCK(unp2);
 647         error = unp_connect2(so1, so2, PRU_CONNECT2);
 648         UNP_PCB_UNLOCK(unp2);
 649         UNP_PCB_UNLOCK(unp);
 650         UNP_LINK_WUNLOCK();
 651         return (error);
 652 }
 653
 654 static void
 655 uipc_detach(struct socket *so)
 656 {
 657         struct unpcb *unp, *unp2;
 658         struct sockaddr_un *saved_unp_addr;
 659         struct vnode *vp;
 660         int freeunp, local_unp_rights;
 661
 662         unp = sotounpcb(so);
 663         KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
 664
 665         vp = NULL;
 666         local_unp_rights = 0;
 667
 668         UNP_LINK_WLOCK();
 669         LIST_REMOVE(unp, unp_link);
 670         unp->unp_gencnt = ++unp_gencnt;
 671         --unp_count;
 672         UNP_PCB_LOCK(unp);
 673         if ((unp->unp_flags & UNP_NASCENT) != 0)
 674                 goto teardown;
 675
 676         if ((vp = unp->unp_vnode) != NULL) {
 677                 VOP_UNP_DETACH(vp);
 678                 unp->unp_vnode = NULL;
 679         }
 680         unp2 = unp->unp_conn;
 681         if (unp2 != NULL) {
 682                 UNP_PCB_LOCK(unp2);
 683                 unp_disconnect(unp, unp2);
 684                 UNP_PCB_UNLOCK(unp2);
 685         }
 686
 687         /*
 688          * We hold the linkage lock exclusively, so it's OK to acquire
 689          * multiple pcb locks at a time.
 690          */
 691         while (!LIST_EMPTY(&unp->unp_refs)) {
 692                 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
 693
 694                 UNP_PCB_LOCK(ref);
 695                 unp_drop(ref);
 696                 UNP_PCB_UNLOCK(ref);
 697         }
 698         local_unp_rights = unp_rights;
 699 teardown:
 700         UNP_LINK_WUNLOCK();
 701         unp->unp_socket->so_pcb = NULL;
 702         saved_unp_addr = unp->unp_addr;
 703         unp->unp_addr = NULL;
 704         unp->unp_refcount--;
 705         freeunp = (unp->unp_refcount == 0);
 706         if (saved_unp_addr != NULL)
 707                 free(saved_unp_addr, M_SONAME);
 708         if (freeunp) {
 709                 UNP_PCB_LOCK_DESTROY(unp);
 710                 uma_zfree(unp_zone, unp);
 711         } else
 712                 UNP_PCB_UNLOCK(unp);
 713         if (vp)
 714                 vrele(vp);
 715         if (local_unp_rights)
 716                 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
 717 }
 718
 719 static int
 720 uipc_disconnect(struct socket *so)
 721 {
 722         struct unpcb *unp, *unp2;
 723
 724         unp = sotounpcb(so);
 725         KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
 726
 727         UNP_LINK_WLOCK();
 728         UNP_PCB_LOCK(unp);
 729         unp2 = unp->unp_conn;
 730         if (unp2 != NULL) {
 731                 UNP_PCB_LOCK(unp2);
 732                 unp_disconnect(unp, unp2);
 733                 UNP_PCB_UNLOCK(unp2);
 734         }
 735         UNP_PCB_UNLOCK(unp);
 736         UNP_LINK_WUNLOCK();
 737         return (0);
 738 }
 739
 740 static int
 741 uipc_listen(struct socket *so, int backlog, struct thread *td)
 742 {
 743         struct unpcb *unp;
 744         int error;
 745
 746         if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
 747                 return (EOPNOTSUPP);
 748
 749         unp = sotounpcb(so);
 750         KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
 751
 752         UNP_PCB_LOCK(unp);
 753         if (unp->unp_vnode == NULL) {
 754                 /* Already connected or not bound to an address. */
 755                 error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ;
 756                 UNP_PCB_UNLOCK(unp);
 757                 return (error);
 758         }
 759
 760         SOCK_LOCK(so);
 761         error = solisten_proto_check(so);
 762         if (error == 0) {
 763                 cru2x(td->td_ucred, &unp->unp_peercred);
 764                 solisten_proto(so, backlog);
 765         }
 766         SOCK_UNLOCK(so);
 767         UNP_PCB_UNLOCK(unp);
 768         return (error);
 769 }
 770
 771 static int
 772 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
 773 {
 774         struct unpcb *unp, *unp2;
 775         const struct sockaddr *sa;
 776
 777         unp = sotounpcb(so);
 778         KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
 779
 780         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 781         UNP_LINK_RLOCK();
 782         /*
 783          * XXX: It seems that this test always fails even when connection is
 784          * established.  So, this else clause is added as workaround to
 785          * return PF_LOCAL sockaddr.
 786          */
 787         unp2 = unp->unp_conn;
 788         if (unp2 != NULL) {
 789                 UNP_PCB_LOCK(unp2);
 790                 if (unp2->unp_addr != NULL)
 791                         sa = (struct sockaddr *) unp2->unp_addr;
 792                 else
 793                         sa = &sun_noname;
 794                 bcopy(sa, *nam, sa->sa_len);
 795                 UNP_PCB_UNLOCK(unp2);
 796         } else {
 797                 sa = &sun_noname;
 798                 bcopy(sa, *nam, sa->sa_len);
 799         }
 800         UNP_LINK_RUNLOCK();
 801         return (0);
 802 }
 803
 804 static int
 805 uipc_rcvd(struct socket *so, int flags)
 806 {
 807         struct unpcb *unp, *unp2;
 808         struct socket *so2;
 809         u_int mbcnt, sbcc;
 810
 811         unp = sotounpcb(so);
 812         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
 813         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
 814             ("%s: socktype %d", __func__, so->so_type));
 815
 816         /*
 817          * Adjust backpressure on sender and wakeup any waiting to write.
 818          *
 819          * The unp lock is acquired to maintain the validity of the unp_conn
 820          * pointer; no lock on unp2 is required as unp2->unp_socket will be
 821          * static as long as we don't permit unp2 to disconnect from unp,
 822          * which is prevented by the lock on unp.  We cache values from
 823          * so_rcv to avoid holding the so_rcv lock over the entire
 824          * transaction on the remote so_snd.
 825          */
 826         SOCKBUF_LOCK(&so->so_rcv);
 827         mbcnt = so->so_rcv.sb_mbcnt;
 828         sbcc = sbavail(&so->so_rcv);
 829         SOCKBUF_UNLOCK(&so->so_rcv);
 830         /*
 831          * There is a benign race condition at this point.  If we're planning to
 832          * clear SB_STOP, but uipc_send is called on the connected socket at
 833          * this instant, it might add data to the sockbuf and set SB_STOP.  Then
 834          * we would erroneously clear SB_STOP below, even though the sockbuf is
 835          * full.  The race is benign because the only ill effect is to allow the
 836          * sockbuf to exceed its size limit, and the size limits are not
 837          * strictly guaranteed anyway.
 838          */
 839         UNP_PCB_LOCK(unp);
 840         unp2 = unp->unp_conn;
 841         if (unp2 == NULL) {
 842                 UNP_PCB_UNLOCK(unp);
 843                 return (0);
 844         }
 845         so2 = unp2->unp_socket;
 846         SOCKBUF_LOCK(&so2->so_snd);
 847         if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
 848                 so2->so_snd.sb_flags &= ~SB_STOP;
 849         sowwakeup_locked(so2);
 850         UNP_PCB_UNLOCK(unp);
 851         return (0);
 852 }
 853
 854 static int
 855 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
 856     struct mbuf *control, struct thread *td)
 857 {
 858         struct unpcb *unp, *unp2;
 859         struct socket *so2;
 860         u_int mbcnt, sbcc;
 861         int error = 0;
 862
 863         unp = sotounpcb(so);
 864         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
 865         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
 866             so->so_type == SOCK_SEQPACKET,
 867             ("%s: socktype %d", __func__, so->so_type));
 868
 869         if (flags & PRUS_OOB) {
 870                 error = EOPNOTSUPP;
 871                 goto release;
 872         }
 873         if (control != NULL && (error = unp_internalize(&control, td)))
 874                 goto release;
 875         if ((nam != NULL) || (flags & PRUS_EOF))
 876                 UNP_LINK_WLOCK();
 877         else
 878                 UNP_LINK_RLOCK();
 879         switch (so->so_type) {
 880         case SOCK_DGRAM:
 881         {
 882                 const struct sockaddr *from;
 883
 884                 unp2 = unp->unp_conn;
 885                 if (nam != NULL) {
 886                         UNP_LINK_WLOCK_ASSERT();
 887                         if (unp2 != NULL) {
 888                                 error = EISCONN;
 889                                 break;
 890                         }
 891                         error = unp_connect(so, nam, td);
 892                         if (error)
 893                                 break;
 894                         unp2 = unp->unp_conn;
 895                 }
 896
 897                 /*
 898                  * Because connect() and send() are non-atomic in a sendto()
 899                  * with a target address, it's possible that the socket will
 900                  * have disconnected before the send() can run.  In that case
 901                  * return the slightly counter-intuitive but otherwise
 902                  * correct error that the socket is not connected.
 903                  */
 904                 if (unp2 == NULL) {
 905                         error = ENOTCONN;
 906                         break;
 907                 }
 908                 /* Lockless read. */
 909                 if (unp2->unp_flags & UNP_WANTCRED)
 910                         control = unp_addsockcred(td, control);
 911                 UNP_PCB_LOCK(unp);
 912                 if (unp->unp_addr != NULL)
 913                         from = (struct sockaddr *)unp->unp_addr;
 914                 else
 915                         from = &sun_noname;
 916                 so2 = unp2->unp_socket;
 917                 SOCKBUF_LOCK(&so2->so_rcv);
 918                 if (sbappendaddr_locked(&so2->so_rcv, from, m,
 919                     control)) {
 920                         sorwakeup_locked(so2);
 921                         m = NULL;
 922                         control = NULL;
 923                 } else {
 924                         SOCKBUF_UNLOCK(&so2->so_rcv);
 925                         error = ENOBUFS;
 926                 }
 927                 if (nam != NULL) {
 928                         UNP_LINK_WLOCK_ASSERT();
 929                         UNP_PCB_LOCK(unp2);
 930                         unp_disconnect(unp, unp2);
 931                         UNP_PCB_UNLOCK(unp2);
 932                 }
 933                 UNP_PCB_UNLOCK(unp);
 934                 break;
 935         }
 936
 937         case SOCK_SEQPACKET:
 938         case SOCK_STREAM:
 939                 if ((so->so_state & SS_ISCONNECTED) == 0) {
 940                         if (nam != NULL) {
 941                                 UNP_LINK_WLOCK_ASSERT();
 942                                 error = unp_connect(so, nam, td);
 943                                 if (error)
 944                                         break;  /* XXX */
 945                         } else {
 946                                 error = ENOTCONN;
 947                                 break;
 948                         }
 949                 }
 950
 951                 /* Lockless read. */
 952                 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
 953                         error = EPIPE;
 954                         break;
 955                 }
 956
 957                 /*
 958                  * Because connect() and send() are non-atomic in a sendto()
 959                  * with a target address, it's possible that the socket will
 960                  * have disconnected before the send() can run.  In that case
 961                  * return the slightly counter-intuitive but otherwise
 962                  * correct error that the socket is not connected.
 963                  *
 964                  * Locking here must be done carefully: the linkage lock
 965                  * prevents interconnections between unpcbs from changing, so
 966                  * we can traverse from unp to unp2 without acquiring unp's
 967                  * lock.  Socket buffer locks follow unpcb locks, so we can
 968                  * acquire both remote and lock socket buffer locks.
 969                  */
 970                 unp2 = unp->unp_conn;
 971                 if (unp2 == NULL) {
 972                         error = ENOTCONN;
 973                         break;
 974                 }
 975                 so2 = unp2->unp_socket;
 976                 UNP_PCB_LOCK(unp2);
 977                 SOCKBUF_LOCK(&so2->so_rcv);
 978                 if (unp2->unp_flags & UNP_WANTCRED) {
 979                         /*
 980                          * Credentials are passed only once on SOCK_STREAM
 981                          * and SOCK_SEQPACKET.
 982                          */
 983                         unp2->unp_flags &= ~UNP_WANTCRED;
 984                         control = unp_addsockcred(td, control);
 985                 }
 986                 /*
 987                  * Send to paired receive port, and then reduce send buffer
 988                  * hiwater marks to maintain backpressure.  Wake up readers.
 989                  */
 990                 switch (so->so_type) {
 991                 case SOCK_STREAM:
 992                         if (control != NULL) {
 993                                 if (sbappendcontrol_locked(&so2->so_rcv, m,
 994                                     control))
 995                                         control = NULL;
 996                         } else
 997                                 sbappend_locked(&so2->so_rcv, m, flags);
 998                         break;
 999
1000                 case SOCK_SEQPACKET: {
1001                         const struct sockaddr *from;
1002
1003                         from = &sun_noname;
1004                         /*
1005                          * Don't check for space available in so2->so_rcv.
1006                          * Unix domain sockets only check for space in the
1007                          * sending sockbuf, and that check is performed one
1008                          * level up the stack.
1009                          */
1010                         if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1011                                 from, m, control))
1012                                 control = NULL;
1013                         break;
1014                         }
1015                 }
1016
1017                 mbcnt = so2->so_rcv.sb_mbcnt;
1018                 sbcc = sbavail(&so2->so_rcv);
1019                 if (sbcc)
1020                         sorwakeup_locked(so2);
1021                 else
1022                         SOCKBUF_UNLOCK(&so2->so_rcv);
1023
1024                 /*
1025                  * The PCB lock on unp2 protects the SB_STOP flag.  Without it,
1026                  * it would be possible for uipc_rcvd to be called at this
1027                  * point, drain the receiving sockbuf, clear SB_STOP, and then
1028                  * we would set SB_STOP below.  That could lead to an empty
1029                  * sockbuf having SB_STOP set
1030                  */
1031                 SOCKBUF_LOCK(&so->so_snd);
1032                 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1033                         so->so_snd.sb_flags |= SB_STOP;
1034                 SOCKBUF_UNLOCK(&so->so_snd);
1035                 UNP_PCB_UNLOCK(unp2);
1036                 m = NULL;
1037                 break;
1038         }
1039
1040         /*
1041          * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1042          */
1043         if (flags & PRUS_EOF) {
1044                 UNP_PCB_LOCK(unp);
1045                 socantsendmore(so);
1046                 unp_shutdown(unp);
1047                 UNP_PCB_UNLOCK(unp);
1048         }
1049
1050         if ((nam != NULL) || (flags & PRUS_EOF))
1051                 UNP_LINK_WUNLOCK();
1052         else
1053                 UNP_LINK_RUNLOCK();
1054
1055         if (control != NULL && error != 0)
1056                 unp_dispose_mbuf(control);
1057
1058 release:
1059         if (control != NULL)
1060                 m_freem(control);
1061         /*
1062          * In case of PRUS_NOTREADY, uipc_ready() is responsible
1063          * for freeing memory.
1064          */
1065         if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1066                 m_freem(m);
1067         return (error);
1068 }
1069
1070 static int
1071 uipc_ready(struct socket *so, struct mbuf *m, int count)
1072 {
1073         struct unpcb *unp, *unp2;
1074         struct socket *so2;
1075         int error;
1076
1077         unp = sotounpcb(so);
1078
1079         UNP_LINK_RLOCK();
1080         if ((unp2 = unp->unp_conn) == NULL) {
1081                 UNP_LINK_RUNLOCK();
1082                 for (int i = 0; i < count; i++)
1083                         m = m_free(m);
1084                 return (ECONNRESET);
1085         }
1086         UNP_PCB_LOCK(unp2);
1087         so2 = unp2->unp_socket;
1088
1089         SOCKBUF_LOCK(&so2->so_rcv);
1090         if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1091                 sorwakeup_locked(so2);
1092         else
1093                 SOCKBUF_UNLOCK(&so2->so_rcv);
1094
1095         UNP_PCB_UNLOCK(unp2);
1096         UNP_LINK_RUNLOCK();
1097
1098         return (error);
1099 }
1100
1101 static int
1102 uipc_sense(struct socket *so, struct stat *sb)
1103 {
1104         struct unpcb *unp;
1105
1106         unp = sotounpcb(so);
1107         KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1108
1109         sb->st_blksize = so->so_snd.sb_hiwat;
1110         UNP_PCB_LOCK(unp);
1111         sb->st_dev = NODEV;
1112         if (unp->unp_ino == 0)
1113                 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1114         sb->st_ino = unp->unp_ino;
1115         UNP_PCB_UNLOCK(unp);
1116         return (0);
1117 }
1118
1119 static int
1120 uipc_shutdown(struct socket *so)
1121 {
1122         struct unpcb *unp;
1123
1124         unp = sotounpcb(so);
1125         KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1126
1127         UNP_LINK_WLOCK();
1128         UNP_PCB_LOCK(unp);
1129         socantsendmore(so);
1130         unp_shutdown(unp);
1131         UNP_PCB_UNLOCK(unp);
1132         UNP_LINK_WUNLOCK();
1133         return (0);
1134 }
1135
1136 static int
1137 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1138 {
1139         struct unpcb *unp;
1140         const struct sockaddr *sa;
1141
1142         unp = sotounpcb(so);
1143         KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1144
1145         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1146         UNP_PCB_LOCK(unp);
1147         if (unp->unp_addr != NULL)
1148                 sa = (struct sockaddr *) unp->unp_addr;
1149         else
1150                 sa = &sun_noname;
1151         bcopy(sa, *nam, sa->sa_len);
1152         UNP_PCB_UNLOCK(unp);
1153         return (0);
1154 }
1155
1156 static struct pr_usrreqs uipc_usrreqs_dgram = {
1157         .pru_abort =            uipc_abort,
1158         .pru_accept =           uipc_accept,
1159         .pru_attach =           uipc_attach,
1160         .pru_bind =             uipc_bind,
1161         .pru_bindat =           uipc_bindat,
1162         .pru_connect =          uipc_connect,
1163         .pru_connectat =        uipc_connectat,
1164         .pru_connect2 =         uipc_connect2,
1165         .pru_detach =           uipc_detach,
1166         .pru_disconnect =       uipc_disconnect,
1167         .pru_listen =           uipc_listen,
1168         .pru_peeraddr =         uipc_peeraddr,
1169         .pru_rcvd =             uipc_rcvd,
1170         .pru_send =             uipc_send,
1171         .pru_sense =            uipc_sense,
1172         .pru_shutdown =         uipc_shutdown,
1173         .pru_sockaddr =         uipc_sockaddr,
1174         .pru_soreceive =        soreceive_dgram,
1175         .pru_close =            uipc_close,
1176 };
1177
1178 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1179         .pru_abort =            uipc_abort,
1180         .pru_accept =           uipc_accept,
1181         .pru_attach =           uipc_attach,
1182         .pru_bind =             uipc_bind,
1183         .pru_bindat =           uipc_bindat,
1184         .pru_connect =          uipc_connect,
1185         .pru_connectat =        uipc_connectat,
1186         .pru_connect2 =         uipc_connect2,
1187         .pru_detach =           uipc_detach,
1188         .pru_disconnect =       uipc_disconnect,
1189         .pru_listen =           uipc_listen,
1190         .pru_peeraddr =         uipc_peeraddr,
1191         .pru_rcvd =             uipc_rcvd,
1192         .pru_send =             uipc_send,
1193         .pru_sense =            uipc_sense,
1194         .pru_shutdown =         uipc_shutdown,
1195         .pru_sockaddr =         uipc_sockaddr,
1196         .pru_soreceive =        soreceive_generic,      /* XXX: or...? */
1197         .pru_close =            uipc_close,
1198 };
1199
1200 static struct pr_usrreqs uipc_usrreqs_stream = {
1201         .pru_abort =            uipc_abort,
1202         .pru_accept =           uipc_accept,
1203         .pru_attach =           uipc_attach,
1204         .pru_bind =             uipc_bind,
1205         .pru_bindat =           uipc_bindat,
1206         .pru_connect =          uipc_connect,
1207         .pru_connectat =        uipc_connectat,
1208         .pru_connect2 =         uipc_connect2,
1209         .pru_detach =           uipc_detach,
1210         .pru_disconnect =       uipc_disconnect,
1211         .pru_listen =           uipc_listen,
1212         .pru_peeraddr =         uipc_peeraddr,
1213         .pru_rcvd =             uipc_rcvd,
1214         .pru_send =             uipc_send,
1215         .pru_ready =            uipc_ready,
1216         .pru_sense =            uipc_sense,
1217         .pru_shutdown =         uipc_shutdown,
1218         .pru_sockaddr =         uipc_sockaddr,
1219         .pru_soreceive =        soreceive_generic,
1220         .pru_close =            uipc_close,
1221 };
1222
1223 static int
1224 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1225 {
1226         struct unpcb *unp;
1227         struct xucred xu;
1228         int error, optval;
1229
1230         if (sopt->sopt_level != 0)
1231                 return (EINVAL);
1232
1233         unp = sotounpcb(so);
1234         KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1235         error = 0;
1236         switch (sopt->sopt_dir) {
1237         case SOPT_GET:
1238                 switch (sopt->sopt_name) {
1239                 case LOCAL_PEERCRED:
1240                         UNP_PCB_LOCK(unp);
1241                         if (unp->unp_flags & UNP_HAVEPC)
1242                                 xu = unp->unp_peercred;
1243                         else {
1244                                 if (so->so_type == SOCK_STREAM)
1245                                         error = ENOTCONN;
1246                                 else
1247                                         error = EINVAL;
1248                         }
1249                         UNP_PCB_UNLOCK(unp);
1250                         if (error == 0)
1251                                 error = sooptcopyout(sopt, &xu, sizeof(xu));
1252                         break;
1253
1254                 case LOCAL_CREDS:
1255                         /* Unlocked read. */
1256                         optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1257                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1258                         break;
1259
1260                 case LOCAL_CONNWAIT:
1261                         /* Unlocked read. */
1262                         optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1263                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1264                         break;
1265
1266                 default:
1267                         error = EOPNOTSUPP;
1268                         break;
1269                 }
1270                 break;
1271
1272         case SOPT_SET:
1273                 switch (sopt->sopt_name) {
1274                 case LOCAL_CREDS:
1275                 case LOCAL_CONNWAIT:
1276                         error = sooptcopyin(sopt, &optval, sizeof(optval),
1277                                             sizeof(optval));
1278                         if (error)
1279                                 break;
1280
1281 #define OPTSET(bit) do {                                                \
1282         UNP_PCB_LOCK(unp);                                              \
1283         if (optval)                                                     \
1284                 unp->unp_flags |= bit;                                  \
1285         else                                                            \
1286                 unp->unp_flags &= ~bit;                                 \
1287         UNP_PCB_UNLOCK(unp);                                            \
1288 } while (0)
1289
1290                         switch (sopt->sopt_name) {
1291                         case LOCAL_CREDS:
1292                                 OPTSET(UNP_WANTCRED);
1293                                 break;
1294
1295                         case LOCAL_CONNWAIT:
1296                                 OPTSET(UNP_CONNWAIT);
1297                                 break;
1298
1299                         default:
1300                                 break;
1301                         }
1302                         break;
1303 #undef  OPTSET
1304                 default:
1305                         error = ENOPROTOOPT;
1306                         break;
1307                 }
1308                 break;
1309
1310         default:
1311                 error = EOPNOTSUPP;
1312                 break;
1313         }
1314         return (error);
1315 }
1316
1317 static int
1318 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1319 {
1320
1321         return (unp_connectat(AT_FDCWD, so, nam, td));
1322 }
1323
1324 static int
1325 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1326     struct thread *td)
1327 {
1328         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1329         struct vnode *vp;
1330         struct socket *so2;
1331         struct unpcb *unp, *unp2, *unp3;
1332         struct nameidata nd;
1333         char buf[SOCK_MAXADDRLEN];
1334         struct sockaddr *sa;
1335         cap_rights_t rights;
1336         int error, len;
1337
1338         if (nam->sa_family != AF_UNIX)
1339                 return (EAFNOSUPPORT);
1340
1341         UNP_LINK_WLOCK_ASSERT();
1342
1343         unp = sotounpcb(so);
1344         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1345
1346         if (nam->sa_len > sizeof(struct sockaddr_un))
1347                 return (EINVAL);
1348         len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1349         if (len <= 0)
1350                 return (EINVAL);
1351         bcopy(soun->sun_path, buf, len);
1352         buf[len] = 0;
1353
1354         UNP_PCB_LOCK(unp);
1355         if (unp->unp_flags & UNP_CONNECTING) {
1356                 UNP_PCB_UNLOCK(unp);
1357                 return (EALREADY);
1358         }
1359         UNP_LINK_WUNLOCK();
1360         unp->unp_flags |= UNP_CONNECTING;
1361         UNP_PCB_UNLOCK(unp);
1362
1363         sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1364         NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1365             UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1366         error = namei(&nd);
1367         if (error)
1368                 vp = NULL;
1369         else
1370                 vp = nd.ni_vp;
1371         ASSERT_VOP_LOCKED(vp, "unp_connect");
1372         NDFREE(&nd, NDF_ONLY_PNBUF);
1373         if (error)
1374                 goto bad;
1375
1376         if (vp->v_type != VSOCK) {
1377                 error = ENOTSOCK;
1378                 goto bad;
1379         }
1380 #ifdef MAC
1381         error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1382         if (error)
1383                 goto bad;
1384 #endif
1385         error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1386         if (error)
1387                 goto bad;
1388
1389         unp = sotounpcb(so);
1390         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1391
1392         /*
1393          * Lock linkage lock for two reasons: make sure v_socket is stable,
1394          * and to protect simultaneous locking of multiple pcbs.
1395          */
1396         UNP_LINK_WLOCK();
1397         VOP_UNP_CONNECT(vp, &unp2);
1398         if (unp2 == NULL) {
1399                 error = ECONNREFUSED;
1400                 goto bad2;
1401         }
1402         so2 = unp2->unp_socket;
1403         if (so->so_type != so2->so_type) {
1404                 error = EPROTOTYPE;
1405                 goto bad2;
1406         }
1407         UNP_PCB_LOCK(unp);
1408         UNP_PCB_LOCK(unp2);
1409         if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1410                 if (so2->so_options & SO_ACCEPTCONN) {
1411                         CURVNET_SET(so2->so_vnet);
1412                         so2 = sonewconn(so2, 0);
1413                         CURVNET_RESTORE();
1414                 } else
1415                         so2 = NULL;
1416                 if (so2 == NULL) {
1417                         error = ECONNREFUSED;
1418                         goto bad3;
1419                 }
1420                 unp3 = sotounpcb(so2);
1421                 UNP_PCB_LOCK(unp3);
1422                 if (unp2->unp_addr != NULL) {
1423                         bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1424                         unp3->unp_addr = (struct sockaddr_un *) sa;
1425                         sa = NULL;
1426                 }
1427
1428                 /*
1429                  * The connector's (client's) credentials are copied from its
1430                  * process structure at the time of connect() (which is now).
1431                  */
1432                 cru2x(td->td_ucred, &unp3->unp_peercred);
1433                 unp3->unp_flags |= UNP_HAVEPC;
1434
1435                 /*
1436                  * The receiver's (server's) credentials are copied from the
1437                  * unp_peercred member of socket on which the former called
1438                  * listen(); uipc_listen() cached that process's credentials
1439                  * at that time so we can use them now.
1440                  */
1441                 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1442                     sizeof(unp->unp_peercred));
1443                 unp->unp_flags |= UNP_HAVEPC;
1444                 if (unp2->unp_flags & UNP_WANTCRED)
1445                         unp3->unp_flags |= UNP_WANTCRED;
1446                 UNP_PCB_UNLOCK(unp2);
1447                 unp2 = unp3;
1448 #ifdef MAC
1449                 mac_socketpeer_set_from_socket(so, so2);
1450                 mac_socketpeer_set_from_socket(so2, so);
1451 #endif
1452         }
1453
1454         KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 &&
1455             sotounpcb(so2) == unp2,
1456             ("%s: unp2 %p so2 %p", __func__, unp2, so2));
1457         error = unp_connect2(so, so2, PRU_CONNECT);
1458 bad3:
1459         UNP_PCB_UNLOCK(unp2);
1460         UNP_PCB_UNLOCK(unp);
1461 bad2:
1462         UNP_LINK_WUNLOCK();
1463 bad:
1464         if (vp != NULL)
1465                 vput(vp);
1466         free(sa, M_SONAME);
1467         UNP_LINK_WLOCK();
1468         UNP_PCB_LOCK(unp);
1469         unp->unp_flags &= ~UNP_CONNECTING;
1470         UNP_PCB_UNLOCK(unp);
1471         return (error);
1472 }
1473
1474 static int
1475 unp_connect2(struct socket *so, struct socket *so2, int req)
1476 {
1477         struct unpcb *unp;
1478         struct unpcb *unp2;
1479
1480         unp = sotounpcb(so);
1481         KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1482         unp2 = sotounpcb(so2);
1483         KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1484
1485         UNP_LINK_WLOCK_ASSERT();
1486         UNP_PCB_LOCK_ASSERT(unp);
1487         UNP_PCB_LOCK_ASSERT(unp2);
1488
1489         if (so2->so_type != so->so_type)
1490                 return (EPROTOTYPE);
1491         unp2->unp_flags &= ~UNP_NASCENT;
1492         unp->unp_conn = unp2;
1493
1494         switch (so->so_type) {
1495         case SOCK_DGRAM:
1496                 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1497                 soisconnected(so);
1498                 break;
1499
1500         case SOCK_STREAM:
1501         case SOCK_SEQPACKET:
1502                 unp2->unp_conn = unp;
1503                 if (req == PRU_CONNECT &&
1504                     ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1505                         soisconnecting(so);
1506                 else
1507                         soisconnected(so);
1508                 soisconnected(so2);
1509                 break;
1510
1511         default:
1512                 panic("unp_connect2");
1513         }
1514         return (0);
1515 }
1516
1517 static void
1518 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1519 {
1520         struct socket *so;
1521
1522         KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1523
1524         UNP_LINK_WLOCK_ASSERT();
1525         UNP_PCB_LOCK_ASSERT(unp);
1526         UNP_PCB_LOCK_ASSERT(unp2);
1527
1528         unp->unp_conn = NULL;
1529         switch (unp->unp_socket->so_type) {
1530         case SOCK_DGRAM:
1531                 LIST_REMOVE(unp, unp_reflink);
1532                 so = unp->unp_socket;
1533                 SOCK_LOCK(so);
1534                 so->so_state &= ~SS_ISCONNECTED;
1535                 SOCK_UNLOCK(so);
1536                 break;
1537
1538         case SOCK_STREAM:
1539         case SOCK_SEQPACKET:
1540                 soisdisconnected(unp->unp_socket);
1541                 unp2->unp_conn = NULL;
1542                 soisdisconnected(unp2->unp_socket);
1543                 break;
1544         }
1545 }
1546
1547 /*
1548  * unp_pcblist() walks the global list of struct unpcb's to generate a
1549  * pointer list, bumping the refcount on each unpcb.  It then copies them out
1550  * sequentially, validating the generation number on each to see if it has
1551  * been detached.  All of this is necessary because copyout() may sleep on
1552  * disk I/O.
1553  */
1554 static int
1555 unp_pcblist(SYSCTL_HANDLER_ARGS)
1556 {
1557         struct unpcb *unp, **unp_list;
1558         unp_gen_t gencnt;
1559         struct xunpgen *xug;
1560         struct unp_head *head;
1561         struct xunpcb *xu;
1562         u_int i;
1563         int error, freeunp, n;
1564
1565         switch ((intptr_t)arg1) {
1566         case SOCK_STREAM:
1567                 head = &unp_shead;
1568                 break;
1569
1570         case SOCK_DGRAM:
1571                 head = &unp_dhead;
1572                 break;
1573
1574         case SOCK_SEQPACKET:
1575                 head = &unp_sphead;
1576                 break;
1577
1578         default:
1579                 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1580         }
1581
1582         /*
1583          * The process of preparing the PCB list is too time-consuming and
1584          * resource-intensive to repeat twice on every request.
1585          */
1586         if (req->oldptr == NULL) {
1587                 n = unp_count;
1588                 req->oldidx = 2 * (sizeof *xug)
1589                         + (n + n/8) * sizeof(struct xunpcb);
1590                 return (0);
1591         }
1592
1593         if (req->newptr != NULL)
1594                 return (EPERM);
1595
1596         /*
1597          * OK, now we're committed to doing something.
1598          */
1599         xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1600         UNP_LINK_RLOCK();
1601         gencnt = unp_gencnt;
1602         n = unp_count;
1603         UNP_LINK_RUNLOCK();
1604
1605         xug->xug_len = sizeof *xug;
1606         xug->xug_count = n;
1607         xug->xug_gen = gencnt;
1608         xug->xug_sogen = so_gencnt;
1609         error = SYSCTL_OUT(req, xug, sizeof *xug);
1610         if (error) {
1611                 free(xug, M_TEMP);
1612                 return (error);
1613         }
1614
1615         unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1616
1617         UNP_LINK_RLOCK();
1618         for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1619              unp = LIST_NEXT(unp, unp_link)) {
1620                 UNP_PCB_LOCK(unp);
1621                 if (unp->unp_gencnt <= gencnt) {
1622                         if (cr_cansee(req->td->td_ucred,
1623                             unp->unp_socket->so_cred)) {
1624                                 UNP_PCB_UNLOCK(unp);
1625                                 continue;
1626                         }
1627                         unp_list[i++] = unp;
1628                         unp->unp_refcount++;
1629                 }
1630                 UNP_PCB_UNLOCK(unp);
1631         }
1632         UNP_LINK_RUNLOCK();
1633         n = i;                  /* In case we lost some during malloc. */
1634
1635         error = 0;
1636         xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1637         for (i = 0; i < n; i++) {
1638                 unp = unp_list[i];
1639                 UNP_PCB_LOCK(unp);
1640                 unp->unp_refcount--;
1641                 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1642                         xu->xu_len = sizeof *xu;
1643                         xu->xu_unpp = unp;
1644                         /*
1645                          * XXX - need more locking here to protect against
1646                          * connect/disconnect races for SMP.
1647                          */
1648                         if (unp->unp_addr != NULL)
1649                                 bcopy(unp->unp_addr, &xu->xu_addr,
1650                                       unp->unp_addr->sun_len);
1651                         else
1652                                 bzero(&xu->xu_addr, sizeof(xu->xu_addr));
1653                         if (unp->unp_conn != NULL &&
1654                             unp->unp_conn->unp_addr != NULL)
1655                                 bcopy(unp->unp_conn->unp_addr,
1656                                       &xu->xu_caddr,
1657                                       unp->unp_conn->unp_addr->sun_len);
1658                         else
1659                                 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr));
1660                         xu->unp_vnode = unp->unp_vnode;
1661                         xu->unp_conn = unp->unp_conn;
1662                         xu->xu_firstref = LIST_FIRST(&unp->unp_refs);
1663                         xu->xu_nextref = LIST_NEXT(unp, unp_reflink);
1664                         xu->unp_gencnt = unp->unp_gencnt;
1665                         sotoxsocket(unp->unp_socket, &xu->xu_socket);
1666                         UNP_PCB_UNLOCK(unp);
1667                         error = SYSCTL_OUT(req, xu, sizeof *xu);
1668                 } else {
1669                         freeunp = (unp->unp_refcount == 0);
1670                         UNP_PCB_UNLOCK(unp);
1671                         if (freeunp) {
1672                                 UNP_PCB_LOCK_DESTROY(unp);
1673                                 uma_zfree(unp_zone, unp);
1674                         }
1675                 }
1676         }
1677         free(xu, M_TEMP);
1678         if (!error) {
1679                 /*
1680                  * Give the user an updated idea of our state.  If the
1681                  * generation differs from what we told her before, she knows
1682                  * that something happened while we were processing this
1683                  * request, and it might be necessary to retry.
1684                  */
1685                 xug->xug_gen = unp_gencnt;
1686                 xug->xug_sogen = so_gencnt;
1687                 xug->xug_count = unp_count;
1688                 error = SYSCTL_OUT(req, xug, sizeof *xug);
1689         }
1690         free(unp_list, M_TEMP);
1691         free(xug, M_TEMP);
1692         return (error);
1693 }
1694
1695 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1696     (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1697     "List of active local datagram sockets");
1698 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1699     (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1700     "List of active local stream sockets");
1701 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1702     CTLTYPE_OPAQUE | CTLFLAG_RD,
1703     (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1704     "List of active local seqpacket sockets");
1705
1706 static void
1707 unp_shutdown(struct unpcb *unp)
1708 {
1709         struct unpcb *unp2;
1710         struct socket *so;
1711
1712         UNP_LINK_WLOCK_ASSERT();
1713         UNP_PCB_LOCK_ASSERT(unp);
1714
1715         unp2 = unp->unp_conn;
1716         if ((unp->unp_socket->so_type == SOCK_STREAM ||
1717             (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1718                 so = unp2->unp_socket;
1719                 if (so != NULL)
1720                         socantrcvmore(so);
1721         }
1722 }
1723
1724 static void
1725 unp_drop(struct unpcb *unp)
1726 {
1727         struct socket *so = unp->unp_socket;
1728         struct unpcb *unp2;
1729
1730         UNP_LINK_WLOCK_ASSERT();
1731         UNP_PCB_LOCK_ASSERT(unp);
1732
1733         /*
1734          * Regardless of whether the socket's peer dropped the connection
1735          * with this socket by aborting or disconnecting, POSIX requires
1736          * that ECONNRESET is returned.
1737          */
1738         so->so_error = ECONNRESET;
1739         unp2 = unp->unp_conn;
1740         if (unp2 == NULL)
1741                 return;
1742         UNP_PCB_LOCK(unp2);
1743         unp_disconnect(unp, unp2);
1744         UNP_PCB_UNLOCK(unp2);
1745 }
1746
1747 static void
1748 unp_freerights(struct filedescent **fdep, int fdcount)
1749 {
1750         struct file *fp;
1751         int i;
1752
1753         KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1754
1755         for (i = 0; i < fdcount; i++) {
1756                 fp = fdep[i]->fde_file;
1757                 filecaps_free(&fdep[i]->fde_caps);
1758                 unp_discard(fp);
1759         }
1760         free(fdep[0], M_FILECAPS);
1761 }
1762
1763 static int
1764 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1765 {
1766         struct thread *td = curthread;          /* XXX */
1767         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1768         int i;
1769         int *fdp;
1770         struct filedesc *fdesc = td->td_proc->p_fd;
1771         struct filedescent **fdep;
1772         void *data;
1773         socklen_t clen = control->m_len, datalen;
1774         int error, newfds;
1775         u_int newlen;
1776
1777         UNP_LINK_UNLOCK_ASSERT();
1778
1779         error = 0;
1780         if (controlp != NULL) /* controlp == NULL => free control messages */
1781                 *controlp = NULL;
1782         while (cm != NULL) {
1783                 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1784                         error = EINVAL;
1785                         break;
1786                 }
1787                 data = CMSG_DATA(cm);
1788                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1789                 if (cm->cmsg_level == SOL_SOCKET
1790                     && cm->cmsg_type == SCM_RIGHTS) {
1791                         newfds = datalen / sizeof(*fdep);
1792                         if (newfds == 0)
1793                                 goto next;
1794                         fdep = data;
1795
1796                         /* If we're not outputting the descriptors free them. */
1797                         if (error || controlp == NULL) {
1798                                 unp_freerights(fdep, newfds);
1799                                 goto next;
1800                         }
1801                         FILEDESC_XLOCK(fdesc);
1802
1803                         /*
1804                          * Now change each pointer to an fd in the global
1805                          * table to an integer that is the index to the local
1806                          * fd table entry that we set up to point to the
1807                          * global one we are transferring.
1808                          */
1809                         newlen = newfds * sizeof(int);
1810                         *controlp = sbcreatecontrol(NULL, newlen,
1811                             SCM_RIGHTS, SOL_SOCKET);
1812                         if (*controlp == NULL) {
1813                                 FILEDESC_XUNLOCK(fdesc);
1814                                 error = E2BIG;
1815                                 unp_freerights(fdep, newfds);
1816                                 goto next;
1817                         }
1818
1819                         fdp = (int *)
1820                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1821                         if (fdallocn(td, 0, fdp, newfds) != 0) {
1822                                 FILEDESC_XUNLOCK(fdesc);
1823                                 error = EMSGSIZE;
1824                                 unp_freerights(fdep, newfds);
1825                                 m_freem(*controlp);
1826                                 *controlp = NULL;
1827                                 goto next;
1828                         }
1829                         for (i = 0; i < newfds; i++, fdp++) {
1830                                 _finstall(fdesc, fdep[i]->fde_file, *fdp,
1831                                     (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0,
1832                                     &fdep[i]->fde_caps);
1833                                 unp_externalize_fp(fdep[i]->fde_file);
1834                         }
1835                         FILEDESC_XUNLOCK(fdesc);
1836                         free(fdep[0], M_FILECAPS);
1837                 } else {
1838                         /* We can just copy anything else across. */
1839                         if (error || controlp == NULL)
1840                                 goto next;
1841                         *controlp = sbcreatecontrol(NULL, datalen,
1842                             cm->cmsg_type, cm->cmsg_level);
1843                         if (*controlp == NULL) {
1844                                 error = ENOBUFS;
1845                                 goto next;
1846                         }
1847                         bcopy(data,
1848                             CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1849                             datalen);
1850                 }
1851                 controlp = &(*controlp)->m_next;
1852
1853 next:
1854                 if (CMSG_SPACE(datalen) < clen) {
1855                         clen -= CMSG_SPACE(datalen);
1856                         cm = (struct cmsghdr *)
1857                             ((caddr_t)cm + CMSG_SPACE(datalen));
1858                 } else {
1859                         clen = 0;
1860                         cm = NULL;
1861                 }
1862         }
1863
1864         m_freem(control);
1865         return (error);
1866 }
1867
1868 static void
1869 unp_zone_change(void *tag)
1870 {
1871
1872         uma_zone_set_max(unp_zone, maxsockets);
1873 }
1874
1875 static void
1876 unp_init(void)
1877 {
1878
1879 #ifdef VIMAGE
1880         if (!IS_DEFAULT_VNET(curvnet))
1881                 return;
1882 #endif
1883         unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1884             NULL, NULL, UMA_ALIGN_PTR, 0);
1885         if (unp_zone == NULL)
1886                 panic("unp_init");
1887         uma_zone_set_max(unp_zone, maxsockets);
1888         uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
1889         EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1890             NULL, EVENTHANDLER_PRI_ANY);
1891         LIST_INIT(&unp_dhead);
1892         LIST_INIT(&unp_shead);
1893         LIST_INIT(&unp_sphead);
1894         SLIST_INIT(&unp_defers);
1895         TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1896         TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1897         UNP_LINK_LOCK_INIT();
1898         UNP_DEFERRED_LOCK_INIT();
1899 }
1900
1901 static int
1902 unp_internalize(struct mbuf **controlp, struct thread *td)
1903 {
1904         struct mbuf *control = *controlp;
1905         struct proc *p = td->td_proc;
1906         struct filedesc *fdesc = p->p_fd;
1907         struct bintime *bt;
1908         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1909         struct cmsgcred *cmcred;
1910         struct filedescent *fde, **fdep, *fdev;
1911         struct file *fp;
1912         struct timeval *tv;
1913         struct timespec *ts;
1914         int i, *fdp;
1915         void *data;
1916         socklen_t clen = control->m_len, datalen;
1917         int error, oldfds;
1918         u_int newlen;
1919
1920         UNP_LINK_UNLOCK_ASSERT();
1921
1922         error = 0;
1923         *controlp = NULL;
1924         while (cm != NULL) {
1925                 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1926                     || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
1927                         error = EINVAL;
1928                         goto out;
1929                 }
1930                 data = CMSG_DATA(cm);
1931                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1932
1933                 switch (cm->cmsg_type) {
1934                 /*
1935                  * Fill in credential information.
1936                  */
1937                 case SCM_CREDS:
1938                         *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1939                             SCM_CREDS, SOL_SOCKET);
1940                         if (*controlp == NULL) {
1941                                 error = ENOBUFS;
1942                                 goto out;
1943                         }
1944                         cmcred = (struct cmsgcred *)
1945                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1946                         cmcred->cmcred_pid = p->p_pid;
1947                         cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1948                         cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1949                         cmcred->cmcred_euid = td->td_ucred->cr_uid;
1950                         cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1951                             CMGROUP_MAX);
1952                         for (i = 0; i < cmcred->cmcred_ngroups; i++)
1953                                 cmcred->cmcred_groups[i] =
1954                                     td->td_ucred->cr_groups[i];
1955                         break;
1956
1957                 case SCM_RIGHTS:
1958                         oldfds = datalen / sizeof (int);
1959                         if (oldfds == 0)
1960                                 break;
1961                         /*
1962                          * Check that all the FDs passed in refer to legal
1963                          * files.  If not, reject the entire operation.
1964                          */
1965                         fdp = data;
1966                         FILEDESC_SLOCK(fdesc);
1967                         for (i = 0; i < oldfds; i++, fdp++) {
1968                                 fp = fget_locked(fdesc, *fdp);
1969                                 if (fp == NULL) {
1970                                         FILEDESC_SUNLOCK(fdesc);
1971                                         error = EBADF;
1972                                         goto out;
1973                                 }
1974                                 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1975                                         FILEDESC_SUNLOCK(fdesc);
1976                                         error = EOPNOTSUPP;
1977                                         goto out;
1978                                 }
1979
1980                         }
1981
1982                         /*
1983                          * Now replace the integer FDs with pointers to the
1984                          * file structure and capability rights.
1985                          */
1986                         newlen = oldfds * sizeof(fdep[0]);
1987                         *controlp = sbcreatecontrol(NULL, newlen,
1988                             SCM_RIGHTS, SOL_SOCKET);
1989                         if (*controlp == NULL) {
1990                                 FILEDESC_SUNLOCK(fdesc);
1991                                 error = E2BIG;
1992                                 goto out;
1993                         }
1994                         fdp = data;
1995                         fdep = (struct filedescent **)
1996                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1997                         fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
1998                             M_WAITOK);
1999                         for (i = 0; i < oldfds; i++, fdev++, fdp++) {
2000                                 fde = &fdesc->fd_ofiles[*fdp];
2001                                 fdep[i] = fdev;
2002                                 fdep[i]->fde_file = fde->fde_file;
2003                                 filecaps_copy(&fde->fde_caps,
2004                                     &fdep[i]->fde_caps, true);
2005                                 unp_internalize_fp(fdep[i]->fde_file);
2006                         }
2007                         FILEDESC_SUNLOCK(fdesc);
2008                         break;
2009
2010                 case SCM_TIMESTAMP:
2011                         *controlp = sbcreatecontrol(NULL, sizeof(*tv),
2012                             SCM_TIMESTAMP, SOL_SOCKET);
2013                         if (*controlp == NULL) {
2014                                 error = ENOBUFS;
2015                                 goto out;
2016                         }
2017                         tv = (struct timeval *)
2018                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2019                         microtime(tv);
2020                         break;
2021
2022                 case SCM_BINTIME:
2023                         *controlp = sbcreatecontrol(NULL, sizeof(*bt),
2024                             SCM_BINTIME, SOL_SOCKET);
2025                         if (*controlp == NULL) {
2026                                 error = ENOBUFS;
2027                                 goto out;
2028                         }
2029                         bt = (struct bintime *)
2030                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2031                         bintime(bt);
2032                         break;
2033
2034                 case SCM_REALTIME:
2035                         *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2036                             SCM_REALTIME, SOL_SOCKET);
2037                         if (*controlp == NULL) {
2038                                 error = ENOBUFS;
2039                                 goto out;
2040                         }
2041                         ts = (struct timespec *)
2042                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2043                         nanotime(ts);
2044                         break;
2045
2046                 case SCM_MONOTONIC:
2047                         *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2048                             SCM_MONOTONIC, SOL_SOCKET);
2049                         if (*controlp == NULL) {
2050                                 error = ENOBUFS;
2051                                 goto out;
2052                         }
2053                         ts = (struct timespec *)
2054                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2055                         nanouptime(ts);
2056                         break;
2057
2058                 default:
2059                         error = EINVAL;
2060                         goto out;
2061                 }
2062
2063                 controlp = &(*controlp)->m_next;
2064                 if (CMSG_SPACE(datalen) < clen) {
2065                         clen -= CMSG_SPACE(datalen);
2066                         cm = (struct cmsghdr *)
2067                             ((caddr_t)cm + CMSG_SPACE(datalen));
2068                 } else {
2069                         clen = 0;
2070                         cm = NULL;
2071                 }
2072         }
2073
2074 out:
2075         m_freem(control);
2076         return (error);
2077 }
2078
2079 static struct mbuf *
2080 unp_addsockcred(struct thread *td, struct mbuf *control)
2081 {
2082         struct mbuf *m, *n, *n_prev;
2083         struct sockcred *sc;
2084         const struct cmsghdr *cm;
2085         int ngroups;
2086         int i;
2087
2088         ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2089         m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2090         if (m == NULL)
2091                 return (control);
2092
2093         sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2094         sc->sc_uid = td->td_ucred->cr_ruid;
2095         sc->sc_euid = td->td_ucred->cr_uid;
2096         sc->sc_gid = td->td_ucred->cr_rgid;
2097         sc->sc_egid = td->td_ucred->cr_gid;
2098         sc->sc_ngroups = ngroups;
2099         for (i = 0; i < sc->sc_ngroups; i++)
2100                 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2101
2102         /*
2103          * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2104          * created SCM_CREDS control message (struct sockcred) has another
2105          * format.
2106          */
2107         if (control != NULL)
2108                 for (n = control, n_prev = NULL; n != NULL;) {
2109                         cm = mtod(n, struct cmsghdr *);
2110                         if (cm->cmsg_level == SOL_SOCKET &&
2111                             cm->cmsg_type == SCM_CREDS) {
2112                                 if (n_prev == NULL)
2113                                         control = n->m_next;
2114                                 else
2115                                         n_prev->m_next = n->m_next;
2116                                 n = m_free(n);
2117                         } else {
2118                                 n_prev = n;
2119                                 n = n->m_next;
2120                         }
2121                 }
2122
2123         /* Prepend it to the head. */
2124         m->m_next = control;
2125         return (m);
2126 }
2127
2128 static struct unpcb *
2129 fptounp(struct file *fp)
2130 {
2131         struct socket *so;
2132
2133         if (fp->f_type != DTYPE_SOCKET)
2134                 return (NULL);
2135         if ((so = fp->f_data) == NULL)
2136                 return (NULL);
2137         if (so->so_proto->pr_domain != &localdomain)
2138                 return (NULL);
2139         return sotounpcb(so);
2140 }
2141
2142 static void
2143 unp_discard(struct file *fp)
2144 {
2145         struct unp_defer *dr;
2146
2147         if (unp_externalize_fp(fp)) {
2148                 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2149                 dr->ud_fp = fp;
2150                 UNP_DEFERRED_LOCK();
2151                 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2152                 UNP_DEFERRED_UNLOCK();
2153                 atomic_add_int(&unp_defers_count, 1);
2154                 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2155         } else
2156                 (void) closef(fp, (struct thread *)NULL);
2157 }
2158
2159 static void
2160 unp_process_defers(void *arg __unused, int pending)
2161 {
2162         struct unp_defer *dr;
2163         SLIST_HEAD(, unp_defer) drl;
2164         int count;
2165
2166         SLIST_INIT(&drl);
2167         for (;;) {
2168                 UNP_DEFERRED_LOCK();
2169                 if (SLIST_FIRST(&unp_defers) == NULL) {
2170                         UNP_DEFERRED_UNLOCK();
2171                         break;
2172                 }
2173                 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2174                 UNP_DEFERRED_UNLOCK();
2175                 count = 0;
2176                 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2177                         SLIST_REMOVE_HEAD(&drl, ud_link);
2178                         closef(dr->ud_fp, NULL);
2179                         free(dr, M_TEMP);
2180                         count++;
2181                 }
2182                 atomic_add_int(&unp_defers_count, -count);
2183         }
2184 }
2185
2186 static void
2187 unp_internalize_fp(struct file *fp)
2188 {
2189         struct unpcb *unp;
2190
2191         UNP_LINK_WLOCK();
2192         if ((unp = fptounp(fp)) != NULL) {
2193                 unp->unp_file = fp;
2194                 unp->unp_msgcount++;
2195         }
2196         fhold(fp);
2197         unp_rights++;
2198         UNP_LINK_WUNLOCK();
2199 }
2200
2201 static int
2202 unp_externalize_fp(struct file *fp)
2203 {
2204         struct unpcb *unp;
2205         int ret;
2206
2207         UNP_LINK_WLOCK();
2208         if ((unp = fptounp(fp)) != NULL) {
2209                 unp->unp_msgcount--;
2210                 ret = 1;
2211         } else
2212                 ret = 0;
2213         unp_rights--;
2214         UNP_LINK_WUNLOCK();
2215         return (ret);
2216 }
2217
2218 /*
2219  * unp_defer indicates whether additional work has been defered for a future
2220  * pass through unp_gc().  It is thread local and does not require explicit
2221  * synchronization.
2222  */
2223 static int      unp_marked;
2224 static int      unp_unreachable;
2225
2226 static void
2227 unp_accessable(struct filedescent **fdep, int fdcount)
2228 {
2229         struct unpcb *unp;
2230         struct file *fp;
2231         int i;
2232
2233         for (i = 0; i < fdcount; i++) {
2234                 fp = fdep[i]->fde_file;
2235                 if ((unp = fptounp(fp)) == NULL)
2236                         continue;
2237                 if (unp->unp_gcflag & UNPGC_REF)
2238                         continue;
2239                 unp->unp_gcflag &= ~UNPGC_DEAD;
2240                 unp->unp_gcflag |= UNPGC_REF;
2241                 unp_marked++;
2242         }
2243 }
2244
2245 static void
2246 unp_gc_process(struct unpcb *unp)
2247 {
2248         struct socket *so, *soa;
2249         struct file *fp;
2250
2251         /* Already processed. */
2252         if (unp->unp_gcflag & UNPGC_SCANNED)
2253                 return;
2254         fp = unp->unp_file;
2255
2256         /*
2257          * Check for a socket potentially in a cycle.  It must be in a
2258          * queue as indicated by msgcount, and this must equal the file
2259          * reference count.  Note that when msgcount is 0 the file is NULL.
2260          */
2261         if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2262             unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2263                 unp->unp_gcflag |= UNPGC_DEAD;
2264                 unp_unreachable++;
2265                 return;
2266         }
2267
2268         so = unp->unp_socket;
2269         SOCK_LOCK(so);
2270         if (SOLISTENING(so)) {
2271                 /*
2272                  * Mark all sockets in our accept queue.
2273                  */
2274                 TAILQ_FOREACH(soa, &so->sol_comp, so_list) {
2275                         if (sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS)
2276                                 continue;
2277                         SOCKBUF_LOCK(&soa->so_rcv);
2278                         unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2279                         SOCKBUF_UNLOCK(&soa->so_rcv);
2280                 }
2281         } else {
2282                 /*
2283                  * Mark all sockets we reference with RIGHTS.
2284                  */
2285                 if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2286                         SOCKBUF_LOCK(&so->so_rcv);
2287                         unp_scan(so->so_rcv.sb_mb, unp_accessable);
2288                         SOCKBUF_UNLOCK(&so->so_rcv);
2289                 }
2290         }
2291         SOCK_UNLOCK(so);
2292         unp->unp_gcflag |= UNPGC_SCANNED;
2293 }
2294
2295 static int unp_recycled;
2296 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2297     "Number of unreachable sockets claimed by the garbage collector.");
2298
2299 static int unp_taskcount;
2300 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2301     "Number of times the garbage collector has run.");
2302
2303 static void
2304 unp_gc(__unused void *arg, int pending)
2305 {
2306         struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2307                                     NULL };
2308         struct unp_head **head;
2309         struct file *f, **unref;
2310         struct unpcb *unp;
2311         int i, total;
2312
2313         unp_taskcount++;
2314         UNP_LINK_RLOCK();
2315         /*
2316          * First clear all gc flags from previous runs, apart from
2317          * UNPGC_IGNORE_RIGHTS.
2318          */
2319         for (head = heads; *head != NULL; head++)
2320                 LIST_FOREACH(unp, *head, unp_link)
2321                         unp->unp_gcflag =
2322                             (unp->unp_gcflag & UNPGC_IGNORE_RIGHTS);
2323
2324         /*
2325          * Scan marking all reachable sockets with UNPGC_REF.  Once a socket
2326          * is reachable all of the sockets it references are reachable.
2327          * Stop the scan once we do a complete loop without discovering
2328          * a new reachable socket.
2329          */
2330         do {
2331                 unp_unreachable = 0;
2332                 unp_marked = 0;
2333                 for (head = heads; *head != NULL; head++)
2334                         LIST_FOREACH(unp, *head, unp_link)
2335                                 unp_gc_process(unp);
2336         } while (unp_marked);
2337         UNP_LINK_RUNLOCK();
2338         if (unp_unreachable == 0)
2339                 return;
2340
2341         /*
2342          * Allocate space for a local list of dead unpcbs.
2343          */
2344         unref = malloc(unp_unreachable * sizeof(struct file *),
2345             M_TEMP, M_WAITOK);
2346
2347         /*
2348          * Iterate looking for sockets which have been specifically marked
2349          * as as unreachable and store them locally.
2350          */
2351         UNP_LINK_RLOCK();
2352         for (total = 0, head = heads; *head != NULL; head++)
2353                 LIST_FOREACH(unp, *head, unp_link)
2354                         if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2355                                 f = unp->unp_file;
2356                                 if (unp->unp_msgcount == 0 || f == NULL ||
2357                                     f->f_count != unp->unp_msgcount)
2358                                         continue;
2359                                 unref[total++] = f;
2360                                 fhold(f);
2361                                 KASSERT(total <= unp_unreachable,
2362                                     ("unp_gc: incorrect unreachable count."));
2363                         }
2364         UNP_LINK_RUNLOCK();
2365
2366         /*
2367          * Now flush all sockets, free'ing rights.  This will free the
2368          * struct files associated with these sockets but leave each socket
2369          * with one remaining ref.
2370          */
2371         for (i = 0; i < total; i++) {
2372                 struct socket *so;
2373
2374                 so = unref[i]->f_data;
2375                 CURVNET_SET(so->so_vnet);
2376                 sorflush(so);
2377                 CURVNET_RESTORE();
2378         }
2379
2380         /*
2381          * And finally release the sockets so they can be reclaimed.
2382          */
2383         for (i = 0; i < total; i++)
2384                 fdrop(unref[i], NULL);
2385         unp_recycled += total;
2386         free(unref, M_TEMP);
2387 }
2388
2389 static void
2390 unp_dispose_mbuf(struct mbuf *m)
2391 {
2392
2393         if (m)
2394                 unp_scan(m, unp_freerights);
2395 }
2396
2397 /*
2398  * Synchronize against unp_gc, which can trip over data as we are freeing it.
2399  */
2400 static void
2401 unp_dispose(struct socket *so)
2402 {
2403         struct unpcb *unp;
2404
2405         unp = sotounpcb(so);
2406         UNP_LINK_WLOCK();
2407         unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2408         UNP_LINK_WUNLOCK();
2409         if (!SOLISTENING(so))
2410                 unp_dispose_mbuf(so->so_rcv.sb_mb);
2411 }
2412
2413 static void
2414 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2415 {
2416         struct mbuf *m;
2417         struct cmsghdr *cm;
2418         void *data;
2419         socklen_t clen, datalen;
2420
2421         while (m0 != NULL) {
2422                 for (m = m0; m; m = m->m_next) {
2423                         if (m->m_type != MT_CONTROL)
2424                                 continue;
2425
2426                         cm = mtod(m, struct cmsghdr *);
2427                         clen = m->m_len;
2428
2429                         while (cm != NULL) {
2430                                 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2431                                         break;
2432
2433                                 data = CMSG_DATA(cm);
2434                                 datalen = (caddr_t)cm + cm->cmsg_len
2435                                     - (caddr_t)data;
2436
2437                                 if (cm->cmsg_level == SOL_SOCKET &&
2438                                     cm->cmsg_type == SCM_RIGHTS) {
2439                                         (*op)(data, datalen /
2440                                             sizeof(struct filedescent *));
2441                                 }
2442
2443                                 if (CMSG_SPACE(datalen) < clen) {
2444                                         clen -= CMSG_SPACE(datalen);
2445                                         cm = (struct cmsghdr *)
2446                                             ((caddr_t)cm + CMSG_SPACE(datalen));
2447                                 } else {
2448                                         clen = 0;
2449                                         cm = NULL;
2450                                 }
2451                         }
2452                 }
2453                 m0 = m0->m_nextpkt;
2454         }
2455 }
2456
2457 /*
2458  * A helper function called by VFS before socket-type vnode reclamation.
2459  * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2460  * use count.
2461  */
2462 void
2463 vfs_unp_reclaim(struct vnode *vp)
2464 {
2465         struct unpcb *unp;
2466         int active;
2467
2468         ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2469         KASSERT(vp->v_type == VSOCK,
2470             ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2471
2472         active = 0;
2473         UNP_LINK_WLOCK();
2474         VOP_UNP_CONNECT(vp, &unp);
2475         if (unp == NULL)
2476                 goto done;
2477         UNP_PCB_LOCK(unp);
2478         if (unp->unp_vnode == vp) {
2479                 VOP_UNP_DETACH(vp);
2480                 unp->unp_vnode = NULL;
2481                 active = 1;
2482         }
2483         UNP_PCB_UNLOCK(unp);
2484 done:
2485         UNP_LINK_WUNLOCK();
2486         if (active)
2487                 vunref(vp);
2488 }
2489
2490 #ifdef DDB
2491 static void
2492 db_print_indent(int indent)
2493 {
2494         int i;
2495
2496         for (i = 0; i < indent; i++)
2497                 db_printf(" ");
2498 }
2499
2500 static void
2501 db_print_unpflags(int unp_flags)
2502 {
2503         int comma;
2504
2505         comma = 0;
2506         if (unp_flags & UNP_HAVEPC) {
2507                 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2508                 comma = 1;
2509         }
2510         if (unp_flags & UNP_WANTCRED) {
2511                 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2512                 comma = 1;
2513         }
2514         if (unp_flags & UNP_CONNWAIT) {
2515                 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2516                 comma = 1;
2517         }
2518         if (unp_flags & UNP_CONNECTING) {
2519                 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2520                 comma = 1;
2521         }
2522         if (unp_flags & UNP_BINDING) {
2523                 db_printf("%sUNP_BINDING", comma ? ", " : "");
2524                 comma = 1;
2525         }
2526 }
2527
2528 static void
2529 db_print_xucred(int indent, struct xucred *xu)
2530 {
2531         int comma, i;
2532
2533         db_print_indent(indent);
2534         db_printf("cr_version: %u   cr_uid: %u   cr_ngroups: %d\n",
2535             xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2536         db_print_indent(indent);
2537         db_printf("cr_groups: ");
2538         comma = 0;
2539         for (i = 0; i < xu->cr_ngroups; i++) {
2540                 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2541                 comma = 1;
2542         }
2543         db_printf("\n");
2544 }
2545
2546 static void
2547 db_print_unprefs(int indent, struct unp_head *uh)
2548 {
2549         struct unpcb *unp;
2550         int counter;
2551
2552         counter = 0;
2553         LIST_FOREACH(unp, uh, unp_reflink) {
2554                 if (counter % 4 == 0)
2555                         db_print_indent(indent);
2556                 db_printf("%p  ", unp);
2557                 if (counter % 4 == 3)
2558                         db_printf("\n");
2559                 counter++;
2560         }
2561         if (counter != 0 && counter % 4 != 0)
2562                 db_printf("\n");
2563 }
2564
2565 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2566 {
2567         struct unpcb *unp;
2568
2569         if (!have_addr) {
2570                 db_printf("usage: show unpcb <addr>\n");
2571                 return;
2572         }
2573         unp = (struct unpcb *)addr;
2574
2575         db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
2576             unp->unp_vnode);
2577
2578         db_printf("unp_ino: %ju   unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2579             unp->unp_conn);
2580
2581         db_printf("unp_refs:\n");
2582         db_print_unprefs(2, &unp->unp_refs);
2583
2584         /* XXXRW: Would be nice to print the full address, if any. */
2585         db_printf("unp_addr: %p\n", unp->unp_addr);
2586
2587         db_printf("unp_gencnt: %llu\n",
2588             (unsigned long long)unp->unp_gencnt);
2589
2590         db_printf("unp_flags: %x (", unp->unp_flags);
2591         db_print_unpflags(unp->unp_flags);
2592         db_printf(")\n");
2593
2594         db_printf("unp_peercred:\n");
2595         db_print_xucred(2, &unp->unp_peercred);
2596
2597         db_printf("unp_refcount: %u\n", unp->unp_refcount);
2598 }
2599 #endif