sys/kern/uipc_usrreq.c

   1 /*-
   2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
   3  *      The Regents of the University of California.
   4  * Copyright (c) 2004-2009 Robert N. M. Watson
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer.
  12  * 2. Redistributions in binary form must reproduce the above copyright
  13  *    notice, this list of conditions and the following disclaimer in the
  14  *    documentation and/or other materials provided with the distribution.
  15  * 3. Neither the name of the University nor the names of its contributors
  16  *    may be used to endorse or promote products derived from this software
  17  *    without specific prior written permission.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  29  * SUCH DAMAGE.
  30  *
  31  *      From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
  32  */
  33
  34 /*
  35  * UNIX Domain (Local) Sockets
  36  *
  37  * This is an implementation of UNIX (local) domain sockets.  Each socket has
  38  * an associated struct unpcb (UNIX protocol control block).  Stream sockets
  39  * may be connected to 0 or 1 other socket.  Datagram sockets may be
  40  * connected to 0, 1, or many other sockets.  Sockets may be created and
  41  * connected in pairs (socketpair(2)), or bound/connected to using the file
  42  * system name space.  For most purposes, only the receive socket buffer is
  43  * used, as sending on one socket delivers directly to the receive socket
  44  * buffer of a second socket.
  45  *
  46  * The implementation is substantially complicated by the fact that
  47  * "ancillary data", such as file descriptors or credentials, may be passed
  48  * across UNIX domain sockets.  The potential for passing UNIX domain sockets
  49  * over other UNIX domain sockets requires the implementation of a simple
  50  * garbage collector to find and tear down cycles of disconnected sockets.
  51  *
  52  * TODO:
  53  *      RDM
  54  *      rethink name space problems
  55  *      need a proper out-of-band
  56  */
  57
  58 #include <sys/cdefs.h>
  59 __FBSDID("$FreeBSD$");
  60
  61 #include "opt_ddb.h"
  62
  63 #include <sys/param.h>
  64 #include <sys/capsicum.h>
  65 #include <sys/domain.h>
  66 #include <sys/fcntl.h>
  67 #include <sys/malloc.h>         /* XXX must be before <sys/file.h> */
  68 #include <sys/eventhandler.h>
  69 #include <sys/file.h>
  70 #include <sys/filedesc.h>
  71 #include <sys/kernel.h>
  72 #include <sys/lock.h>
  73 #include <sys/mbuf.h>
  74 #include <sys/mount.h>
  75 #include <sys/mutex.h>
  76 #include <sys/namei.h>
  77 #include <sys/proc.h>
  78 #include <sys/protosw.h>
  79 #include <sys/queue.h>
  80 #include <sys/resourcevar.h>
  81 #include <sys/rwlock.h>
  82 #include <sys/socket.h>
  83 #include <sys/socketvar.h>
  84 #include <sys/signalvar.h>
  85 #include <sys/stat.h>
  86 #include <sys/sx.h>
  87 #include <sys/sysctl.h>
  88 #include <sys/systm.h>
  89 #include <sys/taskqueue.h>
  90 #include <sys/un.h>
  91 #include <sys/unpcb.h>
  92 #include <sys/vnode.h>
  93
  94 #include <net/vnet.h>
  95
  96 #ifdef DDB
  97 #include <ddb/ddb.h>
  98 #endif
  99
 100 #include <security/mac/mac_framework.h>
 101
 102 #include <vm/uma.h>
 103
 104 MALLOC_DECLARE(M_FILECAPS);
 105
 106 /*
 107  * Locking key:
 108  * (l)  Locked using list lock
 109  * (g)  Locked using linkage lock
 110  */
 111
 112 static uma_zone_t       unp_zone;
 113 static unp_gen_t        unp_gencnt;     /* (l) */
 114 static u_int            unp_count;      /* (l) Count of local sockets. */
 115 static ino_t            unp_ino;        /* Prototype for fake inode numbers. */
 116 static int              unp_rights;     /* (g) File descriptors in flight. */
 117 static struct unp_head  unp_shead;      /* (l) List of stream sockets. */
 118 static struct unp_head  unp_dhead;      /* (l) List of datagram sockets. */
 119 static struct unp_head  unp_sphead;     /* (l) List of seqpacket sockets. */
 120
 121 struct unp_defer {
 122         SLIST_ENTRY(unp_defer) ud_link;
 123         struct file *ud_fp;
 124 };
 125 static SLIST_HEAD(, unp_defer) unp_defers;
 126 static int unp_defers_count;
 127
 128 static const struct sockaddr    sun_noname = { sizeof(sun_noname), AF_LOCAL };
 129
 130 /*
 131  * Garbage collection of cyclic file descriptor/socket references occurs
 132  * asynchronously in a taskqueue context in order to avoid recursion and
 133  * reentrance in the UNIX domain socket, file descriptor, and socket layer
 134  * code.  See unp_gc() for a full description.
 135  */
 136 static struct timeout_task unp_gc_task;
 137
 138 /*
 139  * The close of unix domain sockets attached as SCM_RIGHTS is
 140  * postponed to the taskqueue, to avoid arbitrary recursion depth.
 141  * The attached sockets might have another sockets attached.
 142  */
 143 static struct task      unp_defer_task;
 144
 145 /*
 146  * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
 147  * stream sockets, although the total for sender and receiver is actually
 148  * only PIPSIZ.
 149  *
 150  * Datagram sockets really use the sendspace as the maximum datagram size,
 151  * and don't really want to reserve the sendspace.  Their recvspace should be
 152  * large enough for at least one max-size datagram plus address.
 153  */
 154 #ifndef PIPSIZ
 155 #define PIPSIZ  8192
 156 #endif
 157 static u_long   unpst_sendspace = PIPSIZ;
 158 static u_long   unpst_recvspace = PIPSIZ;
 159 static u_long   unpdg_sendspace = 2*1024;       /* really max datagram size */
 160 static u_long   unpdg_recvspace = 4*1024;
 161 static u_long   unpsp_sendspace = PIPSIZ;       /* really max datagram size */
 162 static u_long   unpsp_recvspace = PIPSIZ;
 163
 164 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
 165 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0,
 166     "SOCK_STREAM");
 167 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
 168 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0,
 169     "SOCK_SEQPACKET");
 170
 171 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
 172            &unpst_sendspace, 0, "Default stream send space.");
 173 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
 174            &unpst_recvspace, 0, "Default stream receive space.");
 175 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
 176            &unpdg_sendspace, 0, "Default datagram send space.");
 177 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
 178            &unpdg_recvspace, 0, "Default datagram receive space.");
 179 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
 180            &unpsp_sendspace, 0, "Default seqpacket send space.");
 181 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
 182            &unpsp_recvspace, 0, "Default seqpacket receive space.");
 183 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
 184     "File descriptors in flight.");
 185 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
 186     &unp_defers_count, 0,
 187     "File descriptors deferred to taskqueue for close.");
 188
 189 /*
 190  * Locking and synchronization:
 191  *
 192  * Three types of locks exit in the local domain socket implementation: a
 193  * global list mutex, a global linkage rwlock, and per-unpcb mutexes.  Of the
 194  * global locks, the list lock protects the socket count, global generation
 195  * number, and stream/datagram global lists.  The linkage lock protects the
 196  * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be
 197  * held exclusively over the acquisition of multiple unpcb locks to prevent
 198  * deadlock.
 199  *
 200  * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
 201  * allocated in pru_attach() and freed in pru_detach().  The validity of that
 202  * pointer is an invariant, so no lock is required to dereference the so_pcb
 203  * pointer if a valid socket reference is held by the caller.  In practice,
 204  * this is always true during operations performed on a socket.  Each unpcb
 205  * has a back-pointer to its socket, unp_socket, which will be stable under
 206  * the same circumstances.
 207  *
 208  * This pointer may only be safely dereferenced as long as a valid reference
 209  * to the unpcb is held.  Typically, this reference will be from the socket,
 210  * or from another unpcb when the referring unpcb's lock is held (in order
 211  * that the reference not be invalidated during use).  For example, to follow
 212  * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
 213  * as unp_socket remains valid as long as the reference to unp_conn is valid.
 214  *
 215  * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx.  Individual
 216  * atomic reads without the lock may be performed "lockless", but more
 217  * complex reads and read-modify-writes require the mutex to be held.  No
 218  * lock order is defined between unpcb locks -- multiple unpcb locks may be
 219  * acquired at the same time only when holding the linkage rwlock
 220  * exclusively, which prevents deadlocks.
 221  *
 222  * Blocking with UNIX domain sockets is a tricky issue: unlike most network
 223  * protocols, bind() is a non-atomic operation, and connect() requires
 224  * potential sleeping in the protocol, due to potentially waiting on local or
 225  * distributed file systems.  We try to separate "lookup" operations, which
 226  * may sleep, and the IPC operations themselves, which typically can occur
 227  * with relative atomicity as locks can be held over the entire operation.
 228  *
 229  * Another tricky issue is simultaneous multi-threaded or multi-process
 230  * access to a single UNIX domain socket.  These are handled by the flags
 231  * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
 232  * binding, both of which involve dropping UNIX domain socket locks in order
 233  * to perform namei() and other file system operations.
 234  */
 235 static struct rwlock    unp_link_rwlock;
 236 static struct mtx       unp_list_lock;
 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
 254 #define UNP_LIST_LOCK_INIT()            mtx_init(&unp_list_lock,        \
 255                                             "unp_list_lock", NULL, MTX_DEF)
 256 #define UNP_LIST_LOCK()                 mtx_lock(&unp_list_lock)
 257 #define UNP_LIST_UNLOCK()               mtx_unlock(&unp_list_lock)
 258
 259 #define UNP_DEFERRED_LOCK_INIT()        mtx_init(&unp_defers_lock, \
 260                                             "unp_defer", NULL, MTX_DEF)
 261 #define UNP_DEFERRED_LOCK()             mtx_lock(&unp_defers_lock)
 262 #define UNP_DEFERRED_UNLOCK()           mtx_unlock(&unp_defers_lock)
 263
 264 #define UNP_PCB_LOCK_INIT(unp)          mtx_init(&(unp)->unp_mtx,       \
 265                                             "unp_mtx", "unp_mtx",       \
 266                                             MTX_DUPOK|MTX_DEF|MTX_RECURSE)
 267 #define UNP_PCB_LOCK_DESTROY(unp)       mtx_destroy(&(unp)->unp_mtx)
 268 #define UNP_PCB_LOCK(unp)               mtx_lock(&(unp)->unp_mtx)
 269 #define UNP_PCB_UNLOCK(unp)             mtx_unlock(&(unp)->unp_mtx)
 270 #define UNP_PCB_LOCK_ASSERT(unp)        mtx_assert(&(unp)->unp_mtx, MA_OWNED)
 271
 272 static int      uipc_connect2(struct socket *, struct socket *);
 273 static int      uipc_ctloutput(struct socket *, struct sockopt *);
 274 static int      unp_connect(struct socket *, struct sockaddr *,
 275                     struct thread *);
 276 static int      unp_connectat(int, struct socket *, struct sockaddr *,
 277                     struct thread *);
 278 static int      unp_connect2(struct socket *so, struct socket *so2, int);
 279 static void     unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
 280 static void     unp_dispose(struct socket *so);
 281 static void     unp_dispose_mbuf(struct mbuf *);
 282 static void     unp_shutdown(struct unpcb *);
 283 static void     unp_drop(struct unpcb *);
 284 static void     unp_gc(__unused void *, int);
 285 static void     unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
 286 static void     unp_discard(struct file *);
 287 static void     unp_freerights(struct filedescent **, int);
 288 static void     unp_init(void);
 289 static int      unp_internalize(struct mbuf **, struct thread *);
 290 static void     unp_internalize_fp(struct file *);
 291 static int      unp_externalize(struct mbuf *, struct mbuf **, int);
 292 static int      unp_externalize_fp(struct file *);
 293 static struct mbuf      *unp_addsockcred(struct thread *, struct mbuf *);
 294 static void     unp_process_defers(void * __unused, int);
 295
 296 /*
 297  * Definitions of protocols supported in the LOCAL domain.
 298  */
 299 static struct domain localdomain;
 300 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
 301 static struct pr_usrreqs uipc_usrreqs_seqpacket;
 302 static struct protosw localsw[] = {
 303 {
 304         .pr_type =              SOCK_STREAM,
 305         .pr_domain =            &localdomain,
 306         .pr_flags =             PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
 307         .pr_ctloutput =         &uipc_ctloutput,
 308         .pr_usrreqs =           &uipc_usrreqs_stream
 309 },
 310 {
 311         .pr_type =              SOCK_DGRAM,
 312         .pr_domain =            &localdomain,
 313         .pr_flags =             PR_ATOMIC|PR_ADDR|PR_RIGHTS,
 314         .pr_ctloutput =         &uipc_ctloutput,
 315         .pr_usrreqs =           &uipc_usrreqs_dgram
 316 },
 317 {
 318         .pr_type =              SOCK_SEQPACKET,
 319         .pr_domain =            &localdomain,
 320
 321         /*
 322          * XXXRW: For now, PR_ADDR because soreceive will bump into them
 323          * due to our use of sbappendaddr.  A new sbappend variants is needed
 324          * that supports both atomic record writes and control data.
 325          */
 326         .pr_flags =             PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
 327                                     PR_RIGHTS,
 328         .pr_ctloutput =         &uipc_ctloutput,
 329         .pr_usrreqs =           &uipc_usrreqs_seqpacket,
 330 },
 331 };
 332
 333 static struct domain localdomain = {
 334         .dom_family =           AF_LOCAL,
 335         .dom_name =             "local",
 336         .dom_init =             unp_init,
 337         .dom_externalize =      unp_externalize,
 338         .dom_dispose =          unp_dispose,
 339         .dom_protosw =          localsw,
 340         .dom_protoswNPROTOSW =  &localsw[nitems(localsw)]
 341 };
 342 DOMAIN_SET(local);
 343
 344 static void
 345 uipc_abort(struct socket *so)
 346 {
 347         struct unpcb *unp, *unp2;
 348
 349         unp = sotounpcb(so);
 350         KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
 351
 352         UNP_LINK_WLOCK();
 353         UNP_PCB_LOCK(unp);
 354         unp2 = unp->unp_conn;
 355         if (unp2 != NULL) {
 356                 UNP_PCB_LOCK(unp2);
 357                 unp_drop(unp2);
 358                 UNP_PCB_UNLOCK(unp2);
 359         }
 360         UNP_PCB_UNLOCK(unp);
 361         UNP_LINK_WUNLOCK();
 362 }
 363
 364 static int
 365 uipc_accept(struct socket *so, struct sockaddr **nam)
 366 {
 367         struct unpcb *unp, *unp2;
 368         const struct sockaddr *sa;
 369
 370         /*
 371          * Pass back name of connected socket, if it was bound and we are
 372          * still connected (our peer may have closed already!).
 373          */
 374         unp = sotounpcb(so);
 375         KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
 376
 377         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 378         UNP_LINK_RLOCK();
 379         unp2 = unp->unp_conn;
 380         if (unp2 != NULL && unp2->unp_addr != NULL) {
 381                 UNP_PCB_LOCK(unp2);
 382                 sa = (struct sockaddr *) unp2->unp_addr;
 383                 bcopy(sa, *nam, sa->sa_len);
 384                 UNP_PCB_UNLOCK(unp2);
 385         } else {
 386                 sa = &sun_noname;
 387                 bcopy(sa, *nam, sa->sa_len);
 388         }
 389         UNP_LINK_RUNLOCK();
 390         return (0);
 391 }
 392
 393 static int
 394 uipc_attach(struct socket *so, int proto, struct thread *td)
 395 {
 396         u_long sendspace, recvspace;
 397         struct unpcb *unp;
 398         int error;
 399
 400         KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
 401         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
 402                 switch (so->so_type) {
 403                 case SOCK_STREAM:
 404                         sendspace = unpst_sendspace;
 405                         recvspace = unpst_recvspace;
 406                         break;
 407
 408                 case SOCK_DGRAM:
 409                         sendspace = unpdg_sendspace;
 410                         recvspace = unpdg_recvspace;
 411                         break;
 412
 413                 case SOCK_SEQPACKET:
 414                         sendspace = unpsp_sendspace;
 415                         recvspace = unpsp_recvspace;
 416                         break;
 417
 418                 default:
 419                         panic("uipc_attach");
 420                 }
 421                 error = soreserve(so, sendspace, recvspace);
 422                 if (error)
 423                         return (error);
 424         }
 425         unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
 426         if (unp == NULL)
 427                 return (ENOBUFS);
 428         LIST_INIT(&unp->unp_refs);
 429         UNP_PCB_LOCK_INIT(unp);
 430         unp->unp_socket = so;
 431         so->so_pcb = unp;
 432         unp->unp_refcount = 1;
 433         if (so->so_head != NULL)
 434                 unp->unp_flags |= UNP_NASCENT;
 435
 436         UNP_LIST_LOCK();
 437         unp->unp_gencnt = ++unp_gencnt;
 438         unp_count++;
 439         switch (so->so_type) {
 440         case SOCK_STREAM:
 441                 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
 442                 break;
 443
 444         case SOCK_DGRAM:
 445                 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
 446                 break;
 447
 448         case SOCK_SEQPACKET:
 449                 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
 450                 break;
 451
 452         default:
 453                 panic("uipc_attach");
 454         }
 455         UNP_LIST_UNLOCK();
 456
 457         return (0);
 458 }
 459
 460 static int
 461 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
 462 {
 463         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
 464         struct vattr vattr;
 465         int error, namelen;
 466         struct nameidata nd;
 467         struct unpcb *unp;
 468         struct vnode *vp;
 469         struct mount *mp;
 470         cap_rights_t rights;
 471         char *buf;
 472
 473         if (nam->sa_family != AF_UNIX)
 474                 return (EAFNOSUPPORT);
 475
 476         unp = sotounpcb(so);
 477         KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
 478
 479         if (soun->sun_len > sizeof(struct sockaddr_un))
 480                 return (EINVAL);
 481         namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
 482         if (namelen <= 0)
 483                 return (EINVAL);
 484
 485         /*
 486          * We don't allow simultaneous bind() calls on a single UNIX domain
 487          * socket, so flag in-progress operations, and return an error if an
 488          * operation is already in progress.
 489          *
 490          * Historically, we have not allowed a socket to be rebound, so this
 491          * also returns an error.  Not allowing re-binding simplifies the
 492          * implementation and avoids a great many possible failure modes.
 493          */
 494         UNP_PCB_LOCK(unp);
 495         if (unp->unp_vnode != NULL) {
 496                 UNP_PCB_UNLOCK(unp);
 497                 return (EINVAL);
 498         }
 499         if (unp->unp_flags & UNP_BINDING) {
 500                 UNP_PCB_UNLOCK(unp);
 501                 return (EALREADY);
 502         }
 503         unp->unp_flags |= UNP_BINDING;
 504         UNP_PCB_UNLOCK(unp);
 505
 506         buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
 507         bcopy(soun->sun_path, buf, namelen);
 508         buf[namelen] = 0;
 509
 510 restart:
 511         NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE,
 512             UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td);
 513 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
 514         error = namei(&nd);
 515         if (error)
 516                 goto error;
 517         vp = nd.ni_vp;
 518         if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
 519                 NDFREE(&nd, NDF_ONLY_PNBUF);
 520                 if (nd.ni_dvp == vp)
 521                         vrele(nd.ni_dvp);
 522                 else
 523                         vput(nd.ni_dvp);
 524                 if (vp != NULL) {
 525                         vrele(vp);
 526                         error = EADDRINUSE;
 527                         goto error;
 528                 }
 529                 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
 530                 if (error)
 531                         goto error;
 532                 goto restart;
 533         }
 534         VATTR_NULL(&vattr);
 535         vattr.va_type = VSOCK;
 536         vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
 537 #ifdef MAC
 538         error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
 539             &vattr);
 540 #endif
 541         if (error == 0)
 542                 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
 543         NDFREE(&nd, NDF_ONLY_PNBUF);
 544         vput(nd.ni_dvp);
 545         if (error) {
 546                 vn_finished_write(mp);
 547                 goto error;
 548         }
 549         vp = nd.ni_vp;
 550         ASSERT_VOP_ELOCKED(vp, "uipc_bind");
 551         soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
 552
 553         UNP_LINK_WLOCK();
 554         UNP_PCB_LOCK(unp);
 555         VOP_UNP_BIND(vp, unp);
 556         unp->unp_vnode = vp;
 557         unp->unp_addr = soun;
 558         unp->unp_flags &= ~UNP_BINDING;
 559         UNP_PCB_UNLOCK(unp);
 560         UNP_LINK_WUNLOCK();
 561         VOP_UNLOCK(vp, 0);
 562         vn_finished_write(mp);
 563         free(buf, M_TEMP);
 564         return (0);
 565
 566 error:
 567         UNP_PCB_LOCK(unp);
 568         unp->unp_flags &= ~UNP_BINDING;
 569         UNP_PCB_UNLOCK(unp);
 570         free(buf, M_TEMP);
 571         return (error);
 572 }
 573
 574 static int
 575 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 576 {
 577
 578         return (uipc_bindat(AT_FDCWD, so, nam, td));
 579 }
 580
 581 static int
 582 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 583 {
 584         int error;
 585
 586         KASSERT(td == curthread, ("uipc_connect: td != curthread"));
 587         UNP_LINK_WLOCK();
 588         error = unp_connect(so, nam, td);
 589         UNP_LINK_WUNLOCK();
 590         return (error);
 591 }
 592
 593 static int
 594 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
 595     struct thread *td)
 596 {
 597         int error;
 598
 599         KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
 600         UNP_LINK_WLOCK();
 601         error = unp_connectat(fd, so, nam, td);
 602         UNP_LINK_WUNLOCK();
 603         return (error);
 604 }
 605
 606 static void
 607 uipc_close(struct socket *so)
 608 {
 609         struct unpcb *unp, *unp2;
 610
 611         unp = sotounpcb(so);
 612         KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
 613
 614         UNP_LINK_WLOCK();
 615         UNP_PCB_LOCK(unp);
 616         unp2 = unp->unp_conn;
 617         if (unp2 != NULL) {
 618                 UNP_PCB_LOCK(unp2);
 619                 unp_disconnect(unp, unp2);
 620                 UNP_PCB_UNLOCK(unp2);
 621         }
 622         UNP_PCB_UNLOCK(unp);
 623         UNP_LINK_WUNLOCK();
 624 }
 625
 626 static int
 627 uipc_connect2(struct socket *so1, struct socket *so2)
 628 {
 629         struct unpcb *unp, *unp2;
 630         int error;
 631
 632         UNP_LINK_WLOCK();
 633         unp = so1->so_pcb;
 634         KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
 635         UNP_PCB_LOCK(unp);
 636         unp2 = so2->so_pcb;
 637         KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
 638         UNP_PCB_LOCK(unp2);
 639         error = unp_connect2(so1, so2, PRU_CONNECT2);
 640         UNP_PCB_UNLOCK(unp2);
 641         UNP_PCB_UNLOCK(unp);
 642         UNP_LINK_WUNLOCK();
 643         return (error);
 644 }
 645
 646 static void
 647 uipc_detach(struct socket *so)
 648 {
 649         struct unpcb *unp, *unp2;
 650         struct sockaddr_un *saved_unp_addr;
 651         struct vnode *vp;
 652         int freeunp, local_unp_rights;
 653
 654         unp = sotounpcb(so);
 655         KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
 656
 657         vp = NULL;
 658         local_unp_rights = 0;
 659
 660         UNP_LIST_LOCK();
 661         LIST_REMOVE(unp, unp_link);
 662         unp->unp_gencnt = ++unp_gencnt;
 663         --unp_count;
 664         UNP_LIST_UNLOCK();
 665
 666         if ((unp->unp_flags & UNP_NASCENT) != 0) {
 667                 UNP_PCB_LOCK(unp);
 668                 goto teardown;
 669         }
 670         UNP_LINK_WLOCK();
 671         UNP_PCB_LOCK(unp);
 672
 673         if ((vp = unp->unp_vnode) != NULL) {
 674                 VOP_UNP_DETACH(vp);
 675                 unp->unp_vnode = NULL;
 676         }
 677         unp2 = unp->unp_conn;
 678         if (unp2 != NULL) {
 679                 UNP_PCB_LOCK(unp2);
 680                 unp_disconnect(unp, unp2);
 681                 UNP_PCB_UNLOCK(unp2);
 682         }
 683
 684         /*
 685          * We hold the linkage lock exclusively, so it's OK to acquire
 686          * multiple pcb locks at a time.
 687          */
 688         while (!LIST_EMPTY(&unp->unp_refs)) {
 689                 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
 690
 691                 UNP_PCB_LOCK(ref);
 692                 unp_drop(ref);
 693                 UNP_PCB_UNLOCK(ref);
 694         }
 695         local_unp_rights = unp_rights;
 696         UNP_LINK_WUNLOCK();
 697 teardown:
 698         unp->unp_socket->so_pcb = NULL;
 699         saved_unp_addr = unp->unp_addr;
 700         unp->unp_addr = NULL;
 701         unp->unp_refcount--;
 702         freeunp = (unp->unp_refcount == 0);
 703         if (saved_unp_addr != NULL)
 704                 free(saved_unp_addr, M_SONAME);
 705         if (freeunp) {
 706                 UNP_PCB_LOCK_DESTROY(unp);
 707                 uma_zfree(unp_zone, unp);
 708         } else
 709                 UNP_PCB_UNLOCK(unp);
 710         if (vp)
 711                 vrele(vp);
 712         if (local_unp_rights)
 713                 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
 714 }
 715
 716 static int
 717 uipc_disconnect(struct socket *so)
 718 {
 719         struct unpcb *unp, *unp2;
 720
 721         unp = sotounpcb(so);
 722         KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
 723
 724         UNP_LINK_WLOCK();
 725         UNP_PCB_LOCK(unp);
 726         unp2 = unp->unp_conn;
 727         if (unp2 != NULL) {
 728                 UNP_PCB_LOCK(unp2);
 729                 unp_disconnect(unp, unp2);
 730                 UNP_PCB_UNLOCK(unp2);
 731         }
 732         UNP_PCB_UNLOCK(unp);
 733         UNP_LINK_WUNLOCK();
 734         return (0);
 735 }
 736
 737 static int
 738 uipc_listen(struct socket *so, int backlog, struct thread *td)
 739 {
 740         struct unpcb *unp;
 741         int error;
 742
 743         if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
 744                 return (EOPNOTSUPP);
 745
 746         unp = sotounpcb(so);
 747         KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
 748
 749         UNP_PCB_LOCK(unp);
 750         if (unp->unp_vnode == NULL) {
 751                 /* Already connected or not bound to an address. */
 752                 error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ;
 753                 UNP_PCB_UNLOCK(unp);
 754                 return (error);
 755         }
 756
 757         SOCK_LOCK(so);
 758         error = solisten_proto_check(so);
 759         if (error == 0) {
 760                 cru2x(td->td_ucred, &unp->unp_peercred);
 761                 solisten_proto(so, backlog);
 762         }
 763         SOCK_UNLOCK(so);
 764         UNP_PCB_UNLOCK(unp);
 765         return (error);
 766 }
 767
 768 static int
 769 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
 770 {
 771         struct unpcb *unp, *unp2;
 772         const struct sockaddr *sa;
 773
 774         unp = sotounpcb(so);
 775         KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
 776
 777         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 778         UNP_LINK_RLOCK();
 779         /*
 780          * XXX: It seems that this test always fails even when connection is
 781          * established.  So, this else clause is added as workaround to
 782          * return PF_LOCAL sockaddr.
 783          */
 784         unp2 = unp->unp_conn;
 785         if (unp2 != NULL) {
 786                 UNP_PCB_LOCK(unp2);
 787                 if (unp2->unp_addr != NULL)
 788                         sa = (struct sockaddr *) unp2->unp_addr;
 789                 else
 790                         sa = &sun_noname;
 791                 bcopy(sa, *nam, sa->sa_len);
 792                 UNP_PCB_UNLOCK(unp2);
 793         } else {
 794                 sa = &sun_noname;
 795                 bcopy(sa, *nam, sa->sa_len);
 796         }
 797         UNP_LINK_RUNLOCK();
 798         return (0);
 799 }
 800
 801 static int
 802 uipc_rcvd(struct socket *so, int flags)
 803 {
 804         struct unpcb *unp, *unp2;
 805         struct socket *so2;
 806         u_int mbcnt, sbcc;
 807
 808         unp = sotounpcb(so);
 809         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
 810         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
 811             ("%s: socktype %d", __func__, so->so_type));
 812
 813         /*
 814          * Adjust backpressure on sender and wakeup any waiting to write.
 815          *
 816          * The unp lock is acquired to maintain the validity of the unp_conn
 817          * pointer; no lock on unp2 is required as unp2->unp_socket will be
 818          * static as long as we don't permit unp2 to disconnect from unp,
 819          * which is prevented by the lock on unp.  We cache values from
 820          * so_rcv to avoid holding the so_rcv lock over the entire
 821          * transaction on the remote so_snd.
 822          */
 823         SOCKBUF_LOCK(&so->so_rcv);
 824         mbcnt = so->so_rcv.sb_mbcnt;
 825         sbcc = sbavail(&so->so_rcv);
 826         SOCKBUF_UNLOCK(&so->so_rcv);
 827         /*
 828          * There is a benign race condition at this point.  If we're planning to
 829          * clear SB_STOP, but uipc_send is called on the connected socket at
 830          * this instant, it might add data to the sockbuf and set SB_STOP.  Then
 831          * we would erroneously clear SB_STOP below, even though the sockbuf is
 832          * full.  The race is benign because the only ill effect is to allow the
 833          * sockbuf to exceed its size limit, and the size limits are not
 834          * strictly guaranteed anyway.
 835          */
 836         UNP_PCB_LOCK(unp);
 837         unp2 = unp->unp_conn;
 838         if (unp2 == NULL) {
 839                 UNP_PCB_UNLOCK(unp);
 840                 return (0);
 841         }
 842         so2 = unp2->unp_socket;
 843         SOCKBUF_LOCK(&so2->so_snd);
 844         if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
 845                 so2->so_snd.sb_flags &= ~SB_STOP;
 846         sowwakeup_locked(so2);
 847         UNP_PCB_UNLOCK(unp);
 848         return (0);
 849 }
 850
 851 static int
 852 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
 853     struct mbuf *control, struct thread *td)
 854 {
 855         struct unpcb *unp, *unp2;
 856         struct socket *so2;
 857         u_int mbcnt, sbcc;
 858         int error = 0;
 859
 860         unp = sotounpcb(so);
 861         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
 862         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
 863             so->so_type == SOCK_SEQPACKET,
 864             ("%s: socktype %d", __func__, so->so_type));
 865
 866         if (flags & PRUS_OOB) {
 867                 error = EOPNOTSUPP;
 868                 goto release;
 869         }
 870         if (control != NULL && (error = unp_internalize(&control, td)))
 871                 goto release;
 872         if ((nam != NULL) || (flags & PRUS_EOF))
 873                 UNP_LINK_WLOCK();
 874         else
 875                 UNP_LINK_RLOCK();
 876         switch (so->so_type) {
 877         case SOCK_DGRAM:
 878         {
 879                 const struct sockaddr *from;
 880
 881                 unp2 = unp->unp_conn;
 882                 if (nam != NULL) {
 883                         UNP_LINK_WLOCK_ASSERT();
 884                         if (unp2 != NULL) {
 885                                 error = EISCONN;
 886                                 break;
 887                         }
 888                         error = unp_connect(so, nam, td);
 889                         if (error)
 890                                 break;
 891                         unp2 = unp->unp_conn;
 892                 }
 893
 894                 /*
 895                  * Because connect() and send() are non-atomic in a sendto()
 896                  * with a target address, it's possible that the socket will
 897                  * have disconnected before the send() can run.  In that case
 898                  * return the slightly counter-intuitive but otherwise
 899                  * correct error that the socket is not connected.
 900                  */
 901                 if (unp2 == NULL) {
 902                         error = ENOTCONN;
 903                         break;
 904                 }
 905                 /* Lockless read. */
 906                 if (unp2->unp_flags & UNP_WANTCRED)
 907                         control = unp_addsockcred(td, control);
 908                 UNP_PCB_LOCK(unp);
 909                 if (unp->unp_addr != NULL)
 910                         from = (struct sockaddr *)unp->unp_addr;
 911                 else
 912                         from = &sun_noname;
 913                 so2 = unp2->unp_socket;
 914                 SOCKBUF_LOCK(&so2->so_rcv);
 915                 if (sbappendaddr_locked(&so2->so_rcv, from, m,
 916                     control)) {
 917                         sorwakeup_locked(so2);
 918                         m = NULL;
 919                         control = NULL;
 920                 } else {
 921                         SOCKBUF_UNLOCK(&so2->so_rcv);
 922                         error = ENOBUFS;
 923                 }
 924                 if (nam != NULL) {
 925                         UNP_LINK_WLOCK_ASSERT();
 926                         UNP_PCB_LOCK(unp2);
 927                         unp_disconnect(unp, unp2);
 928                         UNP_PCB_UNLOCK(unp2);
 929                 }
 930                 UNP_PCB_UNLOCK(unp);
 931                 break;
 932         }
 933
 934         case SOCK_SEQPACKET:
 935         case SOCK_STREAM:
 936                 if ((so->so_state & SS_ISCONNECTED) == 0) {
 937                         if (nam != NULL) {
 938                                 UNP_LINK_WLOCK_ASSERT();
 939                                 error = unp_connect(so, nam, td);
 940                                 if (error)
 941                                         break;  /* XXX */
 942                         } else {
 943                                 error = ENOTCONN;
 944                                 break;
 945                         }
 946                 }
 947
 948                 /* Lockless read. */
 949                 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
 950                         error = EPIPE;
 951                         break;
 952                 }
 953
 954                 /*
 955                  * Because connect() and send() are non-atomic in a sendto()
 956                  * with a target address, it's possible that the socket will
 957                  * have disconnected before the send() can run.  In that case
 958                  * return the slightly counter-intuitive but otherwise
 959                  * correct error that the socket is not connected.
 960                  *
 961                  * Locking here must be done carefully: the linkage lock
 962                  * prevents interconnections between unpcbs from changing, so
 963                  * we can traverse from unp to unp2 without acquiring unp's
 964                  * lock.  Socket buffer locks follow unpcb locks, so we can
 965                  * acquire both remote and lock socket buffer locks.
 966                  */
 967                 unp2 = unp->unp_conn;
 968                 if (unp2 == NULL) {
 969                         error = ENOTCONN;
 970                         break;
 971                 }
 972                 so2 = unp2->unp_socket;
 973                 UNP_PCB_LOCK(unp2);
 974                 SOCKBUF_LOCK(&so2->so_rcv);
 975                 if (unp2->unp_flags & UNP_WANTCRED) {
 976                         /*
 977                          * Credentials are passed only once on SOCK_STREAM
 978                          * and SOCK_SEQPACKET.
 979                          */
 980                         unp2->unp_flags &= ~UNP_WANTCRED;
 981                         control = unp_addsockcred(td, control);
 982                 }
 983                 /*
 984                  * Send to paired receive port, and then reduce send buffer
 985                  * hiwater marks to maintain backpressure.  Wake up readers.
 986                  */
 987                 switch (so->so_type) {
 988                 case SOCK_STREAM:
 989                         if (control != NULL) {
 990                                 if (sbappendcontrol_locked(&so2->so_rcv, m,
 991                                     control))
 992                                         control = NULL;
 993                         } else
 994                                 sbappend_locked(&so2->so_rcv, m, flags);
 995                         break;
 996
 997                 case SOCK_SEQPACKET: {
 998                         const struct sockaddr *from;
 999
1000                         from = &sun_noname;
1001                         /*
1002                          * Don't check for space available in so2->so_rcv.
1003                          * Unix domain sockets only check for space in the
1004                          * sending sockbuf, and that check is performed one
1005                          * level up the stack.
1006                          */
1007                         if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1008                                 from, m, control))
1009                                 control = NULL;
1010                         break;
1011                         }
1012                 }
1013
1014                 mbcnt = so2->so_rcv.sb_mbcnt;
1015                 sbcc = sbavail(&so2->so_rcv);
1016                 if (sbcc)
1017                         sorwakeup_locked(so2);
1018                 else
1019                         SOCKBUF_UNLOCK(&so2->so_rcv);
1020
1021                 /*
1022                  * The PCB lock on unp2 protects the SB_STOP flag.  Without it,
1023                  * it would be possible for uipc_rcvd to be called at this
1024                  * point, drain the receiving sockbuf, clear SB_STOP, and then
1025                  * we would set SB_STOP below.  That could lead to an empty
1026                  * sockbuf having SB_STOP set
1027                  */
1028                 SOCKBUF_LOCK(&so->so_snd);
1029                 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1030                         so->so_snd.sb_flags |= SB_STOP;
1031                 SOCKBUF_UNLOCK(&so->so_snd);
1032                 UNP_PCB_UNLOCK(unp2);
1033                 m = NULL;
1034                 break;
1035         }
1036
1037         /*
1038          * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1039          */
1040         if (flags & PRUS_EOF) {
1041                 UNP_PCB_LOCK(unp);
1042                 socantsendmore(so);
1043                 unp_shutdown(unp);
1044                 UNP_PCB_UNLOCK(unp);
1045         }
1046
1047         if ((nam != NULL) || (flags & PRUS_EOF))
1048                 UNP_LINK_WUNLOCK();
1049         else
1050                 UNP_LINK_RUNLOCK();
1051
1052         if (control != NULL && error != 0)
1053                 unp_dispose_mbuf(control);
1054
1055 release:
1056         if (control != NULL)
1057                 m_freem(control);
1058         if (m != NULL)
1059                 m_freem(m);
1060         return (error);
1061 }
1062
1063 static int
1064 uipc_ready(struct socket *so, struct mbuf *m, int count)
1065 {
1066         struct unpcb *unp, *unp2;
1067         struct socket *so2;
1068         int error;
1069
1070         unp = sotounpcb(so);
1071
1072         UNP_LINK_RLOCK();
1073         unp2 = unp->unp_conn;
1074         UNP_PCB_LOCK(unp2);
1075         so2 = unp2->unp_socket;
1076
1077         SOCKBUF_LOCK(&so2->so_rcv);
1078         if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1079                 sorwakeup_locked(so2);
1080         else
1081                 SOCKBUF_UNLOCK(&so2->so_rcv);
1082
1083         UNP_PCB_UNLOCK(unp2);
1084         UNP_LINK_RUNLOCK();
1085
1086         return (error);
1087 }
1088
1089 static int
1090 uipc_sense(struct socket *so, struct stat *sb)
1091 {
1092         struct unpcb *unp;
1093
1094         unp = sotounpcb(so);
1095         KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1096
1097         sb->st_blksize = so->so_snd.sb_hiwat;
1098         UNP_PCB_LOCK(unp);
1099         sb->st_dev = NODEV;
1100         if (unp->unp_ino == 0)
1101                 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1102         sb->st_ino = unp->unp_ino;
1103         UNP_PCB_UNLOCK(unp);
1104         return (0);
1105 }
1106
1107 static int
1108 uipc_shutdown(struct socket *so)
1109 {
1110         struct unpcb *unp;
1111
1112         unp = sotounpcb(so);
1113         KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1114
1115         UNP_LINK_WLOCK();
1116         UNP_PCB_LOCK(unp);
1117         socantsendmore(so);
1118         unp_shutdown(unp);
1119         UNP_PCB_UNLOCK(unp);
1120         UNP_LINK_WUNLOCK();
1121         return (0);
1122 }
1123
1124 static int
1125 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1126 {
1127         struct unpcb *unp;
1128         const struct sockaddr *sa;
1129
1130         unp = sotounpcb(so);
1131         KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1132
1133         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1134         UNP_PCB_LOCK(unp);
1135         if (unp->unp_addr != NULL)
1136                 sa = (struct sockaddr *) unp->unp_addr;
1137         else
1138                 sa = &sun_noname;
1139         bcopy(sa, *nam, sa->sa_len);
1140         UNP_PCB_UNLOCK(unp);
1141         return (0);
1142 }
1143
1144 static struct pr_usrreqs uipc_usrreqs_dgram = {
1145         .pru_abort =            uipc_abort,
1146         .pru_accept =           uipc_accept,
1147         .pru_attach =           uipc_attach,
1148         .pru_bind =             uipc_bind,
1149         .pru_bindat =           uipc_bindat,
1150         .pru_connect =          uipc_connect,
1151         .pru_connectat =        uipc_connectat,
1152         .pru_connect2 =         uipc_connect2,
1153         .pru_detach =           uipc_detach,
1154         .pru_disconnect =       uipc_disconnect,
1155         .pru_listen =           uipc_listen,
1156         .pru_peeraddr =         uipc_peeraddr,
1157         .pru_rcvd =             uipc_rcvd,
1158         .pru_send =             uipc_send,
1159         .pru_sense =            uipc_sense,
1160         .pru_shutdown =         uipc_shutdown,
1161         .pru_sockaddr =         uipc_sockaddr,
1162         .pru_soreceive =        soreceive_dgram,
1163         .pru_close =            uipc_close,
1164 };
1165
1166 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1167         .pru_abort =            uipc_abort,
1168         .pru_accept =           uipc_accept,
1169         .pru_attach =           uipc_attach,
1170         .pru_bind =             uipc_bind,
1171         .pru_bindat =           uipc_bindat,
1172         .pru_connect =          uipc_connect,
1173         .pru_connectat =        uipc_connectat,
1174         .pru_connect2 =         uipc_connect2,
1175         .pru_detach =           uipc_detach,
1176         .pru_disconnect =       uipc_disconnect,
1177         .pru_listen =           uipc_listen,
1178         .pru_peeraddr =         uipc_peeraddr,
1179         .pru_rcvd =             uipc_rcvd,
1180         .pru_send =             uipc_send,
1181         .pru_sense =            uipc_sense,
1182         .pru_shutdown =         uipc_shutdown,
1183         .pru_sockaddr =         uipc_sockaddr,
1184         .pru_soreceive =        soreceive_generic,      /* XXX: or...? */
1185         .pru_close =            uipc_close,
1186 };
1187
1188 static struct pr_usrreqs uipc_usrreqs_stream = {
1189         .pru_abort =            uipc_abort,
1190         .pru_accept =           uipc_accept,
1191         .pru_attach =           uipc_attach,
1192         .pru_bind =             uipc_bind,
1193         .pru_bindat =           uipc_bindat,
1194         .pru_connect =          uipc_connect,
1195         .pru_connectat =        uipc_connectat,
1196         .pru_connect2 =         uipc_connect2,
1197         .pru_detach =           uipc_detach,
1198         .pru_disconnect =       uipc_disconnect,
1199         .pru_listen =           uipc_listen,
1200         .pru_peeraddr =         uipc_peeraddr,
1201         .pru_rcvd =             uipc_rcvd,
1202         .pru_send =             uipc_send,
1203         .pru_ready =            uipc_ready,
1204         .pru_sense =            uipc_sense,
1205         .pru_shutdown =         uipc_shutdown,
1206         .pru_sockaddr =         uipc_sockaddr,
1207         .pru_soreceive =        soreceive_generic,
1208         .pru_close =            uipc_close,
1209 };
1210
1211 static int
1212 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1213 {
1214         struct unpcb *unp;
1215         struct xucred xu;
1216         int error, optval;
1217
1218         if (sopt->sopt_level != 0)
1219                 return (EINVAL);
1220
1221         unp = sotounpcb(so);
1222         KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1223         error = 0;
1224         switch (sopt->sopt_dir) {
1225         case SOPT_GET:
1226                 switch (sopt->sopt_name) {
1227                 case LOCAL_PEERCRED:
1228                         UNP_PCB_LOCK(unp);
1229                         if (unp->unp_flags & UNP_HAVEPC)
1230                                 xu = unp->unp_peercred;
1231                         else {
1232                                 if (so->so_type == SOCK_STREAM)
1233                                         error = ENOTCONN;
1234                                 else
1235                                         error = EINVAL;
1236                         }
1237                         UNP_PCB_UNLOCK(unp);
1238                         if (error == 0)
1239                                 error = sooptcopyout(sopt, &xu, sizeof(xu));
1240                         break;
1241
1242                 case LOCAL_CREDS:
1243                         /* Unlocked read. */
1244                         optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1245                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1246                         break;
1247
1248                 case LOCAL_CONNWAIT:
1249                         /* Unlocked read. */
1250                         optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1251                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1252                         break;
1253
1254                 default:
1255                         error = EOPNOTSUPP;
1256                         break;
1257                 }
1258                 break;
1259
1260         case SOPT_SET:
1261                 switch (sopt->sopt_name) {
1262                 case LOCAL_CREDS:
1263                 case LOCAL_CONNWAIT:
1264                         error = sooptcopyin(sopt, &optval, sizeof(optval),
1265                                             sizeof(optval));
1266                         if (error)
1267                                 break;
1268
1269 #define OPTSET(bit) do {                                                \
1270         UNP_PCB_LOCK(unp);                                              \
1271         if (optval)                                                     \
1272                 unp->unp_flags |= bit;                                  \
1273         else                                                            \
1274                 unp->unp_flags &= ~bit;                                 \
1275         UNP_PCB_UNLOCK(unp);                                            \
1276 } while (0)
1277
1278                         switch (sopt->sopt_name) {
1279                         case LOCAL_CREDS:
1280                                 OPTSET(UNP_WANTCRED);
1281                                 break;
1282
1283                         case LOCAL_CONNWAIT:
1284                                 OPTSET(UNP_CONNWAIT);
1285                                 break;
1286
1287                         default:
1288                                 break;
1289                         }
1290                         break;
1291 #undef  OPTSET
1292                 default:
1293                         error = ENOPROTOOPT;
1294                         break;
1295                 }
1296                 break;
1297
1298         default:
1299                 error = EOPNOTSUPP;
1300                 break;
1301         }
1302         return (error);
1303 }
1304
1305 static int
1306 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1307 {
1308
1309         return (unp_connectat(AT_FDCWD, so, nam, td));
1310 }
1311
1312 static int
1313 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1314     struct thread *td)
1315 {
1316         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1317         struct vnode *vp;
1318         struct socket *so2, *so3;
1319         struct unpcb *unp, *unp2, *unp3;
1320         struct nameidata nd;
1321         char buf[SOCK_MAXADDRLEN];
1322         struct sockaddr *sa;
1323         cap_rights_t rights;
1324         int error, len;
1325
1326         if (nam->sa_family != AF_UNIX)
1327                 return (EAFNOSUPPORT);
1328
1329         UNP_LINK_WLOCK_ASSERT();
1330
1331         unp = sotounpcb(so);
1332         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1333
1334         if (nam->sa_len > sizeof(struct sockaddr_un))
1335                 return (EINVAL);
1336         len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1337         if (len <= 0)
1338                 return (EINVAL);
1339         bcopy(soun->sun_path, buf, len);
1340         buf[len] = 0;
1341
1342         UNP_PCB_LOCK(unp);
1343         if (unp->unp_flags & UNP_CONNECTING) {
1344                 UNP_PCB_UNLOCK(unp);
1345                 return (EALREADY);
1346         }
1347         UNP_LINK_WUNLOCK();
1348         unp->unp_flags |= UNP_CONNECTING;
1349         UNP_PCB_UNLOCK(unp);
1350
1351         sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1352         NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1353             UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1354         error = namei(&nd);
1355         if (error)
1356                 vp = NULL;
1357         else
1358                 vp = nd.ni_vp;
1359         ASSERT_VOP_LOCKED(vp, "unp_connect");
1360         NDFREE(&nd, NDF_ONLY_PNBUF);
1361         if (error)
1362                 goto bad;
1363
1364         if (vp->v_type != VSOCK) {
1365                 error = ENOTSOCK;
1366                 goto bad;
1367         }
1368 #ifdef MAC
1369         error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1370         if (error)
1371                 goto bad;
1372 #endif
1373         error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1374         if (error)
1375                 goto bad;
1376
1377         unp = sotounpcb(so);
1378         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1379
1380         /*
1381          * Lock linkage lock for two reasons: make sure v_socket is stable,
1382          * and to protect simultaneous locking of multiple pcbs.
1383          */
1384         UNP_LINK_WLOCK();
1385         VOP_UNP_CONNECT(vp, &unp2);
1386         if (unp2 == NULL) {
1387                 error = ECONNREFUSED;
1388                 goto bad2;
1389         }
1390         so2 = unp2->unp_socket;
1391         if (so->so_type != so2->so_type) {
1392                 error = EPROTOTYPE;
1393                 goto bad2;
1394         }
1395         if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1396                 if (so2->so_options & SO_ACCEPTCONN) {
1397                         CURVNET_SET(so2->so_vnet);
1398                         so3 = sonewconn(so2, 0);
1399                         CURVNET_RESTORE();
1400                 } else
1401                         so3 = NULL;
1402                 if (so3 == NULL) {
1403                         error = ECONNREFUSED;
1404                         goto bad2;
1405                 }
1406                 unp = sotounpcb(so);
1407                 unp2 = sotounpcb(so2);
1408                 unp3 = sotounpcb(so3);
1409                 UNP_PCB_LOCK(unp);
1410                 UNP_PCB_LOCK(unp2);
1411                 UNP_PCB_LOCK(unp3);
1412                 if (unp2->unp_addr != NULL) {
1413                         bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1414                         unp3->unp_addr = (struct sockaddr_un *) sa;
1415                         sa = NULL;
1416                 }
1417
1418                 /*
1419                  * The connector's (client's) credentials are copied from its
1420                  * process structure at the time of connect() (which is now).
1421                  */
1422                 cru2x(td->td_ucred, &unp3->unp_peercred);
1423                 unp3->unp_flags |= UNP_HAVEPC;
1424
1425                 /*
1426                  * The receiver's (server's) credentials are copied from the
1427                  * unp_peercred member of socket on which the former called
1428                  * listen(); uipc_listen() cached that process's credentials
1429                  * at that time so we can use them now.
1430                  */
1431                 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1432                     sizeof(unp->unp_peercred));
1433                 unp->unp_flags |= UNP_HAVEPC;
1434                 if (unp2->unp_flags & UNP_WANTCRED)
1435                         unp3->unp_flags |= UNP_WANTCRED;
1436                 UNP_PCB_UNLOCK(unp3);
1437                 UNP_PCB_UNLOCK(unp2);
1438                 UNP_PCB_UNLOCK(unp);
1439 #ifdef MAC
1440                 mac_socketpeer_set_from_socket(so, so3);
1441                 mac_socketpeer_set_from_socket(so3, so);
1442 #endif
1443
1444                 so2 = so3;
1445         }
1446         unp = sotounpcb(so);
1447         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1448         unp2 = sotounpcb(so2);
1449         KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1450         UNP_PCB_LOCK(unp);
1451         UNP_PCB_LOCK(unp2);
1452         error = unp_connect2(so, so2, PRU_CONNECT);
1453         UNP_PCB_UNLOCK(unp2);
1454         UNP_PCB_UNLOCK(unp);
1455 bad2:
1456         UNP_LINK_WUNLOCK();
1457 bad:
1458         if (vp != NULL)
1459                 vput(vp);
1460         free(sa, M_SONAME);
1461         UNP_LINK_WLOCK();
1462         UNP_PCB_LOCK(unp);
1463         unp->unp_flags &= ~UNP_CONNECTING;
1464         UNP_PCB_UNLOCK(unp);
1465         return (error);
1466 }
1467
1468 static int
1469 unp_connect2(struct socket *so, struct socket *so2, int req)
1470 {
1471         struct unpcb *unp;
1472         struct unpcb *unp2;
1473
1474         unp = sotounpcb(so);
1475         KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1476         unp2 = sotounpcb(so2);
1477         KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1478
1479         UNP_LINK_WLOCK_ASSERT();
1480         UNP_PCB_LOCK_ASSERT(unp);
1481         UNP_PCB_LOCK_ASSERT(unp2);
1482
1483         if (so2->so_type != so->so_type)
1484                 return (EPROTOTYPE);
1485         unp2->unp_flags &= ~UNP_NASCENT;
1486         unp->unp_conn = unp2;
1487
1488         switch (so->so_type) {
1489         case SOCK_DGRAM:
1490                 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1491                 soisconnected(so);
1492                 break;
1493
1494         case SOCK_STREAM:
1495         case SOCK_SEQPACKET:
1496                 unp2->unp_conn = unp;
1497                 if (req == PRU_CONNECT &&
1498                     ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1499                         soisconnecting(so);
1500                 else
1501                         soisconnected(so);
1502                 soisconnected(so2);
1503                 break;
1504
1505         default:
1506                 panic("unp_connect2");
1507         }
1508         return (0);
1509 }
1510
1511 static void
1512 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1513 {
1514         struct socket *so;
1515
1516         KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1517
1518         UNP_LINK_WLOCK_ASSERT();
1519         UNP_PCB_LOCK_ASSERT(unp);
1520         UNP_PCB_LOCK_ASSERT(unp2);
1521
1522         unp->unp_conn = NULL;
1523         switch (unp->unp_socket->so_type) {
1524         case SOCK_DGRAM:
1525                 LIST_REMOVE(unp, unp_reflink);
1526                 so = unp->unp_socket;
1527                 SOCK_LOCK(so);
1528                 so->so_state &= ~SS_ISCONNECTED;
1529                 SOCK_UNLOCK(so);
1530                 break;
1531
1532         case SOCK_STREAM:
1533         case SOCK_SEQPACKET:
1534                 soisdisconnected(unp->unp_socket);
1535                 unp2->unp_conn = NULL;
1536                 soisdisconnected(unp2->unp_socket);
1537                 break;
1538         }
1539 }
1540
1541 /*
1542  * unp_pcblist() walks the global list of struct unpcb's to generate a
1543  * pointer list, bumping the refcount on each unpcb.  It then copies them out
1544  * sequentially, validating the generation number on each to see if it has
1545  * been detached.  All of this is necessary because copyout() may sleep on
1546  * disk I/O.
1547  */
1548 static int
1549 unp_pcblist(SYSCTL_HANDLER_ARGS)
1550 {
1551         int error, i, n;
1552         int freeunp;
1553         struct unpcb *unp, **unp_list;
1554         unp_gen_t gencnt;
1555         struct xunpgen *xug;
1556         struct unp_head *head;
1557         struct xunpcb *xu;
1558
1559         switch ((intptr_t)arg1) {
1560         case SOCK_STREAM:
1561                 head = &unp_shead;
1562                 break;
1563
1564         case SOCK_DGRAM:
1565                 head = &unp_dhead;
1566                 break;
1567
1568         case SOCK_SEQPACKET:
1569                 head = &unp_sphead;
1570                 break;
1571
1572         default:
1573                 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1574         }
1575
1576         /*
1577          * The process of preparing the PCB list is too time-consuming and
1578          * resource-intensive to repeat twice on every request.
1579          */
1580         if (req->oldptr == NULL) {
1581                 n = unp_count;
1582                 req->oldidx = 2 * (sizeof *xug)
1583                         + (n + n/8) * sizeof(struct xunpcb);
1584                 return (0);
1585         }
1586
1587         if (req->newptr != NULL)
1588                 return (EPERM);
1589
1590         /*
1591          * OK, now we're committed to doing something.
1592          */
1593         xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1594         UNP_LIST_LOCK();
1595         gencnt = unp_gencnt;
1596         n = unp_count;
1597         UNP_LIST_UNLOCK();
1598
1599         xug->xug_len = sizeof *xug;
1600         xug->xug_count = n;
1601         xug->xug_gen = gencnt;
1602         xug->xug_sogen = so_gencnt;
1603         error = SYSCTL_OUT(req, xug, sizeof *xug);
1604         if (error) {
1605                 free(xug, M_TEMP);
1606                 return (error);
1607         }
1608
1609         unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1610
1611         UNP_LIST_LOCK();
1612         for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1613              unp = LIST_NEXT(unp, unp_link)) {
1614                 UNP_PCB_LOCK(unp);
1615                 if (unp->unp_gencnt <= gencnt) {
1616                         if (cr_cansee(req->td->td_ucred,
1617                             unp->unp_socket->so_cred)) {
1618                                 UNP_PCB_UNLOCK(unp);
1619                                 continue;
1620                         }
1621                         unp_list[i++] = unp;
1622                         unp->unp_refcount++;
1623                 }
1624                 UNP_PCB_UNLOCK(unp);
1625         }
1626         UNP_LIST_UNLOCK();
1627         n = i;                  /* In case we lost some during malloc. */
1628
1629         error = 0;
1630         xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1631         for (i = 0; i < n; i++) {
1632                 unp = unp_list[i];
1633                 UNP_PCB_LOCK(unp);
1634                 unp->unp_refcount--;
1635                 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1636                         xu->xu_len = sizeof *xu;
1637                         xu->xu_unpp = unp;
1638                         /*
1639                          * XXX - need more locking here to protect against
1640                          * connect/disconnect races for SMP.
1641                          */
1642                         if (unp->unp_addr != NULL)
1643                                 bcopy(unp->unp_addr, &xu->xu_addr,
1644                                       unp->unp_addr->sun_len);
1645                         if (unp->unp_conn != NULL &&
1646                             unp->unp_conn->unp_addr != NULL)
1647                                 bcopy(unp->unp_conn->unp_addr,
1648                                       &xu->xu_caddr,
1649                                       unp->unp_conn->unp_addr->sun_len);
1650                         bcopy(unp, &xu->xu_unp, sizeof *unp);
1651                         sotoxsocket(unp->unp_socket, &xu->xu_socket);
1652                         UNP_PCB_UNLOCK(unp);
1653                         error = SYSCTL_OUT(req, xu, sizeof *xu);
1654                 } else {
1655                         freeunp = (unp->unp_refcount == 0);
1656                         UNP_PCB_UNLOCK(unp);
1657                         if (freeunp) {
1658                                 UNP_PCB_LOCK_DESTROY(unp);
1659                                 uma_zfree(unp_zone, unp);
1660                         }
1661                 }
1662         }
1663         free(xu, M_TEMP);
1664         if (!error) {
1665                 /*
1666                  * Give the user an updated idea of our state.  If the
1667                  * generation differs from what we told her before, she knows
1668                  * that something happened while we were processing this
1669                  * request, and it might be necessary to retry.
1670                  */
1671                 xug->xug_gen = unp_gencnt;
1672                 xug->xug_sogen = so_gencnt;
1673                 xug->xug_count = unp_count;
1674                 error = SYSCTL_OUT(req, xug, sizeof *xug);
1675         }
1676         free(unp_list, M_TEMP);
1677         free(xug, M_TEMP);
1678         return (error);
1679 }
1680
1681 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1682     (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1683     "List of active local datagram sockets");
1684 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1685     (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1686     "List of active local stream sockets");
1687 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1688     CTLTYPE_OPAQUE | CTLFLAG_RD,
1689     (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1690     "List of active local seqpacket sockets");
1691
1692 static void
1693 unp_shutdown(struct unpcb *unp)
1694 {
1695         struct unpcb *unp2;
1696         struct socket *so;
1697
1698         UNP_LINK_WLOCK_ASSERT();
1699         UNP_PCB_LOCK_ASSERT(unp);
1700
1701         unp2 = unp->unp_conn;
1702         if ((unp->unp_socket->so_type == SOCK_STREAM ||
1703             (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1704                 so = unp2->unp_socket;
1705                 if (so != NULL)
1706                         socantrcvmore(so);
1707         }
1708 }
1709
1710 static void
1711 unp_drop(struct unpcb *unp)
1712 {
1713         struct socket *so = unp->unp_socket;
1714         struct unpcb *unp2;
1715
1716         UNP_LINK_WLOCK_ASSERT();
1717         UNP_PCB_LOCK_ASSERT(unp);
1718
1719         /*
1720          * Regardless of whether the socket's peer dropped the connection
1721          * with this socket by aborting or disconnecting, POSIX requires
1722          * that ECONNRESET is returned.
1723          */
1724         so->so_error = ECONNRESET;
1725         unp2 = unp->unp_conn;
1726         if (unp2 == NULL)
1727                 return;
1728         UNP_PCB_LOCK(unp2);
1729         unp_disconnect(unp, unp2);
1730         UNP_PCB_UNLOCK(unp2);
1731 }
1732
1733 static void
1734 unp_freerights(struct filedescent **fdep, int fdcount)
1735 {
1736         struct file *fp;
1737         int i;
1738
1739         KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1740
1741         for (i = 0; i < fdcount; i++) {
1742                 fp = fdep[i]->fde_file;
1743                 filecaps_free(&fdep[i]->fde_caps);
1744                 unp_discard(fp);
1745         }
1746         free(fdep[0], M_FILECAPS);
1747 }
1748
1749 static int
1750 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1751 {
1752         struct thread *td = curthread;          /* XXX */
1753         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1754         int i;
1755         int *fdp;
1756         struct filedesc *fdesc = td->td_proc->p_fd;
1757         struct filedescent **fdep;
1758         void *data;
1759         socklen_t clen = control->m_len, datalen;
1760         int error, newfds;
1761         u_int newlen;
1762
1763         UNP_LINK_UNLOCK_ASSERT();
1764
1765         error = 0;
1766         if (controlp != NULL) /* controlp == NULL => free control messages */
1767                 *controlp = NULL;
1768         while (cm != NULL) {
1769                 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1770                         error = EINVAL;
1771                         break;
1772                 }
1773                 data = CMSG_DATA(cm);
1774                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1775                 if (cm->cmsg_level == SOL_SOCKET
1776                     && cm->cmsg_type == SCM_RIGHTS) {
1777                         newfds = datalen / sizeof(*fdep);
1778                         if (newfds == 0)
1779                                 goto next;
1780                         fdep = data;
1781
1782                         /* If we're not outputting the descriptors free them. */
1783                         if (error || controlp == NULL) {
1784                                 unp_freerights(fdep, newfds);
1785                                 goto next;
1786                         }
1787                         FILEDESC_XLOCK(fdesc);
1788
1789                         /*
1790                          * Now change each pointer to an fd in the global
1791                          * table to an integer that is the index to the local
1792                          * fd table entry that we set up to point to the
1793                          * global one we are transferring.
1794                          */
1795                         newlen = newfds * sizeof(int);
1796                         *controlp = sbcreatecontrol(NULL, newlen,
1797                             SCM_RIGHTS, SOL_SOCKET);
1798                         if (*controlp == NULL) {
1799                                 FILEDESC_XUNLOCK(fdesc);
1800                                 error = E2BIG;
1801                                 unp_freerights(fdep, newfds);
1802                                 goto next;
1803                         }
1804
1805                         fdp = (int *)
1806                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1807                         if (fdallocn(td, 0, fdp, newfds) != 0) {
1808                                 FILEDESC_XUNLOCK(fdesc);
1809                                 error = EMSGSIZE;
1810                                 unp_freerights(fdep, newfds);
1811                                 m_freem(*controlp);
1812                                 *controlp = NULL;
1813                                 goto next;
1814                         }
1815                         for (i = 0; i < newfds; i++, fdp++) {
1816                                 _finstall(fdesc, fdep[i]->fde_file, *fdp,
1817                                     (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0,
1818                                     &fdep[i]->fde_caps);
1819                                 unp_externalize_fp(fdep[i]->fde_file);
1820                         }
1821                         FILEDESC_XUNLOCK(fdesc);
1822                         free(fdep[0], M_FILECAPS);
1823                 } else {
1824                         /* We can just copy anything else across. */
1825                         if (error || controlp == NULL)
1826                                 goto next;
1827                         *controlp = sbcreatecontrol(NULL, datalen,
1828                             cm->cmsg_type, cm->cmsg_level);
1829                         if (*controlp == NULL) {
1830                                 error = ENOBUFS;
1831                                 goto next;
1832                         }
1833                         bcopy(data,
1834                             CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1835                             datalen);
1836                 }
1837                 controlp = &(*controlp)->m_next;
1838
1839 next:
1840                 if (CMSG_SPACE(datalen) < clen) {
1841                         clen -= CMSG_SPACE(datalen);
1842                         cm = (struct cmsghdr *)
1843                             ((caddr_t)cm + CMSG_SPACE(datalen));
1844                 } else {
1845                         clen = 0;
1846                         cm = NULL;
1847                 }
1848         }
1849
1850         m_freem(control);
1851         return (error);
1852 }
1853
1854 static void
1855 unp_zone_change(void *tag)
1856 {
1857
1858         uma_zone_set_max(unp_zone, maxsockets);
1859 }
1860
1861 static void
1862 unp_init(void)
1863 {
1864
1865 #ifdef VIMAGE
1866         if (!IS_DEFAULT_VNET(curvnet))
1867                 return;
1868 #endif
1869         unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1870             NULL, NULL, UMA_ALIGN_PTR, 0);
1871         if (unp_zone == NULL)
1872                 panic("unp_init");
1873         uma_zone_set_max(unp_zone, maxsockets);
1874         uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
1875         EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1876             NULL, EVENTHANDLER_PRI_ANY);
1877         LIST_INIT(&unp_dhead);
1878         LIST_INIT(&unp_shead);
1879         LIST_INIT(&unp_sphead);
1880         SLIST_INIT(&unp_defers);
1881         TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1882         TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1883         UNP_LINK_LOCK_INIT();
1884         UNP_LIST_LOCK_INIT();
1885         UNP_DEFERRED_LOCK_INIT();
1886 }
1887
1888 static int
1889 unp_internalize(struct mbuf **controlp, struct thread *td)
1890 {
1891         struct mbuf *control = *controlp;
1892         struct proc *p = td->td_proc;
1893         struct filedesc *fdesc = p->p_fd;
1894         struct bintime *bt;
1895         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1896         struct cmsgcred *cmcred;
1897         struct filedescent *fde, **fdep, *fdev;
1898         struct file *fp;
1899         struct timeval *tv;
1900         struct timespec *ts;
1901         int i, *fdp;
1902         void *data;
1903         socklen_t clen = control->m_len, datalen;
1904         int error, oldfds;
1905         u_int newlen;
1906
1907         UNP_LINK_UNLOCK_ASSERT();
1908
1909         error = 0;
1910         *controlp = NULL;
1911         while (cm != NULL) {
1912                 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1913                     || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
1914                         error = EINVAL;
1915                         goto out;
1916                 }
1917                 data = CMSG_DATA(cm);
1918                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1919
1920                 switch (cm->cmsg_type) {
1921                 /*
1922                  * Fill in credential information.
1923                  */
1924                 case SCM_CREDS:
1925                         *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1926                             SCM_CREDS, SOL_SOCKET);
1927                         if (*controlp == NULL) {
1928                                 error = ENOBUFS;
1929                                 goto out;
1930                         }
1931                         cmcred = (struct cmsgcred *)
1932                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1933                         cmcred->cmcred_pid = p->p_pid;
1934                         cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1935                         cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1936                         cmcred->cmcred_euid = td->td_ucred->cr_uid;
1937                         cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1938                             CMGROUP_MAX);
1939                         for (i = 0; i < cmcred->cmcred_ngroups; i++)
1940                                 cmcred->cmcred_groups[i] =
1941                                     td->td_ucred->cr_groups[i];
1942                         break;
1943
1944                 case SCM_RIGHTS:
1945                         oldfds = datalen / sizeof (int);
1946                         if (oldfds == 0)
1947                                 break;
1948                         /*
1949                          * Check that all the FDs passed in refer to legal
1950                          * files.  If not, reject the entire operation.
1951                          */
1952                         fdp = data;
1953                         FILEDESC_SLOCK(fdesc);
1954                         for (i = 0; i < oldfds; i++, fdp++) {
1955                                 fp = fget_locked(fdesc, *fdp);
1956                                 if (fp == NULL) {
1957                                         FILEDESC_SUNLOCK(fdesc);
1958                                         error = EBADF;
1959                                         goto out;
1960                                 }
1961                                 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1962                                         FILEDESC_SUNLOCK(fdesc);
1963                                         error = EOPNOTSUPP;
1964                                         goto out;
1965                                 }
1966
1967                         }
1968
1969                         /*
1970                          * Now replace the integer FDs with pointers to the
1971                          * file structure and capability rights.
1972                          */
1973                         newlen = oldfds * sizeof(fdep[0]);
1974                         *controlp = sbcreatecontrol(NULL, newlen,
1975                             SCM_RIGHTS, SOL_SOCKET);
1976                         if (*controlp == NULL) {
1977                                 FILEDESC_SUNLOCK(fdesc);
1978                                 error = E2BIG;
1979                                 goto out;
1980                         }
1981                         fdp = data;
1982                         fdep = (struct filedescent **)
1983                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1984                         fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
1985                             M_WAITOK);
1986                         for (i = 0; i < oldfds; i++, fdev++, fdp++) {
1987                                 fde = &fdesc->fd_ofiles[*fdp];
1988                                 fdep[i] = fdev;
1989                                 fdep[i]->fde_file = fde->fde_file;
1990                                 filecaps_copy(&fde->fde_caps,
1991                                     &fdep[i]->fde_caps, true);
1992                                 unp_internalize_fp(fdep[i]->fde_file);
1993                         }
1994                         FILEDESC_SUNLOCK(fdesc);
1995                         break;
1996
1997                 case SCM_TIMESTAMP:
1998                         *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1999                             SCM_TIMESTAMP, SOL_SOCKET);
2000                         if (*controlp == NULL) {
2001                                 error = ENOBUFS;
2002                                 goto out;
2003                         }
2004                         tv = (struct timeval *)
2005                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2006                         microtime(tv);
2007                         break;
2008
2009                 case SCM_BINTIME:
2010                         *controlp = sbcreatecontrol(NULL, sizeof(*bt),
2011                             SCM_BINTIME, SOL_SOCKET);
2012                         if (*controlp == NULL) {
2013                                 error = ENOBUFS;
2014                                 goto out;
2015                         }
2016                         bt = (struct bintime *)
2017                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2018                         bintime(bt);
2019                         break;
2020
2021                 case SCM_REALTIME:
2022                         *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2023                             SCM_REALTIME, SOL_SOCKET);
2024                         if (*controlp == NULL) {
2025                                 error = ENOBUFS;
2026                                 goto out;
2027                         }
2028                         ts = (struct timespec *)
2029                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2030                         nanotime(ts);
2031                         break;
2032
2033                 case SCM_MONOTONIC:
2034                         *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2035                             SCM_MONOTONIC, SOL_SOCKET);
2036                         if (*controlp == NULL) {
2037                                 error = ENOBUFS;
2038                                 goto out;
2039                         }
2040                         ts = (struct timespec *)
2041                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2042                         nanouptime(ts);
2043                         break;
2044
2045                 default:
2046                         error = EINVAL;
2047                         goto out;
2048                 }
2049
2050                 controlp = &(*controlp)->m_next;
2051                 if (CMSG_SPACE(datalen) < clen) {
2052                         clen -= CMSG_SPACE(datalen);
2053                         cm = (struct cmsghdr *)
2054                             ((caddr_t)cm + CMSG_SPACE(datalen));
2055                 } else {
2056                         clen = 0;
2057                         cm = NULL;
2058                 }
2059         }
2060
2061 out:
2062         m_freem(control);
2063         return (error);
2064 }
2065
2066 static struct mbuf *
2067 unp_addsockcred(struct thread *td, struct mbuf *control)
2068 {
2069         struct mbuf *m, *n, *n_prev;
2070         struct sockcred *sc;
2071         const struct cmsghdr *cm;
2072         int ngroups;
2073         int i;
2074
2075         ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2076         m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2077         if (m == NULL)
2078                 return (control);
2079
2080         sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2081         sc->sc_uid = td->td_ucred->cr_ruid;
2082         sc->sc_euid = td->td_ucred->cr_uid;
2083         sc->sc_gid = td->td_ucred->cr_rgid;
2084         sc->sc_egid = td->td_ucred->cr_gid;
2085         sc->sc_ngroups = ngroups;
2086         for (i = 0; i < sc->sc_ngroups; i++)
2087                 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2088
2089         /*
2090          * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2091          * created SCM_CREDS control message (struct sockcred) has another
2092          * format.
2093          */
2094         if (control != NULL)
2095                 for (n = control, n_prev = NULL; n != NULL;) {
2096                         cm = mtod(n, struct cmsghdr *);
2097                         if (cm->cmsg_level == SOL_SOCKET &&
2098                             cm->cmsg_type == SCM_CREDS) {
2099                                 if (n_prev == NULL)
2100                                         control = n->m_next;
2101                                 else
2102                                         n_prev->m_next = n->m_next;
2103                                 n = m_free(n);
2104                         } else {
2105                                 n_prev = n;
2106                                 n = n->m_next;
2107                         }
2108                 }
2109
2110         /* Prepend it to the head. */
2111         m->m_next = control;
2112         return (m);
2113 }
2114
2115 static struct unpcb *
2116 fptounp(struct file *fp)
2117 {
2118         struct socket *so;
2119
2120         if (fp->f_type != DTYPE_SOCKET)
2121                 return (NULL);
2122         if ((so = fp->f_data) == NULL)
2123                 return (NULL);
2124         if (so->so_proto->pr_domain != &localdomain)
2125                 return (NULL);
2126         return sotounpcb(so);
2127 }
2128
2129 static void
2130 unp_discard(struct file *fp)
2131 {
2132         struct unp_defer *dr;
2133
2134         if (unp_externalize_fp(fp)) {
2135                 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2136                 dr->ud_fp = fp;
2137                 UNP_DEFERRED_LOCK();
2138                 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2139                 UNP_DEFERRED_UNLOCK();
2140                 atomic_add_int(&unp_defers_count, 1);
2141                 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2142         } else
2143                 (void) closef(fp, (struct thread *)NULL);
2144 }
2145
2146 static void
2147 unp_process_defers(void *arg __unused, int pending)
2148 {
2149         struct unp_defer *dr;
2150         SLIST_HEAD(, unp_defer) drl;
2151         int count;
2152
2153         SLIST_INIT(&drl);
2154         for (;;) {
2155                 UNP_DEFERRED_LOCK();
2156                 if (SLIST_FIRST(&unp_defers) == NULL) {
2157                         UNP_DEFERRED_UNLOCK();
2158                         break;
2159                 }
2160                 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2161                 UNP_DEFERRED_UNLOCK();
2162                 count = 0;
2163                 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2164                         SLIST_REMOVE_HEAD(&drl, ud_link);
2165                         closef(dr->ud_fp, NULL);
2166                         free(dr, M_TEMP);
2167                         count++;
2168                 }
2169                 atomic_add_int(&unp_defers_count, -count);
2170         }
2171 }
2172
2173 static void
2174 unp_internalize_fp(struct file *fp)
2175 {
2176         struct unpcb *unp;
2177
2178         UNP_LINK_WLOCK();
2179         if ((unp = fptounp(fp)) != NULL) {
2180                 unp->unp_file = fp;
2181                 unp->unp_msgcount++;
2182         }
2183         fhold(fp);
2184         unp_rights++;
2185         UNP_LINK_WUNLOCK();
2186 }
2187
2188 static int
2189 unp_externalize_fp(struct file *fp)
2190 {
2191         struct unpcb *unp;
2192         int ret;
2193
2194         UNP_LINK_WLOCK();
2195         if ((unp = fptounp(fp)) != NULL) {
2196                 unp->unp_msgcount--;
2197                 ret = 1;
2198         } else
2199                 ret = 0;
2200         unp_rights--;
2201         UNP_LINK_WUNLOCK();
2202         return (ret);
2203 }
2204
2205 /*
2206  * unp_defer indicates whether additional work has been defered for a future
2207  * pass through unp_gc().  It is thread local and does not require explicit
2208  * synchronization.
2209  */
2210 static int      unp_marked;
2211 static int      unp_unreachable;
2212
2213 static void
2214 unp_accessable(struct filedescent **fdep, int fdcount)
2215 {
2216         struct unpcb *unp;
2217         struct file *fp;
2218         int i;
2219
2220         for (i = 0; i < fdcount; i++) {
2221                 fp = fdep[i]->fde_file;
2222                 if ((unp = fptounp(fp)) == NULL)
2223                         continue;
2224                 if (unp->unp_gcflag & UNPGC_REF)
2225                         continue;
2226                 unp->unp_gcflag &= ~UNPGC_DEAD;
2227                 unp->unp_gcflag |= UNPGC_REF;
2228                 unp_marked++;
2229         }
2230 }
2231
2232 static void
2233 unp_gc_process(struct unpcb *unp)
2234 {
2235         struct socket *soa;
2236         struct socket *so;
2237         struct file *fp;
2238
2239         /* Already processed. */
2240         if (unp->unp_gcflag & UNPGC_SCANNED)
2241                 return;
2242         fp = unp->unp_file;
2243
2244         /*
2245          * Check for a socket potentially in a cycle.  It must be in a
2246          * queue as indicated by msgcount, and this must equal the file
2247          * reference count.  Note that when msgcount is 0 the file is NULL.
2248          */
2249         if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2250             unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2251                 unp->unp_gcflag |= UNPGC_DEAD;
2252                 unp_unreachable++;
2253                 return;
2254         }
2255
2256         /*
2257          * Mark all sockets we reference with RIGHTS.
2258          */
2259         so = unp->unp_socket;
2260         if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2261                 SOCKBUF_LOCK(&so->so_rcv);
2262                 unp_scan(so->so_rcv.sb_mb, unp_accessable);
2263                 SOCKBUF_UNLOCK(&so->so_rcv);
2264         }
2265
2266         /*
2267          * Mark all sockets in our accept queue.
2268          */
2269         ACCEPT_LOCK();
2270         TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2271                 if ((sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS) != 0)
2272                         continue;
2273                 SOCKBUF_LOCK(&soa->so_rcv);
2274                 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2275                 SOCKBUF_UNLOCK(&soa->so_rcv);
2276         }
2277         ACCEPT_UNLOCK();
2278         unp->unp_gcflag |= UNPGC_SCANNED;
2279 }
2280
2281 static int unp_recycled;
2282 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2283     "Number of unreachable sockets claimed by the garbage collector.");
2284
2285 static int unp_taskcount;
2286 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2287     "Number of times the garbage collector has run.");
2288
2289 static void
2290 unp_gc(__unused void *arg, int pending)
2291 {
2292         struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2293                                     NULL };
2294         struct unp_head **head;
2295         struct file *f, **unref;
2296         struct unpcb *unp;
2297         int i, total;
2298
2299         unp_taskcount++;
2300         UNP_LIST_LOCK();
2301         /*
2302          * First clear all gc flags from previous runs, apart from
2303          * UNPGC_IGNORE_RIGHTS.
2304          */
2305         for (head = heads; *head != NULL; head++)
2306                 LIST_FOREACH(unp, *head, unp_link)
2307                         unp->unp_gcflag =
2308                             (unp->unp_gcflag & UNPGC_IGNORE_RIGHTS);
2309
2310         /*
2311          * Scan marking all reachable sockets with UNPGC_REF.  Once a socket
2312          * is reachable all of the sockets it references are reachable.
2313          * Stop the scan once we do a complete loop without discovering
2314          * a new reachable socket.
2315          */
2316         do {
2317                 unp_unreachable = 0;
2318                 unp_marked = 0;
2319                 for (head = heads; *head != NULL; head++)
2320                         LIST_FOREACH(unp, *head, unp_link)
2321                                 unp_gc_process(unp);
2322         } while (unp_marked);
2323         UNP_LIST_UNLOCK();
2324         if (unp_unreachable == 0)
2325                 return;
2326
2327         /*
2328          * Allocate space for a local list of dead unpcbs.
2329          */
2330         unref = malloc(unp_unreachable * sizeof(struct file *),
2331             M_TEMP, M_WAITOK);
2332
2333         /*
2334          * Iterate looking for sockets which have been specifically marked
2335          * as as unreachable and store them locally.
2336          */
2337         UNP_LINK_RLOCK();
2338         UNP_LIST_LOCK();
2339         for (total = 0, head = heads; *head != NULL; head++)
2340                 LIST_FOREACH(unp, *head, unp_link)
2341                         if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2342                                 f = unp->unp_file;
2343                                 if (unp->unp_msgcount == 0 || f == NULL ||
2344                                     f->f_count != unp->unp_msgcount)
2345                                         continue;
2346                                 unref[total++] = f;
2347                                 fhold(f);
2348                                 KASSERT(total <= unp_unreachable,
2349                                     ("unp_gc: incorrect unreachable count."));
2350                         }
2351         UNP_LIST_UNLOCK();
2352         UNP_LINK_RUNLOCK();
2353
2354         /*
2355          * Now flush all sockets, free'ing rights.  This will free the
2356          * struct files associated with these sockets but leave each socket
2357          * with one remaining ref.
2358          */
2359         for (i = 0; i < total; i++) {
2360                 struct socket *so;
2361
2362                 so = unref[i]->f_data;
2363                 CURVNET_SET(so->so_vnet);
2364                 sorflush(so);
2365                 CURVNET_RESTORE();
2366         }
2367
2368         /*
2369          * And finally release the sockets so they can be reclaimed.
2370          */
2371         for (i = 0; i < total; i++)
2372                 fdrop(unref[i], NULL);
2373         unp_recycled += total;
2374         free(unref, M_TEMP);
2375 }
2376
2377 static void
2378 unp_dispose_mbuf(struct mbuf *m)
2379 {
2380
2381         if (m)
2382                 unp_scan(m, unp_freerights);
2383 }
2384
2385 /*
2386  * Synchronize against unp_gc, which can trip over data as we are freeing it.
2387  */
2388 static void
2389 unp_dispose(struct socket *so)
2390 {
2391         struct unpcb *unp;
2392
2393         unp = sotounpcb(so);
2394         UNP_LIST_LOCK();
2395         unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2396         UNP_LIST_UNLOCK();
2397         unp_dispose_mbuf(so->so_rcv.sb_mb);
2398 }
2399
2400 static void
2401 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2402 {
2403         struct mbuf *m;
2404         struct cmsghdr *cm;
2405         void *data;
2406         socklen_t clen, datalen;
2407
2408         while (m0 != NULL) {
2409                 for (m = m0; m; m = m->m_next) {
2410                         if (m->m_type != MT_CONTROL)
2411                                 continue;
2412
2413                         cm = mtod(m, struct cmsghdr *);
2414                         clen = m->m_len;
2415
2416                         while (cm != NULL) {
2417                                 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2418                                         break;
2419
2420                                 data = CMSG_DATA(cm);
2421                                 datalen = (caddr_t)cm + cm->cmsg_len
2422                                     - (caddr_t)data;
2423
2424                                 if (cm->cmsg_level == SOL_SOCKET &&
2425                                     cm->cmsg_type == SCM_RIGHTS) {
2426                                         (*op)(data, datalen /
2427                                             sizeof(struct filedescent *));
2428                                 }
2429
2430                                 if (CMSG_SPACE(datalen) < clen) {
2431                                         clen -= CMSG_SPACE(datalen);
2432                                         cm = (struct cmsghdr *)
2433                                             ((caddr_t)cm + CMSG_SPACE(datalen));
2434                                 } else {
2435                                         clen = 0;
2436                                         cm = NULL;
2437                                 }
2438                         }
2439                 }
2440                 m0 = m0->m_nextpkt;
2441         }
2442 }
2443
2444 /*
2445  * A helper function called by VFS before socket-type vnode reclamation.
2446  * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2447  * use count.
2448  */
2449 void
2450 vfs_unp_reclaim(struct vnode *vp)
2451 {
2452         struct unpcb *unp;
2453         int active;
2454
2455         ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2456         KASSERT(vp->v_type == VSOCK,
2457             ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2458
2459         active = 0;
2460         UNP_LINK_WLOCK();
2461         VOP_UNP_CONNECT(vp, &unp);
2462         if (unp == NULL)
2463                 goto done;
2464         UNP_PCB_LOCK(unp);
2465         if (unp->unp_vnode == vp) {
2466                 VOP_UNP_DETACH(vp);
2467                 unp->unp_vnode = NULL;
2468                 active = 1;
2469         }
2470         UNP_PCB_UNLOCK(unp);
2471 done:
2472         UNP_LINK_WUNLOCK();
2473         if (active)
2474                 vunref(vp);
2475 }
2476
2477 #ifdef DDB
2478 static void
2479 db_print_indent(int indent)
2480 {
2481         int i;
2482
2483         for (i = 0; i < indent; i++)
2484                 db_printf(" ");
2485 }
2486
2487 static void
2488 db_print_unpflags(int unp_flags)
2489 {
2490         int comma;
2491
2492         comma = 0;
2493         if (unp_flags & UNP_HAVEPC) {
2494                 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2495                 comma = 1;
2496         }
2497         if (unp_flags & UNP_WANTCRED) {
2498                 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2499                 comma = 1;
2500         }
2501         if (unp_flags & UNP_CONNWAIT) {
2502                 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2503                 comma = 1;
2504         }
2505         if (unp_flags & UNP_CONNECTING) {
2506                 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2507                 comma = 1;
2508         }
2509         if (unp_flags & UNP_BINDING) {
2510                 db_printf("%sUNP_BINDING", comma ? ", " : "");
2511                 comma = 1;
2512         }
2513 }
2514
2515 static void
2516 db_print_xucred(int indent, struct xucred *xu)
2517 {
2518         int comma, i;
2519
2520         db_print_indent(indent);
2521         db_printf("cr_version: %u   cr_uid: %u   cr_ngroups: %d\n",
2522             xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2523         db_print_indent(indent);
2524         db_printf("cr_groups: ");
2525         comma = 0;
2526         for (i = 0; i < xu->cr_ngroups; i++) {
2527                 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2528                 comma = 1;
2529         }
2530         db_printf("\n");
2531 }
2532
2533 static void
2534 db_print_unprefs(int indent, struct unp_head *uh)
2535 {
2536         struct unpcb *unp;
2537         int counter;
2538
2539         counter = 0;
2540         LIST_FOREACH(unp, uh, unp_reflink) {
2541                 if (counter % 4 == 0)
2542                         db_print_indent(indent);
2543                 db_printf("%p  ", unp);
2544                 if (counter % 4 == 3)
2545                         db_printf("\n");
2546                 counter++;
2547         }
2548         if (counter != 0 && counter % 4 != 0)
2549                 db_printf("\n");
2550 }
2551
2552 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2553 {
2554         struct unpcb *unp;
2555
2556         if (!have_addr) {
2557                 db_printf("usage: show unpcb <addr>\n");
2558                 return;
2559         }
2560         unp = (struct unpcb *)addr;
2561
2562         db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
2563             unp->unp_vnode);
2564
2565         db_printf("unp_ino: %ju   unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2566             unp->unp_conn);
2567
2568         db_printf("unp_refs:\n");
2569         db_print_unprefs(2, &unp->unp_refs);
2570
2571         /* XXXRW: Would be nice to print the full address, if any. */
2572         db_printf("unp_addr: %p\n", unp->unp_addr);
2573
2574         db_printf("unp_gencnt: %llu\n",
2575             (unsigned long long)unp->unp_gencnt);
2576
2577         db_printf("unp_flags: %x (", unp->unp_flags);
2578         db_print_unpflags(unp->unp_flags);
2579         db_printf(")\n");
2580
2581         db_printf("unp_peercred:\n");
2582         db_print_xucred(2, &unp->unp_peercred);
2583
2584         db_printf("unp_refcount: %u\n", unp->unp_refcount);
2585 }
2586 #endif