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->unp_socket);
 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         /*
 674          * XXXRW: Should assert vp->v_socket == so.
 675          */
 676         if ((vp = unp->unp_vnode) != NULL) {
 677                 VOP_UNP_DETACH(vp);
 678                 unp->unp_vnode = NULL;
 679         }
 680         unp2 = unp->unp_conn;
 681         if (unp2 != NULL) {
 682                 UNP_PCB_LOCK(unp2);
 683                 unp_disconnect(unp, unp2);
 684                 UNP_PCB_UNLOCK(unp2);
 685         }
 686
 687         /*
 688          * We hold the linkage lock exclusively, so it's OK to acquire
 689          * multiple pcb locks at a time.
 690          */
 691         while (!LIST_EMPTY(&unp->unp_refs)) {
 692                 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
 693
 694                 UNP_PCB_LOCK(ref);
 695                 unp_drop(ref);
 696                 UNP_PCB_UNLOCK(ref);
 697         }
 698         local_unp_rights = unp_rights;
 699         UNP_LINK_WUNLOCK();
 700 teardown:
 701         unp->unp_socket->so_pcb = NULL;
 702         saved_unp_addr = unp->unp_addr;
 703         unp->unp_addr = NULL;
 704         unp->unp_refcount--;
 705         freeunp = (unp->unp_refcount == 0);
 706         if (saved_unp_addr != NULL)
 707                 free(saved_unp_addr, M_SONAME);
 708         if (freeunp) {
 709                 UNP_PCB_LOCK_DESTROY(unp);
 710                 uma_zfree(unp_zone, unp);
 711         } else
 712                 UNP_PCB_UNLOCK(unp);
 713         if (vp)
 714                 vrele(vp);
 715         if (local_unp_rights)
 716                 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
 717 }
 718
 719 static int
 720 uipc_disconnect(struct socket *so)
 721 {
 722         struct unpcb *unp, *unp2;
 723
 724         unp = sotounpcb(so);
 725         KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
 726
 727         UNP_LINK_WLOCK();
 728         UNP_PCB_LOCK(unp);
 729         unp2 = unp->unp_conn;
 730         if (unp2 != NULL) {
 731                 UNP_PCB_LOCK(unp2);
 732                 unp_disconnect(unp, unp2);
 733                 UNP_PCB_UNLOCK(unp2);
 734         }
 735         UNP_PCB_UNLOCK(unp);
 736         UNP_LINK_WUNLOCK();
 737         return (0);
 738 }
 739
 740 static int
 741 uipc_listen(struct socket *so, int backlog, struct thread *td)
 742 {
 743         struct unpcb *unp;
 744         int error;
 745
 746         if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
 747                 return (EOPNOTSUPP);
 748
 749         unp = sotounpcb(so);
 750         KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
 751
 752         UNP_PCB_LOCK(unp);
 753         if (unp->unp_vnode == NULL) {
 754                 /* Already connected or not bound to an address. */
 755                 error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ;
 756                 UNP_PCB_UNLOCK(unp);
 757                 return (error);
 758         }
 759
 760         SOCK_LOCK(so);
 761         error = solisten_proto_check(so);
 762         if (error == 0) {
 763                 cru2x(td->td_ucred, &unp->unp_peercred);
 764                 unp->unp_flags |= UNP_HAVEPCCACHED;
 765                 solisten_proto(so, backlog);
 766         }
 767         SOCK_UNLOCK(so);
 768         UNP_PCB_UNLOCK(unp);
 769         return (error);
 770 }
 771
 772 static int
 773 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
 774 {
 775         struct unpcb *unp, *unp2;
 776         const struct sockaddr *sa;
 777
 778         unp = sotounpcb(so);
 779         KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
 780
 781         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 782         UNP_LINK_RLOCK();
 783         /*
 784          * XXX: It seems that this test always fails even when connection is
 785          * established.  So, this else clause is added as workaround to
 786          * return PF_LOCAL sockaddr.
 787          */
 788         unp2 = unp->unp_conn;
 789         if (unp2 != NULL) {
 790                 UNP_PCB_LOCK(unp2);
 791                 if (unp2->unp_addr != NULL)
 792                         sa = (struct sockaddr *) unp2->unp_addr;
 793                 else
 794                         sa = &sun_noname;
 795                 bcopy(sa, *nam, sa->sa_len);
 796                 UNP_PCB_UNLOCK(unp2);
 797         } else {
 798                 sa = &sun_noname;
 799                 bcopy(sa, *nam, sa->sa_len);
 800         }
 801         UNP_LINK_RUNLOCK();
 802         return (0);
 803 }
 804
 805 static int
 806 uipc_rcvd(struct socket *so, int flags)
 807 {
 808         struct unpcb *unp, *unp2;
 809         struct socket *so2;
 810         u_int mbcnt, sbcc;
 811
 812         unp = sotounpcb(so);
 813         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
 814         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
 815             ("%s: socktype %d", __func__, so->so_type));
 816
 817         /*
 818          * Adjust backpressure on sender and wakeup any waiting to write.
 819          *
 820          * The unp lock is acquired to maintain the validity of the unp_conn
 821          * pointer; no lock on unp2 is required as unp2->unp_socket will be
 822          * static as long as we don't permit unp2 to disconnect from unp,
 823          * which is prevented by the lock on unp.  We cache values from
 824          * so_rcv to avoid holding the so_rcv lock over the entire
 825          * transaction on the remote so_snd.
 826          */
 827         SOCKBUF_LOCK(&so->so_rcv);
 828         mbcnt = so->so_rcv.sb_mbcnt;
 829         sbcc = sbavail(&so->so_rcv);
 830         SOCKBUF_UNLOCK(&so->so_rcv);
 831         /*
 832          * There is a benign race condition at this point.  If we're planning to
 833          * clear SB_STOP, but uipc_send is called on the connected socket at
 834          * this instant, it might add data to the sockbuf and set SB_STOP.  Then
 835          * we would erroneously clear SB_STOP below, even though the sockbuf is
 836          * full.  The race is benign because the only ill effect is to allow the
 837          * sockbuf to exceed its size limit, and the size limits are not
 838          * strictly guaranteed anyway.
 839          */
 840         UNP_PCB_LOCK(unp);
 841         unp2 = unp->unp_conn;
 842         if (unp2 == NULL) {
 843                 UNP_PCB_UNLOCK(unp);
 844                 return (0);
 845         }
 846         so2 = unp2->unp_socket;
 847         SOCKBUF_LOCK(&so2->so_snd);
 848         if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
 849                 so2->so_snd.sb_flags &= ~SB_STOP;
 850         sowwakeup_locked(so2);
 851         UNP_PCB_UNLOCK(unp);
 852         return (0);
 853 }
 854
 855 static int
 856 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
 857     struct mbuf *control, struct thread *td)
 858 {
 859         struct unpcb *unp, *unp2;
 860         struct socket *so2;
 861         u_int mbcnt, sbcc;
 862         int error = 0;
 863
 864         unp = sotounpcb(so);
 865         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
 866         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
 867             so->so_type == SOCK_SEQPACKET,
 868             ("%s: socktype %d", __func__, so->so_type));
 869
 870         if (flags & PRUS_OOB) {
 871                 error = EOPNOTSUPP;
 872                 goto release;
 873         }
 874         if (control != NULL && (error = unp_internalize(&control, td)))
 875                 goto release;
 876         if ((nam != NULL) || (flags & PRUS_EOF))
 877                 UNP_LINK_WLOCK();
 878         else
 879                 UNP_LINK_RLOCK();
 880         switch (so->so_type) {
 881         case SOCK_DGRAM:
 882         {
 883                 const struct sockaddr *from;
 884
 885                 unp2 = unp->unp_conn;
 886                 if (nam != NULL) {
 887                         UNP_LINK_WLOCK_ASSERT();
 888                         if (unp2 != NULL) {
 889                                 error = EISCONN;
 890                                 break;
 891                         }
 892                         error = unp_connect(so, nam, td);
 893                         if (error)
 894                                 break;
 895                         unp2 = unp->unp_conn;
 896                 }
 897
 898                 /*
 899                  * Because connect() and send() are non-atomic in a sendto()
 900                  * with a target address, it's possible that the socket will
 901                  * have disconnected before the send() can run.  In that case
 902                  * return the slightly counter-intuitive but otherwise
 903                  * correct error that the socket is not connected.
 904                  */
 905                 if (unp2 == NULL) {
 906                         error = ENOTCONN;
 907                         break;
 908                 }
 909                 /* Lockless read. */
 910                 if (unp2->unp_flags & UNP_WANTCRED)
 911                         control = unp_addsockcred(td, control);
 912                 UNP_PCB_LOCK(unp);
 913                 if (unp->unp_addr != NULL)
 914                         from = (struct sockaddr *)unp->unp_addr;
 915                 else
 916                         from = &sun_noname;
 917                 so2 = unp2->unp_socket;
 918                 SOCKBUF_LOCK(&so2->so_rcv);
 919                 if (sbappendaddr_locked(&so2->so_rcv, from, m,
 920                     control)) {
 921                         sorwakeup_locked(so2);
 922                         m = NULL;
 923                         control = NULL;
 924                 } else {
 925                         SOCKBUF_UNLOCK(&so2->so_rcv);
 926                         error = ENOBUFS;
 927                 }
 928                 if (nam != NULL) {
 929                         UNP_LINK_WLOCK_ASSERT();
 930                         UNP_PCB_LOCK(unp2);
 931                         unp_disconnect(unp, unp2);
 932                         UNP_PCB_UNLOCK(unp2);
 933                 }
 934                 UNP_PCB_UNLOCK(unp);
 935                 break;
 936         }
 937
 938         case SOCK_SEQPACKET:
 939         case SOCK_STREAM:
 940                 if ((so->so_state & SS_ISCONNECTED) == 0) {
 941                         if (nam != NULL) {
 942                                 UNP_LINK_WLOCK_ASSERT();
 943                                 error = unp_connect(so, nam, td);
 944                                 if (error)
 945                                         break;  /* XXX */
 946                         } else {
 947                                 error = ENOTCONN;
 948                                 break;
 949                         }
 950                 }
 951
 952                 /* Lockless read. */
 953                 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
 954                         error = EPIPE;
 955                         break;
 956                 }
 957
 958                 /*
 959                  * Because connect() and send() are non-atomic in a sendto()
 960                  * with a target address, it's possible that the socket will
 961                  * have disconnected before the send() can run.  In that case
 962                  * return the slightly counter-intuitive but otherwise
 963                  * correct error that the socket is not connected.
 964                  *
 965                  * Locking here must be done carefully: the linkage lock
 966                  * prevents interconnections between unpcbs from changing, so
 967                  * we can traverse from unp to unp2 without acquiring unp's
 968                  * lock.  Socket buffer locks follow unpcb locks, so we can
 969                  * acquire both remote and lock socket buffer locks.
 970                  */
 971                 unp2 = unp->unp_conn;
 972                 if (unp2 == NULL) {
 973                         error = ENOTCONN;
 974                         break;
 975                 }
 976                 so2 = unp2->unp_socket;
 977                 UNP_PCB_LOCK(unp2);
 978                 SOCKBUF_LOCK(&so2->so_rcv);
 979                 if (unp2->unp_flags & UNP_WANTCRED) {
 980                         /*
 981                          * Credentials are passed only once on SOCK_STREAM
 982                          * and SOCK_SEQPACKET.
 983                          */
 984                         unp2->unp_flags &= ~UNP_WANTCRED;
 985                         control = unp_addsockcred(td, control);
 986                 }
 987                 /*
 988                  * Send to paired receive port, and then reduce send buffer
 989                  * hiwater marks to maintain backpressure.  Wake up readers.
 990                  */
 991                 switch (so->so_type) {
 992                 case SOCK_STREAM:
 993                         if (control != NULL) {
 994                                 if (sbappendcontrol_locked(&so2->so_rcv, m,
 995                                     control))
 996                                         control = NULL;
 997                         } else
 998                                 sbappend_locked(&so2->so_rcv, m, flags);
 999                         break;
1000
1001                 case SOCK_SEQPACKET: {
1002                         const struct sockaddr *from;
1003
1004                         from = &sun_noname;
1005                         /*
1006                          * Don't check for space available in so2->so_rcv.
1007                          * Unix domain sockets only check for space in the
1008                          * sending sockbuf, and that check is performed one
1009                          * level up the stack.
1010                          */
1011                         if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1012                                 from, m, control))
1013                                 control = NULL;
1014                         break;
1015                         }
1016                 }
1017
1018                 mbcnt = so2->so_rcv.sb_mbcnt;
1019                 sbcc = sbavail(&so2->so_rcv);
1020                 if (sbcc)
1021                         sorwakeup_locked(so2);
1022                 else
1023                         SOCKBUF_UNLOCK(&so2->so_rcv);
1024
1025                 /*
1026                  * The PCB lock on unp2 protects the SB_STOP flag.  Without it,
1027                  * it would be possible for uipc_rcvd to be called at this
1028                  * point, drain the receiving sockbuf, clear SB_STOP, and then
1029                  * we would set SB_STOP below.  That could lead to an empty
1030                  * sockbuf having SB_STOP set
1031                  */
1032                 SOCKBUF_LOCK(&so->so_snd);
1033                 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1034                         so->so_snd.sb_flags |= SB_STOP;
1035                 SOCKBUF_UNLOCK(&so->so_snd);
1036                 UNP_PCB_UNLOCK(unp2);
1037                 m = NULL;
1038                 break;
1039         }
1040
1041         /*
1042          * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1043          */
1044         if (flags & PRUS_EOF) {
1045                 UNP_PCB_LOCK(unp);
1046                 socantsendmore(so);
1047                 unp_shutdown(unp);
1048                 UNP_PCB_UNLOCK(unp);
1049         }
1050
1051         if ((nam != NULL) || (flags & PRUS_EOF))
1052                 UNP_LINK_WUNLOCK();
1053         else
1054                 UNP_LINK_RUNLOCK();
1055
1056         if (control != NULL && error != 0)
1057                 unp_dispose_mbuf(control);
1058
1059 release:
1060         if (control != NULL)
1061                 m_freem(control);
1062         if (m != NULL)
1063                 m_freem(m);
1064         return (error);
1065 }
1066
1067 static int
1068 uipc_ready(struct socket *so, struct mbuf *m, int count)
1069 {
1070         struct unpcb *unp, *unp2;
1071         struct socket *so2;
1072         int error;
1073
1074         unp = sotounpcb(so);
1075
1076         UNP_LINK_RLOCK();
1077         unp2 = unp->unp_conn;
1078         UNP_PCB_LOCK(unp2);
1079         so2 = unp2->unp_socket;
1080
1081         SOCKBUF_LOCK(&so2->so_rcv);
1082         if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1083                 sorwakeup_locked(so2);
1084         else
1085                 SOCKBUF_UNLOCK(&so2->so_rcv);
1086
1087         UNP_PCB_UNLOCK(unp2);
1088         UNP_LINK_RUNLOCK();
1089
1090         return (error);
1091 }
1092
1093 static int
1094 uipc_sense(struct socket *so, struct stat *sb)
1095 {
1096         struct unpcb *unp;
1097
1098         unp = sotounpcb(so);
1099         KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1100
1101         sb->st_blksize = so->so_snd.sb_hiwat;
1102         UNP_PCB_LOCK(unp);
1103         sb->st_dev = NODEV;
1104         if (unp->unp_ino == 0)
1105                 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1106         sb->st_ino = unp->unp_ino;
1107         UNP_PCB_UNLOCK(unp);
1108         return (0);
1109 }
1110
1111 static int
1112 uipc_shutdown(struct socket *so)
1113 {
1114         struct unpcb *unp;
1115
1116         unp = sotounpcb(so);
1117         KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1118
1119         UNP_LINK_WLOCK();
1120         UNP_PCB_LOCK(unp);
1121         socantsendmore(so);
1122         unp_shutdown(unp);
1123         UNP_PCB_UNLOCK(unp);
1124         UNP_LINK_WUNLOCK();
1125         return (0);
1126 }
1127
1128 static int
1129 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1130 {
1131         struct unpcb *unp;
1132         const struct sockaddr *sa;
1133
1134         unp = sotounpcb(so);
1135         KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1136
1137         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1138         UNP_PCB_LOCK(unp);
1139         if (unp->unp_addr != NULL)
1140                 sa = (struct sockaddr *) unp->unp_addr;
1141         else
1142                 sa = &sun_noname;
1143         bcopy(sa, *nam, sa->sa_len);
1144         UNP_PCB_UNLOCK(unp);
1145         return (0);
1146 }
1147
1148 static struct pr_usrreqs uipc_usrreqs_dgram = {
1149         .pru_abort =            uipc_abort,
1150         .pru_accept =           uipc_accept,
1151         .pru_attach =           uipc_attach,
1152         .pru_bind =             uipc_bind,
1153         .pru_bindat =           uipc_bindat,
1154         .pru_connect =          uipc_connect,
1155         .pru_connectat =        uipc_connectat,
1156         .pru_connect2 =         uipc_connect2,
1157         .pru_detach =           uipc_detach,
1158         .pru_disconnect =       uipc_disconnect,
1159         .pru_listen =           uipc_listen,
1160         .pru_peeraddr =         uipc_peeraddr,
1161         .pru_rcvd =             uipc_rcvd,
1162         .pru_send =             uipc_send,
1163         .pru_sense =            uipc_sense,
1164         .pru_shutdown =         uipc_shutdown,
1165         .pru_sockaddr =         uipc_sockaddr,
1166         .pru_soreceive =        soreceive_dgram,
1167         .pru_close =            uipc_close,
1168 };
1169
1170 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1171         .pru_abort =            uipc_abort,
1172         .pru_accept =           uipc_accept,
1173         .pru_attach =           uipc_attach,
1174         .pru_bind =             uipc_bind,
1175         .pru_bindat =           uipc_bindat,
1176         .pru_connect =          uipc_connect,
1177         .pru_connectat =        uipc_connectat,
1178         .pru_connect2 =         uipc_connect2,
1179         .pru_detach =           uipc_detach,
1180         .pru_disconnect =       uipc_disconnect,
1181         .pru_listen =           uipc_listen,
1182         .pru_peeraddr =         uipc_peeraddr,
1183         .pru_rcvd =             uipc_rcvd,
1184         .pru_send =             uipc_send,
1185         .pru_sense =            uipc_sense,
1186         .pru_shutdown =         uipc_shutdown,
1187         .pru_sockaddr =         uipc_sockaddr,
1188         .pru_soreceive =        soreceive_generic,      /* XXX: or...? */
1189         .pru_close =            uipc_close,
1190 };
1191
1192 static struct pr_usrreqs uipc_usrreqs_stream = {
1193         .pru_abort =            uipc_abort,
1194         .pru_accept =           uipc_accept,
1195         .pru_attach =           uipc_attach,
1196         .pru_bind =             uipc_bind,
1197         .pru_bindat =           uipc_bindat,
1198         .pru_connect =          uipc_connect,
1199         .pru_connectat =        uipc_connectat,
1200         .pru_connect2 =         uipc_connect2,
1201         .pru_detach =           uipc_detach,
1202         .pru_disconnect =       uipc_disconnect,
1203         .pru_listen =           uipc_listen,
1204         .pru_peeraddr =         uipc_peeraddr,
1205         .pru_rcvd =             uipc_rcvd,
1206         .pru_send =             uipc_send,
1207         .pru_ready =            uipc_ready,
1208         .pru_sense =            uipc_sense,
1209         .pru_shutdown =         uipc_shutdown,
1210         .pru_sockaddr =         uipc_sockaddr,
1211         .pru_soreceive =        soreceive_generic,
1212         .pru_close =            uipc_close,
1213 };
1214
1215 static int
1216 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1217 {
1218         struct unpcb *unp;
1219         struct xucred xu;
1220         int error, optval;
1221
1222         if (sopt->sopt_level != 0)
1223                 return (EINVAL);
1224
1225         unp = sotounpcb(so);
1226         KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1227         error = 0;
1228         switch (sopt->sopt_dir) {
1229         case SOPT_GET:
1230                 switch (sopt->sopt_name) {
1231                 case LOCAL_PEERCRED:
1232                         UNP_PCB_LOCK(unp);
1233                         if (unp->unp_flags & UNP_HAVEPC)
1234                                 xu = unp->unp_peercred;
1235                         else {
1236                                 if (so->so_type == SOCK_STREAM)
1237                                         error = ENOTCONN;
1238                                 else
1239                                         error = EINVAL;
1240                         }
1241                         UNP_PCB_UNLOCK(unp);
1242                         if (error == 0)
1243                                 error = sooptcopyout(sopt, &xu, sizeof(xu));
1244                         break;
1245
1246                 case LOCAL_CREDS:
1247                         /* Unlocked read. */
1248                         optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1249                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1250                         break;
1251
1252                 case LOCAL_CONNWAIT:
1253                         /* Unlocked read. */
1254                         optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1255                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1256                         break;
1257
1258                 default:
1259                         error = EOPNOTSUPP;
1260                         break;
1261                 }
1262                 break;
1263
1264         case SOPT_SET:
1265                 switch (sopt->sopt_name) {
1266                 case LOCAL_CREDS:
1267                 case LOCAL_CONNWAIT:
1268                         error = sooptcopyin(sopt, &optval, sizeof(optval),
1269                                             sizeof(optval));
1270                         if (error)
1271                                 break;
1272
1273 #define OPTSET(bit) do {                                                \
1274         UNP_PCB_LOCK(unp);                                              \
1275         if (optval)                                                     \
1276                 unp->unp_flags |= bit;                                  \
1277         else                                                            \
1278                 unp->unp_flags &= ~bit;                                 \
1279         UNP_PCB_UNLOCK(unp);                                            \
1280 } while (0)
1281
1282                         switch (sopt->sopt_name) {
1283                         case LOCAL_CREDS:
1284                                 OPTSET(UNP_WANTCRED);
1285                                 break;
1286
1287                         case LOCAL_CONNWAIT:
1288                                 OPTSET(UNP_CONNWAIT);
1289                                 break;
1290
1291                         default:
1292                                 break;
1293                         }
1294                         break;
1295 #undef  OPTSET
1296                 default:
1297                         error = ENOPROTOOPT;
1298                         break;
1299                 }
1300                 break;
1301
1302         default:
1303                 error = EOPNOTSUPP;
1304                 break;
1305         }
1306         return (error);
1307 }
1308
1309 static int
1310 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1311 {
1312
1313         return (unp_connectat(AT_FDCWD, so, nam, td));
1314 }
1315
1316 static int
1317 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1318     struct thread *td)
1319 {
1320         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1321         struct vnode *vp;
1322         struct socket *so2, *so3;
1323         struct unpcb *unp, *unp2, *unp3;
1324         struct nameidata nd;
1325         char buf[SOCK_MAXADDRLEN];
1326         struct sockaddr *sa;
1327         cap_rights_t rights;
1328         int error, len;
1329
1330         if (nam->sa_family != AF_UNIX)
1331                 return (EAFNOSUPPORT);
1332
1333         UNP_LINK_WLOCK_ASSERT();
1334
1335         unp = sotounpcb(so);
1336         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1337
1338         if (nam->sa_len > sizeof(struct sockaddr_un))
1339                 return (EINVAL);
1340         len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1341         if (len <= 0)
1342                 return (EINVAL);
1343         bcopy(soun->sun_path, buf, len);
1344         buf[len] = 0;
1345
1346         UNP_PCB_LOCK(unp);
1347         if (unp->unp_flags & UNP_CONNECTING) {
1348                 UNP_PCB_UNLOCK(unp);
1349                 return (EALREADY);
1350         }
1351         UNP_LINK_WUNLOCK();
1352         unp->unp_flags |= UNP_CONNECTING;
1353         UNP_PCB_UNLOCK(unp);
1354
1355         sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1356         NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1357             UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1358         error = namei(&nd);
1359         if (error)
1360                 vp = NULL;
1361         else
1362                 vp = nd.ni_vp;
1363         ASSERT_VOP_LOCKED(vp, "unp_connect");
1364         NDFREE(&nd, NDF_ONLY_PNBUF);
1365         if (error)
1366                 goto bad;
1367
1368         if (vp->v_type != VSOCK) {
1369                 error = ENOTSOCK;
1370                 goto bad;
1371         }
1372 #ifdef MAC
1373         error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1374         if (error)
1375                 goto bad;
1376 #endif
1377         error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1378         if (error)
1379                 goto bad;
1380
1381         unp = sotounpcb(so);
1382         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1383
1384         /*
1385          * Lock linkage lock for two reasons: make sure v_socket is stable,
1386          * and to protect simultaneous locking of multiple pcbs.
1387          */
1388         UNP_LINK_WLOCK();
1389         VOP_UNP_CONNECT(vp, &so2);
1390         if (so2 == NULL) {
1391                 error = ECONNREFUSED;
1392                 goto bad2;
1393         }
1394         if (so->so_type != so2->so_type) {
1395                 error = EPROTOTYPE;
1396                 goto bad2;
1397         }
1398         if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1399                 if (so2->so_options & SO_ACCEPTCONN) {
1400                         CURVNET_SET(so2->so_vnet);
1401                         so3 = sonewconn(so2, 0);
1402                         CURVNET_RESTORE();
1403                 } else
1404                         so3 = NULL;
1405                 if (so3 == NULL) {
1406                         error = ECONNREFUSED;
1407                         goto bad2;
1408                 }
1409                 unp = sotounpcb(so);
1410                 unp2 = sotounpcb(so2);
1411                 unp3 = sotounpcb(so3);
1412                 UNP_PCB_LOCK(unp);
1413                 UNP_PCB_LOCK(unp2);
1414                 UNP_PCB_LOCK(unp3);
1415                 if (unp2->unp_addr != NULL) {
1416                         bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1417                         unp3->unp_addr = (struct sockaddr_un *) sa;
1418                         sa = NULL;
1419                 }
1420
1421                 /*
1422                  * The connector's (client's) credentials are copied from its
1423                  * process structure at the time of connect() (which is now).
1424                  */
1425                 cru2x(td->td_ucred, &unp3->unp_peercred);
1426                 unp3->unp_flags |= UNP_HAVEPC;
1427
1428                 /*
1429                  * The receiver's (server's) credentials are copied from the
1430                  * unp_peercred member of socket on which the former called
1431                  * listen(); uipc_listen() cached that process's credentials
1432                  * at that time so we can use them now.
1433                  */
1434                 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1435                     ("unp_connect: listener without cached peercred"));
1436                 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1437                     sizeof(unp->unp_peercred));
1438                 unp->unp_flags |= UNP_HAVEPC;
1439                 if (unp2->unp_flags & UNP_WANTCRED)
1440                         unp3->unp_flags |= UNP_WANTCRED;
1441                 UNP_PCB_UNLOCK(unp3);
1442                 UNP_PCB_UNLOCK(unp2);
1443                 UNP_PCB_UNLOCK(unp);
1444 #ifdef MAC
1445                 mac_socketpeer_set_from_socket(so, so3);
1446                 mac_socketpeer_set_from_socket(so3, so);
1447 #endif
1448
1449                 so2 = so3;
1450         }
1451         unp = sotounpcb(so);
1452         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1453         unp2 = sotounpcb(so2);
1454         KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1455         UNP_PCB_LOCK(unp);
1456         UNP_PCB_LOCK(unp2);
1457         error = unp_connect2(so, so2, PRU_CONNECT);
1458         UNP_PCB_UNLOCK(unp2);
1459         UNP_PCB_UNLOCK(unp);
1460 bad2:
1461         UNP_LINK_WUNLOCK();
1462 bad:
1463         if (vp != NULL)
1464                 vput(vp);
1465         free(sa, M_SONAME);
1466         UNP_LINK_WLOCK();
1467         UNP_PCB_LOCK(unp);
1468         unp->unp_flags &= ~UNP_CONNECTING;
1469         UNP_PCB_UNLOCK(unp);
1470         return (error);
1471 }
1472
1473 static int
1474 unp_connect2(struct socket *so, struct socket *so2, int req)
1475 {
1476         struct unpcb *unp;
1477         struct unpcb *unp2;
1478
1479         unp = sotounpcb(so);
1480         KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1481         unp2 = sotounpcb(so2);
1482         KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1483
1484         UNP_LINK_WLOCK_ASSERT();
1485         UNP_PCB_LOCK_ASSERT(unp);
1486         UNP_PCB_LOCK_ASSERT(unp2);
1487
1488         if (so2->so_type != so->so_type)
1489                 return (EPROTOTYPE);
1490         unp2->unp_flags &= ~UNP_NASCENT;
1491         unp->unp_conn = unp2;
1492
1493         switch (so->so_type) {
1494         case SOCK_DGRAM:
1495                 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1496                 soisconnected(so);
1497                 break;
1498
1499         case SOCK_STREAM:
1500         case SOCK_SEQPACKET:
1501                 unp2->unp_conn = unp;
1502                 if (req == PRU_CONNECT &&
1503                     ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1504                         soisconnecting(so);
1505                 else
1506                         soisconnected(so);
1507                 soisconnected(so2);
1508                 break;
1509
1510         default:
1511                 panic("unp_connect2");
1512         }
1513         return (0);
1514 }
1515
1516 static void
1517 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1518 {
1519         struct socket *so;
1520
1521         KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1522
1523         UNP_LINK_WLOCK_ASSERT();
1524         UNP_PCB_LOCK_ASSERT(unp);
1525         UNP_PCB_LOCK_ASSERT(unp2);
1526
1527         unp->unp_conn = NULL;
1528         switch (unp->unp_socket->so_type) {
1529         case SOCK_DGRAM:
1530                 LIST_REMOVE(unp, unp_reflink);
1531                 so = unp->unp_socket;
1532                 SOCK_LOCK(so);
1533                 so->so_state &= ~SS_ISCONNECTED;
1534                 SOCK_UNLOCK(so);
1535                 break;
1536
1537         case SOCK_STREAM:
1538         case SOCK_SEQPACKET:
1539                 soisdisconnected(unp->unp_socket);
1540                 unp2->unp_conn = NULL;
1541                 soisdisconnected(unp2->unp_socket);
1542                 break;
1543         }
1544 }
1545
1546 /*
1547  * unp_pcblist() walks the global list of struct unpcb's to generate a
1548  * pointer list, bumping the refcount on each unpcb.  It then copies them out
1549  * sequentially, validating the generation number on each to see if it has
1550  * been detached.  All of this is necessary because copyout() may sleep on
1551  * disk I/O.
1552  */
1553 static int
1554 unp_pcblist(SYSCTL_HANDLER_ARGS)
1555 {
1556         int error, i, n;
1557         int freeunp;
1558         struct unpcb *unp, **unp_list;
1559         unp_gen_t gencnt;
1560         struct xunpgen *xug;
1561         struct unp_head *head;
1562         struct xunpcb *xu;
1563
1564         switch ((intptr_t)arg1) {
1565         case SOCK_STREAM:
1566                 head = &unp_shead;
1567                 break;
1568
1569         case SOCK_DGRAM:
1570                 head = &unp_dhead;
1571                 break;
1572
1573         case SOCK_SEQPACKET:
1574                 head = &unp_sphead;
1575                 break;
1576
1577         default:
1578                 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1579         }
1580
1581         /*
1582          * The process of preparing the PCB list is too time-consuming and
1583          * resource-intensive to repeat twice on every request.
1584          */
1585         if (req->oldptr == NULL) {
1586                 n = unp_count;
1587                 req->oldidx = 2 * (sizeof *xug)
1588                         + (n + n/8) * sizeof(struct xunpcb);
1589                 return (0);
1590         }
1591
1592         if (req->newptr != NULL)
1593                 return (EPERM);
1594
1595         /*
1596          * OK, now we're committed to doing something.
1597          */
1598         xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1599         UNP_LIST_LOCK();
1600         gencnt = unp_gencnt;
1601         n = unp_count;
1602         UNP_LIST_UNLOCK();
1603
1604         xug->xug_len = sizeof *xug;
1605         xug->xug_count = n;
1606         xug->xug_gen = gencnt;
1607         xug->xug_sogen = so_gencnt;
1608         error = SYSCTL_OUT(req, xug, sizeof *xug);
1609         if (error) {
1610                 free(xug, M_TEMP);
1611                 return (error);
1612         }
1613
1614         unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1615
1616         UNP_LIST_LOCK();
1617         for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1618              unp = LIST_NEXT(unp, unp_link)) {
1619                 UNP_PCB_LOCK(unp);
1620                 if (unp->unp_gencnt <= gencnt) {
1621                         if (cr_cansee(req->td->td_ucred,
1622                             unp->unp_socket->so_cred)) {
1623                                 UNP_PCB_UNLOCK(unp);
1624                                 continue;
1625                         }
1626                         unp_list[i++] = unp;
1627                         unp->unp_refcount++;
1628                 }
1629                 UNP_PCB_UNLOCK(unp);
1630         }
1631         UNP_LIST_UNLOCK();
1632         n = i;                  /* In case we lost some during malloc. */
1633
1634         error = 0;
1635         xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1636         for (i = 0; i < n; i++) {
1637                 unp = unp_list[i];
1638                 UNP_PCB_LOCK(unp);
1639                 unp->unp_refcount--;
1640                 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1641                         xu->xu_len = sizeof *xu;
1642                         xu->xu_unpp = unp;
1643                         /*
1644                          * XXX - need more locking here to protect against
1645                          * connect/disconnect races for SMP.
1646                          */
1647                         if (unp->unp_addr != NULL)
1648                                 bcopy(unp->unp_addr, &xu->xu_addr,
1649                                       unp->unp_addr->sun_len);
1650                         if (unp->unp_conn != NULL &&
1651                             unp->unp_conn->unp_addr != NULL)
1652                                 bcopy(unp->unp_conn->unp_addr,
1653                                       &xu->xu_caddr,
1654                                       unp->unp_conn->unp_addr->sun_len);
1655                         bcopy(unp, &xu->xu_unp, sizeof *unp);
1656                         sotoxsocket(unp->unp_socket, &xu->xu_socket);
1657                         UNP_PCB_UNLOCK(unp);
1658                         error = SYSCTL_OUT(req, xu, sizeof *xu);
1659                 } else {
1660                         freeunp = (unp->unp_refcount == 0);
1661                         UNP_PCB_UNLOCK(unp);
1662                         if (freeunp) {
1663                                 UNP_PCB_LOCK_DESTROY(unp);
1664                                 uma_zfree(unp_zone, unp);
1665                         }
1666                 }
1667         }
1668         free(xu, M_TEMP);
1669         if (!error) {
1670                 /*
1671                  * Give the user an updated idea of our state.  If the
1672                  * generation differs from what we told her before, she knows
1673                  * that something happened while we were processing this
1674                  * request, and it might be necessary to retry.
1675                  */
1676                 xug->xug_gen = unp_gencnt;
1677                 xug->xug_sogen = so_gencnt;
1678                 xug->xug_count = unp_count;
1679                 error = SYSCTL_OUT(req, xug, sizeof *xug);
1680         }
1681         free(unp_list, M_TEMP);
1682         free(xug, M_TEMP);
1683         return (error);
1684 }
1685
1686 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1687     (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1688     "List of active local datagram sockets");
1689 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1690     (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1691     "List of active local stream sockets");
1692 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1693     CTLTYPE_OPAQUE | CTLFLAG_RD,
1694     (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1695     "List of active local seqpacket sockets");
1696
1697 static void
1698 unp_shutdown(struct unpcb *unp)
1699 {
1700         struct unpcb *unp2;
1701         struct socket *so;
1702
1703         UNP_LINK_WLOCK_ASSERT();
1704         UNP_PCB_LOCK_ASSERT(unp);
1705
1706         unp2 = unp->unp_conn;
1707         if ((unp->unp_socket->so_type == SOCK_STREAM ||
1708             (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1709                 so = unp2->unp_socket;
1710                 if (so != NULL)
1711                         socantrcvmore(so);
1712         }
1713 }
1714
1715 static void
1716 unp_drop(struct unpcb *unp)
1717 {
1718         struct socket *so = unp->unp_socket;
1719         struct unpcb *unp2;
1720
1721         UNP_LINK_WLOCK_ASSERT();
1722         UNP_PCB_LOCK_ASSERT(unp);
1723
1724         /*
1725          * Regardless of whether the socket's peer dropped the connection
1726          * with this socket by aborting or disconnecting, POSIX requires
1727          * that ECONNRESET is returned.
1728          */
1729         so->so_error = ECONNRESET;
1730         unp2 = unp->unp_conn;
1731         if (unp2 == NULL)
1732                 return;
1733         UNP_PCB_LOCK(unp2);
1734         unp_disconnect(unp, unp2);
1735         UNP_PCB_UNLOCK(unp2);
1736 }
1737
1738 static void
1739 unp_freerights(struct filedescent **fdep, int fdcount)
1740 {
1741         struct file *fp;
1742         int i;
1743
1744         KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1745
1746         for (i = 0; i < fdcount; i++) {
1747                 fp = fdep[i]->fde_file;
1748                 filecaps_free(&fdep[i]->fde_caps);
1749                 unp_discard(fp);
1750         }
1751         free(fdep[0], M_FILECAPS);
1752 }
1753
1754 static int
1755 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1756 {
1757         struct thread *td = curthread;          /* XXX */
1758         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1759         int i;
1760         int *fdp;
1761         struct filedesc *fdesc = td->td_proc->p_fd;
1762         struct filedescent **fdep;
1763         void *data;
1764         socklen_t clen = control->m_len, datalen;
1765         int error, newfds;
1766         u_int newlen;
1767
1768         UNP_LINK_UNLOCK_ASSERT();
1769
1770         error = 0;
1771         if (controlp != NULL) /* controlp == NULL => free control messages */
1772                 *controlp = NULL;
1773         while (cm != NULL) {
1774                 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1775                         error = EINVAL;
1776                         break;
1777                 }
1778                 data = CMSG_DATA(cm);
1779                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1780                 if (cm->cmsg_level == SOL_SOCKET
1781                     && cm->cmsg_type == SCM_RIGHTS) {
1782                         newfds = datalen / sizeof(*fdep);
1783                         if (newfds == 0)
1784                                 goto next;
1785                         fdep = data;
1786
1787                         /* If we're not outputting the descriptors free them. */
1788                         if (error || controlp == NULL) {
1789                                 unp_freerights(fdep, newfds);
1790                                 goto next;
1791                         }
1792                         FILEDESC_XLOCK(fdesc);
1793
1794                         /*
1795                          * Now change each pointer to an fd in the global
1796                          * table to an integer that is the index to the local
1797                          * fd table entry that we set up to point to the
1798                          * global one we are transferring.
1799                          */
1800                         newlen = newfds * sizeof(int);
1801                         *controlp = sbcreatecontrol(NULL, newlen,
1802                             SCM_RIGHTS, SOL_SOCKET);
1803                         if (*controlp == NULL) {
1804                                 FILEDESC_XUNLOCK(fdesc);
1805                                 error = E2BIG;
1806                                 unp_freerights(fdep, newfds);
1807                                 goto next;
1808                         }
1809
1810                         fdp = (int *)
1811                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1812                         if (fdallocn(td, 0, fdp, newfds) != 0) {
1813                                 FILEDESC_XUNLOCK(fdesc);
1814                                 error = EMSGSIZE;
1815                                 unp_freerights(fdep, newfds);
1816                                 m_freem(*controlp);
1817                                 *controlp = NULL;
1818                                 goto next;
1819                         }
1820                         for (i = 0; i < newfds; i++, fdp++) {
1821                                 _finstall(fdesc, fdep[i]->fde_file, *fdp,
1822                                     (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0,
1823                                     &fdep[i]->fde_caps);
1824                                 unp_externalize_fp(fdep[i]->fde_file);
1825                         }
1826                         FILEDESC_XUNLOCK(fdesc);
1827                         free(fdep[0], M_FILECAPS);
1828                 } else {
1829                         /* We can just copy anything else across. */
1830                         if (error || controlp == NULL)
1831                                 goto next;
1832                         *controlp = sbcreatecontrol(NULL, datalen,
1833                             cm->cmsg_type, cm->cmsg_level);
1834                         if (*controlp == NULL) {
1835                                 error = ENOBUFS;
1836                                 goto next;
1837                         }
1838                         bcopy(data,
1839                             CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1840                             datalen);
1841                 }
1842                 controlp = &(*controlp)->m_next;
1843
1844 next:
1845                 if (CMSG_SPACE(datalen) < clen) {
1846                         clen -= CMSG_SPACE(datalen);
1847                         cm = (struct cmsghdr *)
1848                             ((caddr_t)cm + CMSG_SPACE(datalen));
1849                 } else {
1850                         clen = 0;
1851                         cm = NULL;
1852                 }
1853         }
1854
1855         m_freem(control);
1856         return (error);
1857 }
1858
1859 static void
1860 unp_zone_change(void *tag)
1861 {
1862
1863         uma_zone_set_max(unp_zone, maxsockets);
1864 }
1865
1866 static void
1867 unp_init(void)
1868 {
1869
1870 #ifdef VIMAGE
1871         if (!IS_DEFAULT_VNET(curvnet))
1872                 return;
1873 #endif
1874         unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1875             NULL, NULL, UMA_ALIGN_PTR, 0);
1876         if (unp_zone == NULL)
1877                 panic("unp_init");
1878         uma_zone_set_max(unp_zone, maxsockets);
1879         uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
1880         EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1881             NULL, EVENTHANDLER_PRI_ANY);
1882         LIST_INIT(&unp_dhead);
1883         LIST_INIT(&unp_shead);
1884         LIST_INIT(&unp_sphead);
1885         SLIST_INIT(&unp_defers);
1886         TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1887         TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1888         UNP_LINK_LOCK_INIT();
1889         UNP_LIST_LOCK_INIT();
1890         UNP_DEFERRED_LOCK_INIT();
1891 }
1892
1893 static int
1894 unp_internalize(struct mbuf **controlp, struct thread *td)
1895 {
1896         struct mbuf *control = *controlp;
1897         struct proc *p = td->td_proc;
1898         struct filedesc *fdesc = p->p_fd;
1899         struct bintime *bt;
1900         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1901         struct cmsgcred *cmcred;
1902         struct filedescent *fde, **fdep, *fdev;
1903         struct file *fp;
1904         struct timeval *tv;
1905         struct timespec *ts;
1906         int i, *fdp;
1907         void *data;
1908         socklen_t clen = control->m_len, datalen;
1909         int error, oldfds;
1910         u_int newlen;
1911
1912         UNP_LINK_UNLOCK_ASSERT();
1913
1914         error = 0;
1915         *controlp = NULL;
1916         while (cm != NULL) {
1917                 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1918                     || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
1919                         error = EINVAL;
1920                         goto out;
1921                 }
1922                 data = CMSG_DATA(cm);
1923                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1924
1925                 switch (cm->cmsg_type) {
1926                 /*
1927                  * Fill in credential information.
1928                  */
1929                 case SCM_CREDS:
1930                         *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1931                             SCM_CREDS, SOL_SOCKET);
1932                         if (*controlp == NULL) {
1933                                 error = ENOBUFS;
1934                                 goto out;
1935                         }
1936                         cmcred = (struct cmsgcred *)
1937                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1938                         cmcred->cmcred_pid = p->p_pid;
1939                         cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1940                         cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1941                         cmcred->cmcred_euid = td->td_ucred->cr_uid;
1942                         cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1943                             CMGROUP_MAX);
1944                         for (i = 0; i < cmcred->cmcred_ngroups; i++)
1945                                 cmcred->cmcred_groups[i] =
1946                                     td->td_ucred->cr_groups[i];
1947                         break;
1948
1949                 case SCM_RIGHTS:
1950                         oldfds = datalen / sizeof (int);
1951                         if (oldfds == 0)
1952                                 break;
1953                         /*
1954                          * Check that all the FDs passed in refer to legal
1955                          * files.  If not, reject the entire operation.
1956                          */
1957                         fdp = data;
1958                         FILEDESC_SLOCK(fdesc);
1959                         for (i = 0; i < oldfds; i++, fdp++) {
1960                                 fp = fget_locked(fdesc, *fdp);
1961                                 if (fp == NULL) {
1962                                         FILEDESC_SUNLOCK(fdesc);
1963                                         error = EBADF;
1964                                         goto out;
1965                                 }
1966                                 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1967                                         FILEDESC_SUNLOCK(fdesc);
1968                                         error = EOPNOTSUPP;
1969                                         goto out;
1970                                 }
1971
1972                         }
1973
1974                         /*
1975                          * Now replace the integer FDs with pointers to the
1976                          * file structure and capability rights.
1977                          */
1978                         newlen = oldfds * sizeof(fdep[0]);
1979                         *controlp = sbcreatecontrol(NULL, newlen,
1980                             SCM_RIGHTS, SOL_SOCKET);
1981                         if (*controlp == NULL) {
1982                                 FILEDESC_SUNLOCK(fdesc);
1983                                 error = E2BIG;
1984                                 goto out;
1985                         }
1986                         fdp = data;
1987                         fdep = (struct filedescent **)
1988                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1989                         fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
1990                             M_WAITOK);
1991                         for (i = 0; i < oldfds; i++, fdev++, fdp++) {
1992                                 fde = &fdesc->fd_ofiles[*fdp];
1993                                 fdep[i] = fdev;
1994                                 fdep[i]->fde_file = fde->fde_file;
1995                                 filecaps_copy(&fde->fde_caps,
1996                                     &fdep[i]->fde_caps, true);
1997                                 unp_internalize_fp(fdep[i]->fde_file);
1998                         }
1999                         FILEDESC_SUNLOCK(fdesc);
2000                         break;
2001
2002                 case SCM_TIMESTAMP:
2003                         *controlp = sbcreatecontrol(NULL, sizeof(*tv),
2004                             SCM_TIMESTAMP, SOL_SOCKET);
2005                         if (*controlp == NULL) {
2006                                 error = ENOBUFS;
2007                                 goto out;
2008                         }
2009                         tv = (struct timeval *)
2010                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2011                         microtime(tv);
2012                         break;
2013
2014                 case SCM_BINTIME:
2015                         *controlp = sbcreatecontrol(NULL, sizeof(*bt),
2016                             SCM_BINTIME, SOL_SOCKET);
2017                         if (*controlp == NULL) {
2018                                 error = ENOBUFS;
2019                                 goto out;
2020                         }
2021                         bt = (struct bintime *)
2022                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2023                         bintime(bt);
2024                         break;
2025
2026                 case SCM_REALTIME:
2027                         *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2028                             SCM_REALTIME, SOL_SOCKET);
2029                         if (*controlp == NULL) {
2030                                 error = ENOBUFS;
2031                                 goto out;
2032                         }
2033                         ts = (struct timespec *)
2034                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2035                         nanotime(ts);
2036                         break;
2037
2038                 case SCM_MONOTONIC:
2039                         *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2040                             SCM_MONOTONIC, SOL_SOCKET);
2041                         if (*controlp == NULL) {
2042                                 error = ENOBUFS;
2043                                 goto out;
2044                         }
2045                         ts = (struct timespec *)
2046                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2047                         nanouptime(ts);
2048                         break;
2049
2050                 default:
2051                         error = EINVAL;
2052                         goto out;
2053                 }
2054
2055                 controlp = &(*controlp)->m_next;
2056                 if (CMSG_SPACE(datalen) < clen) {
2057                         clen -= CMSG_SPACE(datalen);
2058                         cm = (struct cmsghdr *)
2059                             ((caddr_t)cm + CMSG_SPACE(datalen));
2060                 } else {
2061                         clen = 0;
2062                         cm = NULL;
2063                 }
2064         }
2065
2066 out:
2067         m_freem(control);
2068         return (error);
2069 }
2070
2071 static struct mbuf *
2072 unp_addsockcred(struct thread *td, struct mbuf *control)
2073 {
2074         struct mbuf *m, *n, *n_prev;
2075         struct sockcred *sc;
2076         const struct cmsghdr *cm;
2077         int ngroups;
2078         int i;
2079
2080         ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2081         m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2082         if (m == NULL)
2083                 return (control);
2084
2085         sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2086         sc->sc_uid = td->td_ucred->cr_ruid;
2087         sc->sc_euid = td->td_ucred->cr_uid;
2088         sc->sc_gid = td->td_ucred->cr_rgid;
2089         sc->sc_egid = td->td_ucred->cr_gid;
2090         sc->sc_ngroups = ngroups;
2091         for (i = 0; i < sc->sc_ngroups; i++)
2092                 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2093
2094         /*
2095          * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2096          * created SCM_CREDS control message (struct sockcred) has another
2097          * format.
2098          */
2099         if (control != NULL)
2100                 for (n = control, n_prev = NULL; n != NULL;) {
2101                         cm = mtod(n, struct cmsghdr *);
2102                         if (cm->cmsg_level == SOL_SOCKET &&
2103                             cm->cmsg_type == SCM_CREDS) {
2104                                 if (n_prev == NULL)
2105                                         control = n->m_next;
2106                                 else
2107                                         n_prev->m_next = n->m_next;
2108                                 n = m_free(n);
2109                         } else {
2110                                 n_prev = n;
2111                                 n = n->m_next;
2112                         }
2113                 }
2114
2115         /* Prepend it to the head. */
2116         m->m_next = control;
2117         return (m);
2118 }
2119
2120 static struct unpcb *
2121 fptounp(struct file *fp)
2122 {
2123         struct socket *so;
2124
2125         if (fp->f_type != DTYPE_SOCKET)
2126                 return (NULL);
2127         if ((so = fp->f_data) == NULL)
2128                 return (NULL);
2129         if (so->so_proto->pr_domain != &localdomain)
2130                 return (NULL);
2131         return sotounpcb(so);
2132 }
2133
2134 static void
2135 unp_discard(struct file *fp)
2136 {
2137         struct unp_defer *dr;
2138
2139         if (unp_externalize_fp(fp)) {
2140                 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2141                 dr->ud_fp = fp;
2142                 UNP_DEFERRED_LOCK();
2143                 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2144                 UNP_DEFERRED_UNLOCK();
2145                 atomic_add_int(&unp_defers_count, 1);
2146                 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2147         } else
2148                 (void) closef(fp, (struct thread *)NULL);
2149 }
2150
2151 static void
2152 unp_process_defers(void *arg __unused, int pending)
2153 {
2154         struct unp_defer *dr;
2155         SLIST_HEAD(, unp_defer) drl;
2156         int count;
2157
2158         SLIST_INIT(&drl);
2159         for (;;) {
2160                 UNP_DEFERRED_LOCK();
2161                 if (SLIST_FIRST(&unp_defers) == NULL) {
2162                         UNP_DEFERRED_UNLOCK();
2163                         break;
2164                 }
2165                 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2166                 UNP_DEFERRED_UNLOCK();
2167                 count = 0;
2168                 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2169                         SLIST_REMOVE_HEAD(&drl, ud_link);
2170                         closef(dr->ud_fp, NULL);
2171                         free(dr, M_TEMP);
2172                         count++;
2173                 }
2174                 atomic_add_int(&unp_defers_count, -count);
2175         }
2176 }
2177
2178 static void
2179 unp_internalize_fp(struct file *fp)
2180 {
2181         struct unpcb *unp;
2182
2183         UNP_LINK_WLOCK();
2184         if ((unp = fptounp(fp)) != NULL) {
2185                 unp->unp_file = fp;
2186                 unp->unp_msgcount++;
2187         }
2188         fhold(fp);
2189         unp_rights++;
2190         UNP_LINK_WUNLOCK();
2191 }
2192
2193 static int
2194 unp_externalize_fp(struct file *fp)
2195 {
2196         struct unpcb *unp;
2197         int ret;
2198
2199         UNP_LINK_WLOCK();
2200         if ((unp = fptounp(fp)) != NULL) {
2201                 unp->unp_msgcount--;
2202                 ret = 1;
2203         } else
2204                 ret = 0;
2205         unp_rights--;
2206         UNP_LINK_WUNLOCK();
2207         return (ret);
2208 }
2209
2210 /*
2211  * unp_defer indicates whether additional work has been defered for a future
2212  * pass through unp_gc().  It is thread local and does not require explicit
2213  * synchronization.
2214  */
2215 static int      unp_marked;
2216 static int      unp_unreachable;
2217
2218 static void
2219 unp_accessable(struct filedescent **fdep, int fdcount)
2220 {
2221         struct unpcb *unp;
2222         struct file *fp;
2223         int i;
2224
2225         for (i = 0; i < fdcount; i++) {
2226                 fp = fdep[i]->fde_file;
2227                 if ((unp = fptounp(fp)) == NULL)
2228                         continue;
2229                 if (unp->unp_gcflag & UNPGC_REF)
2230                         continue;
2231                 unp->unp_gcflag &= ~UNPGC_DEAD;
2232                 unp->unp_gcflag |= UNPGC_REF;
2233                 unp_marked++;
2234         }
2235 }
2236
2237 static void
2238 unp_gc_process(struct unpcb *unp)
2239 {
2240         struct socket *soa;
2241         struct socket *so;
2242         struct file *fp;
2243
2244         /* Already processed. */
2245         if (unp->unp_gcflag & UNPGC_SCANNED)
2246                 return;
2247         fp = unp->unp_file;
2248
2249         /*
2250          * Check for a socket potentially in a cycle.  It must be in a
2251          * queue as indicated by msgcount, and this must equal the file
2252          * reference count.  Note that when msgcount is 0 the file is NULL.
2253          */
2254         if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2255             unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2256                 unp->unp_gcflag |= UNPGC_DEAD;
2257                 unp_unreachable++;
2258                 return;
2259         }
2260
2261         /*
2262          * Mark all sockets we reference with RIGHTS.
2263          */
2264         so = unp->unp_socket;
2265         if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2266                 SOCKBUF_LOCK(&so->so_rcv);
2267                 unp_scan(so->so_rcv.sb_mb, unp_accessable);
2268                 SOCKBUF_UNLOCK(&so->so_rcv);
2269         }
2270
2271         /*
2272          * Mark all sockets in our accept queue.
2273          */
2274         ACCEPT_LOCK();
2275         TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2276                 if ((sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS) != 0)
2277                         continue;
2278                 SOCKBUF_LOCK(&soa->so_rcv);
2279                 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2280                 SOCKBUF_UNLOCK(&soa->so_rcv);
2281         }
2282         ACCEPT_UNLOCK();
2283         unp->unp_gcflag |= UNPGC_SCANNED;
2284 }
2285
2286 static int unp_recycled;
2287 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2288     "Number of unreachable sockets claimed by the garbage collector.");
2289
2290 static int unp_taskcount;
2291 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2292     "Number of times the garbage collector has run.");
2293
2294 static void
2295 unp_gc(__unused void *arg, int pending)
2296 {
2297         struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2298                                     NULL };
2299         struct unp_head **head;
2300         struct file *f, **unref;
2301         struct unpcb *unp;
2302         int i, total;
2303
2304         unp_taskcount++;
2305         UNP_LIST_LOCK();
2306         /*
2307          * First clear all gc flags from previous runs, apart from
2308          * UNPGC_IGNORE_RIGHTS.
2309          */
2310         for (head = heads; *head != NULL; head++)
2311                 LIST_FOREACH(unp, *head, unp_link)
2312                         unp->unp_gcflag =
2313                             (unp->unp_gcflag & UNPGC_IGNORE_RIGHTS);
2314
2315         /*
2316          * Scan marking all reachable sockets with UNPGC_REF.  Once a socket
2317          * is reachable all of the sockets it references are reachable.
2318          * Stop the scan once we do a complete loop without discovering
2319          * a new reachable socket.
2320          */
2321         do {
2322                 unp_unreachable = 0;
2323                 unp_marked = 0;
2324                 for (head = heads; *head != NULL; head++)
2325                         LIST_FOREACH(unp, *head, unp_link)
2326                                 unp_gc_process(unp);
2327         } while (unp_marked);
2328         UNP_LIST_UNLOCK();
2329         if (unp_unreachable == 0)
2330                 return;
2331
2332         /*
2333          * Allocate space for a local list of dead unpcbs.
2334          */
2335         unref = malloc(unp_unreachable * sizeof(struct file *),
2336             M_TEMP, M_WAITOK);
2337
2338         /*
2339          * Iterate looking for sockets which have been specifically marked
2340          * as as unreachable and store them locally.
2341          */
2342         UNP_LINK_RLOCK();
2343         UNP_LIST_LOCK();
2344         for (total = 0, head = heads; *head != NULL; head++)
2345                 LIST_FOREACH(unp, *head, unp_link)
2346                         if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2347                                 f = unp->unp_file;
2348                                 if (unp->unp_msgcount == 0 || f == NULL ||
2349                                     f->f_count != unp->unp_msgcount)
2350                                         continue;
2351                                 unref[total++] = f;
2352                                 fhold(f);
2353                                 KASSERT(total <= unp_unreachable,
2354                                     ("unp_gc: incorrect unreachable count."));
2355                         }
2356         UNP_LIST_UNLOCK();
2357         UNP_LINK_RUNLOCK();
2358
2359         /*
2360          * Now flush all sockets, free'ing rights.  This will free the
2361          * struct files associated with these sockets but leave each socket
2362          * with one remaining ref.
2363          */
2364         for (i = 0; i < total; i++) {
2365                 struct socket *so;
2366
2367                 so = unref[i]->f_data;
2368                 CURVNET_SET(so->so_vnet);
2369                 sorflush(so);
2370                 CURVNET_RESTORE();
2371         }
2372
2373         /*
2374          * And finally release the sockets so they can be reclaimed.
2375          */
2376         for (i = 0; i < total; i++)
2377                 fdrop(unref[i], NULL);
2378         unp_recycled += total;
2379         free(unref, M_TEMP);
2380 }
2381
2382 static void
2383 unp_dispose_mbuf(struct mbuf *m)
2384 {
2385
2386         if (m)
2387                 unp_scan(m, unp_freerights);
2388 }
2389
2390 /*
2391  * Synchronize against unp_gc, which can trip over data as we are freeing it.
2392  */
2393 static void
2394 unp_dispose(struct socket *so)
2395 {
2396         struct unpcb *unp;
2397
2398         unp = sotounpcb(so);
2399         UNP_LIST_LOCK();
2400         unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2401         UNP_LIST_UNLOCK();
2402         unp_dispose_mbuf(so->so_rcv.sb_mb);
2403 }
2404
2405 static void
2406 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2407 {
2408         struct mbuf *m;
2409         struct cmsghdr *cm;
2410         void *data;
2411         socklen_t clen, datalen;
2412
2413         while (m0 != NULL) {
2414                 for (m = m0; m; m = m->m_next) {
2415                         if (m->m_type != MT_CONTROL)
2416                                 continue;
2417
2418                         cm = mtod(m, struct cmsghdr *);
2419                         clen = m->m_len;
2420
2421                         while (cm != NULL) {
2422                                 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2423                                         break;
2424
2425                                 data = CMSG_DATA(cm);
2426                                 datalen = (caddr_t)cm + cm->cmsg_len
2427                                     - (caddr_t)data;
2428
2429                                 if (cm->cmsg_level == SOL_SOCKET &&
2430                                     cm->cmsg_type == SCM_RIGHTS) {
2431                                         (*op)(data, datalen /
2432                                             sizeof(struct filedescent *));
2433                                 }
2434
2435                                 if (CMSG_SPACE(datalen) < clen) {
2436                                         clen -= CMSG_SPACE(datalen);
2437                                         cm = (struct cmsghdr *)
2438                                             ((caddr_t)cm + CMSG_SPACE(datalen));
2439                                 } else {
2440                                         clen = 0;
2441                                         cm = NULL;
2442                                 }
2443                         }
2444                 }
2445                 m0 = m0->m_nextpkt;
2446         }
2447 }
2448
2449 /*
2450  * A helper function called by VFS before socket-type vnode reclamation.
2451  * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2452  * use count.
2453  */
2454 void
2455 vfs_unp_reclaim(struct vnode *vp)
2456 {
2457         struct socket *so;
2458         struct unpcb *unp;
2459         int active;
2460
2461         ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2462         KASSERT(vp->v_type == VSOCK,
2463             ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2464
2465         active = 0;
2466         UNP_LINK_WLOCK();
2467         VOP_UNP_CONNECT(vp, &so);
2468         if (so == NULL)
2469                 goto done;
2470         unp = sotounpcb(so);
2471         if (unp == NULL)
2472                 goto done;
2473         UNP_PCB_LOCK(unp);
2474         if (unp->unp_vnode == vp) {
2475                 VOP_UNP_DETACH(vp);
2476                 unp->unp_vnode = NULL;
2477                 active = 1;
2478         }
2479         UNP_PCB_UNLOCK(unp);
2480 done:
2481         UNP_LINK_WUNLOCK();
2482         if (active)
2483                 vunref(vp);
2484 }
2485
2486 #ifdef DDB
2487 static void
2488 db_print_indent(int indent)
2489 {
2490         int i;
2491
2492         for (i = 0; i < indent; i++)
2493                 db_printf(" ");
2494 }
2495
2496 static void
2497 db_print_unpflags(int unp_flags)
2498 {
2499         int comma;
2500
2501         comma = 0;
2502         if (unp_flags & UNP_HAVEPC) {
2503                 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2504                 comma = 1;
2505         }
2506         if (unp_flags & UNP_HAVEPCCACHED) {
2507                 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2508                 comma = 1;
2509         }
2510         if (unp_flags & UNP_WANTCRED) {
2511                 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2512                 comma = 1;
2513         }
2514         if (unp_flags & UNP_CONNWAIT) {
2515                 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2516                 comma = 1;
2517         }
2518         if (unp_flags & UNP_CONNECTING) {
2519                 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2520                 comma = 1;
2521         }
2522         if (unp_flags & UNP_BINDING) {
2523                 db_printf("%sUNP_BINDING", comma ? ", " : "");
2524                 comma = 1;
2525         }
2526 }
2527
2528 static void
2529 db_print_xucred(int indent, struct xucred *xu)
2530 {
2531         int comma, i;
2532
2533         db_print_indent(indent);
2534         db_printf("cr_version: %u   cr_uid: %u   cr_ngroups: %d\n",
2535             xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2536         db_print_indent(indent);
2537         db_printf("cr_groups: ");
2538         comma = 0;
2539         for (i = 0; i < xu->cr_ngroups; i++) {
2540                 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2541                 comma = 1;
2542         }
2543         db_printf("\n");
2544 }
2545
2546 static void
2547 db_print_unprefs(int indent, struct unp_head *uh)
2548 {
2549         struct unpcb *unp;
2550         int counter;
2551
2552         counter = 0;
2553         LIST_FOREACH(unp, uh, unp_reflink) {
2554                 if (counter % 4 == 0)
2555                         db_print_indent(indent);
2556                 db_printf("%p  ", unp);
2557                 if (counter % 4 == 3)
2558                         db_printf("\n");
2559                 counter++;
2560         }
2561         if (counter != 0 && counter % 4 != 0)
2562                 db_printf("\n");
2563 }
2564
2565 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2566 {
2567         struct unpcb *unp;
2568
2569         if (!have_addr) {
2570                 db_printf("usage: show unpcb <addr>\n");
2571                 return;
2572         }
2573         unp = (struct unpcb *)addr;
2574
2575         db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
2576             unp->unp_vnode);
2577
2578         db_printf("unp_ino: %ju   unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2579             unp->unp_conn);
2580
2581         db_printf("unp_refs:\n");
2582         db_print_unprefs(2, &unp->unp_refs);
2583
2584         /* XXXRW: Would be nice to print the full address, if any. */
2585         db_printf("unp_addr: %p\n", unp->unp_addr);
2586
2587         db_printf("unp_gencnt: %llu\n",
2588             (unsigned long long)unp->unp_gencnt);
2589
2590         db_printf("unp_flags: %x (", unp->unp_flags);
2591         db_print_unpflags(unp->unp_flags);
2592         db_printf(")\n");
2593
2594         db_printf("unp_peercred:\n");
2595         db_print_xucred(2, &unp->unp_peercred);
2596
2597         db_printf("unp_refcount: %u\n", unp->unp_refcount);
2598 }
2599 #endif