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  * 4. Neither the name of the University nor the names of its contributors
  16  *    may be used to endorse or promote products derived from this software
  17  *    without specific prior written permission.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  29  * SUCH DAMAGE.
  30  *
  31  *      From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
  32  */
  33
  34 /*
  35  * UNIX Domain (Local) Sockets
  36  *
  37  * This is an implementation of UNIX (local) domain sockets.  Each socket has
  38  * an associated struct unpcb (UNIX protocol control block).  Stream sockets
  39  * may be connected to 0 or 1 other socket.  Datagram sockets may be
  40  * connected to 0, 1, or many other sockets.  Sockets may be created and
  41  * connected in pairs (socketpair(2)), or bound/connected to using the file
  42  * system name space.  For most purposes, only the receive socket buffer is
  43  * used, as sending on one socket delivers directly to the receive socket
  44  * buffer of a second socket.
  45  *
  46  * The implementation is substantially complicated by the fact that
  47  * "ancillary data", such as file descriptors or credentials, may be passed
  48  * across UNIX domain sockets.  The potential for passing UNIX domain sockets
  49  * over other UNIX domain sockets requires the implementation of a simple
  50  * garbage collector to find and tear down cycles of disconnected sockets.
  51  *
  52  * TODO:
  53  *      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 mbuf *);
 281 static void     unp_dispose_so(struct socket *so);
 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_so,
 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         unp = sotounpcb(so);
 747         KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
 748
 749         UNP_PCB_LOCK(unp);
 750         if (unp->unp_vnode == NULL) {
 751                 /* Already connected or not bound to an address. */
 752                 error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ;
 753                 UNP_PCB_UNLOCK(unp);
 754                 return (error);
 755         }
 756
 757         SOCK_LOCK(so);
 758         error = solisten_proto_check(so);
 759         if (error == 0) {
 760                 cru2x(td->td_ucred, &unp->unp_peercred);
 761                 unp->unp_flags |= UNP_HAVEPCCACHED;
 762                 solisten_proto(so, backlog);
 763         }
 764         SOCK_UNLOCK(so);
 765         UNP_PCB_UNLOCK(unp);
 766         return (error);
 767 }
 768
 769 static int
 770 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
 771 {
 772         struct unpcb *unp, *unp2;
 773         const struct sockaddr *sa;
 774
 775         unp = sotounpcb(so);
 776         KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
 777
 778         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 779         UNP_LINK_RLOCK();
 780         /*
 781          * XXX: It seems that this test always fails even when connection is
 782          * established.  So, this else clause is added as workaround to
 783          * return PF_LOCAL sockaddr.
 784          */
 785         unp2 = unp->unp_conn;
 786         if (unp2 != NULL) {
 787                 UNP_PCB_LOCK(unp2);
 788                 if (unp2->unp_addr != NULL)
 789                         sa = (struct sockaddr *) unp2->unp_addr;
 790                 else
 791                         sa = &sun_noname;
 792                 bcopy(sa, *nam, sa->sa_len);
 793                 UNP_PCB_UNLOCK(unp2);
 794         } else {
 795                 sa = &sun_noname;
 796                 bcopy(sa, *nam, sa->sa_len);
 797         }
 798         UNP_LINK_RUNLOCK();
 799         return (0);
 800 }
 801
 802 static int
 803 uipc_rcvd(struct socket *so, int flags)
 804 {
 805         struct unpcb *unp, *unp2;
 806         struct socket *so2;
 807         u_int mbcnt, sbcc;
 808
 809         unp = sotounpcb(so);
 810         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
 811         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
 812             ("%s: socktype %d", __func__, so->so_type));
 813
 814         /*
 815          * Adjust backpressure on sender and wakeup any waiting to write.
 816          *
 817          * The unp lock is acquired to maintain the validity of the unp_conn
 818          * pointer; no lock on unp2 is required as unp2->unp_socket will be
 819          * static as long as we don't permit unp2 to disconnect from unp,
 820          * which is prevented by the lock on unp.  We cache values from
 821          * so_rcv to avoid holding the so_rcv lock over the entire
 822          * transaction on the remote so_snd.
 823          */
 824         SOCKBUF_LOCK(&so->so_rcv);
 825         mbcnt = so->so_rcv.sb_mbcnt;
 826         sbcc = sbavail(&so->so_rcv);
 827         SOCKBUF_UNLOCK(&so->so_rcv);
 828         /*
 829          * There is a benign race condition at this point.  If we're planning to
 830          * clear SB_STOP, but uipc_send is called on the connected socket at
 831          * this instant, it might add data to the sockbuf and set SB_STOP.  Then
 832          * we would erroneously clear SB_STOP below, even though the sockbuf is
 833          * full.  The race is benign because the only ill effect is to allow the
 834          * sockbuf to exceed its size limit, and the size limits are not
 835          * strictly guaranteed anyway.
 836          */
 837         UNP_PCB_LOCK(unp);
 838         unp2 = unp->unp_conn;
 839         if (unp2 == NULL) {
 840                 UNP_PCB_UNLOCK(unp);
 841                 return (0);
 842         }
 843         so2 = unp2->unp_socket;
 844         SOCKBUF_LOCK(&so2->so_snd);
 845         if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
 846                 so2->so_snd.sb_flags &= ~SB_STOP;
 847         sowwakeup_locked(so2);
 848         UNP_PCB_UNLOCK(unp);
 849         return (0);
 850 }
 851
 852 static int
 853 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
 854     struct mbuf *control, struct thread *td)
 855 {
 856         struct unpcb *unp, *unp2;
 857         struct socket *so2;
 858         u_int mbcnt, sbcc;
 859         int error = 0;
 860
 861         unp = sotounpcb(so);
 862         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
 863         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
 864             so->so_type == SOCK_SEQPACKET,
 865             ("%s: socktype %d", __func__, so->so_type));
 866
 867         if (flags & PRUS_OOB) {
 868                 error = EOPNOTSUPP;
 869                 goto release;
 870         }
 871         if (control != NULL && (error = unp_internalize(&control, td)))
 872                 goto release;
 873         if ((nam != NULL) || (flags & PRUS_EOF))
 874                 UNP_LINK_WLOCK();
 875         else
 876                 UNP_LINK_RLOCK();
 877         switch (so->so_type) {
 878         case SOCK_DGRAM:
 879         {
 880                 const struct sockaddr *from;
 881
 882                 unp2 = unp->unp_conn;
 883                 if (nam != NULL) {
 884                         UNP_LINK_WLOCK_ASSERT();
 885                         if (unp2 != NULL) {
 886                                 error = EISCONN;
 887                                 break;
 888                         }
 889                         error = unp_connect(so, nam, td);
 890                         if (error)
 891                                 break;
 892                         unp2 = unp->unp_conn;
 893                 }
 894
 895                 /*
 896                  * Because connect() and send() are non-atomic in a sendto()
 897                  * with a target address, it's possible that the socket will
 898                  * have disconnected before the send() can run.  In that case
 899                  * return the slightly counter-intuitive but otherwise
 900                  * correct error that the socket is not connected.
 901                  */
 902                 if (unp2 == NULL) {
 903                         error = ENOTCONN;
 904                         break;
 905                 }
 906                 /* Lockless read. */
 907                 if (unp2->unp_flags & UNP_WANTCRED)
 908                         control = unp_addsockcred(td, control);
 909                 UNP_PCB_LOCK(unp);
 910                 if (unp->unp_addr != NULL)
 911                         from = (struct sockaddr *)unp->unp_addr;
 912                 else
 913                         from = &sun_noname;
 914                 so2 = unp2->unp_socket;
 915                 SOCKBUF_LOCK(&so2->so_rcv);
 916                 if (sbappendaddr_locked(&so2->so_rcv, from, m,
 917                     control)) {
 918                         sorwakeup_locked(so2);
 919                         m = NULL;
 920                         control = NULL;
 921                 } else {
 922                         SOCKBUF_UNLOCK(&so2->so_rcv);
 923                         error = ENOBUFS;
 924                 }
 925                 if (nam != NULL) {
 926                         UNP_LINK_WLOCK_ASSERT();
 927                         UNP_PCB_LOCK(unp2);
 928                         unp_disconnect(unp, unp2);
 929                         UNP_PCB_UNLOCK(unp2);
 930                 }
 931                 UNP_PCB_UNLOCK(unp);
 932                 break;
 933         }
 934
 935         case SOCK_SEQPACKET:
 936         case SOCK_STREAM:
 937                 if ((so->so_state & SS_ISCONNECTED) == 0) {
 938                         if (nam != NULL) {
 939                                 UNP_LINK_WLOCK_ASSERT();
 940                                 error = unp_connect(so, nam, td);
 941                                 if (error)
 942                                         break;  /* XXX */
 943                         } else {
 944                                 error = ENOTCONN;
 945                                 break;
 946                         }
 947                 }
 948
 949                 /* Lockless read. */
 950                 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
 951                         error = EPIPE;
 952                         break;
 953                 }
 954
 955                 /*
 956                  * Because connect() and send() are non-atomic in a sendto()
 957                  * with a target address, it's possible that the socket will
 958                  * have disconnected before the send() can run.  In that case
 959                  * return the slightly counter-intuitive but otherwise
 960                  * correct error that the socket is not connected.
 961                  *
 962                  * Locking here must be done carefully: the linkage lock
 963                  * prevents interconnections between unpcbs from changing, so
 964                  * we can traverse from unp to unp2 without acquiring unp's
 965                  * lock.  Socket buffer locks follow unpcb locks, so we can
 966                  * acquire both remote and lock socket buffer locks.
 967                  */
 968                 unp2 = unp->unp_conn;
 969                 if (unp2 == NULL) {
 970                         error = ENOTCONN;
 971                         break;
 972                 }
 973                 so2 = unp2->unp_socket;
 974                 UNP_PCB_LOCK(unp2);
 975                 SOCKBUF_LOCK(&so2->so_rcv);
 976                 if (unp2->unp_flags & UNP_WANTCRED) {
 977                         /*
 978                          * Credentials are passed only once on SOCK_STREAM
 979                          * and SOCK_SEQPACKET.
 980                          */
 981                         unp2->unp_flags &= ~UNP_WANTCRED;
 982                         control = unp_addsockcred(td, control);
 983                 }
 984                 /*
 985                  * Send to paired receive port, and then reduce send buffer
 986                  * hiwater marks to maintain backpressure.  Wake up readers.
 987                  */
 988                 switch (so->so_type) {
 989                 case SOCK_STREAM:
 990                         if (control != NULL) {
 991                                 if (sbappendcontrol_locked(&so2->so_rcv, m,
 992                                     control))
 993                                         control = NULL;
 994                         } else
 995                                 sbappend_locked(&so2->so_rcv, m, flags);
 996                         break;
 997
 998                 case SOCK_SEQPACKET: {
 999                         const struct sockaddr *from;
1000
1001                         from = &sun_noname;
1002                         /*
1003                          * Don't check for space available in so2->so_rcv.
1004                          * Unix domain sockets only check for space in the
1005                          * sending sockbuf, and that check is performed one
1006                          * level up the stack.
1007                          */
1008                         if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1009                                 from, m, control))
1010                                 control = NULL;
1011                         break;
1012                         }
1013                 }
1014
1015                 mbcnt = so2->so_rcv.sb_mbcnt;
1016                 sbcc = sbavail(&so2->so_rcv);
1017                 if (sbcc)
1018                         sorwakeup_locked(so2);
1019                 else
1020                         SOCKBUF_UNLOCK(&so2->so_rcv);
1021
1022                 /*
1023                  * The PCB lock on unp2 protects the SB_STOP flag.  Without it,
1024                  * it would be possible for uipc_rcvd to be called at this
1025                  * point, drain the receiving sockbuf, clear SB_STOP, and then
1026                  * we would set SB_STOP below.  That could lead to an empty
1027                  * sockbuf having SB_STOP set
1028                  */
1029                 SOCKBUF_LOCK(&so->so_snd);
1030                 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1031                         so->so_snd.sb_flags |= SB_STOP;
1032                 SOCKBUF_UNLOCK(&so->so_snd);
1033                 UNP_PCB_UNLOCK(unp2);
1034                 m = NULL;
1035                 break;
1036         }
1037
1038         /*
1039          * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1040          */
1041         if (flags & PRUS_EOF) {
1042                 UNP_PCB_LOCK(unp);
1043                 socantsendmore(so);
1044                 unp_shutdown(unp);
1045                 UNP_PCB_UNLOCK(unp);
1046         }
1047
1048         if ((nam != NULL) || (flags & PRUS_EOF))
1049                 UNP_LINK_WUNLOCK();
1050         else
1051                 UNP_LINK_RUNLOCK();
1052
1053         if (control != NULL && error != 0)
1054                 unp_dispose(control);
1055
1056 release:
1057         if (control != NULL)
1058                 m_freem(control);
1059         if (m != NULL)
1060                 m_freem(m);
1061         return (error);
1062 }
1063
1064 static int
1065 uipc_ready(struct socket *so, struct mbuf *m, int count)
1066 {
1067         struct unpcb *unp, *unp2;
1068         struct socket *so2;
1069         int error;
1070
1071         unp = sotounpcb(so);
1072
1073         UNP_LINK_RLOCK();
1074         unp2 = unp->unp_conn;
1075         UNP_PCB_LOCK(unp2);
1076         so2 = unp2->unp_socket;
1077
1078         SOCKBUF_LOCK(&so2->so_rcv);
1079         if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1080                 sorwakeup_locked(so2);
1081         else
1082                 SOCKBUF_UNLOCK(&so2->so_rcv);
1083
1084         UNP_PCB_UNLOCK(unp2);
1085         UNP_LINK_RUNLOCK();
1086
1087         return (error);
1088 }
1089
1090 static int
1091 uipc_sense(struct socket *so, struct stat *sb)
1092 {
1093         struct unpcb *unp;
1094
1095         unp = sotounpcb(so);
1096         KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1097
1098         sb->st_blksize = so->so_snd.sb_hiwat;
1099         UNP_PCB_LOCK(unp);
1100         sb->st_dev = NODEV;
1101         if (unp->unp_ino == 0)
1102                 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1103         sb->st_ino = unp->unp_ino;
1104         UNP_PCB_UNLOCK(unp);
1105         return (0);
1106 }
1107
1108 static int
1109 uipc_shutdown(struct socket *so)
1110 {
1111         struct unpcb *unp;
1112
1113         unp = sotounpcb(so);
1114         KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1115
1116         UNP_LINK_WLOCK();
1117         UNP_PCB_LOCK(unp);
1118         socantsendmore(so);
1119         unp_shutdown(unp);
1120         UNP_PCB_UNLOCK(unp);
1121         UNP_LINK_WUNLOCK();
1122         return (0);
1123 }
1124
1125 static int
1126 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1127 {
1128         struct unpcb *unp;
1129         const struct sockaddr *sa;
1130
1131         unp = sotounpcb(so);
1132         KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1133
1134         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1135         UNP_PCB_LOCK(unp);
1136         if (unp->unp_addr != NULL)
1137                 sa = (struct sockaddr *) unp->unp_addr;
1138         else
1139                 sa = &sun_noname;
1140         bcopy(sa, *nam, sa->sa_len);
1141         UNP_PCB_UNLOCK(unp);
1142         return (0);
1143 }
1144
1145 static struct pr_usrreqs uipc_usrreqs_dgram = {
1146         .pru_abort =            uipc_abort,
1147         .pru_accept =           uipc_accept,
1148         .pru_attach =           uipc_attach,
1149         .pru_bind =             uipc_bind,
1150         .pru_bindat =           uipc_bindat,
1151         .pru_connect =          uipc_connect,
1152         .pru_connectat =        uipc_connectat,
1153         .pru_connect2 =         uipc_connect2,
1154         .pru_detach =           uipc_detach,
1155         .pru_disconnect =       uipc_disconnect,
1156         .pru_listen =           uipc_listen,
1157         .pru_peeraddr =         uipc_peeraddr,
1158         .pru_rcvd =             uipc_rcvd,
1159         .pru_send =             uipc_send,
1160         .pru_sense =            uipc_sense,
1161         .pru_shutdown =         uipc_shutdown,
1162         .pru_sockaddr =         uipc_sockaddr,
1163         .pru_soreceive =        soreceive_dgram,
1164         .pru_close =            uipc_close,
1165 };
1166
1167 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1168         .pru_abort =            uipc_abort,
1169         .pru_accept =           uipc_accept,
1170         .pru_attach =           uipc_attach,
1171         .pru_bind =             uipc_bind,
1172         .pru_bindat =           uipc_bindat,
1173         .pru_connect =          uipc_connect,
1174         .pru_connectat =        uipc_connectat,
1175         .pru_connect2 =         uipc_connect2,
1176         .pru_detach =           uipc_detach,
1177         .pru_disconnect =       uipc_disconnect,
1178         .pru_listen =           uipc_listen,
1179         .pru_peeraddr =         uipc_peeraddr,
1180         .pru_rcvd =             uipc_rcvd,
1181         .pru_send =             uipc_send,
1182         .pru_sense =            uipc_sense,
1183         .pru_shutdown =         uipc_shutdown,
1184         .pru_sockaddr =         uipc_sockaddr,
1185         .pru_soreceive =        soreceive_generic,      /* XXX: or...? */
1186         .pru_close =            uipc_close,
1187 };
1188
1189 static struct pr_usrreqs uipc_usrreqs_stream = {
1190         .pru_abort =            uipc_abort,
1191         .pru_accept =           uipc_accept,
1192         .pru_attach =           uipc_attach,
1193         .pru_bind =             uipc_bind,
1194         .pru_bindat =           uipc_bindat,
1195         .pru_connect =          uipc_connect,
1196         .pru_connectat =        uipc_connectat,
1197         .pru_connect2 =         uipc_connect2,
1198         .pru_detach =           uipc_detach,
1199         .pru_disconnect =       uipc_disconnect,
1200         .pru_listen =           uipc_listen,
1201         .pru_peeraddr =         uipc_peeraddr,
1202         .pru_rcvd =             uipc_rcvd,
1203         .pru_send =             uipc_send,
1204         .pru_ready =            uipc_ready,
1205         .pru_sense =            uipc_sense,
1206         .pru_shutdown =         uipc_shutdown,
1207         .pru_sockaddr =         uipc_sockaddr,
1208         .pru_soreceive =        soreceive_generic,
1209         .pru_close =            uipc_close,
1210 };
1211
1212 static int
1213 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1214 {
1215         struct unpcb *unp;
1216         struct xucred xu;
1217         int error, optval;
1218
1219         if (sopt->sopt_level != 0)
1220                 return (EINVAL);
1221
1222         unp = sotounpcb(so);
1223         KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1224         error = 0;
1225         switch (sopt->sopt_dir) {
1226         case SOPT_GET:
1227                 switch (sopt->sopt_name) {
1228                 case LOCAL_PEERCRED:
1229                         UNP_PCB_LOCK(unp);
1230                         if (unp->unp_flags & UNP_HAVEPC)
1231                                 xu = unp->unp_peercred;
1232                         else {
1233                                 if (so->so_type == SOCK_STREAM)
1234                                         error = ENOTCONN;
1235                                 else
1236                                         error = EINVAL;
1237                         }
1238                         UNP_PCB_UNLOCK(unp);
1239                         if (error == 0)
1240                                 error = sooptcopyout(sopt, &xu, sizeof(xu));
1241                         break;
1242
1243                 case LOCAL_CREDS:
1244                         /* Unlocked read. */
1245                         optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1246                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1247                         break;
1248
1249                 case LOCAL_CONNWAIT:
1250                         /* Unlocked read. */
1251                         optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1252                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1253                         break;
1254
1255                 default:
1256                         error = EOPNOTSUPP;
1257                         break;
1258                 }
1259                 break;
1260
1261         case SOPT_SET:
1262                 switch (sopt->sopt_name) {
1263                 case LOCAL_CREDS:
1264                 case LOCAL_CONNWAIT:
1265                         error = sooptcopyin(sopt, &optval, sizeof(optval),
1266                                             sizeof(optval));
1267                         if (error)
1268                                 break;
1269
1270 #define OPTSET(bit) do {                                                \
1271         UNP_PCB_LOCK(unp);                                              \
1272         if (optval)                                                     \
1273                 unp->unp_flags |= bit;                                  \
1274         else                                                            \
1275                 unp->unp_flags &= ~bit;                                 \
1276         UNP_PCB_UNLOCK(unp);                                            \
1277 } while (0)
1278
1279                         switch (sopt->sopt_name) {
1280                         case LOCAL_CREDS:
1281                                 OPTSET(UNP_WANTCRED);
1282                                 break;
1283
1284                         case LOCAL_CONNWAIT:
1285                                 OPTSET(UNP_CONNWAIT);
1286                                 break;
1287
1288                         default:
1289                                 break;
1290                         }
1291                         break;
1292 #undef  OPTSET
1293                 default:
1294                         error = ENOPROTOOPT;
1295                         break;
1296                 }
1297                 break;
1298
1299         default:
1300                 error = EOPNOTSUPP;
1301                 break;
1302         }
1303         return (error);
1304 }
1305
1306 static int
1307 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1308 {
1309
1310         return (unp_connectat(AT_FDCWD, so, nam, td));
1311 }
1312
1313 static int
1314 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1315     struct thread *td)
1316 {
1317         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1318         struct vnode *vp;
1319         struct socket *so2, *so3;
1320         struct unpcb *unp, *unp2, *unp3;
1321         struct nameidata nd;
1322         char buf[SOCK_MAXADDRLEN];
1323         struct sockaddr *sa;
1324         cap_rights_t rights;
1325         int error, len;
1326
1327         if (nam->sa_family != AF_UNIX)
1328                 return (EAFNOSUPPORT);
1329
1330         UNP_LINK_WLOCK_ASSERT();
1331
1332         unp = sotounpcb(so);
1333         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1334
1335         if (nam->sa_len > sizeof(struct sockaddr_un))
1336                 return (EINVAL);
1337         len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1338         if (len <= 0)
1339                 return (EINVAL);
1340         bcopy(soun->sun_path, buf, len);
1341         buf[len] = 0;
1342
1343         UNP_PCB_LOCK(unp);
1344         if (unp->unp_flags & UNP_CONNECTING) {
1345                 UNP_PCB_UNLOCK(unp);
1346                 return (EALREADY);
1347         }
1348         UNP_LINK_WUNLOCK();
1349         unp->unp_flags |= UNP_CONNECTING;
1350         UNP_PCB_UNLOCK(unp);
1351
1352         sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1353         NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1354             UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1355         error = namei(&nd);
1356         if (error)
1357                 vp = NULL;
1358         else
1359                 vp = nd.ni_vp;
1360         ASSERT_VOP_LOCKED(vp, "unp_connect");
1361         NDFREE(&nd, NDF_ONLY_PNBUF);
1362         if (error)
1363                 goto bad;
1364
1365         if (vp->v_type != VSOCK) {
1366                 error = ENOTSOCK;
1367                 goto bad;
1368         }
1369 #ifdef MAC
1370         error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1371         if (error)
1372                 goto bad;
1373 #endif
1374         error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1375         if (error)
1376                 goto bad;
1377
1378         unp = sotounpcb(so);
1379         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1380
1381         /*
1382          * Lock linkage lock for two reasons: make sure v_socket is stable,
1383          * and to protect simultaneous locking of multiple pcbs.
1384          */
1385         UNP_LINK_WLOCK();
1386         VOP_UNP_CONNECT(vp, &so2);
1387         if (so2 == NULL) {
1388                 error = ECONNREFUSED;
1389                 goto bad2;
1390         }
1391         if (so->so_type != so2->so_type) {
1392                 error = EPROTOTYPE;
1393                 goto bad2;
1394         }
1395         if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1396                 if (so2->so_options & SO_ACCEPTCONN) {
1397                         CURVNET_SET(so2->so_vnet);
1398                         so3 = sonewconn(so2, 0);
1399                         CURVNET_RESTORE();
1400                 } else
1401                         so3 = NULL;
1402                 if (so3 == NULL) {
1403                         error = ECONNREFUSED;
1404                         goto bad2;
1405                 }
1406                 unp = sotounpcb(so);
1407                 unp2 = sotounpcb(so2);
1408                 unp3 = sotounpcb(so3);
1409                 UNP_PCB_LOCK(unp);
1410                 UNP_PCB_LOCK(unp2);
1411                 UNP_PCB_LOCK(unp3);
1412                 if (unp2->unp_addr != NULL) {
1413                         bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1414                         unp3->unp_addr = (struct sockaddr_un *) sa;
1415                         sa = NULL;
1416                 }
1417
1418                 /*
1419                  * The connector's (client's) credentials are copied from its
1420                  * process structure at the time of connect() (which is now).
1421                  */
1422                 cru2x(td->td_ucred, &unp3->unp_peercred);
1423                 unp3->unp_flags |= UNP_HAVEPC;
1424
1425                 /*
1426                  * The receiver's (server's) credentials are copied from the
1427                  * unp_peercred member of socket on which the former called
1428                  * listen(); uipc_listen() cached that process's credentials
1429                  * at that time so we can use them now.
1430                  */
1431                 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1432                     ("unp_connect: listener without cached peercred"));
1433                 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1434                     sizeof(unp->unp_peercred));
1435                 unp->unp_flags |= UNP_HAVEPC;
1436                 if (unp2->unp_flags & UNP_WANTCRED)
1437                         unp3->unp_flags |= UNP_WANTCRED;
1438                 UNP_PCB_UNLOCK(unp3);
1439                 UNP_PCB_UNLOCK(unp2);
1440                 UNP_PCB_UNLOCK(unp);
1441 #ifdef MAC
1442                 mac_socketpeer_set_from_socket(so, so3);
1443                 mac_socketpeer_set_from_socket(so3, so);
1444 #endif
1445
1446                 so2 = so3;
1447         }
1448         unp = sotounpcb(so);
1449         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1450         unp2 = sotounpcb(so2);
1451         KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1452         UNP_PCB_LOCK(unp);
1453         UNP_PCB_LOCK(unp2);
1454         error = unp_connect2(so, so2, PRU_CONNECT);
1455         UNP_PCB_UNLOCK(unp2);
1456         UNP_PCB_UNLOCK(unp);
1457 bad2:
1458         UNP_LINK_WUNLOCK();
1459 bad:
1460         if (vp != NULL)
1461                 vput(vp);
1462         free(sa, M_SONAME);
1463         UNP_LINK_WLOCK();
1464         UNP_PCB_LOCK(unp);
1465         unp->unp_flags &= ~UNP_CONNECTING;
1466         UNP_PCB_UNLOCK(unp);
1467         return (error);
1468 }
1469
1470 static int
1471 unp_connect2(struct socket *so, struct socket *so2, int req)
1472 {
1473         struct unpcb *unp;
1474         struct unpcb *unp2;
1475
1476         unp = sotounpcb(so);
1477         KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1478         unp2 = sotounpcb(so2);
1479         KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1480
1481         UNP_LINK_WLOCK_ASSERT();
1482         UNP_PCB_LOCK_ASSERT(unp);
1483         UNP_PCB_LOCK_ASSERT(unp2);
1484
1485         if (so2->so_type != so->so_type)
1486                 return (EPROTOTYPE);
1487         unp2->unp_flags &= ~UNP_NASCENT;
1488         unp->unp_conn = unp2;
1489
1490         switch (so->so_type) {
1491         case SOCK_DGRAM:
1492                 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1493                 soisconnected(so);
1494                 break;
1495
1496         case SOCK_STREAM:
1497         case SOCK_SEQPACKET:
1498                 unp2->unp_conn = unp;
1499                 if (req == PRU_CONNECT &&
1500                     ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1501                         soisconnecting(so);
1502                 else
1503                         soisconnected(so);
1504                 soisconnected(so2);
1505                 break;
1506
1507         default:
1508                 panic("unp_connect2");
1509         }
1510         return (0);
1511 }
1512
1513 static void
1514 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1515 {
1516         struct socket *so;
1517
1518         KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1519
1520         UNP_LINK_WLOCK_ASSERT();
1521         UNP_PCB_LOCK_ASSERT(unp);
1522         UNP_PCB_LOCK_ASSERT(unp2);
1523
1524         unp->unp_conn = NULL;
1525         switch (unp->unp_socket->so_type) {
1526         case SOCK_DGRAM:
1527                 LIST_REMOVE(unp, unp_reflink);
1528                 so = unp->unp_socket;
1529                 SOCK_LOCK(so);
1530                 so->so_state &= ~SS_ISCONNECTED;
1531                 SOCK_UNLOCK(so);
1532                 break;
1533
1534         case SOCK_STREAM:
1535         case SOCK_SEQPACKET:
1536                 soisdisconnected(unp->unp_socket);
1537                 unp2->unp_conn = NULL;
1538                 soisdisconnected(unp2->unp_socket);
1539                 break;
1540         }
1541 }
1542
1543 /*
1544  * unp_pcblist() walks the global list of struct unpcb's to generate a
1545  * pointer list, bumping the refcount on each unpcb.  It then copies them out
1546  * sequentially, validating the generation number on each to see if it has
1547  * been detached.  All of this is necessary because copyout() may sleep on
1548  * disk I/O.
1549  */
1550 static int
1551 unp_pcblist(SYSCTL_HANDLER_ARGS)
1552 {
1553         int error, i, n;
1554         int freeunp;
1555         struct unpcb *unp, **unp_list;
1556         unp_gen_t gencnt;
1557         struct xunpgen *xug;
1558         struct unp_head *head;
1559         struct xunpcb *xu;
1560
1561         switch ((intptr_t)arg1) {
1562         case SOCK_STREAM:
1563                 head = &unp_shead;
1564                 break;
1565
1566         case SOCK_DGRAM:
1567                 head = &unp_dhead;
1568                 break;
1569
1570         case SOCK_SEQPACKET:
1571                 head = &unp_sphead;
1572                 break;
1573
1574         default:
1575                 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1576         }
1577
1578         /*
1579          * The process of preparing the PCB list is too time-consuming and
1580          * resource-intensive to repeat twice on every request.
1581          */
1582         if (req->oldptr == NULL) {
1583                 n = unp_count;
1584                 req->oldidx = 2 * (sizeof *xug)
1585                         + (n + n/8) * sizeof(struct xunpcb);
1586                 return (0);
1587         }
1588
1589         if (req->newptr != NULL)
1590                 return (EPERM);
1591
1592         /*
1593          * OK, now we're committed to doing something.
1594          */
1595         xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1596         UNP_LIST_LOCK();
1597         gencnt = unp_gencnt;
1598         n = unp_count;
1599         UNP_LIST_UNLOCK();
1600
1601         xug->xug_len = sizeof *xug;
1602         xug->xug_count = n;
1603         xug->xug_gen = gencnt;
1604         xug->xug_sogen = so_gencnt;
1605         error = SYSCTL_OUT(req, xug, sizeof *xug);
1606         if (error) {
1607                 free(xug, M_TEMP);
1608                 return (error);
1609         }
1610
1611         unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1612
1613         UNP_LIST_LOCK();
1614         for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1615              unp = LIST_NEXT(unp, unp_link)) {
1616                 UNP_PCB_LOCK(unp);
1617                 if (unp->unp_gencnt <= gencnt) {
1618                         if (cr_cansee(req->td->td_ucred,
1619                             unp->unp_socket->so_cred)) {
1620                                 UNP_PCB_UNLOCK(unp);
1621                                 continue;
1622                         }
1623                         unp_list[i++] = unp;
1624                         unp->unp_refcount++;
1625                 }
1626                 UNP_PCB_UNLOCK(unp);
1627         }
1628         UNP_LIST_UNLOCK();
1629         n = i;                  /* In case we lost some during malloc. */
1630
1631         error = 0;
1632         xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1633         for (i = 0; i < n; i++) {
1634                 unp = unp_list[i];
1635                 UNP_PCB_LOCK(unp);
1636                 unp->unp_refcount--;
1637                 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1638                         xu->xu_len = sizeof *xu;
1639                         xu->xu_unpp = unp;
1640                         /*
1641                          * XXX - need more locking here to protect against
1642                          * connect/disconnect races for SMP.
1643                          */
1644                         if (unp->unp_addr != NULL)
1645                                 bcopy(unp->unp_addr, &xu->xu_addr,
1646                                       unp->unp_addr->sun_len);
1647                         if (unp->unp_conn != NULL &&
1648                             unp->unp_conn->unp_addr != NULL)
1649                                 bcopy(unp->unp_conn->unp_addr,
1650                                       &xu->xu_caddr,
1651                                       unp->unp_conn->unp_addr->sun_len);
1652                         bcopy(unp, &xu->xu_unp, sizeof *unp);
1653                         sotoxsocket(unp->unp_socket, &xu->xu_socket);
1654                         UNP_PCB_UNLOCK(unp);
1655                         error = SYSCTL_OUT(req, xu, sizeof *xu);
1656                 } else {
1657                         freeunp = (unp->unp_refcount == 0);
1658                         UNP_PCB_UNLOCK(unp);
1659                         if (freeunp) {
1660                                 UNP_PCB_LOCK_DESTROY(unp);
1661                                 uma_zfree(unp_zone, unp);
1662                         }
1663                 }
1664         }
1665         free(xu, M_TEMP);
1666         if (!error) {
1667                 /*
1668                  * Give the user an updated idea of our state.  If the
1669                  * generation differs from what we told her before, she knows
1670                  * that something happened while we were processing this
1671                  * request, and it might be necessary to retry.
1672                  */
1673                 xug->xug_gen = unp_gencnt;
1674                 xug->xug_sogen = so_gencnt;
1675                 xug->xug_count = unp_count;
1676                 error = SYSCTL_OUT(req, xug, sizeof *xug);
1677         }
1678         free(unp_list, M_TEMP);
1679         free(xug, M_TEMP);
1680         return (error);
1681 }
1682
1683 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1684     (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1685     "List of active local datagram sockets");
1686 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1687     (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1688     "List of active local stream sockets");
1689 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1690     CTLTYPE_OPAQUE | CTLFLAG_RD,
1691     (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1692     "List of active local seqpacket sockets");
1693
1694 static void
1695 unp_shutdown(struct unpcb *unp)
1696 {
1697         struct unpcb *unp2;
1698         struct socket *so;
1699
1700         UNP_LINK_WLOCK_ASSERT();
1701         UNP_PCB_LOCK_ASSERT(unp);
1702
1703         unp2 = unp->unp_conn;
1704         if ((unp->unp_socket->so_type == SOCK_STREAM ||
1705             (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1706                 so = unp2->unp_socket;
1707                 if (so != NULL)
1708                         socantrcvmore(so);
1709         }
1710 }
1711
1712 static void
1713 unp_drop(struct unpcb *unp)
1714 {
1715         struct socket *so = unp->unp_socket;
1716         struct unpcb *unp2;
1717
1718         UNP_LINK_WLOCK_ASSERT();
1719         UNP_PCB_LOCK_ASSERT(unp);
1720
1721         /*
1722          * Regardless of whether the socket's peer dropped the connection
1723          * with this socket by aborting or disconnecting, POSIX requires
1724          * that ECONNRESET is returned.
1725          */
1726         so->so_error = ECONNRESET;
1727         unp2 = unp->unp_conn;
1728         if (unp2 == NULL)
1729                 return;
1730         UNP_PCB_LOCK(unp2);
1731         unp_disconnect(unp, unp2);
1732         UNP_PCB_UNLOCK(unp2);
1733 }
1734
1735 static void
1736 unp_freerights(struct filedescent **fdep, int fdcount)
1737 {
1738         struct file *fp;
1739         int i;
1740
1741         KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1742
1743         for (i = 0; i < fdcount; i++) {
1744                 fp = fdep[i]->fde_file;
1745                 filecaps_free(&fdep[i]->fde_caps);
1746                 unp_discard(fp);
1747         }
1748         free(fdep[0], M_FILECAPS);
1749 }
1750
1751 static int
1752 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1753 {
1754         struct thread *td = curthread;          /* XXX */
1755         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1756         int i;
1757         int *fdp;
1758         struct filedesc *fdesc = td->td_proc->p_fd;
1759         struct filedescent **fdep;
1760         void *data;
1761         socklen_t clen = control->m_len, datalen;
1762         int error, newfds;
1763         u_int newlen;
1764
1765         UNP_LINK_UNLOCK_ASSERT();
1766
1767         error = 0;
1768         if (controlp != NULL) /* controlp == NULL => free control messages */
1769                 *controlp = NULL;
1770         while (cm != NULL) {
1771                 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1772                         error = EINVAL;
1773                         break;
1774                 }
1775                 data = CMSG_DATA(cm);
1776                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1777                 if (cm->cmsg_level == SOL_SOCKET
1778                     && cm->cmsg_type == SCM_RIGHTS) {
1779                         newfds = datalen / sizeof(*fdep);
1780                         if (newfds == 0)
1781                                 goto next;
1782                         fdep = data;
1783
1784                         /* If we're not outputting the descriptors free them. */
1785                         if (error || controlp == NULL) {
1786                                 unp_freerights(fdep, newfds);
1787                                 goto next;
1788                         }
1789                         FILEDESC_XLOCK(fdesc);
1790
1791                         /*
1792                          * Now change each pointer to an fd in the global
1793                          * table to an integer that is the index to the local
1794                          * fd table entry that we set up to point to the
1795                          * global one we are transferring.
1796                          */
1797                         newlen = newfds * sizeof(int);
1798                         *controlp = sbcreatecontrol(NULL, newlen,
1799                             SCM_RIGHTS, SOL_SOCKET);
1800                         if (*controlp == NULL) {
1801                                 FILEDESC_XUNLOCK(fdesc);
1802                                 error = E2BIG;
1803                                 unp_freerights(fdep, newfds);
1804                                 goto next;
1805                         }
1806
1807                         fdp = (int *)
1808                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1809                         if (fdallocn(td, 0, fdp, newfds) != 0) {
1810                                 FILEDESC_XUNLOCK(fdesc);
1811                                 error = EMSGSIZE;
1812                                 unp_freerights(fdep, newfds);
1813                                 m_freem(*controlp);
1814                                 *controlp = NULL;
1815                                 goto next;
1816                         }
1817                         for (i = 0; i < newfds; i++, fdp++) {
1818                                 _finstall(fdesc, fdep[i]->fde_file, *fdp,
1819                                     (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0,
1820                                     &fdep[i]->fde_caps);
1821                                 unp_externalize_fp(fdep[i]->fde_file);
1822                         }
1823                         FILEDESC_XUNLOCK(fdesc);
1824                         free(fdep[0], M_FILECAPS);
1825                 } else {
1826                         /* We can just copy anything else across. */
1827                         if (error || controlp == NULL)
1828                                 goto next;
1829                         *controlp = sbcreatecontrol(NULL, datalen,
1830                             cm->cmsg_type, cm->cmsg_level);
1831                         if (*controlp == NULL) {
1832                                 error = ENOBUFS;
1833                                 goto next;
1834                         }
1835                         bcopy(data,
1836                             CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1837                             datalen);
1838                 }
1839                 controlp = &(*controlp)->m_next;
1840
1841 next:
1842                 if (CMSG_SPACE(datalen) < clen) {
1843                         clen -= CMSG_SPACE(datalen);
1844                         cm = (struct cmsghdr *)
1845                             ((caddr_t)cm + CMSG_SPACE(datalen));
1846                 } else {
1847                         clen = 0;
1848                         cm = NULL;
1849                 }
1850         }
1851
1852         m_freem(control);
1853         return (error);
1854 }
1855
1856 static void
1857 unp_zone_change(void *tag)
1858 {
1859
1860         uma_zone_set_max(unp_zone, maxsockets);
1861 }
1862
1863 static void
1864 unp_init(void)
1865 {
1866
1867 #ifdef VIMAGE
1868         if (!IS_DEFAULT_VNET(curvnet))
1869                 return;
1870 #endif
1871         unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1872             NULL, NULL, UMA_ALIGN_PTR, 0);
1873         if (unp_zone == NULL)
1874                 panic("unp_init");
1875         uma_zone_set_max(unp_zone, maxsockets);
1876         uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
1877         EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1878             NULL, EVENTHANDLER_PRI_ANY);
1879         LIST_INIT(&unp_dhead);
1880         LIST_INIT(&unp_shead);
1881         LIST_INIT(&unp_sphead);
1882         SLIST_INIT(&unp_defers);
1883         TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1884         TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1885         UNP_LINK_LOCK_INIT();
1886         UNP_LIST_LOCK_INIT();
1887         UNP_DEFERRED_LOCK_INIT();
1888 }
1889
1890 static int
1891 unp_internalize(struct mbuf **controlp, struct thread *td)
1892 {
1893         struct mbuf *control = *controlp;
1894         struct proc *p = td->td_proc;
1895         struct filedesc *fdesc = p->p_fd;
1896         struct bintime *bt;
1897         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1898         struct cmsgcred *cmcred;
1899         struct filedescent *fde, **fdep, *fdev;
1900         struct file *fp;
1901         struct timeval *tv;
1902         int i, *fdp;
1903         void *data;
1904         socklen_t clen = control->m_len, datalen;
1905         int error, oldfds;
1906         u_int newlen;
1907
1908         UNP_LINK_UNLOCK_ASSERT();
1909
1910         error = 0;
1911         *controlp = NULL;
1912         while (cm != NULL) {
1913                 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1914                     || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
1915                         error = EINVAL;
1916                         goto out;
1917                 }
1918                 data = CMSG_DATA(cm);
1919                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1920
1921                 switch (cm->cmsg_type) {
1922                 /*
1923                  * Fill in credential information.
1924                  */
1925                 case SCM_CREDS:
1926                         *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1927                             SCM_CREDS, SOL_SOCKET);
1928                         if (*controlp == NULL) {
1929                                 error = ENOBUFS;
1930                                 goto out;
1931                         }
1932                         cmcred = (struct cmsgcred *)
1933                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1934                         cmcred->cmcred_pid = p->p_pid;
1935                         cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1936                         cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1937                         cmcred->cmcred_euid = td->td_ucred->cr_uid;
1938                         cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1939                             CMGROUP_MAX);
1940                         for (i = 0; i < cmcred->cmcred_ngroups; i++)
1941                                 cmcred->cmcred_groups[i] =
1942                                     td->td_ucred->cr_groups[i];
1943                         break;
1944
1945                 case SCM_RIGHTS:
1946                         oldfds = datalen / sizeof (int);
1947                         if (oldfds == 0)
1948                                 break;
1949                         /*
1950                          * Check that all the FDs passed in refer to legal
1951                          * files.  If not, reject the entire operation.
1952                          */
1953                         fdp = data;
1954                         FILEDESC_SLOCK(fdesc);
1955                         for (i = 0; i < oldfds; i++, fdp++) {
1956                                 fp = fget_locked(fdesc, *fdp);
1957                                 if (fp == NULL) {
1958                                         FILEDESC_SUNLOCK(fdesc);
1959                                         error = EBADF;
1960                                         goto out;
1961                                 }
1962                                 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1963                                         FILEDESC_SUNLOCK(fdesc);
1964                                         error = EOPNOTSUPP;
1965                                         goto out;
1966                                 }
1967
1968                         }
1969
1970                         /*
1971                          * Now replace the integer FDs with pointers to the
1972                          * file structure and capability rights.
1973                          */
1974                         newlen = oldfds * sizeof(fdep[0]);
1975                         *controlp = sbcreatecontrol(NULL, newlen,
1976                             SCM_RIGHTS, SOL_SOCKET);
1977                         if (*controlp == NULL) {
1978                                 FILEDESC_SUNLOCK(fdesc);
1979                                 error = E2BIG;
1980                                 goto out;
1981                         }
1982                         fdp = data;
1983                         fdep = (struct filedescent **)
1984                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1985                         fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
1986                             M_WAITOK);
1987                         for (i = 0; i < oldfds; i++, fdev++, fdp++) {
1988                                 fde = &fdesc->fd_ofiles[*fdp];
1989                                 fdep[i] = fdev;
1990                                 fdep[i]->fde_file = fde->fde_file;
1991                                 filecaps_copy(&fde->fde_caps,
1992                                     &fdep[i]->fde_caps, true);
1993                                 unp_internalize_fp(fdep[i]->fde_file);
1994                         }
1995                         FILEDESC_SUNLOCK(fdesc);
1996                         break;
1997
1998                 case SCM_TIMESTAMP:
1999                         *controlp = sbcreatecontrol(NULL, sizeof(*tv),
2000                             SCM_TIMESTAMP, SOL_SOCKET);
2001                         if (*controlp == NULL) {
2002                                 error = ENOBUFS;
2003                                 goto out;
2004                         }
2005                         tv = (struct timeval *)
2006                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2007                         microtime(tv);
2008                         break;
2009
2010                 case SCM_BINTIME:
2011                         *controlp = sbcreatecontrol(NULL, sizeof(*bt),
2012                             SCM_BINTIME, SOL_SOCKET);
2013                         if (*controlp == NULL) {
2014                                 error = ENOBUFS;
2015                                 goto out;
2016                         }
2017                         bt = (struct bintime *)
2018                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2019                         bintime(bt);
2020                         break;
2021
2022                 default:
2023                         error = EINVAL;
2024                         goto out;
2025                 }
2026
2027                 controlp = &(*controlp)->m_next;
2028                 if (CMSG_SPACE(datalen) < clen) {
2029                         clen -= CMSG_SPACE(datalen);
2030                         cm = (struct cmsghdr *)
2031                             ((caddr_t)cm + CMSG_SPACE(datalen));
2032                 } else {
2033                         clen = 0;
2034                         cm = NULL;
2035                 }
2036         }
2037
2038 out:
2039         m_freem(control);
2040         return (error);
2041 }
2042
2043 static struct mbuf *
2044 unp_addsockcred(struct thread *td, struct mbuf *control)
2045 {
2046         struct mbuf *m, *n, *n_prev;
2047         struct sockcred *sc;
2048         const struct cmsghdr *cm;
2049         int ngroups;
2050         int i;
2051
2052         ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2053         m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2054         if (m == NULL)
2055                 return (control);
2056
2057         sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2058         sc->sc_uid = td->td_ucred->cr_ruid;
2059         sc->sc_euid = td->td_ucred->cr_uid;
2060         sc->sc_gid = td->td_ucred->cr_rgid;
2061         sc->sc_egid = td->td_ucred->cr_gid;
2062         sc->sc_ngroups = ngroups;
2063         for (i = 0; i < sc->sc_ngroups; i++)
2064                 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2065
2066         /*
2067          * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2068          * created SCM_CREDS control message (struct sockcred) has another
2069          * format.
2070          */
2071         if (control != NULL)
2072                 for (n = control, n_prev = NULL; n != NULL;) {
2073                         cm = mtod(n, struct cmsghdr *);
2074                         if (cm->cmsg_level == SOL_SOCKET &&
2075                             cm->cmsg_type == SCM_CREDS) {
2076                                 if (n_prev == NULL)
2077                                         control = n->m_next;
2078                                 else
2079                                         n_prev->m_next = n->m_next;
2080                                 n = m_free(n);
2081                         } else {
2082                                 n_prev = n;
2083                                 n = n->m_next;
2084                         }
2085                 }
2086
2087         /* Prepend it to the head. */
2088         m->m_next = control;
2089         return (m);
2090 }
2091
2092 static struct unpcb *
2093 fptounp(struct file *fp)
2094 {
2095         struct socket *so;
2096
2097         if (fp->f_type != DTYPE_SOCKET)
2098                 return (NULL);
2099         if ((so = fp->f_data) == NULL)
2100                 return (NULL);
2101         if (so->so_proto->pr_domain != &localdomain)
2102                 return (NULL);
2103         return sotounpcb(so);
2104 }
2105
2106 static void
2107 unp_discard(struct file *fp)
2108 {
2109         struct unp_defer *dr;
2110
2111         if (unp_externalize_fp(fp)) {
2112                 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2113                 dr->ud_fp = fp;
2114                 UNP_DEFERRED_LOCK();
2115                 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2116                 UNP_DEFERRED_UNLOCK();
2117                 atomic_add_int(&unp_defers_count, 1);
2118                 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2119         } else
2120                 (void) closef(fp, (struct thread *)NULL);
2121 }
2122
2123 static void
2124 unp_process_defers(void *arg __unused, int pending)
2125 {
2126         struct unp_defer *dr;
2127         SLIST_HEAD(, unp_defer) drl;
2128         int count;
2129
2130         SLIST_INIT(&drl);
2131         for (;;) {
2132                 UNP_DEFERRED_LOCK();
2133                 if (SLIST_FIRST(&unp_defers) == NULL) {
2134                         UNP_DEFERRED_UNLOCK();
2135                         break;
2136                 }
2137                 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2138                 UNP_DEFERRED_UNLOCK();
2139                 count = 0;
2140                 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2141                         SLIST_REMOVE_HEAD(&drl, ud_link);
2142                         closef(dr->ud_fp, NULL);
2143                         free(dr, M_TEMP);
2144                         count++;
2145                 }
2146                 atomic_add_int(&unp_defers_count, -count);
2147         }
2148 }
2149
2150 static void
2151 unp_internalize_fp(struct file *fp)
2152 {
2153         struct unpcb *unp;
2154
2155         UNP_LINK_WLOCK();
2156         if ((unp = fptounp(fp)) != NULL) {
2157                 unp->unp_file = fp;
2158                 unp->unp_msgcount++;
2159         }
2160         fhold(fp);
2161         unp_rights++;
2162         UNP_LINK_WUNLOCK();
2163 }
2164
2165 static int
2166 unp_externalize_fp(struct file *fp)
2167 {
2168         struct unpcb *unp;
2169         int ret;
2170
2171         UNP_LINK_WLOCK();
2172         if ((unp = fptounp(fp)) != NULL) {
2173                 unp->unp_msgcount--;
2174                 ret = 1;
2175         } else
2176                 ret = 0;
2177         unp_rights--;
2178         UNP_LINK_WUNLOCK();
2179         return (ret);
2180 }
2181
2182 /*
2183  * unp_defer indicates whether additional work has been defered for a future
2184  * pass through unp_gc().  It is thread local and does not require explicit
2185  * synchronization.
2186  */
2187 static int      unp_marked;
2188 static int      unp_unreachable;
2189
2190 static void
2191 unp_accessable(struct filedescent **fdep, int fdcount)
2192 {
2193         struct unpcb *unp;
2194         struct file *fp;
2195         int i;
2196
2197         for (i = 0; i < fdcount; i++) {
2198                 fp = fdep[i]->fde_file;
2199                 if ((unp = fptounp(fp)) == NULL)
2200                         continue;
2201                 if (unp->unp_gcflag & UNPGC_REF)
2202                         continue;
2203                 unp->unp_gcflag &= ~UNPGC_DEAD;
2204                 unp->unp_gcflag |= UNPGC_REF;
2205                 unp_marked++;
2206         }
2207 }
2208
2209 static void
2210 unp_gc_process(struct unpcb *unp)
2211 {
2212         struct socket *soa;
2213         struct socket *so;
2214         struct file *fp;
2215
2216         /* Already processed. */
2217         if (unp->unp_gcflag & UNPGC_SCANNED)
2218                 return;
2219         fp = unp->unp_file;
2220
2221         /*
2222          * Check for a socket potentially in a cycle.  It must be in a
2223          * queue as indicated by msgcount, and this must equal the file
2224          * reference count.  Note that when msgcount is 0 the file is NULL.
2225          */
2226         if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2227             unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2228                 unp->unp_gcflag |= UNPGC_DEAD;
2229                 unp_unreachable++;
2230                 return;
2231         }
2232
2233         /*
2234          * Mark all sockets we reference with RIGHTS.
2235          */
2236         so = unp->unp_socket;
2237         if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2238                 SOCKBUF_LOCK(&so->so_rcv);
2239                 unp_scan(so->so_rcv.sb_mb, unp_accessable);
2240                 SOCKBUF_UNLOCK(&so->so_rcv);
2241         }
2242
2243         /*
2244          * Mark all sockets in our accept queue.
2245          */
2246         ACCEPT_LOCK();
2247         TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2248                 if ((sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS) != 0)
2249                         continue;
2250                 SOCKBUF_LOCK(&soa->so_rcv);
2251                 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2252                 SOCKBUF_UNLOCK(&soa->so_rcv);
2253         }
2254         ACCEPT_UNLOCK();
2255         unp->unp_gcflag |= UNPGC_SCANNED;
2256 }
2257
2258 static int unp_recycled;
2259 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2260     "Number of unreachable sockets claimed by the garbage collector.");
2261
2262 static int unp_taskcount;
2263 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2264     "Number of times the garbage collector has run.");
2265
2266 static void
2267 unp_gc(__unused void *arg, int pending)
2268 {
2269         struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2270                                     NULL };
2271         struct unp_head **head;
2272         struct file *f, **unref;
2273         struct unpcb *unp;
2274         int i, total;
2275
2276         unp_taskcount++;
2277         UNP_LIST_LOCK();
2278         /*
2279          * First clear all gc flags from previous runs, apart from
2280          * UNPGC_IGNORE_RIGHTS.
2281          */
2282         for (head = heads; *head != NULL; head++)
2283                 LIST_FOREACH(unp, *head, unp_link)
2284                         unp->unp_gcflag =
2285                             (unp->unp_gcflag & UNPGC_IGNORE_RIGHTS);
2286
2287         /*
2288          * Scan marking all reachable sockets with UNPGC_REF.  Once a socket
2289          * is reachable all of the sockets it references are reachable.
2290          * Stop the scan once we do a complete loop without discovering
2291          * a new reachable socket.
2292          */
2293         do {
2294                 unp_unreachable = 0;
2295                 unp_marked = 0;
2296                 for (head = heads; *head != NULL; head++)
2297                         LIST_FOREACH(unp, *head, unp_link)
2298                                 unp_gc_process(unp);
2299         } while (unp_marked);
2300         UNP_LIST_UNLOCK();
2301         if (unp_unreachable == 0)
2302                 return;
2303
2304         /*
2305          * Allocate space for a local list of dead unpcbs.
2306          */
2307         unref = malloc(unp_unreachable * sizeof(struct file *),
2308             M_TEMP, M_WAITOK);
2309
2310         /*
2311          * Iterate looking for sockets which have been specifically marked
2312          * as as unreachable and store them locally.
2313          */
2314         UNP_LINK_RLOCK();
2315         UNP_LIST_LOCK();
2316         for (total = 0, head = heads; *head != NULL; head++)
2317                 LIST_FOREACH(unp, *head, unp_link)
2318                         if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2319                                 f = unp->unp_file;
2320                                 if (unp->unp_msgcount == 0 || f == NULL ||
2321                                     f->f_count != unp->unp_msgcount)
2322                                         continue;
2323                                 unref[total++] = f;
2324                                 fhold(f);
2325                                 KASSERT(total <= unp_unreachable,
2326                                     ("unp_gc: incorrect unreachable count."));
2327                         }
2328         UNP_LIST_UNLOCK();
2329         UNP_LINK_RUNLOCK();
2330
2331         /*
2332          * Now flush all sockets, free'ing rights.  This will free the
2333          * struct files associated with these sockets but leave each socket
2334          * with one remaining ref.
2335          */
2336         for (i = 0; i < total; i++) {
2337                 struct socket *so;
2338
2339                 so = unref[i]->f_data;
2340                 CURVNET_SET(so->so_vnet);
2341                 sorflush(so);
2342                 CURVNET_RESTORE();
2343         }
2344
2345         /*
2346          * And finally release the sockets so they can be reclaimed.
2347          */
2348         for (i = 0; i < total; i++)
2349                 fdrop(unref[i], NULL);
2350         unp_recycled += total;
2351         free(unref, M_TEMP);
2352 }
2353
2354 static void
2355 unp_dispose(struct mbuf *m)
2356 {
2357
2358         if (m)
2359                 unp_scan(m, unp_freerights);
2360 }
2361
2362 /*
2363  * Synchronize against unp_gc, which can trip over data as we are freeing it.
2364  */
2365 static void
2366 unp_dispose_so(struct socket *so)
2367 {
2368         struct unpcb *unp;
2369
2370         unp = sotounpcb(so);
2371         UNP_LIST_LOCK();
2372         unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2373         UNP_LIST_UNLOCK();
2374         unp_dispose(so->so_rcv.sb_mb);
2375 }
2376
2377 static void
2378 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2379 {
2380         struct mbuf *m;
2381         struct cmsghdr *cm;
2382         void *data;
2383         socklen_t clen, datalen;
2384
2385         while (m0 != NULL) {
2386                 for (m = m0; m; m = m->m_next) {
2387                         if (m->m_type != MT_CONTROL)
2388                                 continue;
2389
2390                         cm = mtod(m, struct cmsghdr *);
2391                         clen = m->m_len;
2392
2393                         while (cm != NULL) {
2394                                 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2395                                         break;
2396
2397                                 data = CMSG_DATA(cm);
2398                                 datalen = (caddr_t)cm + cm->cmsg_len
2399                                     - (caddr_t)data;
2400
2401                                 if (cm->cmsg_level == SOL_SOCKET &&
2402                                     cm->cmsg_type == SCM_RIGHTS) {
2403                                         (*op)(data, datalen /
2404                                             sizeof(struct filedescent *));
2405                                 }
2406
2407                                 if (CMSG_SPACE(datalen) < clen) {
2408                                         clen -= CMSG_SPACE(datalen);
2409                                         cm = (struct cmsghdr *)
2410                                             ((caddr_t)cm + CMSG_SPACE(datalen));
2411                                 } else {
2412                                         clen = 0;
2413                                         cm = NULL;
2414                                 }
2415                         }
2416                 }
2417                 m0 = m0->m_nextpkt;
2418         }
2419 }
2420
2421 /*
2422  * A helper function called by VFS before socket-type vnode reclamation.
2423  * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2424  * use count.
2425  */
2426 void
2427 vfs_unp_reclaim(struct vnode *vp)
2428 {
2429         struct socket *so;
2430         struct unpcb *unp;
2431         int active;
2432
2433         ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2434         KASSERT(vp->v_type == VSOCK,
2435             ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2436
2437         active = 0;
2438         UNP_LINK_WLOCK();
2439         VOP_UNP_CONNECT(vp, &so);
2440         if (so == NULL)
2441                 goto done;
2442         unp = sotounpcb(so);
2443         if (unp == NULL)
2444                 goto done;
2445         UNP_PCB_LOCK(unp);
2446         if (unp->unp_vnode == vp) {
2447                 VOP_UNP_DETACH(vp);
2448                 unp->unp_vnode = NULL;
2449                 active = 1;
2450         }
2451         UNP_PCB_UNLOCK(unp);
2452 done:
2453         UNP_LINK_WUNLOCK();
2454         if (active)
2455                 vunref(vp);
2456 }
2457
2458 #ifdef DDB
2459 static void
2460 db_print_indent(int indent)
2461 {
2462         int i;
2463
2464         for (i = 0; i < indent; i++)
2465                 db_printf(" ");
2466 }
2467
2468 static void
2469 db_print_unpflags(int unp_flags)
2470 {
2471         int comma;
2472
2473         comma = 0;
2474         if (unp_flags & UNP_HAVEPC) {
2475                 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2476                 comma = 1;
2477         }
2478         if (unp_flags & UNP_HAVEPCCACHED) {
2479                 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2480                 comma = 1;
2481         }
2482         if (unp_flags & UNP_WANTCRED) {
2483                 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2484                 comma = 1;
2485         }
2486         if (unp_flags & UNP_CONNWAIT) {
2487                 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2488                 comma = 1;
2489         }
2490         if (unp_flags & UNP_CONNECTING) {
2491                 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2492                 comma = 1;
2493         }
2494         if (unp_flags & UNP_BINDING) {
2495                 db_printf("%sUNP_BINDING", comma ? ", " : "");
2496                 comma = 1;
2497         }
2498 }
2499
2500 static void
2501 db_print_xucred(int indent, struct xucred *xu)
2502 {
2503         int comma, i;
2504
2505         db_print_indent(indent);
2506         db_printf("cr_version: %u   cr_uid: %u   cr_ngroups: %d\n",
2507             xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2508         db_print_indent(indent);
2509         db_printf("cr_groups: ");
2510         comma = 0;
2511         for (i = 0; i < xu->cr_ngroups; i++) {
2512                 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2513                 comma = 1;
2514         }
2515         db_printf("\n");
2516 }
2517
2518 static void
2519 db_print_unprefs(int indent, struct unp_head *uh)
2520 {
2521         struct unpcb *unp;
2522         int counter;
2523
2524         counter = 0;
2525         LIST_FOREACH(unp, uh, unp_reflink) {
2526                 if (counter % 4 == 0)
2527                         db_print_indent(indent);
2528                 db_printf("%p  ", unp);
2529                 if (counter % 4 == 3)
2530                         db_printf("\n");
2531                 counter++;
2532         }
2533         if (counter != 0 && counter % 4 != 0)
2534                 db_printf("\n");
2535 }
2536
2537 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2538 {
2539         struct unpcb *unp;
2540
2541         if (!have_addr) {
2542                 db_printf("usage: show unpcb <addr>\n");
2543                 return;
2544         }
2545         unp = (struct unpcb *)addr;
2546
2547         db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
2548             unp->unp_vnode);
2549
2550         db_printf("unp_ino: %ju   unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2551             unp->unp_conn);
2552
2553         db_printf("unp_refs:\n");
2554         db_print_unprefs(2, &unp->unp_refs);
2555
2556         /* XXXRW: Would be nice to print the full address, if any. */
2557         db_printf("unp_addr: %p\n", unp->unp_addr);
2558
2559         db_printf("unp_gencnt: %llu\n",
2560             (unsigned long long)unp->unp_gencnt);
2561
2562         db_printf("unp_flags: %x (", unp->unp_flags);
2563         db_print_unpflags(unp->unp_flags);
2564         db_printf(")\n");
2565
2566         db_printf("unp_peercred:\n");
2567         db_print_xucred(2, &unp->unp_peercred);
2568
2569         db_printf("unp_refcount: %u\n", unp->unp_refcount);
2570 }
2571 #endif