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  *      distinguish datagram size limits from flow control limits in SEQPACKET
  55  *      rethink name space problems
  56  *      need a proper out-of-band
  57  */
  58
  59 #include <sys/cdefs.h>
  60 __FBSDID("$FreeBSD$");
  61
  62 #include "opt_ddb.h"
  63
  64 #include <sys/param.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 /*
 105  * Locking key:
 106  * (l)  Locked using list lock
 107  * (g)  Locked using linkage lock
 108  */
 109
 110 static uma_zone_t       unp_zone;
 111 static unp_gen_t        unp_gencnt;     /* (l) */
 112 static u_int            unp_count;      /* (l) Count of local sockets. */
 113 static ino_t            unp_ino;        /* Prototype for fake inode numbers. */
 114 static int              unp_rights;     /* (g) File descriptors in flight. */
 115 static struct unp_head  unp_shead;      /* (l) List of stream sockets. */
 116 static struct unp_head  unp_dhead;      /* (l) List of datagram sockets. */
 117 static struct unp_head  unp_sphead;     /* (l) List of seqpacket sockets. */
 118
 119 struct unp_defer {
 120         SLIST_ENTRY(unp_defer) ud_link;
 121         struct file *ud_fp;
 122 };
 123 static SLIST_HEAD(, unp_defer) unp_defers;
 124 static int unp_defers_count;
 125
 126 static const struct sockaddr    sun_noname = { sizeof(sun_noname), AF_LOCAL };
 127
 128 /*
 129  * Garbage collection of cyclic file descriptor/socket references occurs
 130  * asynchronously in a taskqueue context in order to avoid recursion and
 131  * reentrance in the UNIX domain socket, file descriptor, and socket layer
 132  * code.  See unp_gc() for a full description.
 133  */
 134 static struct task      unp_gc_task;
 135
 136 /*
 137  * The close of unix domain sockets attached as SCM_RIGHTS is
 138  * postponed to the taskqueue, to avoid arbitrary recursion depth.
 139  * The attached sockets might have another sockets attached.
 140  */
 141 static struct task      unp_defer_task;
 142
 143 /*
 144  * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
 145  * stream sockets, although the total for sender and receiver is actually
 146  * only PIPSIZ.
 147  *
 148  * Datagram sockets really use the sendspace as the maximum datagram size,
 149  * and don't really want to reserve the sendspace.  Their recvspace should be
 150  * large enough for at least one max-size datagram plus address.
 151  */
 152 #ifndef PIPSIZ
 153 #define PIPSIZ  8192
 154 #endif
 155 static u_long   unpst_sendspace = PIPSIZ;
 156 static u_long   unpst_recvspace = PIPSIZ;
 157 static u_long   unpdg_sendspace = 2*1024;       /* really max datagram size */
 158 static u_long   unpdg_recvspace = 4*1024;
 159 static u_long   unpsp_sendspace = PIPSIZ;       /* really max datagram size */
 160 static u_long   unpsp_recvspace = PIPSIZ;
 161
 162 SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
 163 SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, "SOCK_STREAM");
 164 SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
 165 SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0,
 166     "SOCK_SEQPACKET");
 167
 168 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
 169            &unpst_sendspace, 0, "Default stream send space.");
 170 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
 171            &unpst_recvspace, 0, "Default stream receive space.");
 172 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
 173            &unpdg_sendspace, 0, "Default datagram send space.");
 174 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
 175            &unpdg_recvspace, 0, "Default datagram receive space.");
 176 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
 177            &unpsp_sendspace, 0, "Default seqpacket send space.");
 178 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
 179            &unpsp_recvspace, 0, "Default seqpacket receive space.");
 180 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
 181     "File descriptors in flight.");
 182 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
 183     &unp_defers_count, 0,
 184     "File descriptors deferred to taskqueue for close.");
 185
 186 /*
 187  * Locking and synchronization:
 188  *
 189  * Three types of locks exit in the local domain socket implementation: a
 190  * global list mutex, a global linkage rwlock, and per-unpcb mutexes.  Of the
 191  * global locks, the list lock protects the socket count, global generation
 192  * number, and stream/datagram global lists.  The linkage lock protects the
 193  * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be
 194  * held exclusively over the acquisition of multiple unpcb locks to prevent
 195  * deadlock.
 196  *
 197  * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
 198  * allocated in pru_attach() and freed in pru_detach().  The validity of that
 199  * pointer is an invariant, so no lock is required to dereference the so_pcb
 200  * pointer if a valid socket reference is held by the caller.  In practice,
 201  * this is always true during operations performed on a socket.  Each unpcb
 202  * has a back-pointer to its socket, unp_socket, which will be stable under
 203  * the same circumstances.
 204  *
 205  * This pointer may only be safely dereferenced as long as a valid reference
 206  * to the unpcb is held.  Typically, this reference will be from the socket,
 207  * or from another unpcb when the referring unpcb's lock is held (in order
 208  * that the reference not be invalidated during use).  For example, to follow
 209  * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
 210  * as unp_socket remains valid as long as the reference to unp_conn is valid.
 211  *
 212  * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx.  Individual
 213  * atomic reads without the lock may be performed "lockless", but more
 214  * complex reads and read-modify-writes require the mutex to be held.  No
 215  * lock order is defined between unpcb locks -- multiple unpcb locks may be
 216  * acquired at the same time only when holding the linkage rwlock
 217  * exclusively, which prevents deadlocks.
 218  *
 219  * Blocking with UNIX domain sockets is a tricky issue: unlike most network
 220  * protocols, bind() is a non-atomic operation, and connect() requires
 221  * potential sleeping in the protocol, due to potentially waiting on local or
 222  * distributed file systems.  We try to separate "lookup" operations, which
 223  * may sleep, and the IPC operations themselves, which typically can occur
 224  * with relative atomicity as locks can be held over the entire operation.
 225  *
 226  * Another tricky issue is simultaneous multi-threaded or multi-process
 227  * access to a single UNIX domain socket.  These are handled by the flags
 228  * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
 229  * binding, both of which involve dropping UNIX domain socket locks in order
 230  * to perform namei() and other file system operations.
 231  */
 232 static struct rwlock    unp_link_rwlock;
 233 static struct mtx       unp_list_lock;
 234 static struct mtx       unp_defers_lock;
 235
 236 #define UNP_LINK_LOCK_INIT()            rw_init(&unp_link_rwlock,       \
 237                                             "unp_link_rwlock")
 238
 239 #define UNP_LINK_LOCK_ASSERT()  rw_assert(&unp_link_rwlock,     \
 240                                             RA_LOCKED)
 241 #define UNP_LINK_UNLOCK_ASSERT()        rw_assert(&unp_link_rwlock,     \
 242                                             RA_UNLOCKED)
 243
 244 #define UNP_LINK_RLOCK()                rw_rlock(&unp_link_rwlock)
 245 #define UNP_LINK_RUNLOCK()              rw_runlock(&unp_link_rwlock)
 246 #define UNP_LINK_WLOCK()                rw_wlock(&unp_link_rwlock)
 247 #define UNP_LINK_WUNLOCK()              rw_wunlock(&unp_link_rwlock)
 248 #define UNP_LINK_WLOCK_ASSERT()         rw_assert(&unp_link_rwlock,     \
 249                                             RA_WLOCKED)
 250
 251 #define UNP_LIST_LOCK_INIT()            mtx_init(&unp_list_lock,        \
 252                                             "unp_list_lock", NULL, MTX_DEF)
 253 #define UNP_LIST_LOCK()                 mtx_lock(&unp_list_lock)
 254 #define UNP_LIST_UNLOCK()               mtx_unlock(&unp_list_lock)
 255
 256 #define UNP_DEFERRED_LOCK_INIT()        mtx_init(&unp_defers_lock, \
 257                                             "unp_defer", NULL, MTX_DEF)
 258 #define UNP_DEFERRED_LOCK()             mtx_lock(&unp_defers_lock)
 259 #define UNP_DEFERRED_UNLOCK()           mtx_unlock(&unp_defers_lock)
 260
 261 #define UNP_PCB_LOCK_INIT(unp)          mtx_init(&(unp)->unp_mtx,       \
 262                                             "unp_mtx", "unp_mtx",       \
 263                                             MTX_DUPOK|MTX_DEF|MTX_RECURSE)
 264 #define UNP_PCB_LOCK_DESTROY(unp)       mtx_destroy(&(unp)->unp_mtx)
 265 #define UNP_PCB_LOCK(unp)               mtx_lock(&(unp)->unp_mtx)
 266 #define UNP_PCB_UNLOCK(unp)             mtx_unlock(&(unp)->unp_mtx)
 267 #define UNP_PCB_LOCK_ASSERT(unp)        mtx_assert(&(unp)->unp_mtx, MA_OWNED)
 268
 269 static int      uipc_connect2(struct socket *, struct socket *);
 270 static int      uipc_ctloutput(struct socket *, struct sockopt *);
 271 static int      unp_connect(struct socket *, struct sockaddr *,
 272                     struct thread *);
 273 static int      unp_connect2(struct socket *so, struct socket *so2, int);
 274 static void     unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
 275 static void     unp_dispose(struct mbuf *);
 276 static void     unp_shutdown(struct unpcb *);
 277 static void     unp_drop(struct unpcb *, int);
 278 static void     unp_gc(__unused void *, int);
 279 static void     unp_scan(struct mbuf *, void (*)(struct file *));
 280 static void     unp_discard(struct file *);
 281 static void     unp_freerights(struct file **, int);
 282 static void     unp_init(void);
 283 static int      unp_internalize(struct mbuf **, struct thread *);
 284 static void     unp_internalize_fp(struct file *);
 285 static int      unp_externalize(struct mbuf *, struct mbuf **);
 286 static int      unp_externalize_fp(struct file *);
 287 static struct mbuf      *unp_addsockcred(struct thread *, struct mbuf *);
 288 static void     unp_process_defers(void * __unused, int);
 289
 290 /*
 291  * Definitions of protocols supported in the LOCAL domain.
 292  */
 293 static struct domain localdomain;
 294 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
 295 static struct pr_usrreqs uipc_usrreqs_seqpacket;
 296 static struct protosw localsw[] = {
 297 {
 298         .pr_type =              SOCK_STREAM,
 299         .pr_domain =            &localdomain,
 300         .pr_flags =             PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
 301         .pr_ctloutput =         &uipc_ctloutput,
 302         .pr_usrreqs =           &uipc_usrreqs_stream
 303 },
 304 {
 305         .pr_type =              SOCK_DGRAM,
 306         .pr_domain =            &localdomain,
 307         .pr_flags =             PR_ATOMIC|PR_ADDR|PR_RIGHTS,
 308         .pr_usrreqs =           &uipc_usrreqs_dgram
 309 },
 310 {
 311         .pr_type =              SOCK_SEQPACKET,
 312         .pr_domain =            &localdomain,
 313
 314         /*
 315          * XXXRW: For now, PR_ADDR because soreceive will bump into them
 316          * due to our use of sbappendaddr.  A new sbappend variants is needed
 317          * that supports both atomic record writes and control data.
 318          */
 319         .pr_flags =             PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
 320                                     PR_RIGHTS,
 321         .pr_usrreqs =           &uipc_usrreqs_seqpacket,
 322 },
 323 };
 324
 325 static struct domain localdomain = {
 326         .dom_family =           AF_LOCAL,
 327         .dom_name =             "local",
 328         .dom_init =             unp_init,
 329         .dom_externalize =      unp_externalize,
 330         .dom_dispose =          unp_dispose,
 331         .dom_protosw =          localsw,
 332         .dom_protoswNPROTOSW =  &localsw[sizeof(localsw)/sizeof(localsw[0])]
 333 };
 334 DOMAIN_SET(local);
 335
 336 static void
 337 uipc_abort(struct socket *so)
 338 {
 339         struct unpcb *unp, *unp2;
 340
 341         unp = sotounpcb(so);
 342         KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
 343
 344         UNP_LINK_WLOCK();
 345         UNP_PCB_LOCK(unp);
 346         unp2 = unp->unp_conn;
 347         if (unp2 != NULL) {
 348                 UNP_PCB_LOCK(unp2);
 349                 unp_drop(unp2, ECONNABORTED);
 350                 UNP_PCB_UNLOCK(unp2);
 351         }
 352         UNP_PCB_UNLOCK(unp);
 353         UNP_LINK_WUNLOCK();
 354 }
 355
 356 static int
 357 uipc_accept(struct socket *so, struct sockaddr **nam)
 358 {
 359         struct unpcb *unp, *unp2;
 360         const struct sockaddr *sa;
 361
 362         /*
 363          * Pass back name of connected socket, if it was bound and we are
 364          * still connected (our peer may have closed already!).
 365          */
 366         unp = sotounpcb(so);
 367         KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
 368
 369         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 370         UNP_LINK_RLOCK();
 371         unp2 = unp->unp_conn;
 372         if (unp2 != NULL && unp2->unp_addr != NULL) {
 373                 UNP_PCB_LOCK(unp2);
 374                 sa = (struct sockaddr *) unp2->unp_addr;
 375                 bcopy(sa, *nam, sa->sa_len);
 376                 UNP_PCB_UNLOCK(unp2);
 377         } else {
 378                 sa = &sun_noname;
 379                 bcopy(sa, *nam, sa->sa_len);
 380         }
 381         UNP_LINK_RUNLOCK();
 382         return (0);
 383 }
 384
 385 static int
 386 uipc_attach(struct socket *so, int proto, struct thread *td)
 387 {
 388         u_long sendspace, recvspace;
 389         struct unpcb *unp;
 390         int error;
 391
 392         KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
 393         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
 394                 switch (so->so_type) {
 395                 case SOCK_STREAM:
 396                         sendspace = unpst_sendspace;
 397                         recvspace = unpst_recvspace;
 398                         break;
 399
 400                 case SOCK_DGRAM:
 401                         sendspace = unpdg_sendspace;
 402                         recvspace = unpdg_recvspace;
 403                         break;
 404
 405                 case SOCK_SEQPACKET:
 406                         sendspace = unpsp_sendspace;
 407                         recvspace = unpsp_recvspace;
 408                         break;
 409
 410                 default:
 411                         panic("uipc_attach");
 412                 }
 413                 error = soreserve(so, sendspace, recvspace);
 414                 if (error)
 415                         return (error);
 416         }
 417         unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
 418         if (unp == NULL)
 419                 return (ENOBUFS);
 420         LIST_INIT(&unp->unp_refs);
 421         UNP_PCB_LOCK_INIT(unp);
 422         unp->unp_socket = so;
 423         so->so_pcb = unp;
 424         unp->unp_refcount = 1;
 425
 426         UNP_LIST_LOCK();
 427         unp->unp_gencnt = ++unp_gencnt;
 428         unp_count++;
 429         switch (so->so_type) {
 430         case SOCK_STREAM:
 431                 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
 432                 break;
 433
 434         case SOCK_DGRAM:
 435                 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
 436                 break;
 437
 438         case SOCK_SEQPACKET:
 439                 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
 440                 break;
 441
 442         default:
 443                 panic("uipc_attach");
 444         }
 445         UNP_LIST_UNLOCK();
 446
 447         return (0);
 448 }
 449
 450 static int
 451 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 452 {
 453         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
 454         struct vattr vattr;
 455         int error, namelen, vfslocked;
 456         struct nameidata nd;
 457         struct unpcb *unp;
 458         struct vnode *vp;
 459         struct mount *mp;
 460         char *buf;
 461
 462         unp = sotounpcb(so);
 463         KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
 464
 465         namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
 466         if (namelen <= 0)
 467                 return (EINVAL);
 468
 469         /*
 470          * We don't allow simultaneous bind() calls on a single UNIX domain
 471          * socket, so flag in-progress operations, and return an error if an
 472          * operation is already in progress.
 473          *
 474          * Historically, we have not allowed a socket to be rebound, so this
 475          * also returns an error.  Not allowing re-binding simplifies the
 476          * implementation and avoids a great many possible failure modes.
 477          */
 478         UNP_PCB_LOCK(unp);
 479         if (unp->unp_vnode != NULL) {
 480                 UNP_PCB_UNLOCK(unp);
 481                 return (EINVAL);
 482         }
 483         if (unp->unp_flags & UNP_BINDING) {
 484                 UNP_PCB_UNLOCK(unp);
 485                 return (EALREADY);
 486         }
 487         unp->unp_flags |= UNP_BINDING;
 488         UNP_PCB_UNLOCK(unp);
 489
 490         buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
 491         bcopy(soun->sun_path, buf, namelen);
 492         buf[namelen] = 0;
 493
 494 restart:
 495         vfslocked = 0;
 496         NDINIT(&nd, CREATE, MPSAFE | NOFOLLOW | LOCKPARENT | SAVENAME,
 497             UIO_SYSSPACE, buf, td);
 498 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
 499         error = namei(&nd);
 500         if (error)
 501                 goto error;
 502         vp = nd.ni_vp;
 503         vfslocked = NDHASGIANT(&nd);
 504         if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
 505                 NDFREE(&nd, NDF_ONLY_PNBUF);
 506                 if (nd.ni_dvp == vp)
 507                         vrele(nd.ni_dvp);
 508                 else
 509                         vput(nd.ni_dvp);
 510                 if (vp != NULL) {
 511                         vrele(vp);
 512                         error = EADDRINUSE;
 513                         goto error;
 514                 }
 515                 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
 516                 if (error)
 517                         goto error;
 518                 VFS_UNLOCK_GIANT(vfslocked);
 519                 goto restart;
 520         }
 521         VATTR_NULL(&vattr);
 522         vattr.va_type = VSOCK;
 523         vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
 524 #ifdef MAC
 525         error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
 526             &vattr);
 527 #endif
 528         if (error == 0)
 529                 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
 530         NDFREE(&nd, NDF_ONLY_PNBUF);
 531         vput(nd.ni_dvp);
 532         if (error) {
 533                 vn_finished_write(mp);
 534                 goto error;
 535         }
 536         vp = nd.ni_vp;
 537         ASSERT_VOP_ELOCKED(vp, "uipc_bind");
 538         soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
 539
 540         UNP_LINK_WLOCK();
 541         UNP_PCB_LOCK(unp);
 542         vp->v_socket = unp->unp_socket;
 543         unp->unp_vnode = vp;
 544         unp->unp_addr = soun;
 545         unp->unp_flags &= ~UNP_BINDING;
 546         UNP_PCB_UNLOCK(unp);
 547         UNP_LINK_WUNLOCK();
 548         VOP_UNLOCK(vp, 0);
 549         vn_finished_write(mp);
 550         VFS_UNLOCK_GIANT(vfslocked);
 551         free(buf, M_TEMP);
 552         return (0);
 553
 554 error:
 555         VFS_UNLOCK_GIANT(vfslocked);
 556         UNP_PCB_LOCK(unp);
 557         unp->unp_flags &= ~UNP_BINDING;
 558         UNP_PCB_UNLOCK(unp);
 559         free(buf, M_TEMP);
 560         return (error);
 561 }
 562
 563 static int
 564 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 565 {
 566         int error;
 567
 568         KASSERT(td == curthread, ("uipc_connect: td != curthread"));
 569         UNP_LINK_WLOCK();
 570         error = unp_connect(so, nam, td);
 571         UNP_LINK_WUNLOCK();
 572         return (error);
 573 }
 574
 575 static void
 576 uipc_close(struct socket *so)
 577 {
 578         struct unpcb *unp, *unp2;
 579
 580         unp = sotounpcb(so);
 581         KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
 582
 583         UNP_LINK_WLOCK();
 584         UNP_PCB_LOCK(unp);
 585         unp2 = unp->unp_conn;
 586         if (unp2 != NULL) {
 587                 UNP_PCB_LOCK(unp2);
 588                 unp_disconnect(unp, unp2);
 589                 UNP_PCB_UNLOCK(unp2);
 590         }
 591         UNP_PCB_UNLOCK(unp);
 592         UNP_LINK_WUNLOCK();
 593 }
 594
 595 static int
 596 uipc_connect2(struct socket *so1, struct socket *so2)
 597 {
 598         struct unpcb *unp, *unp2;
 599         int error;
 600
 601         UNP_LINK_WLOCK();
 602         unp = so1->so_pcb;
 603         KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
 604         UNP_PCB_LOCK(unp);
 605         unp2 = so2->so_pcb;
 606         KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
 607         UNP_PCB_LOCK(unp2);
 608         error = unp_connect2(so1, so2, PRU_CONNECT2);
 609         UNP_PCB_UNLOCK(unp2);
 610         UNP_PCB_UNLOCK(unp);
 611         UNP_LINK_WUNLOCK();
 612         return (error);
 613 }
 614
 615 static void
 616 uipc_detach(struct socket *so)
 617 {
 618         struct unpcb *unp, *unp2;
 619         struct sockaddr_un *saved_unp_addr;
 620         struct vnode *vp;
 621         int freeunp, local_unp_rights;
 622
 623         unp = sotounpcb(so);
 624         KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
 625
 626         UNP_LINK_WLOCK();
 627         UNP_LIST_LOCK();
 628         UNP_PCB_LOCK(unp);
 629         LIST_REMOVE(unp, unp_link);
 630         unp->unp_gencnt = ++unp_gencnt;
 631         --unp_count;
 632         UNP_LIST_UNLOCK();
 633
 634         /*
 635          * XXXRW: Should assert vp->v_socket == so.
 636          */
 637         if ((vp = unp->unp_vnode) != NULL) {
 638                 unp->unp_vnode->v_socket = NULL;
 639                 unp->unp_vnode = NULL;
 640         }
 641         unp2 = unp->unp_conn;
 642         if (unp2 != NULL) {
 643                 UNP_PCB_LOCK(unp2);
 644                 unp_disconnect(unp, unp2);
 645                 UNP_PCB_UNLOCK(unp2);
 646         }
 647
 648         /*
 649          * We hold the linkage lock exclusively, so it's OK to acquire
 650          * multiple pcb locks at a time.
 651          */
 652         while (!LIST_EMPTY(&unp->unp_refs)) {
 653                 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
 654
 655                 UNP_PCB_LOCK(ref);
 656                 unp_drop(ref, ECONNRESET);
 657                 UNP_PCB_UNLOCK(ref);
 658         }
 659         local_unp_rights = unp_rights;
 660         UNP_LINK_WUNLOCK();
 661         unp->unp_socket->so_pcb = NULL;
 662         saved_unp_addr = unp->unp_addr;
 663         unp->unp_addr = NULL;
 664         unp->unp_refcount--;
 665         freeunp = (unp->unp_refcount == 0);
 666         if (saved_unp_addr != NULL)
 667                 free(saved_unp_addr, M_SONAME);
 668         if (freeunp) {
 669                 UNP_PCB_LOCK_DESTROY(unp);
 670                 uma_zfree(unp_zone, unp);
 671         } else
 672                 UNP_PCB_UNLOCK(unp);
 673         if (vp) {
 674                 int vfslocked;
 675
 676                 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 677                 vrele(vp);
 678                 VFS_UNLOCK_GIANT(vfslocked);
 679         }
 680         if (local_unp_rights)
 681                 taskqueue_enqueue(taskqueue_thread, &unp_gc_task);
 682 }
 683
 684 static int
 685 uipc_disconnect(struct socket *so)
 686 {
 687         struct unpcb *unp, *unp2;
 688
 689         unp = sotounpcb(so);
 690         KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
 691
 692         UNP_LINK_WLOCK();
 693         UNP_PCB_LOCK(unp);
 694         unp2 = unp->unp_conn;
 695         if (unp2 != NULL) {
 696                 UNP_PCB_LOCK(unp2);
 697                 unp_disconnect(unp, unp2);
 698                 UNP_PCB_UNLOCK(unp2);
 699         }
 700         UNP_PCB_UNLOCK(unp);
 701         UNP_LINK_WUNLOCK();
 702         return (0);
 703 }
 704
 705 static int
 706 uipc_listen(struct socket *so, int backlog, struct thread *td)
 707 {
 708         struct unpcb *unp;
 709         int error;
 710
 711         unp = sotounpcb(so);
 712         KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
 713
 714         UNP_PCB_LOCK(unp);
 715         if (unp->unp_vnode == NULL) {
 716                 UNP_PCB_UNLOCK(unp);
 717                 return (EINVAL);
 718         }
 719
 720         SOCK_LOCK(so);
 721         error = solisten_proto_check(so);
 722         if (error == 0) {
 723                 cru2x(td->td_ucred, &unp->unp_peercred);
 724                 unp->unp_flags |= UNP_HAVEPCCACHED;
 725                 solisten_proto(so, backlog);
 726         }
 727         SOCK_UNLOCK(so);
 728         UNP_PCB_UNLOCK(unp);
 729         return (error);
 730 }
 731
 732 static int
 733 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
 734 {
 735         struct unpcb *unp, *unp2;
 736         const struct sockaddr *sa;
 737
 738         unp = sotounpcb(so);
 739         KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
 740
 741         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 742         UNP_LINK_RLOCK();
 743         /*
 744          * XXX: It seems that this test always fails even when connection is
 745          * established.  So, this else clause is added as workaround to
 746          * return PF_LOCAL sockaddr.
 747          */
 748         unp2 = unp->unp_conn;
 749         if (unp2 != NULL) {
 750                 UNP_PCB_LOCK(unp2);
 751                 if (unp2->unp_addr != NULL)
 752                         sa = (struct sockaddr *) unp2->unp_addr;
 753                 else
 754                         sa = &sun_noname;
 755                 bcopy(sa, *nam, sa->sa_len);
 756                 UNP_PCB_UNLOCK(unp2);
 757         } else {
 758                 sa = &sun_noname;
 759                 bcopy(sa, *nam, sa->sa_len);
 760         }
 761         UNP_LINK_RUNLOCK();
 762         return (0);
 763 }
 764
 765 static int
 766 uipc_rcvd(struct socket *so, int flags)
 767 {
 768         struct unpcb *unp, *unp2;
 769         struct socket *so2;
 770         u_int mbcnt, sbcc;
 771         u_long newhiwat;
 772
 773         unp = sotounpcb(so);
 774         KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
 775
 776         if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
 777                 panic("uipc_rcvd socktype %d", so->so_type);
 778
 779         /*
 780          * Adjust backpressure on sender and wakeup any waiting to write.
 781          *
 782          * The unp lock is acquired to maintain the validity of the unp_conn
 783          * pointer; no lock on unp2 is required as unp2->unp_socket will be
 784          * static as long as we don't permit unp2 to disconnect from unp,
 785          * which is prevented by the lock on unp.  We cache values from
 786          * so_rcv to avoid holding the so_rcv lock over the entire
 787          * transaction on the remote so_snd.
 788          */
 789         SOCKBUF_LOCK(&so->so_rcv);
 790         mbcnt = so->so_rcv.sb_mbcnt;
 791         sbcc = so->so_rcv.sb_cc;
 792         SOCKBUF_UNLOCK(&so->so_rcv);
 793         UNP_PCB_LOCK(unp);
 794         unp2 = unp->unp_conn;
 795         if (unp2 == NULL) {
 796                 UNP_PCB_UNLOCK(unp);
 797                 return (0);
 798         }
 799         so2 = unp2->unp_socket;
 800         SOCKBUF_LOCK(&so2->so_snd);
 801         so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt;
 802         newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc;
 803         (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
 804             newhiwat, RLIM_INFINITY);
 805         sowwakeup_locked(so2);
 806         unp->unp_mbcnt = mbcnt;
 807         unp->unp_cc = sbcc;
 808         UNP_PCB_UNLOCK(unp);
 809         return (0);
 810 }
 811
 812 static int
 813 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
 814     struct mbuf *control, struct thread *td)
 815 {
 816         struct unpcb *unp, *unp2;
 817         struct socket *so2;
 818         u_int mbcnt_delta, sbcc;
 819         u_int newhiwat;
 820         int error = 0;
 821
 822         unp = sotounpcb(so);
 823         KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
 824
 825         if (flags & PRUS_OOB) {
 826                 error = EOPNOTSUPP;
 827                 goto release;
 828         }
 829         if (control != NULL && (error = unp_internalize(&control, td)))
 830                 goto release;
 831         if ((nam != NULL) || (flags & PRUS_EOF))
 832                 UNP_LINK_WLOCK();
 833         else
 834                 UNP_LINK_RLOCK();
 835         switch (so->so_type) {
 836         case SOCK_DGRAM:
 837         {
 838                 const struct sockaddr *from;
 839
 840                 unp2 = unp->unp_conn;
 841                 if (nam != NULL) {
 842                         UNP_LINK_WLOCK_ASSERT();
 843                         if (unp2 != NULL) {
 844                                 error = EISCONN;
 845                                 break;
 846                         }
 847                         error = unp_connect(so, nam, td);
 848                         if (error)
 849                                 break;
 850                         unp2 = unp->unp_conn;
 851                 }
 852
 853                 /*
 854                  * Because connect() and send() are non-atomic in a sendto()
 855                  * with a target address, it's possible that the socket will
 856                  * have disconnected before the send() can run.  In that case
 857                  * return the slightly counter-intuitive but otherwise
 858                  * correct error that the socket is not connected.
 859                  */
 860                 if (unp2 == NULL) {
 861                         error = ENOTCONN;
 862                         break;
 863                 }
 864                 /* Lockless read. */
 865                 if (unp2->unp_flags & UNP_WANTCRED)
 866                         control = unp_addsockcred(td, control);
 867                 UNP_PCB_LOCK(unp);
 868                 if (unp->unp_addr != NULL)
 869                         from = (struct sockaddr *)unp->unp_addr;
 870                 else
 871                         from = &sun_noname;
 872                 so2 = unp2->unp_socket;
 873                 SOCKBUF_LOCK(&so2->so_rcv);
 874                 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
 875                         sorwakeup_locked(so2);
 876                         m = NULL;
 877                         control = NULL;
 878                 } else {
 879                         SOCKBUF_UNLOCK(&so2->so_rcv);
 880                         error = ENOBUFS;
 881                 }
 882                 if (nam != NULL) {
 883                         UNP_LINK_WLOCK_ASSERT();
 884                         UNP_PCB_LOCK(unp2);
 885                         unp_disconnect(unp, unp2);
 886                         UNP_PCB_UNLOCK(unp2);
 887                 }
 888                 UNP_PCB_UNLOCK(unp);
 889                 break;
 890         }
 891
 892         case SOCK_SEQPACKET:
 893         case SOCK_STREAM:
 894                 if ((so->so_state & SS_ISCONNECTED) == 0) {
 895                         if (nam != NULL) {
 896                                 UNP_LINK_WLOCK_ASSERT();
 897                                 error = unp_connect(so, nam, td);
 898                                 if (error)
 899                                         break;  /* XXX */
 900                         } else {
 901                                 error = ENOTCONN;
 902                                 break;
 903                         }
 904                 }
 905
 906                 /* Lockless read. */
 907                 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
 908                         error = EPIPE;
 909                         break;
 910                 }
 911
 912                 /*
 913                  * Because connect() and send() are non-atomic in a sendto()
 914                  * with a target address, it's possible that the socket will
 915                  * have disconnected before the send() can run.  In that case
 916                  * return the slightly counter-intuitive but otherwise
 917                  * correct error that the socket is not connected.
 918                  *
 919                  * Locking here must be done carefully: the linkage lock
 920                  * prevents interconnections between unpcbs from changing, so
 921                  * we can traverse from unp to unp2 without acquiring unp's
 922                  * lock.  Socket buffer locks follow unpcb locks, so we can
 923                  * acquire both remote and lock socket buffer locks.
 924                  */
 925                 unp2 = unp->unp_conn;
 926                 if (unp2 == NULL) {
 927                         error = ENOTCONN;
 928                         break;
 929                 }
 930                 so2 = unp2->unp_socket;
 931                 UNP_PCB_LOCK(unp2);
 932                 SOCKBUF_LOCK(&so2->so_rcv);
 933                 if (unp2->unp_flags & UNP_WANTCRED) {
 934                         /*
 935                          * Credentials are passed only once on SOCK_STREAM.
 936                          */
 937                         unp2->unp_flags &= ~UNP_WANTCRED;
 938                         control = unp_addsockcred(td, control);
 939                 }
 940                 /*
 941                  * Send to paired receive port, and then reduce send buffer
 942                  * hiwater marks to maintain backpressure.  Wake up readers.
 943                  */
 944                 switch (so->so_type) {
 945                 case SOCK_STREAM:
 946                         if (control != NULL) {
 947                                 if (sbappendcontrol_locked(&so2->so_rcv, m,
 948                                     control))
 949                                         control = NULL;
 950                         } else
 951                                 sbappend_locked(&so2->so_rcv, m);
 952                         break;
 953
 954                 case SOCK_SEQPACKET: {
 955                         const struct sockaddr *from;
 956
 957                         from = &sun_noname;
 958                         if (sbappendaddr_locked(&so2->so_rcv, from, m,
 959                             control))
 960                                 control = NULL;
 961                         break;
 962                         }
 963                 }
 964
 965                 /*
 966                  * XXXRW: While fine for SOCK_STREAM, this conflates maximum
 967                  * datagram size and back-pressure for SOCK_SEQPACKET, which
 968                  * can lead to undesired return of EMSGSIZE on send instead
 969                  * of more desirable blocking.
 970                  */
 971                 mbcnt_delta = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt;
 972                 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt;
 973                 sbcc = so2->so_rcv.sb_cc;
 974                 sorwakeup_locked(so2);
 975
 976                 SOCKBUF_LOCK(&so->so_snd);
 977                 if ((int)so->so_snd.sb_hiwat >= (int)(sbcc - unp2->unp_cc))
 978                         newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc);
 979                 else
 980                         newhiwat = 0;
 981                 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
 982                     newhiwat, RLIM_INFINITY);
 983                 so->so_snd.sb_mbmax -= mbcnt_delta;
 984                 SOCKBUF_UNLOCK(&so->so_snd);
 985                 unp2->unp_cc = sbcc;
 986                 UNP_PCB_UNLOCK(unp2);
 987                 m = NULL;
 988                 break;
 989
 990         default:
 991                 panic("uipc_send unknown socktype");
 992         }
 993
 994         /*
 995          * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
 996          */
 997         if (flags & PRUS_EOF) {
 998                 UNP_PCB_LOCK(unp);
 999                 socantsendmore(so);
1000                 unp_shutdown(unp);
1001                 UNP_PCB_UNLOCK(unp);
1002         }
1003
1004         if ((nam != NULL) || (flags & PRUS_EOF))
1005                 UNP_LINK_WUNLOCK();
1006         else
1007                 UNP_LINK_RUNLOCK();
1008
1009         if (control != NULL && error != 0)
1010                 unp_dispose(control);
1011
1012 release:
1013         if (control != NULL)
1014                 m_freem(control);
1015         if (m != NULL)
1016                 m_freem(m);
1017         return (error);
1018 }
1019
1020 static int
1021 uipc_sense(struct socket *so, struct stat *sb)
1022 {
1023         struct unpcb *unp, *unp2;
1024         struct socket *so2;
1025
1026         unp = sotounpcb(so);
1027         KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1028
1029         sb->st_blksize = so->so_snd.sb_hiwat;
1030         UNP_LINK_RLOCK();
1031         UNP_PCB_LOCK(unp);
1032         unp2 = unp->unp_conn;
1033         if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) &&
1034             unp2 != NULL) {
1035                 so2 = unp2->unp_socket;
1036                 sb->st_blksize += so2->so_rcv.sb_cc;
1037         }
1038         sb->st_dev = NODEV;
1039         if (unp->unp_ino == 0)
1040                 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1041         sb->st_ino = unp->unp_ino;
1042         UNP_PCB_UNLOCK(unp);
1043         UNP_LINK_RUNLOCK();
1044         return (0);
1045 }
1046
1047 static int
1048 uipc_shutdown(struct socket *so)
1049 {
1050         struct unpcb *unp;
1051
1052         unp = sotounpcb(so);
1053         KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1054
1055         UNP_LINK_WLOCK();
1056         UNP_PCB_LOCK(unp);
1057         socantsendmore(so);
1058         unp_shutdown(unp);
1059         UNP_PCB_UNLOCK(unp);
1060         UNP_LINK_WUNLOCK();
1061         return (0);
1062 }
1063
1064 static int
1065 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1066 {
1067         struct unpcb *unp;
1068         const struct sockaddr *sa;
1069
1070         unp = sotounpcb(so);
1071         KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1072
1073         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1074         UNP_PCB_LOCK(unp);
1075         if (unp->unp_addr != NULL)
1076                 sa = (struct sockaddr *) unp->unp_addr;
1077         else
1078                 sa = &sun_noname;
1079         bcopy(sa, *nam, sa->sa_len);
1080         UNP_PCB_UNLOCK(unp);
1081         return (0);
1082 }
1083
1084 static struct pr_usrreqs uipc_usrreqs_dgram = {
1085         .pru_abort =            uipc_abort,
1086         .pru_accept =           uipc_accept,
1087         .pru_attach =           uipc_attach,
1088         .pru_bind =             uipc_bind,
1089         .pru_connect =          uipc_connect,
1090         .pru_connect2 =         uipc_connect2,
1091         .pru_detach =           uipc_detach,
1092         .pru_disconnect =       uipc_disconnect,
1093         .pru_listen =           uipc_listen,
1094         .pru_peeraddr =         uipc_peeraddr,
1095         .pru_rcvd =             uipc_rcvd,
1096         .pru_send =             uipc_send,
1097         .pru_sense =            uipc_sense,
1098         .pru_shutdown =         uipc_shutdown,
1099         .pru_sockaddr =         uipc_sockaddr,
1100         .pru_soreceive =        soreceive_dgram,
1101         .pru_close =            uipc_close,
1102 };
1103
1104 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1105         .pru_abort =            uipc_abort,
1106         .pru_accept =           uipc_accept,
1107         .pru_attach =           uipc_attach,
1108         .pru_bind =             uipc_bind,
1109         .pru_connect =          uipc_connect,
1110         .pru_connect2 =         uipc_connect2,
1111         .pru_detach =           uipc_detach,
1112         .pru_disconnect =       uipc_disconnect,
1113         .pru_listen =           uipc_listen,
1114         .pru_peeraddr =         uipc_peeraddr,
1115         .pru_rcvd =             uipc_rcvd,
1116         .pru_send =             uipc_send,
1117         .pru_sense =            uipc_sense,
1118         .pru_shutdown =         uipc_shutdown,
1119         .pru_sockaddr =         uipc_sockaddr,
1120         .pru_soreceive =        soreceive_generic,      /* XXX: or...? */
1121         .pru_close =            uipc_close,
1122 };
1123
1124 static struct pr_usrreqs uipc_usrreqs_stream = {
1125         .pru_abort =            uipc_abort,
1126         .pru_accept =           uipc_accept,
1127         .pru_attach =           uipc_attach,
1128         .pru_bind =             uipc_bind,
1129         .pru_connect =          uipc_connect,
1130         .pru_connect2 =         uipc_connect2,
1131         .pru_detach =           uipc_detach,
1132         .pru_disconnect =       uipc_disconnect,
1133         .pru_listen =           uipc_listen,
1134         .pru_peeraddr =         uipc_peeraddr,
1135         .pru_rcvd =             uipc_rcvd,
1136         .pru_send =             uipc_send,
1137         .pru_sense =            uipc_sense,
1138         .pru_shutdown =         uipc_shutdown,
1139         .pru_sockaddr =         uipc_sockaddr,
1140         .pru_soreceive =        soreceive_generic,
1141         .pru_close =            uipc_close,
1142 };
1143
1144 static int
1145 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1146 {
1147         struct unpcb *unp;
1148         struct xucred xu;
1149         int error, optval;
1150
1151         if (sopt->sopt_level != 0)
1152                 return (EINVAL);
1153
1154         unp = sotounpcb(so);
1155         KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1156         error = 0;
1157         switch (sopt->sopt_dir) {
1158         case SOPT_GET:
1159                 switch (sopt->sopt_name) {
1160                 case LOCAL_PEERCRED:
1161                         UNP_PCB_LOCK(unp);
1162                         if (unp->unp_flags & UNP_HAVEPC)
1163                                 xu = unp->unp_peercred;
1164                         else {
1165                                 if (so->so_type == SOCK_STREAM)
1166                                         error = ENOTCONN;
1167                                 else
1168                                         error = EINVAL;
1169                         }
1170                         UNP_PCB_UNLOCK(unp);
1171                         if (error == 0)
1172                                 error = sooptcopyout(sopt, &xu, sizeof(xu));
1173                         break;
1174
1175                 case LOCAL_CREDS:
1176                         /* Unlocked read. */
1177                         optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1178                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1179                         break;
1180
1181                 case LOCAL_CONNWAIT:
1182                         /* Unlocked read. */
1183                         optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1184                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1185                         break;
1186
1187                 default:
1188                         error = EOPNOTSUPP;
1189                         break;
1190                 }
1191                 break;
1192
1193         case SOPT_SET:
1194                 switch (sopt->sopt_name) {
1195                 case LOCAL_CREDS:
1196                 case LOCAL_CONNWAIT:
1197                         error = sooptcopyin(sopt, &optval, sizeof(optval),
1198                                             sizeof(optval));
1199                         if (error)
1200                                 break;
1201
1202 #define OPTSET(bit) do {                                                \
1203         UNP_PCB_LOCK(unp);                                              \
1204         if (optval)                                                     \
1205                 unp->unp_flags |= bit;                                  \
1206         else                                                            \
1207                 unp->unp_flags &= ~bit;                                 \
1208         UNP_PCB_UNLOCK(unp);                                            \
1209 } while (0)
1210
1211                         switch (sopt->sopt_name) {
1212                         case LOCAL_CREDS:
1213                                 OPTSET(UNP_WANTCRED);
1214                                 break;
1215
1216                         case LOCAL_CONNWAIT:
1217                                 OPTSET(UNP_CONNWAIT);
1218                                 break;
1219
1220                         default:
1221                                 break;
1222                         }
1223                         break;
1224 #undef  OPTSET
1225                 default:
1226                         error = ENOPROTOOPT;
1227                         break;
1228                 }
1229                 break;
1230
1231         default:
1232                 error = EOPNOTSUPP;
1233                 break;
1234         }
1235         return (error);
1236 }
1237
1238 static int
1239 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1240 {
1241         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1242         struct vnode *vp;
1243         struct socket *so2, *so3;
1244         struct unpcb *unp, *unp2, *unp3;
1245         int error, len, vfslocked;
1246         struct nameidata nd;
1247         char buf[SOCK_MAXADDRLEN];
1248         struct sockaddr *sa;
1249
1250         UNP_LINK_WLOCK_ASSERT();
1251
1252         unp = sotounpcb(so);
1253         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1254
1255         len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1256         if (len <= 0)
1257                 return (EINVAL);
1258         bcopy(soun->sun_path, buf, len);
1259         buf[len] = 0;
1260
1261         UNP_PCB_LOCK(unp);
1262         if (unp->unp_flags & UNP_CONNECTING) {
1263                 UNP_PCB_UNLOCK(unp);
1264                 return (EALREADY);
1265         }
1266         UNP_LINK_WUNLOCK();
1267         unp->unp_flags |= UNP_CONNECTING;
1268         UNP_PCB_UNLOCK(unp);
1269
1270         sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1271         NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf,
1272             td);
1273         error = namei(&nd);
1274         if (error)
1275                 vp = NULL;
1276         else
1277                 vp = nd.ni_vp;
1278         ASSERT_VOP_LOCKED(vp, "unp_connect");
1279         vfslocked = NDHASGIANT(&nd);
1280         NDFREE(&nd, NDF_ONLY_PNBUF);
1281         if (error)
1282                 goto bad;
1283
1284         if (vp->v_type != VSOCK) {
1285                 error = ENOTSOCK;
1286                 goto bad;
1287         }
1288 #ifdef MAC
1289         error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1290         if (error)
1291                 goto bad;
1292 #endif
1293         error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1294         if (error)
1295                 goto bad;
1296         VFS_UNLOCK_GIANT(vfslocked);
1297
1298         unp = sotounpcb(so);
1299         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1300
1301         /*
1302          * Lock linkage lock for two reasons: make sure v_socket is stable,
1303          * and to protect simultaneous locking of multiple pcbs.
1304          */
1305         UNP_LINK_WLOCK();
1306         so2 = vp->v_socket;
1307         if (so2 == NULL) {
1308                 error = ECONNREFUSED;
1309                 goto bad2;
1310         }
1311         if (so->so_type != so2->so_type) {
1312                 error = EPROTOTYPE;
1313                 goto bad2;
1314         }
1315         if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1316                 if (so2->so_options & SO_ACCEPTCONN) {
1317                         CURVNET_SET(so2->so_vnet);
1318                         so3 = sonewconn(so2, 0);
1319                         CURVNET_RESTORE();
1320                 } else
1321                         so3 = NULL;
1322                 if (so3 == NULL) {
1323                         error = ECONNREFUSED;
1324                         goto bad2;
1325                 }
1326                 unp = sotounpcb(so);
1327                 unp2 = sotounpcb(so2);
1328                 unp3 = sotounpcb(so3);
1329                 UNP_PCB_LOCK(unp);
1330                 UNP_PCB_LOCK(unp2);
1331                 UNP_PCB_LOCK(unp3);
1332                 if (unp2->unp_addr != NULL) {
1333                         bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1334                         unp3->unp_addr = (struct sockaddr_un *) sa;
1335                         sa = NULL;
1336                 }
1337
1338                 /*
1339                  * The connecter's (client's) credentials are copied from its
1340                  * process structure at the time of connect() (which is now).
1341                  */
1342                 cru2x(td->td_ucred, &unp3->unp_peercred);
1343                 unp3->unp_flags |= UNP_HAVEPC;
1344
1345                 /*
1346                  * The receiver's (server's) credentials are copied from the
1347                  * unp_peercred member of socket on which the former called
1348                  * listen(); uipc_listen() cached that process's credentials
1349                  * at that time so we can use them now.
1350                  */
1351                 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1352                     ("unp_connect: listener without cached peercred"));
1353                 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1354                     sizeof(unp->unp_peercred));
1355                 unp->unp_flags |= UNP_HAVEPC;
1356                 if (unp2->unp_flags & UNP_WANTCRED)
1357                         unp3->unp_flags |= UNP_WANTCRED;
1358                 UNP_PCB_UNLOCK(unp3);
1359                 UNP_PCB_UNLOCK(unp2);
1360                 UNP_PCB_UNLOCK(unp);
1361 #ifdef MAC
1362                 mac_socketpeer_set_from_socket(so, so3);
1363                 mac_socketpeer_set_from_socket(so3, so);
1364 #endif
1365
1366                 so2 = so3;
1367         }
1368         unp = sotounpcb(so);
1369         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1370         unp2 = sotounpcb(so2);
1371         KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1372         UNP_PCB_LOCK(unp);
1373         UNP_PCB_LOCK(unp2);
1374         error = unp_connect2(so, so2, PRU_CONNECT);
1375         UNP_PCB_UNLOCK(unp2);
1376         UNP_PCB_UNLOCK(unp);
1377 bad2:
1378         UNP_LINK_WUNLOCK();
1379         if (vfslocked)
1380                 /*
1381                  * Giant has been previously acquired. This means filesystem
1382                  * isn't MPSAFE.  Do it once again.
1383                  */
1384                 mtx_lock(&Giant);
1385 bad:
1386         if (vp != NULL)
1387                 vput(vp);
1388         VFS_UNLOCK_GIANT(vfslocked);
1389         free(sa, M_SONAME);
1390         UNP_LINK_WLOCK();
1391         UNP_PCB_LOCK(unp);
1392         unp->unp_flags &= ~UNP_CONNECTING;
1393         UNP_PCB_UNLOCK(unp);
1394         return (error);
1395 }
1396
1397 static int
1398 unp_connect2(struct socket *so, struct socket *so2, int req)
1399 {
1400         struct unpcb *unp;
1401         struct unpcb *unp2;
1402
1403         unp = sotounpcb(so);
1404         KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1405         unp2 = sotounpcb(so2);
1406         KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1407
1408         UNP_LINK_WLOCK_ASSERT();
1409         UNP_PCB_LOCK_ASSERT(unp);
1410         UNP_PCB_LOCK_ASSERT(unp2);
1411
1412         if (so2->so_type != so->so_type)
1413                 return (EPROTOTYPE);
1414         unp->unp_conn = unp2;
1415
1416         switch (so->so_type) {
1417         case SOCK_DGRAM:
1418                 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1419                 soisconnected(so);
1420                 break;
1421
1422         case SOCK_STREAM:
1423         case SOCK_SEQPACKET:
1424                 unp2->unp_conn = unp;
1425                 if (req == PRU_CONNECT &&
1426                     ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1427                         soisconnecting(so);
1428                 else
1429                         soisconnected(so);
1430                 soisconnected(so2);
1431                 break;
1432
1433         default:
1434                 panic("unp_connect2");
1435         }
1436         return (0);
1437 }
1438
1439 static void
1440 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1441 {
1442         struct socket *so;
1443
1444         KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1445
1446         UNP_LINK_WLOCK_ASSERT();
1447         UNP_PCB_LOCK_ASSERT(unp);
1448         UNP_PCB_LOCK_ASSERT(unp2);
1449
1450         unp->unp_conn = NULL;
1451         switch (unp->unp_socket->so_type) {
1452         case SOCK_DGRAM:
1453                 LIST_REMOVE(unp, unp_reflink);
1454                 so = unp->unp_socket;
1455                 SOCK_LOCK(so);
1456                 so->so_state &= ~SS_ISCONNECTED;
1457                 SOCK_UNLOCK(so);
1458                 break;
1459
1460         case SOCK_STREAM:
1461         case SOCK_SEQPACKET:
1462                 soisdisconnected(unp->unp_socket);
1463                 unp2->unp_conn = NULL;
1464                 soisdisconnected(unp2->unp_socket);
1465                 break;
1466         }
1467 }
1468
1469 /*
1470  * unp_pcblist() walks the global list of struct unpcb's to generate a
1471  * pointer list, bumping the refcount on each unpcb.  It then copies them out
1472  * sequentially, validating the generation number on each to see if it has
1473  * been detached.  All of this is necessary because copyout() may sleep on
1474  * disk I/O.
1475  */
1476 static int
1477 unp_pcblist(SYSCTL_HANDLER_ARGS)
1478 {
1479         int error, i, n;
1480         int freeunp;
1481         struct unpcb *unp, **unp_list;
1482         unp_gen_t gencnt;
1483         struct xunpgen *xug;
1484         struct unp_head *head;
1485         struct xunpcb *xu;
1486
1487         switch ((intptr_t)arg1) {
1488         case SOCK_STREAM:
1489                 head = &unp_shead;
1490                 break;
1491
1492         case SOCK_DGRAM:
1493                 head = &unp_dhead;
1494                 break;
1495
1496         case SOCK_SEQPACKET:
1497                 head = &unp_sphead;
1498                 break;
1499
1500         default:
1501                 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1502         }
1503
1504         /*
1505          * The process of preparing the PCB list is too time-consuming and
1506          * resource-intensive to repeat twice on every request.
1507          */
1508         if (req->oldptr == NULL) {
1509                 n = unp_count;
1510                 req->oldidx = 2 * (sizeof *xug)
1511                         + (n + n/8) * sizeof(struct xunpcb);
1512                 return (0);
1513         }
1514
1515         if (req->newptr != NULL)
1516                 return (EPERM);
1517
1518         /*
1519          * OK, now we're committed to doing something.
1520          */
1521         xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1522         UNP_LIST_LOCK();
1523         gencnt = unp_gencnt;
1524         n = unp_count;
1525         UNP_LIST_UNLOCK();
1526
1527         xug->xug_len = sizeof *xug;
1528         xug->xug_count = n;
1529         xug->xug_gen = gencnt;
1530         xug->xug_sogen = so_gencnt;
1531         error = SYSCTL_OUT(req, xug, sizeof *xug);
1532         if (error) {
1533                 free(xug, M_TEMP);
1534                 return (error);
1535         }
1536
1537         unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1538
1539         UNP_LIST_LOCK();
1540         for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1541              unp = LIST_NEXT(unp, unp_link)) {
1542                 UNP_PCB_LOCK(unp);
1543                 if (unp->unp_gencnt <= gencnt) {
1544                         if (cr_cansee(req->td->td_ucred,
1545                             unp->unp_socket->so_cred)) {
1546                                 UNP_PCB_UNLOCK(unp);
1547                                 continue;
1548                         }
1549                         unp_list[i++] = unp;
1550                         unp->unp_refcount++;
1551                 }
1552                 UNP_PCB_UNLOCK(unp);
1553         }
1554         UNP_LIST_UNLOCK();
1555         n = i;                  /* In case we lost some during malloc. */
1556
1557         error = 0;
1558         xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1559         for (i = 0; i < n; i++) {
1560                 unp = unp_list[i];
1561                 UNP_PCB_LOCK(unp);
1562                 unp->unp_refcount--;
1563                 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1564                         xu->xu_len = sizeof *xu;
1565                         xu->xu_unpp = unp;
1566                         /*
1567                          * XXX - need more locking here to protect against
1568                          * connect/disconnect races for SMP.
1569                          */
1570                         if (unp->unp_addr != NULL)
1571                                 bcopy(unp->unp_addr, &xu->xu_addr,
1572                                       unp->unp_addr->sun_len);
1573                         if (unp->unp_conn != NULL &&
1574                             unp->unp_conn->unp_addr != NULL)
1575                                 bcopy(unp->unp_conn->unp_addr,
1576                                       &xu->xu_caddr,
1577                                       unp->unp_conn->unp_addr->sun_len);
1578                         bcopy(unp, &xu->xu_unp, sizeof *unp);
1579                         sotoxsocket(unp->unp_socket, &xu->xu_socket);
1580                         UNP_PCB_UNLOCK(unp);
1581                         error = SYSCTL_OUT(req, xu, sizeof *xu);
1582                 } else {
1583                         freeunp = (unp->unp_refcount == 0);
1584                         UNP_PCB_UNLOCK(unp);
1585                         if (freeunp) {
1586                                 UNP_PCB_LOCK_DESTROY(unp);
1587                                 uma_zfree(unp_zone, unp);
1588                         }
1589                 }
1590         }
1591         free(xu, M_TEMP);
1592         if (!error) {
1593                 /*
1594                  * Give the user an updated idea of our state.  If the
1595                  * generation differs from what we told her before, she knows
1596                  * that something happened while we were processing this
1597                  * request, and it might be necessary to retry.
1598                  */
1599                 xug->xug_gen = unp_gencnt;
1600                 xug->xug_sogen = so_gencnt;
1601                 xug->xug_count = unp_count;
1602                 error = SYSCTL_OUT(req, xug, sizeof *xug);
1603         }
1604         free(unp_list, M_TEMP);
1605         free(xug, M_TEMP);
1606         return (error);
1607 }
1608
1609 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1610     (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1611     "List of active local datagram sockets");
1612 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1613     (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1614     "List of active local stream sockets");
1615 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1616     CTLTYPE_OPAQUE | CTLFLAG_RD,
1617     (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1618     "List of active local seqpacket sockets");
1619
1620 static void
1621 unp_shutdown(struct unpcb *unp)
1622 {
1623         struct unpcb *unp2;
1624         struct socket *so;
1625
1626         UNP_LINK_WLOCK_ASSERT();
1627         UNP_PCB_LOCK_ASSERT(unp);
1628
1629         unp2 = unp->unp_conn;
1630         if ((unp->unp_socket->so_type == SOCK_STREAM ||
1631             (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1632                 so = unp2->unp_socket;
1633                 if (so != NULL)
1634                         socantrcvmore(so);
1635         }
1636 }
1637
1638 static void
1639 unp_drop(struct unpcb *unp, int errno)
1640 {
1641         struct socket *so = unp->unp_socket;
1642         struct unpcb *unp2;
1643
1644         UNP_LINK_WLOCK_ASSERT();
1645         UNP_PCB_LOCK_ASSERT(unp);
1646
1647         so->so_error = errno;
1648         unp2 = unp->unp_conn;
1649         if (unp2 == NULL)
1650                 return;
1651         UNP_PCB_LOCK(unp2);
1652         unp_disconnect(unp, unp2);
1653         UNP_PCB_UNLOCK(unp2);
1654 }
1655
1656 static void
1657 unp_freerights(struct file **rp, int fdcount)
1658 {
1659         int i;
1660         struct file *fp;
1661
1662         for (i = 0; i < fdcount; i++) {
1663                 fp = *rp;
1664                 *rp++ = NULL;
1665                 unp_discard(fp);
1666         }
1667 }
1668
1669 static int
1670 unp_externalize(struct mbuf *control, struct mbuf **controlp)
1671 {
1672         struct thread *td = curthread;          /* XXX */
1673         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1674         int i;
1675         int *fdp;
1676         struct file **rp;
1677         struct file *fp;
1678         void *data;
1679         socklen_t clen = control->m_len, datalen;
1680         int error, newfds;
1681         int f;
1682         u_int newlen;
1683
1684         UNP_LINK_UNLOCK_ASSERT();
1685
1686         error = 0;
1687         if (controlp != NULL) /* controlp == NULL => free control messages */
1688                 *controlp = NULL;
1689         while (cm != NULL) {
1690                 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1691                         error = EINVAL;
1692                         break;
1693                 }
1694                 data = CMSG_DATA(cm);
1695                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1696                 if (cm->cmsg_level == SOL_SOCKET
1697                     && cm->cmsg_type == SCM_RIGHTS) {
1698                         newfds = datalen / sizeof(struct file *);
1699                         rp = data;
1700
1701                         /* If we're not outputting the descriptors free them. */
1702                         if (error || controlp == NULL) {
1703                                 unp_freerights(rp, newfds);
1704                                 goto next;
1705                         }
1706                         FILEDESC_XLOCK(td->td_proc->p_fd);
1707                         /* if the new FD's will not fit free them.  */
1708                         if (!fdavail(td, newfds)) {
1709                                 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1710                                 error = EMSGSIZE;
1711                                 unp_freerights(rp, newfds);
1712                                 goto next;
1713                         }
1714
1715                         /*
1716                          * Now change each pointer to an fd in the global
1717                          * table to an integer that is the index to the local
1718                          * fd table entry that we set up to point to the
1719                          * global one we are transferring.
1720                          */
1721                         newlen = newfds * sizeof(int);
1722                         *controlp = sbcreatecontrol(NULL, newlen,
1723                             SCM_RIGHTS, SOL_SOCKET);
1724                         if (*controlp == NULL) {
1725                                 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1726                                 error = E2BIG;
1727                                 unp_freerights(rp, newfds);
1728                                 goto next;
1729                         }
1730
1731                         fdp = (int *)
1732                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1733                         for (i = 0; i < newfds; i++) {
1734                                 if (fdalloc(td, 0, &f))
1735                                         panic("unp_externalize fdalloc failed");
1736                                 fp = *rp++;
1737                                 td->td_proc->p_fd->fd_ofiles[f] = fp;
1738                                 unp_externalize_fp(fp);
1739                                 *fdp++ = f;
1740                         }
1741                         FILEDESC_XUNLOCK(td->td_proc->p_fd);
1742                 } else {
1743                         /* We can just copy anything else across. */
1744                         if (error || controlp == NULL)
1745                                 goto next;
1746                         *controlp = sbcreatecontrol(NULL, datalen,
1747                             cm->cmsg_type, cm->cmsg_level);
1748                         if (*controlp == NULL) {
1749                                 error = ENOBUFS;
1750                                 goto next;
1751                         }
1752                         bcopy(data,
1753                             CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1754                             datalen);
1755                 }
1756                 controlp = &(*controlp)->m_next;
1757
1758 next:
1759                 if (CMSG_SPACE(datalen) < clen) {
1760                         clen -= CMSG_SPACE(datalen);
1761                         cm = (struct cmsghdr *)
1762                             ((caddr_t)cm + CMSG_SPACE(datalen));
1763                 } else {
1764                         clen = 0;
1765                         cm = NULL;
1766                 }
1767         }
1768
1769         m_freem(control);
1770         return (error);
1771 }
1772
1773 static void
1774 unp_zone_change(void *tag)
1775 {
1776
1777         uma_zone_set_max(unp_zone, maxsockets);
1778 }
1779
1780 static void
1781 unp_init(void)
1782 {
1783
1784 #ifdef VIMAGE
1785         if (!IS_DEFAULT_VNET(curvnet))
1786                 return;
1787 #endif
1788         unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1789             NULL, NULL, UMA_ALIGN_PTR, 0);
1790         if (unp_zone == NULL)
1791                 panic("unp_init");
1792         uma_zone_set_max(unp_zone, maxsockets);
1793         EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1794             NULL, EVENTHANDLER_PRI_ANY);
1795         LIST_INIT(&unp_dhead);
1796         LIST_INIT(&unp_shead);
1797         LIST_INIT(&unp_sphead);
1798         SLIST_INIT(&unp_defers);
1799         TASK_INIT(&unp_gc_task, 0, unp_gc, NULL);
1800         TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1801         UNP_LINK_LOCK_INIT();
1802         UNP_LIST_LOCK_INIT();
1803         UNP_DEFERRED_LOCK_INIT();
1804 }
1805
1806 static int
1807 unp_internalize(struct mbuf **controlp, struct thread *td)
1808 {
1809         struct mbuf *control = *controlp;
1810         struct proc *p = td->td_proc;
1811         struct filedesc *fdescp = p->p_fd;
1812         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1813         struct cmsgcred *cmcred;
1814         struct file **rp;
1815         struct file *fp;
1816         struct timeval *tv;
1817         int i, fd, *fdp;
1818         void *data;
1819         socklen_t clen = control->m_len, datalen;
1820         int error, oldfds;
1821         u_int newlen;
1822
1823         UNP_LINK_UNLOCK_ASSERT();
1824
1825         error = 0;
1826         *controlp = NULL;
1827         while (cm != NULL) {
1828                 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1829                     || cm->cmsg_len > clen) {
1830                         error = EINVAL;
1831                         goto out;
1832                 }
1833                 data = CMSG_DATA(cm);
1834                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1835
1836                 switch (cm->cmsg_type) {
1837                 /*
1838                  * Fill in credential information.
1839                  */
1840                 case SCM_CREDS:
1841                         *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1842                             SCM_CREDS, SOL_SOCKET);
1843                         if (*controlp == NULL) {
1844                                 error = ENOBUFS;
1845                                 goto out;
1846                         }
1847                         cmcred = (struct cmsgcred *)
1848                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1849                         cmcred->cmcred_pid = p->p_pid;
1850                         cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1851                         cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1852                         cmcred->cmcred_euid = td->td_ucred->cr_uid;
1853                         cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1854                             CMGROUP_MAX);
1855                         for (i = 0; i < cmcred->cmcred_ngroups; i++)
1856                                 cmcred->cmcred_groups[i] =
1857                                     td->td_ucred->cr_groups[i];
1858                         break;
1859
1860                 case SCM_RIGHTS:
1861                         oldfds = datalen / sizeof (int);
1862                         /*
1863                          * Check that all the FDs passed in refer to legal
1864                          * files.  If not, reject the entire operation.
1865                          */
1866                         fdp = data;
1867                         FILEDESC_SLOCK(fdescp);
1868                         for (i = 0; i < oldfds; i++) {
1869                                 fd = *fdp++;
1870                                 if ((unsigned)fd >= fdescp->fd_nfiles ||
1871                                     fdescp->fd_ofiles[fd] == NULL) {
1872                                         FILEDESC_SUNLOCK(fdescp);
1873                                         error = EBADF;
1874                                         goto out;
1875                                 }
1876                                 fp = fdescp->fd_ofiles[fd];
1877                                 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1878                                         FILEDESC_SUNLOCK(fdescp);
1879                                         error = EOPNOTSUPP;
1880                                         goto out;
1881                                 }
1882
1883                         }
1884
1885                         /*
1886                          * Now replace the integer FDs with pointers to the
1887                          * associated global file table entry..
1888                          */
1889                         newlen = oldfds * sizeof(struct file *);
1890                         *controlp = sbcreatecontrol(NULL, newlen,
1891                             SCM_RIGHTS, SOL_SOCKET);
1892                         if (*controlp == NULL) {
1893                                 FILEDESC_SUNLOCK(fdescp);
1894                                 error = E2BIG;
1895                                 goto out;
1896                         }
1897                         fdp = data;
1898                         rp = (struct file **)
1899                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1900                         for (i = 0; i < oldfds; i++) {
1901                                 fp = fdescp->fd_ofiles[*fdp++];
1902                                 *rp++ = fp;
1903                                 unp_internalize_fp(fp);
1904                         }
1905                         FILEDESC_SUNLOCK(fdescp);
1906                         break;
1907
1908                 case SCM_TIMESTAMP:
1909                         *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1910                             SCM_TIMESTAMP, SOL_SOCKET);
1911                         if (*controlp == NULL) {
1912                                 error = ENOBUFS;
1913                                 goto out;
1914                         }
1915                         tv = (struct timeval *)
1916                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1917                         microtime(tv);
1918                         break;
1919
1920                 default:
1921                         error = EINVAL;
1922                         goto out;
1923                 }
1924
1925                 controlp = &(*controlp)->m_next;
1926                 if (CMSG_SPACE(datalen) < clen) {
1927                         clen -= CMSG_SPACE(datalen);
1928                         cm = (struct cmsghdr *)
1929                             ((caddr_t)cm + CMSG_SPACE(datalen));
1930                 } else {
1931                         clen = 0;
1932                         cm = NULL;
1933                 }
1934         }
1935
1936 out:
1937         m_freem(control);
1938         return (error);
1939 }
1940
1941 static struct mbuf *
1942 unp_addsockcred(struct thread *td, struct mbuf *control)
1943 {
1944         struct mbuf *m, *n, *n_prev;
1945         struct sockcred *sc;
1946         const struct cmsghdr *cm;
1947         int ngroups;
1948         int i;
1949
1950         ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
1951         m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
1952         if (m == NULL)
1953                 return (control);
1954
1955         sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
1956         sc->sc_uid = td->td_ucred->cr_ruid;
1957         sc->sc_euid = td->td_ucred->cr_uid;
1958         sc->sc_gid = td->td_ucred->cr_rgid;
1959         sc->sc_egid = td->td_ucred->cr_gid;
1960         sc->sc_ngroups = ngroups;
1961         for (i = 0; i < sc->sc_ngroups; i++)
1962                 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
1963
1964         /*
1965          * Unlink SCM_CREDS control messages (struct cmsgcred), since just
1966          * created SCM_CREDS control message (struct sockcred) has another
1967          * format.
1968          */
1969         if (control != NULL)
1970                 for (n = control, n_prev = NULL; n != NULL;) {
1971                         cm = mtod(n, struct cmsghdr *);
1972                         if (cm->cmsg_level == SOL_SOCKET &&
1973                             cm->cmsg_type == SCM_CREDS) {
1974                                 if (n_prev == NULL)
1975                                         control = n->m_next;
1976                                 else
1977                                         n_prev->m_next = n->m_next;
1978                                 n = m_free(n);
1979                         } else {
1980                                 n_prev = n;
1981                                 n = n->m_next;
1982                         }
1983                 }
1984
1985         /* Prepend it to the head. */
1986         m->m_next = control;
1987         return (m);
1988 }
1989
1990 static struct unpcb *
1991 fptounp(struct file *fp)
1992 {
1993         struct socket *so;
1994
1995         if (fp->f_type != DTYPE_SOCKET)
1996                 return (NULL);
1997         if ((so = fp->f_data) == NULL)
1998                 return (NULL);
1999         if (so->so_proto->pr_domain != &localdomain)
2000                 return (NULL);
2001         return sotounpcb(so);
2002 }
2003
2004 static void
2005 unp_discard(struct file *fp)
2006 {
2007         struct unp_defer *dr;
2008
2009         if (unp_externalize_fp(fp)) {
2010                 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2011                 dr->ud_fp = fp;
2012                 UNP_DEFERRED_LOCK();
2013                 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2014                 UNP_DEFERRED_UNLOCK();
2015                 atomic_add_int(&unp_defers_count, 1);
2016                 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2017         } else
2018                 (void) closef(fp, (struct thread *)NULL);
2019 }
2020
2021 static void
2022 unp_process_defers(void *arg __unused, int pending)
2023 {
2024         struct unp_defer *dr;
2025         SLIST_HEAD(, unp_defer) drl;
2026         int count;
2027
2028         SLIST_INIT(&drl);
2029         for (;;) {
2030                 UNP_DEFERRED_LOCK();
2031                 if (SLIST_FIRST(&unp_defers) == NULL) {
2032                         UNP_DEFERRED_UNLOCK();
2033                         break;
2034                 }
2035                 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2036                 UNP_DEFERRED_UNLOCK();
2037                 count = 0;
2038                 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2039                         SLIST_REMOVE_HEAD(&drl, ud_link);
2040                         closef(dr->ud_fp, NULL);
2041                         free(dr, M_TEMP);
2042                         count++;
2043                 }
2044                 atomic_add_int(&unp_defers_count, -count);
2045         }
2046 }
2047
2048 static void
2049 unp_internalize_fp(struct file *fp)
2050 {
2051         struct unpcb *unp;
2052
2053         UNP_LINK_WLOCK();
2054         if ((unp = fptounp(fp)) != NULL) {
2055                 unp->unp_file = fp;
2056                 unp->unp_msgcount++;
2057         }
2058         fhold(fp);
2059         unp_rights++;
2060         UNP_LINK_WUNLOCK();
2061 }
2062
2063 static int
2064 unp_externalize_fp(struct file *fp)
2065 {
2066         struct unpcb *unp;
2067         int ret;
2068
2069         UNP_LINK_WLOCK();
2070         if ((unp = fptounp(fp)) != NULL) {
2071                 unp->unp_msgcount--;
2072                 ret = 1;
2073         } else
2074                 ret = 0;
2075         unp_rights--;
2076         UNP_LINK_WUNLOCK();
2077         return (ret);
2078 }
2079
2080 /*
2081  * unp_defer indicates whether additional work has been defered for a future
2082  * pass through unp_gc().  It is thread local and does not require explicit
2083  * synchronization.
2084  */
2085 static int      unp_marked;
2086 static int      unp_unreachable;
2087
2088 static void
2089 unp_accessable(struct file *fp)
2090 {
2091         struct unpcb *unp;
2092
2093         if ((unp = fptounp(fp)) == NULL)
2094                 return;
2095         if (unp->unp_gcflag & UNPGC_REF)
2096                 return;
2097         unp->unp_gcflag &= ~UNPGC_DEAD;
2098         unp->unp_gcflag |= UNPGC_REF;
2099         unp_marked++;
2100 }
2101
2102 static void
2103 unp_gc_process(struct unpcb *unp)
2104 {
2105         struct socket *soa;
2106         struct socket *so;
2107         struct file *fp;
2108
2109         /* Already processed. */
2110         if (unp->unp_gcflag & UNPGC_SCANNED)
2111                 return;
2112         fp = unp->unp_file;
2113
2114         /*
2115          * Check for a socket potentially in a cycle.  It must be in a
2116          * queue as indicated by msgcount, and this must equal the file
2117          * reference count.  Note that when msgcount is 0 the file is NULL.
2118          */
2119         if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2120             unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2121                 unp->unp_gcflag |= UNPGC_DEAD;
2122                 unp_unreachable++;
2123                 return;
2124         }
2125
2126         /*
2127          * Mark all sockets we reference with RIGHTS.
2128          */
2129         so = unp->unp_socket;
2130         SOCKBUF_LOCK(&so->so_rcv);
2131         unp_scan(so->so_rcv.sb_mb, unp_accessable);
2132         SOCKBUF_UNLOCK(&so->so_rcv);
2133
2134         /*
2135          * Mark all sockets in our accept queue.
2136          */
2137         ACCEPT_LOCK();
2138         TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2139                 SOCKBUF_LOCK(&soa->so_rcv);
2140                 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2141                 SOCKBUF_UNLOCK(&soa->so_rcv);
2142         }
2143         ACCEPT_UNLOCK();
2144         unp->unp_gcflag |= UNPGC_SCANNED;
2145 }
2146
2147 static int unp_recycled;
2148 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2149     "Number of unreachable sockets claimed by the garbage collector.");
2150
2151 static int unp_taskcount;
2152 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2153     "Number of times the garbage collector has run.");
2154
2155 static void
2156 unp_gc(__unused void *arg, int pending)
2157 {
2158         struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2159                                     NULL };
2160         struct unp_head **head;
2161         struct file *f, **unref;
2162         struct unpcb *unp;
2163         int i, total;
2164
2165         unp_taskcount++;
2166         UNP_LIST_LOCK();
2167         /*
2168          * First clear all gc flags from previous runs.
2169          */
2170         for (head = heads; *head != NULL; head++)
2171                 LIST_FOREACH(unp, *head, unp_link)
2172                         unp->unp_gcflag = 0;
2173
2174         /*
2175          * Scan marking all reachable sockets with UNPGC_REF.  Once a socket
2176          * is reachable all of the sockets it references are reachable.
2177          * Stop the scan once we do a complete loop without discovering
2178          * a new reachable socket.
2179          */
2180         do {
2181                 unp_unreachable = 0;
2182                 unp_marked = 0;
2183                 for (head = heads; *head != NULL; head++)
2184                         LIST_FOREACH(unp, *head, unp_link)
2185                                 unp_gc_process(unp);
2186         } while (unp_marked);
2187         UNP_LIST_UNLOCK();
2188         if (unp_unreachable == 0)
2189                 return;
2190
2191         /*
2192          * Allocate space for a local list of dead unpcbs.
2193          */
2194         unref = malloc(unp_unreachable * sizeof(struct file *),
2195             M_TEMP, M_WAITOK);
2196
2197         /*
2198          * Iterate looking for sockets which have been specifically marked
2199          * as as unreachable and store them locally.
2200          */
2201         UNP_LINK_RLOCK();
2202         UNP_LIST_LOCK();
2203         for (total = 0, head = heads; *head != NULL; head++)
2204                 LIST_FOREACH(unp, *head, unp_link)
2205                         if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2206                                 f = unp->unp_file;
2207                                 if (unp->unp_msgcount == 0 || f == NULL ||
2208                                     f->f_count != unp->unp_msgcount)
2209                                         continue;
2210                                 unref[total++] = f;
2211                                 fhold(f);
2212                                 KASSERT(total <= unp_unreachable,
2213                                     ("unp_gc: incorrect unreachable count."));
2214                         }
2215         UNP_LIST_UNLOCK();
2216         UNP_LINK_RUNLOCK();
2217
2218         /*
2219          * Now flush all sockets, free'ing rights.  This will free the
2220          * struct files associated with these sockets but leave each socket
2221          * with one remaining ref.
2222          */
2223         for (i = 0; i < total; i++) {
2224                 struct socket *so;
2225
2226                 so = unref[i]->f_data;
2227                 CURVNET_SET(so->so_vnet);
2228                 sorflush(so);
2229                 CURVNET_RESTORE();
2230         }
2231
2232         /*
2233          * And finally release the sockets so they can be reclaimed.
2234          */
2235         for (i = 0; i < total; i++)
2236                 fdrop(unref[i], NULL);
2237         unp_recycled += total;
2238         free(unref, M_TEMP);
2239 }
2240
2241 static void
2242 unp_dispose(struct mbuf *m)
2243 {
2244
2245         if (m)
2246                 unp_scan(m, unp_discard);
2247 }
2248
2249 static void
2250 unp_scan(struct mbuf *m0, void (*op)(struct file *))
2251 {
2252         struct mbuf *m;
2253         struct file **rp;
2254         struct cmsghdr *cm;
2255         void *data;
2256         int i;
2257         socklen_t clen, datalen;
2258         int qfds;
2259
2260         while (m0 != NULL) {
2261                 for (m = m0; m; m = m->m_next) {
2262                         if (m->m_type != MT_CONTROL)
2263                                 continue;
2264
2265                         cm = mtod(m, struct cmsghdr *);
2266                         clen = m->m_len;
2267
2268                         while (cm != NULL) {
2269                                 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2270                                         break;
2271
2272                                 data = CMSG_DATA(cm);
2273                                 datalen = (caddr_t)cm + cm->cmsg_len
2274                                     - (caddr_t)data;
2275
2276                                 if (cm->cmsg_level == SOL_SOCKET &&
2277                                     cm->cmsg_type == SCM_RIGHTS) {
2278                                         qfds = datalen / sizeof (struct file *);
2279                                         rp = data;
2280                                         for (i = 0; i < qfds; i++)
2281                                                 (*op)(*rp++);
2282                                 }
2283
2284                                 if (CMSG_SPACE(datalen) < clen) {
2285                                         clen -= CMSG_SPACE(datalen);
2286                                         cm = (struct cmsghdr *)
2287                                             ((caddr_t)cm + CMSG_SPACE(datalen));
2288                                 } else {
2289                                         clen = 0;
2290                                         cm = NULL;
2291                                 }
2292                         }
2293                 }
2294                 m0 = m0->m_act;
2295         }
2296 }
2297
2298 #ifdef DDB
2299 static void
2300 db_print_indent(int indent)
2301 {
2302         int i;
2303
2304         for (i = 0; i < indent; i++)
2305                 db_printf(" ");
2306 }
2307
2308 static void
2309 db_print_unpflags(int unp_flags)
2310 {
2311         int comma;
2312
2313         comma = 0;
2314         if (unp_flags & UNP_HAVEPC) {
2315                 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2316                 comma = 1;
2317         }
2318         if (unp_flags & UNP_HAVEPCCACHED) {
2319                 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2320                 comma = 1;
2321         }
2322         if (unp_flags & UNP_WANTCRED) {
2323                 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2324                 comma = 1;
2325         }
2326         if (unp_flags & UNP_CONNWAIT) {
2327                 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2328                 comma = 1;
2329         }
2330         if (unp_flags & UNP_CONNECTING) {
2331                 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2332                 comma = 1;
2333         }
2334         if (unp_flags & UNP_BINDING) {
2335                 db_printf("%sUNP_BINDING", comma ? ", " : "");
2336                 comma = 1;
2337         }
2338 }
2339
2340 static void
2341 db_print_xucred(int indent, struct xucred *xu)
2342 {
2343         int comma, i;
2344
2345         db_print_indent(indent);
2346         db_printf("cr_version: %u   cr_uid: %u   cr_ngroups: %d\n",
2347             xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2348         db_print_indent(indent);
2349         db_printf("cr_groups: ");
2350         comma = 0;
2351         for (i = 0; i < xu->cr_ngroups; i++) {
2352                 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2353                 comma = 1;
2354         }
2355         db_printf("\n");
2356 }
2357
2358 static void
2359 db_print_unprefs(int indent, struct unp_head *uh)
2360 {
2361         struct unpcb *unp;
2362         int counter;
2363
2364         counter = 0;
2365         LIST_FOREACH(unp, uh, unp_reflink) {
2366                 if (counter % 4 == 0)
2367                         db_print_indent(indent);
2368                 db_printf("%p  ", unp);
2369                 if (counter % 4 == 3)
2370                         db_printf("\n");
2371                 counter++;
2372         }
2373         if (counter != 0 && counter % 4 != 0)
2374                 db_printf("\n");
2375 }
2376
2377 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2378 {
2379         struct unpcb *unp;
2380
2381         if (!have_addr) {
2382                 db_printf("usage: show unpcb <addr>\n");
2383                 return;
2384         }
2385         unp = (struct unpcb *)addr;
2386
2387         db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
2388             unp->unp_vnode);
2389
2390         db_printf("unp_ino: %d   unp_conn: %p\n", unp->unp_ino,
2391             unp->unp_conn);
2392
2393         db_printf("unp_refs:\n");
2394         db_print_unprefs(2, &unp->unp_refs);
2395
2396         /* XXXRW: Would be nice to print the full address, if any. */
2397         db_printf("unp_addr: %p\n", unp->unp_addr);
2398
2399         db_printf("unp_cc: %d   unp_mbcnt: %d   unp_gencnt: %llu\n",
2400             unp->unp_cc, unp->unp_mbcnt,
2401             (unsigned long long)unp->unp_gencnt);
2402
2403         db_printf("unp_flags: %x (", unp->unp_flags);
2404         db_print_unpflags(unp->unp_flags);
2405         db_printf(")\n");
2406
2407         db_printf("unp_peercred:\n");
2408         db_print_xucred(2, &unp->unp_peercred);
2409
2410         db_printf("unp_refcount: %u\n", unp->unp_refcount);
2411 }
2412 #endif