sys/kern/uipc_usrreq.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-3-Clause
   3  *
   4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
   5  *      The Regents of the University of California. All Rights Reserved.
   6  * Copyright (c) 2004-2009 Robert N. M. Watson All Rights Reserved.
   7  * Copyright (c) 2018 Matthew Macy
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in the
  16  *    documentation and/or other materials provided with the distribution.
  17  * 3. Neither the name of the University nor the names of its contributors
  18  *    may be used to endorse or promote products derived from this software
  19  *    without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  *
  33  *      From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
  34  */
  35
  36 /*
  37  * UNIX Domain (Local) Sockets
  38  *
  39  * This is an implementation of UNIX (local) domain sockets.  Each socket has
  40  * an associated struct unpcb (UNIX protocol control block).  Stream sockets
  41  * may be connected to 0 or 1 other socket.  Datagram sockets may be
  42  * connected to 0, 1, or many other sockets.  Sockets may be created and
  43  * connected in pairs (socketpair(2)), or bound/connected to using the file
  44  * system name space.  For most purposes, only the receive socket buffer is
  45  * used, as sending on one socket delivers directly to the receive socket
  46  * buffer of a second socket.
  47  *
  48  * The implementation is substantially complicated by the fact that
  49  * "ancillary data", such as file descriptors or credentials, may be passed
  50  * across UNIX domain sockets.  The potential for passing UNIX domain sockets
  51  * over other UNIX domain sockets requires the implementation of a simple
  52  * garbage collector to find and tear down cycles of disconnected sockets.
  53  *
  54  * TODO:
  55  *      RDM
  56  *      rethink name space problems
  57  *      need a proper out-of-band
  58  */
  59
  60 #include <sys/cdefs.h>
  61 __FBSDID("$FreeBSD$");
  62
  63 #include "opt_ddb.h"
  64
  65 #include <sys/param.h>
  66 #include <sys/capsicum.h>
  67 #include <sys/domain.h>
  68 #include <sys/eventhandler.h>
  69 #include <sys/fcntl.h>
  70 #include <sys/file.h>
  71 #include <sys/filedesc.h>
  72 #include <sys/kernel.h>
  73 #include <sys/lock.h>
  74 #include <sys/malloc.h>
  75 #include <sys/mbuf.h>
  76 #include <sys/mount.h>
  77 #include <sys/mutex.h>
  78 #include <sys/namei.h>
  79 #include <sys/proc.h>
  80 #include <sys/protosw.h>
  81 #include <sys/queue.h>
  82 #include <sys/resourcevar.h>
  83 #include <sys/rwlock.h>
  84 #include <sys/socket.h>
  85 #include <sys/socketvar.h>
  86 #include <sys/signalvar.h>
  87 #include <sys/stat.h>
  88 #include <sys/sx.h>
  89 #include <sys/sysctl.h>
  90 #include <sys/systm.h>
  91 #include <sys/taskqueue.h>
  92 #include <sys/un.h>
  93 #include <sys/unpcb.h>
  94 #include <sys/vnode.h>
  95
  96 #include <net/vnet.h>
  97
  98 #ifdef DDB
  99 #include <ddb/ddb.h>
 100 #endif
 101
 102 #include <security/mac/mac_framework.h>
 103
 104 #include <vm/uma.h>
 105
 106 MALLOC_DECLARE(M_FILECAPS);
 107
 108 /*
 109  * See unpcb.h for the locking key.
 110  */
 111
 112 static uma_zone_t       unp_zone;
 113 static unp_gen_t        unp_gencnt;     /* (l) */
 114 static u_int            unp_count;      /* (l) Count of local sockets. */
 115 static ino_t            unp_ino;        /* Prototype for fake inode numbers. */
 116 static int              unp_rights;     /* (g) File descriptors in flight. */
 117 static struct unp_head  unp_shead;      /* (l) List of stream sockets. */
 118 static struct unp_head  unp_dhead;      /* (l) List of datagram sockets. */
 119 static struct unp_head  unp_sphead;     /* (l) List of seqpacket sockets. */
 120
 121 struct unp_defer {
 122         SLIST_ENTRY(unp_defer) ud_link;
 123         struct file *ud_fp;
 124 };
 125 static SLIST_HEAD(, unp_defer) unp_defers;
 126 static int unp_defers_count;
 127
 128 static const struct sockaddr    sun_noname = { sizeof(sun_noname), AF_LOCAL };
 129
 130 /*
 131  * Garbage collection of cyclic file descriptor/socket references occurs
 132  * asynchronously in a taskqueue context in order to avoid recursion and
 133  * reentrance in the UNIX domain socket, file descriptor, and socket layer
 134  * code.  See unp_gc() for a full description.
 135  */
 136 static struct timeout_task unp_gc_task;
 137
 138 /*
 139  * The close of unix domain sockets attached as SCM_RIGHTS is
 140  * postponed to the taskqueue, to avoid arbitrary recursion depth.
 141  * The attached sockets might have another sockets attached.
 142  */
 143 static struct task      unp_defer_task;
 144
 145 /*
 146  * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
 147  * stream sockets, although the total for sender and receiver is actually
 148  * only PIPSIZ.
 149  *
 150  * Datagram sockets really use the sendspace as the maximum datagram size,
 151  * and don't really want to reserve the sendspace.  Their recvspace should be
 152  * large enough for at least one max-size datagram plus address.
 153  */
 154 #ifndef PIPSIZ
 155 #define PIPSIZ  8192
 156 #endif
 157 static u_long   unpst_sendspace = PIPSIZ;
 158 static u_long   unpst_recvspace = PIPSIZ;
 159 static u_long   unpdg_sendspace = 2*1024;       /* really max datagram size */
 160 static u_long   unpdg_recvspace = 4*1024;
 161 static u_long   unpsp_sendspace = PIPSIZ;       /* really max datagram size */
 162 static u_long   unpsp_recvspace = PIPSIZ;
 163
 164 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 165     "Local domain");
 166 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream,
 167     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 168     "SOCK_STREAM");
 169 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram,
 170     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 171     "SOCK_DGRAM");
 172 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket,
 173     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 174     "SOCK_SEQPACKET");
 175
 176 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
 177            &unpst_sendspace, 0, "Default stream send space.");
 178 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
 179            &unpst_recvspace, 0, "Default stream receive space.");
 180 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
 181            &unpdg_sendspace, 0, "Default datagram send space.");
 182 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
 183            &unpdg_recvspace, 0, "Default datagram receive space.");
 184 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
 185            &unpsp_sendspace, 0, "Default seqpacket send space.");
 186 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
 187            &unpsp_recvspace, 0, "Default seqpacket receive space.");
 188 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
 189     "File descriptors in flight.");
 190 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
 191     &unp_defers_count, 0,
 192     "File descriptors deferred to taskqueue for close.");
 193
 194 /*
 195  * Locking and synchronization:
 196  *
 197  * Several types of locks exist in the local domain socket implementation:
 198  * - a global linkage lock
 199  * - a global connection list lock
 200  * - the mtxpool lock
 201  * - per-unpcb mutexes
 202  *
 203  * The linkage lock protects the global socket lists, the generation number
 204  * counter and garbage collector state.
 205  *
 206  * The connection list lock protects the list of referring sockets in a datagram
 207  * socket PCB.  This lock is also overloaded to protect a global list of
 208  * sockets whose buffers contain socket references in the form of SCM_RIGHTS
 209  * messages.  To avoid recursion, such references are released by a dedicated
 210  * thread.
 211  *
 212  * The mtxpool lock protects the vnode from being modified while referenced.
 213  * Lock ordering rules require that it be acquired before any PCB locks.
 214  *
 215  * The unpcb lock (unp_mtx) protects the most commonly referenced fields in the
 216  * unpcb.  This includes the unp_conn field, which either links two connected
 217  * PCBs together (for connected socket types) or points at the destination
 218  * socket (for connectionless socket types).  The operations of creating or
 219  * destroying a connection therefore involve locking multiple PCBs.  To avoid
 220  * lock order reversals, in some cases this involves dropping a PCB lock and
 221  * using a reference counter to maintain liveness.
 222  *
 223  * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
 224  * allocated in pru_attach() and freed in pru_detach().  The validity of that
 225  * pointer is an invariant, so no lock is required to dereference the so_pcb
 226  * pointer if a valid socket reference is held by the caller.  In practice,
 227  * this is always true during operations performed on a socket.  Each unpcb
 228  * has a back-pointer to its socket, unp_socket, which will be stable under
 229  * the same circumstances.
 230  *
 231  * This pointer may only be safely dereferenced as long as a valid reference
 232  * to the unpcb is held.  Typically, this reference will be from the socket,
 233  * or from another unpcb when the referring unpcb's lock is held (in order
 234  * that the reference not be invalidated during use).  For example, to follow
 235  * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee
 236  * that detach is not run clearing unp_socket.
 237  *
 238  * Blocking with UNIX domain sockets is a tricky issue: unlike most network
 239  * protocols, bind() is a non-atomic operation, and connect() requires
 240  * potential sleeping in the protocol, due to potentially waiting on local or
 241  * distributed file systems.  We try to separate "lookup" operations, which
 242  * may sleep, and the IPC operations themselves, which typically can occur
 243  * with relative atomicity as locks can be held over the entire operation.
 244  *
 245  * Another tricky issue is simultaneous multi-threaded or multi-process
 246  * access to a single UNIX domain socket.  These are handled by the flags
 247  * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
 248  * binding, both of which involve dropping UNIX domain socket locks in order
 249  * to perform namei() and other file system operations.
 250  */
 251 static struct rwlock    unp_link_rwlock;
 252 static struct mtx       unp_defers_lock;
 253
 254 #define UNP_LINK_LOCK_INIT()            rw_init(&unp_link_rwlock,       \
 255                                             "unp_link_rwlock")
 256
 257 #define UNP_LINK_LOCK_ASSERT()          rw_assert(&unp_link_rwlock,     \
 258                                             RA_LOCKED)
 259 #define UNP_LINK_UNLOCK_ASSERT()        rw_assert(&unp_link_rwlock,     \
 260                                             RA_UNLOCKED)
 261
 262 #define UNP_LINK_RLOCK()                rw_rlock(&unp_link_rwlock)
 263 #define UNP_LINK_RUNLOCK()              rw_runlock(&unp_link_rwlock)
 264 #define UNP_LINK_WLOCK()                rw_wlock(&unp_link_rwlock)
 265 #define UNP_LINK_WUNLOCK()              rw_wunlock(&unp_link_rwlock)
 266 #define UNP_LINK_WLOCK_ASSERT()         rw_assert(&unp_link_rwlock,     \
 267                                             RA_WLOCKED)
 268 #define UNP_LINK_WOWNED()               rw_wowned(&unp_link_rwlock)
 269
 270 #define UNP_DEFERRED_LOCK_INIT()        mtx_init(&unp_defers_lock, \
 271                                             "unp_defer", NULL, MTX_DEF)
 272 #define UNP_DEFERRED_LOCK()             mtx_lock(&unp_defers_lock)
 273 #define UNP_DEFERRED_UNLOCK()           mtx_unlock(&unp_defers_lock)
 274
 275 #define UNP_REF_LIST_LOCK()             UNP_DEFERRED_LOCK();
 276 #define UNP_REF_LIST_UNLOCK()           UNP_DEFERRED_UNLOCK();
 277
 278 #define UNP_PCB_LOCK_INIT(unp)          mtx_init(&(unp)->unp_mtx,       \
 279                                             "unp", "unp",       \
 280                                             MTX_DUPOK|MTX_DEF)
 281 #define UNP_PCB_LOCK_DESTROY(unp)       mtx_destroy(&(unp)->unp_mtx)
 282 #define UNP_PCB_LOCKPTR(unp)            (&(unp)->unp_mtx)
 283 #define UNP_PCB_LOCK(unp)               mtx_lock(&(unp)->unp_mtx)
 284 #define UNP_PCB_TRYLOCK(unp)            mtx_trylock(&(unp)->unp_mtx)
 285 #define UNP_PCB_UNLOCK(unp)             mtx_unlock(&(unp)->unp_mtx)
 286 #define UNP_PCB_OWNED(unp)              mtx_owned(&(unp)->unp_mtx)
 287 #define UNP_PCB_LOCK_ASSERT(unp)        mtx_assert(&(unp)->unp_mtx, MA_OWNED)
 288 #define UNP_PCB_UNLOCK_ASSERT(unp)      mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED)
 289
 290 static int      uipc_connect2(struct socket *, struct socket *);
 291 static int      uipc_ctloutput(struct socket *, struct sockopt *);
 292 static int      unp_connect(struct socket *, struct sockaddr *,
 293                     struct thread *);
 294 static int      unp_connectat(int, struct socket *, struct sockaddr *,
 295                     struct thread *);
 296 static int      unp_connect2(struct socket *so, struct socket *so2, int);
 297 static void     unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
 298 static void     unp_dispose(struct socket *so);
 299 static void     unp_dispose_mbuf(struct mbuf *);
 300 static void     unp_shutdown(struct unpcb *);
 301 static void     unp_drop(struct unpcb *);
 302 static void     unp_gc(__unused void *, int);
 303 static void     unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
 304 static void     unp_discard(struct file *);
 305 static void     unp_freerights(struct filedescent **, int);
 306 static void     unp_init(void);
 307 static int      unp_internalize(struct mbuf **, struct thread *);
 308 static void     unp_internalize_fp(struct file *);
 309 static int      unp_externalize(struct mbuf *, struct mbuf **, int);
 310 static int      unp_externalize_fp(struct file *);
 311 static struct mbuf      *unp_addsockcred(struct thread *, struct mbuf *, int);
 312 static void     unp_process_defers(void * __unused, int);
 313
 314 static void
 315 unp_pcb_hold(struct unpcb *unp)
 316 {
 317         u_int old __unused;
 318
 319         old = refcount_acquire(&unp->unp_refcount);
 320         KASSERT(old > 0, ("%s: unpcb %p has no references", __func__, unp));
 321 }
 322
 323 static __result_use_check bool
 324 unp_pcb_rele(struct unpcb *unp)
 325 {
 326         bool ret;
 327
 328         UNP_PCB_LOCK_ASSERT(unp);
 329
 330         if ((ret = refcount_release(&unp->unp_refcount))) {
 331                 UNP_PCB_UNLOCK(unp);
 332                 UNP_PCB_LOCK_DESTROY(unp);
 333                 uma_zfree(unp_zone, unp);
 334         }
 335         return (ret);
 336 }
 337
 338 static void
 339 unp_pcb_rele_notlast(struct unpcb *unp)
 340 {
 341         bool ret __unused;
 342
 343         ret = refcount_release(&unp->unp_refcount);
 344         KASSERT(!ret, ("%s: unpcb %p has no references", __func__, unp));
 345 }
 346
 347 static void
 348 unp_pcb_lock_pair(struct unpcb *unp, struct unpcb *unp2)
 349 {
 350         UNP_PCB_UNLOCK_ASSERT(unp);
 351         UNP_PCB_UNLOCK_ASSERT(unp2);
 352
 353         if (unp == unp2) {
 354                 UNP_PCB_LOCK(unp);
 355         } else if ((uintptr_t)unp2 > (uintptr_t)unp) {
 356                 UNP_PCB_LOCK(unp);
 357                 UNP_PCB_LOCK(unp2);
 358         } else {
 359                 UNP_PCB_LOCK(unp2);
 360                 UNP_PCB_LOCK(unp);
 361         }
 362 }
 363
 364 static void
 365 unp_pcb_unlock_pair(struct unpcb *unp, struct unpcb *unp2)
 366 {
 367         UNP_PCB_UNLOCK(unp);
 368         if (unp != unp2)
 369                 UNP_PCB_UNLOCK(unp2);
 370 }
 371
 372 /*
 373  * Try to lock the connected peer of an already locked socket.  In some cases
 374  * this requires that we unlock the current socket.  The pairbusy counter is
 375  * used to block concurrent connection attempts while the lock is dropped.  The
 376  * caller must be careful to revalidate PCB state.
 377  */
 378 static struct unpcb *
 379 unp_pcb_lock_peer(struct unpcb *unp)
 380 {
 381         struct unpcb *unp2;
 382
 383         UNP_PCB_LOCK_ASSERT(unp);
 384         unp2 = unp->unp_conn;
 385         if (unp2 == NULL)
 386                 return (NULL);
 387         if (__predict_false(unp == unp2))
 388                 return (unp);
 389
 390         UNP_PCB_UNLOCK_ASSERT(unp2);
 391
 392         if (__predict_true(UNP_PCB_TRYLOCK(unp2)))
 393                 return (unp2);
 394         if ((uintptr_t)unp2 > (uintptr_t)unp) {
 395                 UNP_PCB_LOCK(unp2);
 396                 return (unp2);
 397         }
 398         unp->unp_pairbusy++;
 399         unp_pcb_hold(unp2);
 400         UNP_PCB_UNLOCK(unp);
 401
 402         UNP_PCB_LOCK(unp2);
 403         UNP_PCB_LOCK(unp);
 404         KASSERT(unp->unp_conn == unp2 || unp->unp_conn == NULL,
 405             ("%s: socket %p was reconnected", __func__, unp));
 406         if (--unp->unp_pairbusy == 0 && (unp->unp_flags & UNP_WAITING) != 0) {
 407                 unp->unp_flags &= ~UNP_WAITING;
 408                 wakeup(unp);
 409         }
 410         if (unp_pcb_rele(unp2)) {
 411                 /* unp2 is unlocked. */
 412                 return (NULL);
 413         }
 414         if (unp->unp_conn == NULL) {
 415                 UNP_PCB_UNLOCK(unp2);
 416                 return (NULL);
 417         }
 418         return (unp2);
 419 }
 420
 421 /*
 422  * Definitions of protocols supported in the LOCAL domain.
 423  */
 424 static struct domain localdomain;
 425 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
 426 static struct pr_usrreqs uipc_usrreqs_seqpacket;
 427 static struct protosw localsw[] = {
 428 {
 429         .pr_type =              SOCK_STREAM,
 430         .pr_domain =            &localdomain,
 431         .pr_flags =             PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
 432         .pr_ctloutput =         &uipc_ctloutput,
 433         .pr_usrreqs =           &uipc_usrreqs_stream
 434 },
 435 {
 436         .pr_type =              SOCK_DGRAM,
 437         .pr_domain =            &localdomain,
 438         .pr_flags =             PR_ATOMIC|PR_ADDR|PR_RIGHTS,
 439         .pr_ctloutput =         &uipc_ctloutput,
 440         .pr_usrreqs =           &uipc_usrreqs_dgram
 441 },
 442 {
 443         .pr_type =              SOCK_SEQPACKET,
 444         .pr_domain =            &localdomain,
 445
 446         /*
 447          * XXXRW: For now, PR_ADDR because soreceive will bump into them
 448          * due to our use of sbappendaddr.  A new sbappend variants is needed
 449          * that supports both atomic record writes and control data.
 450          */
 451         .pr_flags =             PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
 452                                     PR_RIGHTS,
 453         .pr_ctloutput =         &uipc_ctloutput,
 454         .pr_usrreqs =           &uipc_usrreqs_seqpacket,
 455 },
 456 };
 457
 458 static struct domain localdomain = {
 459         .dom_family =           AF_LOCAL,
 460         .dom_name =             "local",
 461         .dom_init =             unp_init,
 462         .dom_externalize =      unp_externalize,
 463         .dom_dispose =          unp_dispose,
 464         .dom_protosw =          localsw,
 465         .dom_protoswNPROTOSW =  &localsw[nitems(localsw)]
 466 };
 467 DOMAIN_SET(local);
 468
 469 static void
 470 uipc_abort(struct socket *so)
 471 {
 472         struct unpcb *unp, *unp2;
 473
 474         unp = sotounpcb(so);
 475         KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
 476         UNP_PCB_UNLOCK_ASSERT(unp);
 477
 478         UNP_PCB_LOCK(unp);
 479         unp2 = unp->unp_conn;
 480         if (unp2 != NULL) {
 481                 unp_pcb_hold(unp2);
 482                 UNP_PCB_UNLOCK(unp);
 483                 unp_drop(unp2);
 484         } else
 485                 UNP_PCB_UNLOCK(unp);
 486 }
 487
 488 static int
 489 uipc_accept(struct socket *so, struct sockaddr **nam)
 490 {
 491         struct unpcb *unp, *unp2;
 492         const struct sockaddr *sa;
 493
 494         /*
 495          * Pass back name of connected socket, if it was bound and we are
 496          * still connected (our peer may have closed already!).
 497          */
 498         unp = sotounpcb(so);
 499         KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
 500
 501         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 502         UNP_PCB_LOCK(unp);
 503         unp2 = unp_pcb_lock_peer(unp);
 504         if (unp2 != NULL && unp2->unp_addr != NULL)
 505                 sa = (struct sockaddr *)unp2->unp_addr;
 506         else
 507                 sa = &sun_noname;
 508         bcopy(sa, *nam, sa->sa_len);
 509         if (unp2 != NULL)
 510                 unp_pcb_unlock_pair(unp, unp2);
 511         else
 512                 UNP_PCB_UNLOCK(unp);
 513         return (0);
 514 }
 515
 516 static int
 517 uipc_attach(struct socket *so, int proto, struct thread *td)
 518 {
 519         u_long sendspace, recvspace;
 520         struct unpcb *unp;
 521         int error;
 522         bool locked;
 523
 524         KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
 525         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
 526                 switch (so->so_type) {
 527                 case SOCK_STREAM:
 528                         sendspace = unpst_sendspace;
 529                         recvspace = unpst_recvspace;
 530                         break;
 531
 532                 case SOCK_DGRAM:
 533                         sendspace = unpdg_sendspace;
 534                         recvspace = unpdg_recvspace;
 535                         break;
 536
 537                 case SOCK_SEQPACKET:
 538                         sendspace = unpsp_sendspace;
 539                         recvspace = unpsp_recvspace;
 540                         break;
 541
 542                 default:
 543                         panic("uipc_attach");
 544                 }
 545                 error = soreserve(so, sendspace, recvspace);
 546                 if (error)
 547                         return (error);
 548         }
 549         unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
 550         if (unp == NULL)
 551                 return (ENOBUFS);
 552         LIST_INIT(&unp->unp_refs);
 553         UNP_PCB_LOCK_INIT(unp);
 554         unp->unp_socket = so;
 555         so->so_pcb = unp;
 556         refcount_init(&unp->unp_refcount, 1);
 557
 558         if ((locked = UNP_LINK_WOWNED()) == false)
 559                 UNP_LINK_WLOCK();
 560
 561         unp->unp_gencnt = ++unp_gencnt;
 562         unp->unp_ino = ++unp_ino;
 563         unp_count++;
 564         switch (so->so_type) {
 565         case SOCK_STREAM:
 566                 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
 567                 break;
 568
 569         case SOCK_DGRAM:
 570                 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
 571                 break;
 572
 573         case SOCK_SEQPACKET:
 574                 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
 575                 break;
 576
 577         default:
 578                 panic("uipc_attach");
 579         }
 580
 581         if (locked == false)
 582                 UNP_LINK_WUNLOCK();
 583
 584         return (0);
 585 }
 586
 587 static int
 588 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
 589 {
 590         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
 591         struct vattr vattr;
 592         int error, namelen;
 593         struct nameidata nd;
 594         struct unpcb *unp;
 595         struct vnode *vp;
 596         struct mount *mp;
 597         cap_rights_t rights;
 598         char *buf;
 599
 600         if (nam->sa_family != AF_UNIX)
 601                 return (EAFNOSUPPORT);
 602
 603         unp = sotounpcb(so);
 604         KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
 605
 606         if (soun->sun_len > sizeof(struct sockaddr_un))
 607                 return (EINVAL);
 608         namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
 609         if (namelen <= 0)
 610                 return (EINVAL);
 611
 612         /*
 613          * We don't allow simultaneous bind() calls on a single UNIX domain
 614          * socket, so flag in-progress operations, and return an error if an
 615          * operation is already in progress.
 616          *
 617          * Historically, we have not allowed a socket to be rebound, so this
 618          * also returns an error.  Not allowing re-binding simplifies the
 619          * implementation and avoids a great many possible failure modes.
 620          */
 621         UNP_PCB_LOCK(unp);
 622         if (unp->unp_vnode != NULL) {
 623                 UNP_PCB_UNLOCK(unp);
 624                 return (EINVAL);
 625         }
 626         if (unp->unp_flags & UNP_BINDING) {
 627                 UNP_PCB_UNLOCK(unp);
 628                 return (EALREADY);
 629         }
 630         unp->unp_flags |= UNP_BINDING;
 631         UNP_PCB_UNLOCK(unp);
 632
 633         buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
 634         bcopy(soun->sun_path, buf, namelen);
 635         buf[namelen] = 0;
 636
 637 restart:
 638         NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE,
 639             UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_BINDAT),
 640             td);
 641 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
 642         error = namei(&nd);
 643         if (error)
 644                 goto error;
 645         vp = nd.ni_vp;
 646         if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
 647                 NDFREE(&nd, NDF_ONLY_PNBUF);
 648                 if (nd.ni_dvp == vp)
 649                         vrele(nd.ni_dvp);
 650                 else
 651                         vput(nd.ni_dvp);
 652                 if (vp != NULL) {
 653                         vrele(vp);
 654                         error = EADDRINUSE;
 655                         goto error;
 656                 }
 657                 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
 658                 if (error)
 659                         goto error;
 660                 goto restart;
 661         }
 662         VATTR_NULL(&vattr);
 663         vattr.va_type = VSOCK;
 664         vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_pd->pd_cmask);
 665 #ifdef MAC
 666         error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
 667             &vattr);
 668 #endif
 669         if (error == 0)
 670                 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
 671         NDFREE(&nd, NDF_ONLY_PNBUF);
 672         vput(nd.ni_dvp);
 673         if (error) {
 674                 vn_finished_write(mp);
 675                 if (error == ERELOOKUP)
 676                         goto restart;
 677                 goto error;
 678         }
 679         vp = nd.ni_vp;
 680         ASSERT_VOP_ELOCKED(vp, "uipc_bind");
 681         soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
 682
 683         UNP_PCB_LOCK(unp);
 684         VOP_UNP_BIND(vp, unp);
 685         unp->unp_vnode = vp;
 686         unp->unp_addr = soun;
 687         unp->unp_flags &= ~UNP_BINDING;
 688         UNP_PCB_UNLOCK(unp);
 689         VOP_UNLOCK(vp);
 690         vn_finished_write(mp);
 691         free(buf, M_TEMP);
 692         return (0);
 693
 694 error:
 695         UNP_PCB_LOCK(unp);
 696         unp->unp_flags &= ~UNP_BINDING;
 697         UNP_PCB_UNLOCK(unp);
 698         free(buf, M_TEMP);
 699         return (error);
 700 }
 701
 702 static int
 703 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 704 {
 705
 706         return (uipc_bindat(AT_FDCWD, so, nam, td));
 707 }
 708
 709 static int
 710 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 711 {
 712         int error;
 713
 714         KASSERT(td == curthread, ("uipc_connect: td != curthread"));
 715         error = unp_connect(so, nam, td);
 716         return (error);
 717 }
 718
 719 static int
 720 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
 721     struct thread *td)
 722 {
 723         int error;
 724
 725         KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
 726         error = unp_connectat(fd, so, nam, td);
 727         return (error);
 728 }
 729
 730 static void
 731 uipc_close(struct socket *so)
 732 {
 733         struct unpcb *unp, *unp2;
 734         struct vnode *vp = NULL;
 735         struct mtx *vplock;
 736
 737         unp = sotounpcb(so);
 738         KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
 739
 740         vplock = NULL;
 741         if ((vp = unp->unp_vnode) != NULL) {
 742                 vplock = mtx_pool_find(mtxpool_sleep, vp);
 743                 mtx_lock(vplock);
 744         }
 745         UNP_PCB_LOCK(unp);
 746         if (vp && unp->unp_vnode == NULL) {
 747                 mtx_unlock(vplock);
 748                 vp = NULL;
 749         }
 750         if (vp != NULL) {
 751                 VOP_UNP_DETACH(vp);
 752                 unp->unp_vnode = NULL;
 753         }
 754         if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
 755                 unp_disconnect(unp, unp2);
 756         else
 757                 UNP_PCB_UNLOCK(unp);
 758         if (vp) {
 759                 mtx_unlock(vplock);
 760                 vrele(vp);
 761         }
 762 }
 763
 764 static int
 765 uipc_connect2(struct socket *so1, struct socket *so2)
 766 {
 767         struct unpcb *unp, *unp2;
 768         int error;
 769
 770         unp = so1->so_pcb;
 771         KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
 772         unp2 = so2->so_pcb;
 773         KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
 774         unp_pcb_lock_pair(unp, unp2);
 775         error = unp_connect2(so1, so2, PRU_CONNECT2);
 776         unp_pcb_unlock_pair(unp, unp2);
 777         return (error);
 778 }
 779
 780 static void
 781 uipc_detach(struct socket *so)
 782 {
 783         struct unpcb *unp, *unp2;
 784         struct mtx *vplock;
 785         struct vnode *vp;
 786         int local_unp_rights;
 787
 788         unp = sotounpcb(so);
 789         KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
 790
 791         vp = NULL;
 792         vplock = NULL;
 793
 794         SOCK_LOCK(so);
 795         if (!SOLISTENING(so)) {
 796                 /*
 797                  * Once the socket is removed from the global lists,
 798                  * uipc_ready() will not be able to locate its socket buffer, so
 799                  * clear the buffer now.  At this point internalized rights have
 800                  * already been disposed of.
 801                  */
 802                 sbrelease(&so->so_rcv, so);
 803         }
 804         SOCK_UNLOCK(so);
 805
 806         UNP_LINK_WLOCK();
 807         LIST_REMOVE(unp, unp_link);
 808         if (unp->unp_gcflag & UNPGC_DEAD)
 809                 LIST_REMOVE(unp, unp_dead);
 810         unp->unp_gencnt = ++unp_gencnt;
 811         --unp_count;
 812         UNP_LINK_WUNLOCK();
 813
 814         UNP_PCB_UNLOCK_ASSERT(unp);
 815  restart:
 816         if ((vp = unp->unp_vnode) != NULL) {
 817                 vplock = mtx_pool_find(mtxpool_sleep, vp);
 818                 mtx_lock(vplock);
 819         }
 820         UNP_PCB_LOCK(unp);
 821         if (unp->unp_vnode != vp && unp->unp_vnode != NULL) {
 822                 if (vplock)
 823                         mtx_unlock(vplock);
 824                 UNP_PCB_UNLOCK(unp);
 825                 goto restart;
 826         }
 827         if ((vp = unp->unp_vnode) != NULL) {
 828                 VOP_UNP_DETACH(vp);
 829                 unp->unp_vnode = NULL;
 830         }
 831         if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
 832                 unp_disconnect(unp, unp2);
 833         else
 834                 UNP_PCB_UNLOCK(unp);
 835
 836         UNP_REF_LIST_LOCK();
 837         while (!LIST_EMPTY(&unp->unp_refs)) {
 838                 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
 839
 840                 unp_pcb_hold(ref);
 841                 UNP_REF_LIST_UNLOCK();
 842
 843                 MPASS(ref != unp);
 844                 UNP_PCB_UNLOCK_ASSERT(ref);
 845                 unp_drop(ref);
 846                 UNP_REF_LIST_LOCK();
 847         }
 848         UNP_REF_LIST_UNLOCK();
 849
 850         UNP_PCB_LOCK(unp);
 851         local_unp_rights = unp_rights;
 852         unp->unp_socket->so_pcb = NULL;
 853         unp->unp_socket = NULL;
 854         free(unp->unp_addr, M_SONAME);
 855         unp->unp_addr = NULL;
 856         if (!unp_pcb_rele(unp))
 857                 UNP_PCB_UNLOCK(unp);
 858         if (vp) {
 859                 mtx_unlock(vplock);
 860                 vrele(vp);
 861         }
 862         if (local_unp_rights)
 863                 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
 864 }
 865
 866 static int
 867 uipc_disconnect(struct socket *so)
 868 {
 869         struct unpcb *unp, *unp2;
 870
 871         unp = sotounpcb(so);
 872         KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
 873
 874         UNP_PCB_LOCK(unp);
 875         if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
 876                 unp_disconnect(unp, unp2);
 877         else
 878                 UNP_PCB_UNLOCK(unp);
 879         return (0);
 880 }
 881
 882 static int
 883 uipc_listen(struct socket *so, int backlog, struct thread *td)
 884 {
 885         struct unpcb *unp;
 886         int error;
 887
 888         if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
 889                 return (EOPNOTSUPP);
 890
 891         unp = sotounpcb(so);
 892         KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
 893
 894         UNP_PCB_LOCK(unp);
 895         if (unp->unp_vnode == NULL) {
 896                 /* Already connected or not bound to an address. */
 897                 error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ;
 898                 UNP_PCB_UNLOCK(unp);
 899                 return (error);
 900         }
 901
 902         SOCK_LOCK(so);
 903         error = solisten_proto_check(so);
 904         if (error == 0) {
 905                 cru2xt(td, &unp->unp_peercred);
 906                 solisten_proto(so, backlog);
 907         }
 908         SOCK_UNLOCK(so);
 909         UNP_PCB_UNLOCK(unp);
 910         return (error);
 911 }
 912
 913 static int
 914 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
 915 {
 916         struct unpcb *unp, *unp2;
 917         const struct sockaddr *sa;
 918
 919         unp = sotounpcb(so);
 920         KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
 921
 922         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 923         UNP_LINK_RLOCK();
 924         /*
 925          * XXX: It seems that this test always fails even when connection is
 926          * established.  So, this else clause is added as workaround to
 927          * return PF_LOCAL sockaddr.
 928          */
 929         unp2 = unp->unp_conn;
 930         if (unp2 != NULL) {
 931                 UNP_PCB_LOCK(unp2);
 932                 if (unp2->unp_addr != NULL)
 933                         sa = (struct sockaddr *) unp2->unp_addr;
 934                 else
 935                         sa = &sun_noname;
 936                 bcopy(sa, *nam, sa->sa_len);
 937                 UNP_PCB_UNLOCK(unp2);
 938         } else {
 939                 sa = &sun_noname;
 940                 bcopy(sa, *nam, sa->sa_len);
 941         }
 942         UNP_LINK_RUNLOCK();
 943         return (0);
 944 }
 945
 946 static int
 947 uipc_rcvd(struct socket *so, int flags)
 948 {
 949         struct unpcb *unp, *unp2;
 950         struct socket *so2;
 951         u_int mbcnt, sbcc;
 952
 953         unp = sotounpcb(so);
 954         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
 955         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
 956             ("%s: socktype %d", __func__, so->so_type));
 957
 958         /*
 959          * Adjust backpressure on sender and wakeup any waiting to write.
 960          *
 961          * The unp lock is acquired to maintain the validity of the unp_conn
 962          * pointer; no lock on unp2 is required as unp2->unp_socket will be
 963          * static as long as we don't permit unp2 to disconnect from unp,
 964          * which is prevented by the lock on unp.  We cache values from
 965          * so_rcv to avoid holding the so_rcv lock over the entire
 966          * transaction on the remote so_snd.
 967          */
 968         SOCKBUF_LOCK(&so->so_rcv);
 969         mbcnt = so->so_rcv.sb_mbcnt;
 970         sbcc = sbavail(&so->so_rcv);
 971         SOCKBUF_UNLOCK(&so->so_rcv);
 972         /*
 973          * There is a benign race condition at this point.  If we're planning to
 974          * clear SB_STOP, but uipc_send is called on the connected socket at
 975          * this instant, it might add data to the sockbuf and set SB_STOP.  Then
 976          * we would erroneously clear SB_STOP below, even though the sockbuf is
 977          * full.  The race is benign because the only ill effect is to allow the
 978          * sockbuf to exceed its size limit, and the size limits are not
 979          * strictly guaranteed anyway.
 980          */
 981         UNP_PCB_LOCK(unp);
 982         unp2 = unp->unp_conn;
 983         if (unp2 == NULL) {
 984                 UNP_PCB_UNLOCK(unp);
 985                 return (0);
 986         }
 987         so2 = unp2->unp_socket;
 988         SOCKBUF_LOCK(&so2->so_snd);
 989         if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
 990                 so2->so_snd.sb_flags &= ~SB_STOP;
 991         sowwakeup_locked(so2);
 992         UNP_PCB_UNLOCK(unp);
 993         return (0);
 994 }
 995
 996 static int
 997 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
 998     struct mbuf *control, struct thread *td)
 999 {
1000         struct unpcb *unp, *unp2;
1001         struct socket *so2;
1002         u_int mbcnt, sbcc;
1003         int freed, error;
1004
1005         unp = sotounpcb(so);
1006         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
1007         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
1008             so->so_type == SOCK_SEQPACKET,
1009             ("%s: socktype %d", __func__, so->so_type));
1010
1011         freed = error = 0;
1012         if (flags & PRUS_OOB) {
1013                 error = EOPNOTSUPP;
1014                 goto release;
1015         }
1016         if (control != NULL && (error = unp_internalize(&control, td)))
1017                 goto release;
1018
1019         unp2 = NULL;
1020         switch (so->so_type) {
1021         case SOCK_DGRAM:
1022         {
1023                 const struct sockaddr *from;
1024
1025                 if (nam != NULL) {
1026                         error = unp_connect(so, nam, td);
1027                         if (error != 0)
1028                                 break;
1029                 }
1030                 UNP_PCB_LOCK(unp);
1031
1032                 /*
1033                  * Because connect() and send() are non-atomic in a sendto()
1034                  * with a target address, it's possible that the socket will
1035                  * have disconnected before the send() can run.  In that case
1036                  * return the slightly counter-intuitive but otherwise
1037                  * correct error that the socket is not connected.
1038                  */
1039                 unp2 = unp_pcb_lock_peer(unp);
1040                 if (unp2 == NULL) {
1041                         UNP_PCB_UNLOCK(unp);
1042                         error = ENOTCONN;
1043                         break;
1044                 }
1045
1046                 if (unp2->unp_flags & UNP_WANTCRED_MASK)
1047                         control = unp_addsockcred(td, control,
1048                             unp2->unp_flags);
1049                 if (unp->unp_addr != NULL)
1050                         from = (struct sockaddr *)unp->unp_addr;
1051                 else
1052                         from = &sun_noname;
1053                 so2 = unp2->unp_socket;
1054                 SOCKBUF_LOCK(&so2->so_rcv);
1055                 if (sbappendaddr_locked(&so2->so_rcv, from, m,
1056                     control)) {
1057                         sorwakeup_locked(so2);
1058                         m = NULL;
1059                         control = NULL;
1060                 } else {
1061                         SOCKBUF_UNLOCK(&so2->so_rcv);
1062                         error = ENOBUFS;
1063                 }
1064                 if (nam != NULL)
1065                         unp_disconnect(unp, unp2);
1066                 else
1067                         unp_pcb_unlock_pair(unp, unp2);
1068                 break;
1069         }
1070
1071         case SOCK_SEQPACKET:
1072         case SOCK_STREAM:
1073                 if ((so->so_state & SS_ISCONNECTED) == 0) {
1074                         if (nam != NULL) {
1075                                 error = unp_connect(so, nam, td);
1076                                 if (error != 0)
1077                                         break;
1078                         } else {
1079                                 error = ENOTCONN;
1080                                 break;
1081                         }
1082                 }
1083
1084                 UNP_PCB_LOCK(unp);
1085                 if ((unp2 = unp_pcb_lock_peer(unp)) == NULL) {
1086                         UNP_PCB_UNLOCK(unp);
1087                         error = ENOTCONN;
1088                         break;
1089                 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1090                         unp_pcb_unlock_pair(unp, unp2);
1091                         error = EPIPE;
1092                         break;
1093                 }
1094                 UNP_PCB_UNLOCK(unp);
1095                 if ((so2 = unp2->unp_socket) == NULL) {
1096                         UNP_PCB_UNLOCK(unp2);
1097                         error = ENOTCONN;
1098                         break;
1099                 }
1100                 SOCKBUF_LOCK(&so2->so_rcv);
1101                 if (unp2->unp_flags & UNP_WANTCRED_MASK) {
1102                         /*
1103                          * Credentials are passed only once on SOCK_STREAM and
1104                          * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or
1105                          * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS).
1106                          */
1107                         control = unp_addsockcred(td, control, unp2->unp_flags);
1108                         unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT;
1109                 }
1110
1111                 /*
1112                  * Send to paired receive port and wake up readers.  Don't
1113                  * check for space available in the receive buffer if we're
1114                  * attaching ancillary data; Unix domain sockets only check
1115                  * for space in the sending sockbuf, and that check is
1116                  * performed one level up the stack.  At that level we cannot
1117                  * precisely account for the amount of buffer space used
1118                  * (e.g., because control messages are not yet internalized).
1119                  */
1120                 switch (so->so_type) {
1121                 case SOCK_STREAM:
1122                         if (control != NULL) {
1123                                 sbappendcontrol_locked(&so2->so_rcv, m,
1124                                     control, flags);
1125                                 control = NULL;
1126                         } else
1127                                 sbappend_locked(&so2->so_rcv, m, flags);
1128                         break;
1129
1130                 case SOCK_SEQPACKET:
1131                         if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1132                             &sun_noname, m, control))
1133                                 control = NULL;
1134                         break;
1135                 }
1136
1137                 mbcnt = so2->so_rcv.sb_mbcnt;
1138                 sbcc = sbavail(&so2->so_rcv);
1139                 if (sbcc)
1140                         sorwakeup_locked(so2);
1141                 else
1142                         SOCKBUF_UNLOCK(&so2->so_rcv);
1143
1144                 /*
1145                  * The PCB lock on unp2 protects the SB_STOP flag.  Without it,
1146                  * it would be possible for uipc_rcvd to be called at this
1147                  * point, drain the receiving sockbuf, clear SB_STOP, and then
1148                  * we would set SB_STOP below.  That could lead to an empty
1149                  * sockbuf having SB_STOP set
1150                  */
1151                 SOCKBUF_LOCK(&so->so_snd);
1152                 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1153                         so->so_snd.sb_flags |= SB_STOP;
1154                 SOCKBUF_UNLOCK(&so->so_snd);
1155                 UNP_PCB_UNLOCK(unp2);
1156                 m = NULL;
1157                 break;
1158         }
1159
1160         /*
1161          * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1162          */
1163         if (flags & PRUS_EOF) {
1164                 UNP_PCB_LOCK(unp);
1165                 socantsendmore(so);
1166                 unp_shutdown(unp);
1167                 UNP_PCB_UNLOCK(unp);
1168         }
1169         if (control != NULL && error != 0)
1170                 unp_dispose_mbuf(control);
1171
1172 release:
1173         if (control != NULL)
1174                 m_freem(control);
1175         /*
1176          * In case of PRUS_NOTREADY, uipc_ready() is responsible
1177          * for freeing memory.
1178          */
1179         if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1180                 m_freem(m);
1181         return (error);
1182 }
1183
1184 static bool
1185 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp)
1186 {
1187         struct mbuf *mb, *n;
1188         struct sockbuf *sb;
1189
1190         SOCK_LOCK(so);
1191         if (SOLISTENING(so)) {
1192                 SOCK_UNLOCK(so);
1193                 return (false);
1194         }
1195         mb = NULL;
1196         sb = &so->so_rcv;
1197         SOCKBUF_LOCK(sb);
1198         if (sb->sb_fnrdy != NULL) {
1199                 for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) {
1200                         if (mb == m) {
1201                                 *errorp = sbready(sb, m, count);
1202                                 break;
1203                         }
1204                         mb = mb->m_next;
1205                         if (mb == NULL) {
1206                                 mb = n;
1207                                 if (mb != NULL)
1208                                         n = mb->m_nextpkt;
1209                         }
1210                 }
1211         }
1212         SOCKBUF_UNLOCK(sb);
1213         SOCK_UNLOCK(so);
1214         return (mb != NULL);
1215 }
1216
1217 static int
1218 uipc_ready(struct socket *so, struct mbuf *m, int count)
1219 {
1220         struct unpcb *unp, *unp2;
1221         struct socket *so2;
1222         int error, i;
1223
1224         unp = sotounpcb(so);
1225
1226         KASSERT(so->so_type == SOCK_STREAM,
1227             ("%s: unexpected socket type for %p", __func__, so));
1228
1229         UNP_PCB_LOCK(unp);
1230         if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) {
1231                 UNP_PCB_UNLOCK(unp);
1232                 so2 = unp2->unp_socket;
1233                 SOCKBUF_LOCK(&so2->so_rcv);
1234                 if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1235                         sorwakeup_locked(so2);
1236                 else
1237                         SOCKBUF_UNLOCK(&so2->so_rcv);
1238                 UNP_PCB_UNLOCK(unp2);
1239                 return (error);
1240         }
1241         UNP_PCB_UNLOCK(unp);
1242
1243         /*
1244          * The receiving socket has been disconnected, but may still be valid.
1245          * In this case, the now-ready mbufs are still present in its socket
1246          * buffer, so perform an exhaustive search before giving up and freeing
1247          * the mbufs.
1248          */
1249         UNP_LINK_RLOCK();
1250         LIST_FOREACH(unp, &unp_shead, unp_link) {
1251                 if (uipc_ready_scan(unp->unp_socket, m, count, &error))
1252                         break;
1253         }
1254         UNP_LINK_RUNLOCK();
1255
1256         if (unp == NULL) {
1257                 for (i = 0; i < count; i++)
1258                         m = m_free(m);
1259                 error = ECONNRESET;
1260         }
1261         return (error);
1262 }
1263
1264 static int
1265 uipc_sense(struct socket *so, struct stat *sb)
1266 {
1267         struct unpcb *unp;
1268
1269         unp = sotounpcb(so);
1270         KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1271
1272         sb->st_blksize = so->so_snd.sb_hiwat;
1273         sb->st_dev = NODEV;
1274         sb->st_ino = unp->unp_ino;
1275         return (0);
1276 }
1277
1278 static int
1279 uipc_shutdown(struct socket *so)
1280 {
1281         struct unpcb *unp;
1282
1283         unp = sotounpcb(so);
1284         KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1285
1286         UNP_PCB_LOCK(unp);
1287         socantsendmore(so);
1288         unp_shutdown(unp);
1289         UNP_PCB_UNLOCK(unp);
1290         return (0);
1291 }
1292
1293 static int
1294 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1295 {
1296         struct unpcb *unp;
1297         const struct sockaddr *sa;
1298
1299         unp = sotounpcb(so);
1300         KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1301
1302         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1303         UNP_PCB_LOCK(unp);
1304         if (unp->unp_addr != NULL)
1305                 sa = (struct sockaddr *) unp->unp_addr;
1306         else
1307                 sa = &sun_noname;
1308         bcopy(sa, *nam, sa->sa_len);
1309         UNP_PCB_UNLOCK(unp);
1310         return (0);
1311 }
1312
1313 static struct pr_usrreqs uipc_usrreqs_dgram = {
1314         .pru_abort =            uipc_abort,
1315         .pru_accept =           uipc_accept,
1316         .pru_attach =           uipc_attach,
1317         .pru_bind =             uipc_bind,
1318         .pru_bindat =           uipc_bindat,
1319         .pru_connect =          uipc_connect,
1320         .pru_connectat =        uipc_connectat,
1321         .pru_connect2 =         uipc_connect2,
1322         .pru_detach =           uipc_detach,
1323         .pru_disconnect =       uipc_disconnect,
1324         .pru_listen =           uipc_listen,
1325         .pru_peeraddr =         uipc_peeraddr,
1326         .pru_rcvd =             uipc_rcvd,
1327         .pru_send =             uipc_send,
1328         .pru_sense =            uipc_sense,
1329         .pru_shutdown =         uipc_shutdown,
1330         .pru_sockaddr =         uipc_sockaddr,
1331         .pru_soreceive =        soreceive_dgram,
1332         .pru_close =            uipc_close,
1333 };
1334
1335 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1336         .pru_abort =            uipc_abort,
1337         .pru_accept =           uipc_accept,
1338         .pru_attach =           uipc_attach,
1339         .pru_bind =             uipc_bind,
1340         .pru_bindat =           uipc_bindat,
1341         .pru_connect =          uipc_connect,
1342         .pru_connectat =        uipc_connectat,
1343         .pru_connect2 =         uipc_connect2,
1344         .pru_detach =           uipc_detach,
1345         .pru_disconnect =       uipc_disconnect,
1346         .pru_listen =           uipc_listen,
1347         .pru_peeraddr =         uipc_peeraddr,
1348         .pru_rcvd =             uipc_rcvd,
1349         .pru_send =             uipc_send,
1350         .pru_sense =            uipc_sense,
1351         .pru_shutdown =         uipc_shutdown,
1352         .pru_sockaddr =         uipc_sockaddr,
1353         .pru_soreceive =        soreceive_generic,      /* XXX: or...? */
1354         .pru_close =            uipc_close,
1355 };
1356
1357 static struct pr_usrreqs uipc_usrreqs_stream = {
1358         .pru_abort =            uipc_abort,
1359         .pru_accept =           uipc_accept,
1360         .pru_attach =           uipc_attach,
1361         .pru_bind =             uipc_bind,
1362         .pru_bindat =           uipc_bindat,
1363         .pru_connect =          uipc_connect,
1364         .pru_connectat =        uipc_connectat,
1365         .pru_connect2 =         uipc_connect2,
1366         .pru_detach =           uipc_detach,
1367         .pru_disconnect =       uipc_disconnect,
1368         .pru_listen =           uipc_listen,
1369         .pru_peeraddr =         uipc_peeraddr,
1370         .pru_rcvd =             uipc_rcvd,
1371         .pru_send =             uipc_send,
1372         .pru_ready =            uipc_ready,
1373         .pru_sense =            uipc_sense,
1374         .pru_shutdown =         uipc_shutdown,
1375         .pru_sockaddr =         uipc_sockaddr,
1376         .pru_soreceive =        soreceive_generic,
1377         .pru_close =            uipc_close,
1378 };
1379
1380 static int
1381 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1382 {
1383         struct unpcb *unp;
1384         struct xucred xu;
1385         int error, optval;
1386
1387         if (sopt->sopt_level != SOL_LOCAL)
1388                 return (EINVAL);
1389
1390         unp = sotounpcb(so);
1391         KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1392         error = 0;
1393         switch (sopt->sopt_dir) {
1394         case SOPT_GET:
1395                 switch (sopt->sopt_name) {
1396                 case LOCAL_PEERCRED:
1397                         UNP_PCB_LOCK(unp);
1398                         if (unp->unp_flags & UNP_HAVEPC)
1399                                 xu = unp->unp_peercred;
1400                         else {
1401                                 if (so->so_type == SOCK_STREAM)
1402                                         error = ENOTCONN;
1403                                 else
1404                                         error = EINVAL;
1405                         }
1406                         UNP_PCB_UNLOCK(unp);
1407                         if (error == 0)
1408                                 error = sooptcopyout(sopt, &xu, sizeof(xu));
1409                         break;
1410
1411                 case LOCAL_CREDS:
1412                         /* Unlocked read. */
1413                         optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0;
1414                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1415                         break;
1416
1417                 case LOCAL_CREDS_PERSISTENT:
1418                         /* Unlocked read. */
1419                         optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0;
1420                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1421                         break;
1422
1423                 case LOCAL_CONNWAIT:
1424                         /* Unlocked read. */
1425                         optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1426                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1427                         break;
1428
1429                 default:
1430                         error = EOPNOTSUPP;
1431                         break;
1432                 }
1433                 break;
1434
1435         case SOPT_SET:
1436                 switch (sopt->sopt_name) {
1437                 case LOCAL_CREDS:
1438                 case LOCAL_CREDS_PERSISTENT:
1439                 case LOCAL_CONNWAIT:
1440                         error = sooptcopyin(sopt, &optval, sizeof(optval),
1441                                             sizeof(optval));
1442                         if (error)
1443                                 break;
1444
1445 #define OPTSET(bit, exclusive) do {                                     \
1446         UNP_PCB_LOCK(unp);                                              \
1447         if (optval) {                                                   \
1448                 if ((unp->unp_flags & (exclusive)) != 0) {              \
1449                         UNP_PCB_UNLOCK(unp);                            \
1450                         error = EINVAL;                                 \
1451                         break;                                          \
1452                 }                                                       \
1453                 unp->unp_flags |= (bit);                                \
1454         } else                                                          \
1455                 unp->unp_flags &= ~(bit);                               \
1456         UNP_PCB_UNLOCK(unp);                                            \
1457 } while (0)
1458
1459                         switch (sopt->sopt_name) {
1460                         case LOCAL_CREDS:
1461                                 OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS);
1462                                 break;
1463
1464                         case LOCAL_CREDS_PERSISTENT:
1465                                 OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT);
1466                                 break;
1467
1468                         case LOCAL_CONNWAIT:
1469                                 OPTSET(UNP_CONNWAIT, 0);
1470                                 break;
1471
1472                         default:
1473                                 break;
1474                         }
1475                         break;
1476 #undef  OPTSET
1477                 default:
1478                         error = ENOPROTOOPT;
1479                         break;
1480                 }
1481                 break;
1482
1483         default:
1484                 error = EOPNOTSUPP;
1485                 break;
1486         }
1487         return (error);
1488 }
1489
1490 static int
1491 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1492 {
1493
1494         return (unp_connectat(AT_FDCWD, so, nam, td));
1495 }
1496
1497 static int
1498 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1499     struct thread *td)
1500 {
1501         struct mtx *vplock;
1502         struct sockaddr_un *soun;
1503         struct vnode *vp;
1504         struct socket *so2;
1505         struct unpcb *unp, *unp2, *unp3;
1506         struct nameidata nd;
1507         char buf[SOCK_MAXADDRLEN];
1508         struct sockaddr *sa;
1509         cap_rights_t rights;
1510         int error, len;
1511         bool connreq;
1512
1513         if (nam->sa_family != AF_UNIX)
1514                 return (EAFNOSUPPORT);
1515         if (nam->sa_len > sizeof(struct sockaddr_un))
1516                 return (EINVAL);
1517         len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1518         if (len <= 0)
1519                 return (EINVAL);
1520         soun = (struct sockaddr_un *)nam;
1521         bcopy(soun->sun_path, buf, len);
1522         buf[len] = 0;
1523
1524         unp = sotounpcb(so);
1525         UNP_PCB_LOCK(unp);
1526         for (;;) {
1527                 /*
1528                  * Wait for connection state to stabilize.  If a connection
1529                  * already exists, give up.  For datagram sockets, which permit
1530                  * multiple consecutive connect(2) calls, upper layers are
1531                  * responsible for disconnecting in advance of a subsequent
1532                  * connect(2), but this is not synchronized with PCB connection
1533                  * state.
1534                  *
1535                  * Also make sure that no threads are currently attempting to
1536                  * lock the peer socket, to ensure that unp_conn cannot
1537                  * transition between two valid sockets while locks are dropped.
1538                  */
1539                 if (unp->unp_conn != NULL) {
1540                         UNP_PCB_UNLOCK(unp);
1541                         return (EISCONN);
1542                 }
1543                 if ((unp->unp_flags & UNP_CONNECTING) != 0) {
1544                         UNP_PCB_UNLOCK(unp);
1545                         return (EALREADY);
1546                 }
1547                 if (unp->unp_pairbusy > 0) {
1548                         unp->unp_flags |= UNP_WAITING;
1549                         mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0);
1550                         continue;
1551                 }
1552                 break;
1553         }
1554         unp->unp_flags |= UNP_CONNECTING;
1555         UNP_PCB_UNLOCK(unp);
1556
1557         connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0;
1558         if (connreq)
1559                 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1560         else
1561                 sa = NULL;
1562         NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1563             UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT),
1564             td);
1565         error = namei(&nd);
1566         if (error)
1567                 vp = NULL;
1568         else
1569                 vp = nd.ni_vp;
1570         ASSERT_VOP_LOCKED(vp, "unp_connect");
1571         NDFREE(&nd, NDF_ONLY_PNBUF);
1572         if (error)
1573                 goto bad;
1574
1575         if (vp->v_type != VSOCK) {
1576                 error = ENOTSOCK;
1577                 goto bad;
1578         }
1579 #ifdef MAC
1580         error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1581         if (error)
1582                 goto bad;
1583 #endif
1584         error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1585         if (error)
1586                 goto bad;
1587
1588         unp = sotounpcb(so);
1589         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1590
1591         vplock = mtx_pool_find(mtxpool_sleep, vp);
1592         mtx_lock(vplock);
1593         VOP_UNP_CONNECT(vp, &unp2);
1594         if (unp2 == NULL) {
1595                 error = ECONNREFUSED;
1596                 goto bad2;
1597         }
1598         so2 = unp2->unp_socket;
1599         if (so->so_type != so2->so_type) {
1600                 error = EPROTOTYPE;
1601                 goto bad2;
1602         }
1603         if (connreq) {
1604                 if (so2->so_options & SO_ACCEPTCONN) {
1605                         CURVNET_SET(so2->so_vnet);
1606                         so2 = sonewconn(so2, 0);
1607                         CURVNET_RESTORE();
1608                 } else
1609                         so2 = NULL;
1610                 if (so2 == NULL) {
1611                         error = ECONNREFUSED;
1612                         goto bad2;
1613                 }
1614                 unp3 = sotounpcb(so2);
1615                 unp_pcb_lock_pair(unp2, unp3);
1616                 if (unp2->unp_addr != NULL) {
1617                         bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1618                         unp3->unp_addr = (struct sockaddr_un *) sa;
1619                         sa = NULL;
1620                 }
1621
1622                 unp_copy_peercred(td, unp3, unp, unp2);
1623
1624                 UNP_PCB_UNLOCK(unp2);
1625                 unp2 = unp3;
1626
1627                 /*
1628                  * It is safe to block on the PCB lock here since unp2 is
1629                  * nascent and cannot be connected to any other sockets.
1630                  */
1631                 UNP_PCB_LOCK(unp);
1632 #ifdef MAC
1633                 mac_socketpeer_set_from_socket(so, so2);
1634                 mac_socketpeer_set_from_socket(so2, so);
1635 #endif
1636         } else {
1637                 unp_pcb_lock_pair(unp, unp2);
1638         }
1639         KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 &&
1640             sotounpcb(so2) == unp2,
1641             ("%s: unp2 %p so2 %p", __func__, unp2, so2));
1642         error = unp_connect2(so, so2, PRU_CONNECT);
1643         unp_pcb_unlock_pair(unp, unp2);
1644 bad2:
1645         mtx_unlock(vplock);
1646 bad:
1647         if (vp != NULL) {
1648                 vput(vp);
1649         }
1650         free(sa, M_SONAME);
1651         UNP_PCB_LOCK(unp);
1652         KASSERT((unp->unp_flags & UNP_CONNECTING) != 0,
1653             ("%s: unp %p has UNP_CONNECTING clear", __func__, unp));
1654         unp->unp_flags &= ~UNP_CONNECTING;
1655         UNP_PCB_UNLOCK(unp);
1656         return (error);
1657 }
1658
1659 /*
1660  * Set socket peer credentials at connection time.
1661  *
1662  * The client's PCB credentials are copied from its process structure.  The
1663  * server's PCB credentials are copied from the socket on which it called
1664  * listen(2).  uipc_listen cached that process's credentials at the time.
1665  */
1666 void
1667 unp_copy_peercred(struct thread *td, struct unpcb *client_unp,
1668     struct unpcb *server_unp, struct unpcb *listen_unp)
1669 {
1670         cru2xt(td, &client_unp->unp_peercred);
1671         client_unp->unp_flags |= UNP_HAVEPC;
1672
1673         memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred,
1674             sizeof(server_unp->unp_peercred));
1675         server_unp->unp_flags |= UNP_HAVEPC;
1676         client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK);
1677 }
1678
1679 static int
1680 unp_connect2(struct socket *so, struct socket *so2, int req)
1681 {
1682         struct unpcb *unp;
1683         struct unpcb *unp2;
1684
1685         unp = sotounpcb(so);
1686         KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1687         unp2 = sotounpcb(so2);
1688         KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1689
1690         UNP_PCB_LOCK_ASSERT(unp);
1691         UNP_PCB_LOCK_ASSERT(unp2);
1692         KASSERT(unp->unp_conn == NULL,
1693             ("%s: socket %p is already connected", __func__, unp));
1694
1695         if (so2->so_type != so->so_type)
1696                 return (EPROTOTYPE);
1697         unp->unp_conn = unp2;
1698         unp_pcb_hold(unp2);
1699         unp_pcb_hold(unp);
1700         switch (so->so_type) {
1701         case SOCK_DGRAM:
1702                 UNP_REF_LIST_LOCK();
1703                 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1704                 UNP_REF_LIST_UNLOCK();
1705                 soisconnected(so);
1706                 break;
1707
1708         case SOCK_STREAM:
1709         case SOCK_SEQPACKET:
1710                 KASSERT(unp2->unp_conn == NULL,
1711                     ("%s: socket %p is already connected", __func__, unp2));
1712                 unp2->unp_conn = unp;
1713                 if (req == PRU_CONNECT &&
1714                     ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1715                         soisconnecting(so);
1716                 else
1717                         soisconnected(so);
1718                 soisconnected(so2);
1719                 break;
1720
1721         default:
1722                 panic("unp_connect2");
1723         }
1724         return (0);
1725 }
1726
1727 static void
1728 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1729 {
1730         struct socket *so, *so2;
1731 #ifdef INVARIANTS
1732         struct unpcb *unptmp;
1733 #endif
1734
1735         UNP_PCB_LOCK_ASSERT(unp);
1736         UNP_PCB_LOCK_ASSERT(unp2);
1737         KASSERT(unp->unp_conn == unp2,
1738             ("%s: unpcb %p is not connected to %p", __func__, unp, unp2));
1739
1740         unp->unp_conn = NULL;
1741         so = unp->unp_socket;
1742         so2 = unp2->unp_socket;
1743         switch (unp->unp_socket->so_type) {
1744         case SOCK_DGRAM:
1745                 UNP_REF_LIST_LOCK();
1746 #ifdef INVARIANTS
1747                 LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) {
1748                         if (unptmp == unp)
1749                                 break;
1750                 }
1751                 KASSERT(unptmp != NULL,
1752                     ("%s: %p not found in reflist of %p", __func__, unp, unp2));
1753 #endif
1754                 LIST_REMOVE(unp, unp_reflink);
1755                 UNP_REF_LIST_UNLOCK();
1756                 if (so) {
1757                         SOCK_LOCK(so);
1758                         so->so_state &= ~SS_ISCONNECTED;
1759                         SOCK_UNLOCK(so);
1760                 }
1761                 break;
1762
1763         case SOCK_STREAM:
1764         case SOCK_SEQPACKET:
1765                 if (so)
1766                         soisdisconnected(so);
1767                 MPASS(unp2->unp_conn == unp);
1768                 unp2->unp_conn = NULL;
1769                 if (so2)
1770                         soisdisconnected(so2);
1771                 break;
1772         }
1773
1774         if (unp == unp2) {
1775                 unp_pcb_rele_notlast(unp);
1776                 if (!unp_pcb_rele(unp))
1777                         UNP_PCB_UNLOCK(unp);
1778         } else {
1779                 if (!unp_pcb_rele(unp))
1780                         UNP_PCB_UNLOCK(unp);
1781                 if (!unp_pcb_rele(unp2))
1782                         UNP_PCB_UNLOCK(unp2);
1783         }
1784 }
1785
1786 /*
1787  * unp_pcblist() walks the global list of struct unpcb's to generate a
1788  * pointer list, bumping the refcount on each unpcb.  It then copies them out
1789  * sequentially, validating the generation number on each to see if it has
1790  * been detached.  All of this is necessary because copyout() may sleep on
1791  * disk I/O.
1792  */
1793 static int
1794 unp_pcblist(SYSCTL_HANDLER_ARGS)
1795 {
1796         struct unpcb *unp, **unp_list;
1797         unp_gen_t gencnt;
1798         struct xunpgen *xug;
1799         struct unp_head *head;
1800         struct xunpcb *xu;
1801         u_int i;
1802         int error, n;
1803
1804         switch ((intptr_t)arg1) {
1805         case SOCK_STREAM:
1806                 head = &unp_shead;
1807                 break;
1808
1809         case SOCK_DGRAM:
1810                 head = &unp_dhead;
1811                 break;
1812
1813         case SOCK_SEQPACKET:
1814                 head = &unp_sphead;
1815                 break;
1816
1817         default:
1818                 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1819         }
1820
1821         /*
1822          * The process of preparing the PCB list is too time-consuming and
1823          * resource-intensive to repeat twice on every request.
1824          */
1825         if (req->oldptr == NULL) {
1826                 n = unp_count;
1827                 req->oldidx = 2 * (sizeof *xug)
1828                         + (n + n/8) * sizeof(struct xunpcb);
1829                 return (0);
1830         }
1831
1832         if (req->newptr != NULL)
1833                 return (EPERM);
1834
1835         /*
1836          * OK, now we're committed to doing something.
1837          */
1838         xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO);
1839         UNP_LINK_RLOCK();
1840         gencnt = unp_gencnt;
1841         n = unp_count;
1842         UNP_LINK_RUNLOCK();
1843
1844         xug->xug_len = sizeof *xug;
1845         xug->xug_count = n;
1846         xug->xug_gen = gencnt;
1847         xug->xug_sogen = so_gencnt;
1848         error = SYSCTL_OUT(req, xug, sizeof *xug);
1849         if (error) {
1850                 free(xug, M_TEMP);
1851                 return (error);
1852         }
1853
1854         unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1855
1856         UNP_LINK_RLOCK();
1857         for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1858              unp = LIST_NEXT(unp, unp_link)) {
1859                 UNP_PCB_LOCK(unp);
1860                 if (unp->unp_gencnt <= gencnt) {
1861                         if (cr_cansee(req->td->td_ucred,
1862                             unp->unp_socket->so_cred)) {
1863                                 UNP_PCB_UNLOCK(unp);
1864                                 continue;
1865                         }
1866                         unp_list[i++] = unp;
1867                         unp_pcb_hold(unp);
1868                 }
1869                 UNP_PCB_UNLOCK(unp);
1870         }
1871         UNP_LINK_RUNLOCK();
1872         n = i;                  /* In case we lost some during malloc. */
1873
1874         error = 0;
1875         xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1876         for (i = 0; i < n; i++) {
1877                 unp = unp_list[i];
1878                 UNP_PCB_LOCK(unp);
1879                 if (unp_pcb_rele(unp))
1880                         continue;
1881
1882                 if (unp->unp_gencnt <= gencnt) {
1883                         xu->xu_len = sizeof *xu;
1884                         xu->xu_unpp = (uintptr_t)unp;
1885                         /*
1886                          * XXX - need more locking here to protect against
1887                          * connect/disconnect races for SMP.
1888                          */
1889                         if (unp->unp_addr != NULL)
1890                                 bcopy(unp->unp_addr, &xu->xu_addr,
1891                                       unp->unp_addr->sun_len);
1892                         else
1893                                 bzero(&xu->xu_addr, sizeof(xu->xu_addr));
1894                         if (unp->unp_conn != NULL &&
1895                             unp->unp_conn->unp_addr != NULL)
1896                                 bcopy(unp->unp_conn->unp_addr,
1897                                       &xu->xu_caddr,
1898                                       unp->unp_conn->unp_addr->sun_len);
1899                         else
1900                                 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr));
1901                         xu->unp_vnode = (uintptr_t)unp->unp_vnode;
1902                         xu->unp_conn = (uintptr_t)unp->unp_conn;
1903                         xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs);
1904                         xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink);
1905                         xu->unp_gencnt = unp->unp_gencnt;
1906                         sotoxsocket(unp->unp_socket, &xu->xu_socket);
1907                         UNP_PCB_UNLOCK(unp);
1908                         error = SYSCTL_OUT(req, xu, sizeof *xu);
1909                 } else {
1910                         UNP_PCB_UNLOCK(unp);
1911                 }
1912         }
1913         free(xu, M_TEMP);
1914         if (!error) {
1915                 /*
1916                  * Give the user an updated idea of our state.  If the
1917                  * generation differs from what we told her before, she knows
1918                  * that something happened while we were processing this
1919                  * request, and it might be necessary to retry.
1920                  */
1921                 xug->xug_gen = unp_gencnt;
1922                 xug->xug_sogen = so_gencnt;
1923                 xug->xug_count = unp_count;
1924                 error = SYSCTL_OUT(req, xug, sizeof *xug);
1925         }
1926         free(unp_list, M_TEMP);
1927         free(xug, M_TEMP);
1928         return (error);
1929 }
1930
1931 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist,
1932     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1933     (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1934     "List of active local datagram sockets");
1935 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist,
1936     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1937     (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1938     "List of active local stream sockets");
1939 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1940     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1941     (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1942     "List of active local seqpacket sockets");
1943
1944 static void
1945 unp_shutdown(struct unpcb *unp)
1946 {
1947         struct unpcb *unp2;
1948         struct socket *so;
1949
1950         UNP_PCB_LOCK_ASSERT(unp);
1951
1952         unp2 = unp->unp_conn;
1953         if ((unp->unp_socket->so_type == SOCK_STREAM ||
1954             (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1955                 so = unp2->unp_socket;
1956                 if (so != NULL)
1957                         socantrcvmore(so);
1958         }
1959 }
1960
1961 static void
1962 unp_drop(struct unpcb *unp)
1963 {
1964         struct socket *so = unp->unp_socket;
1965         struct unpcb *unp2;
1966
1967         /*
1968          * Regardless of whether the socket's peer dropped the connection
1969          * with this socket by aborting or disconnecting, POSIX requires
1970          * that ECONNRESET is returned.
1971          */
1972
1973         UNP_PCB_LOCK(unp);
1974         if (so)
1975                 so->so_error = ECONNRESET;
1976         if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) {
1977                 /* Last reference dropped in unp_disconnect(). */
1978                 unp_pcb_rele_notlast(unp);
1979                 unp_disconnect(unp, unp2);
1980         } else if (!unp_pcb_rele(unp)) {
1981                 UNP_PCB_UNLOCK(unp);
1982         }
1983 }
1984
1985 static void
1986 unp_freerights(struct filedescent **fdep, int fdcount)
1987 {
1988         struct file *fp;
1989         int i;
1990
1991         KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1992
1993         for (i = 0; i < fdcount; i++) {
1994                 fp = fdep[i]->fde_file;
1995                 filecaps_free(&fdep[i]->fde_caps);
1996                 unp_discard(fp);
1997         }
1998         free(fdep[0], M_FILECAPS);
1999 }
2000
2001 static int
2002 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
2003 {
2004         struct thread *td = curthread;          /* XXX */
2005         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
2006         int i;
2007         int *fdp;
2008         struct filedesc *fdesc = td->td_proc->p_fd;
2009         struct filedescent **fdep;
2010         void *data;
2011         socklen_t clen = control->m_len, datalen;
2012         int error, newfds;
2013         u_int newlen;
2014
2015         UNP_LINK_UNLOCK_ASSERT();
2016
2017         error = 0;
2018         if (controlp != NULL) /* controlp == NULL => free control messages */
2019                 *controlp = NULL;
2020         while (cm != NULL) {
2021                 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
2022                         error = EINVAL;
2023                         break;
2024                 }
2025                 data = CMSG_DATA(cm);
2026                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
2027                 if (cm->cmsg_level == SOL_SOCKET
2028                     && cm->cmsg_type == SCM_RIGHTS) {
2029                         newfds = datalen / sizeof(*fdep);
2030                         if (newfds == 0)
2031                                 goto next;
2032                         fdep = data;
2033
2034                         /* If we're not outputting the descriptors free them. */
2035                         if (error || controlp == NULL) {
2036                                 unp_freerights(fdep, newfds);
2037                                 goto next;
2038                         }
2039                         FILEDESC_XLOCK(fdesc);
2040
2041                         /*
2042                          * Now change each pointer to an fd in the global
2043                          * table to an integer that is the index to the local
2044                          * fd table entry that we set up to point to the
2045                          * global one we are transferring.
2046                          */
2047                         newlen = newfds * sizeof(int);
2048                         *controlp = sbcreatecontrol(NULL, newlen,
2049                             SCM_RIGHTS, SOL_SOCKET);
2050                         if (*controlp == NULL) {
2051                                 FILEDESC_XUNLOCK(fdesc);
2052                                 error = E2BIG;
2053                                 unp_freerights(fdep, newfds);
2054                                 goto next;
2055                         }
2056
2057                         fdp = (int *)
2058                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2059                         if (fdallocn(td, 0, fdp, newfds) != 0) {
2060                                 FILEDESC_XUNLOCK(fdesc);
2061                                 error = EMSGSIZE;
2062                                 unp_freerights(fdep, newfds);
2063                                 m_freem(*controlp);
2064                                 *controlp = NULL;
2065                                 goto next;
2066                         }
2067                         for (i = 0; i < newfds; i++, fdp++) {
2068                                 _finstall(fdesc, fdep[i]->fde_file, *fdp,
2069                                     (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0,
2070                                     &fdep[i]->fde_caps);
2071                                 unp_externalize_fp(fdep[i]->fde_file);
2072                         }
2073
2074                         /*
2075                          * The new type indicates that the mbuf data refers to
2076                          * kernel resources that may need to be released before
2077                          * the mbuf is freed.
2078                          */
2079                         m_chtype(*controlp, MT_EXTCONTROL);
2080                         FILEDESC_XUNLOCK(fdesc);
2081                         free(fdep[0], M_FILECAPS);
2082                 } else {
2083                         /* We can just copy anything else across. */
2084                         if (error || controlp == NULL)
2085                                 goto next;
2086                         *controlp = sbcreatecontrol(NULL, datalen,
2087                             cm->cmsg_type, cm->cmsg_level);
2088                         if (*controlp == NULL) {
2089                                 error = ENOBUFS;
2090                                 goto next;
2091                         }
2092                         bcopy(data,
2093                             CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
2094                             datalen);
2095                 }
2096                 controlp = &(*controlp)->m_next;
2097
2098 next:
2099                 if (CMSG_SPACE(datalen) < clen) {
2100                         clen -= CMSG_SPACE(datalen);
2101                         cm = (struct cmsghdr *)
2102                             ((caddr_t)cm + CMSG_SPACE(datalen));
2103                 } else {
2104                         clen = 0;
2105                         cm = NULL;
2106                 }
2107         }
2108
2109         m_freem(control);
2110         return (error);
2111 }
2112
2113 static void
2114 unp_zone_change(void *tag)
2115 {
2116
2117         uma_zone_set_max(unp_zone, maxsockets);
2118 }
2119
2120 #ifdef INVARIANTS
2121 static void
2122 unp_zdtor(void *mem, int size __unused, void *arg __unused)
2123 {
2124         struct unpcb *unp;
2125
2126         unp = mem;
2127
2128         KASSERT(LIST_EMPTY(&unp->unp_refs),
2129             ("%s: unpcb %p has lingering refs", __func__, unp));
2130         KASSERT(unp->unp_socket == NULL,
2131             ("%s: unpcb %p has socket backpointer", __func__, unp));
2132         KASSERT(unp->unp_vnode == NULL,
2133             ("%s: unpcb %p has vnode references", __func__, unp));
2134         KASSERT(unp->unp_conn == NULL,
2135             ("%s: unpcb %p is still connected", __func__, unp));
2136         KASSERT(unp->unp_addr == NULL,
2137             ("%s: unpcb %p has leaked addr", __func__, unp));
2138 }
2139 #endif
2140
2141 static void
2142 unp_init(void)
2143 {
2144         uma_dtor dtor;
2145
2146 #ifdef VIMAGE
2147         if (!IS_DEFAULT_VNET(curvnet))
2148                 return;
2149 #endif
2150
2151 #ifdef INVARIANTS
2152         dtor = unp_zdtor;
2153 #else
2154         dtor = NULL;
2155 #endif
2156         unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor,
2157             NULL, NULL, UMA_ALIGN_CACHE, 0);
2158         uma_zone_set_max(unp_zone, maxsockets);
2159         uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
2160         EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
2161             NULL, EVENTHANDLER_PRI_ANY);
2162         LIST_INIT(&unp_dhead);
2163         LIST_INIT(&unp_shead);
2164         LIST_INIT(&unp_sphead);
2165         SLIST_INIT(&unp_defers);
2166         TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
2167         TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
2168         UNP_LINK_LOCK_INIT();
2169         UNP_DEFERRED_LOCK_INIT();
2170 }
2171
2172 static void
2173 unp_internalize_cleanup_rights(struct mbuf *control)
2174 {
2175         struct cmsghdr *cp;
2176         struct mbuf *m;
2177         void *data;
2178         socklen_t datalen;
2179
2180         for (m = control; m != NULL; m = m->m_next) {
2181                 cp = mtod(m, struct cmsghdr *);
2182                 if (cp->cmsg_level != SOL_SOCKET ||
2183                     cp->cmsg_type != SCM_RIGHTS)
2184                         continue;
2185                 data = CMSG_DATA(cp);
2186                 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data;
2187                 unp_freerights(data, datalen / sizeof(struct filedesc *));
2188         }
2189 }
2190
2191 static int
2192 unp_internalize(struct mbuf **controlp, struct thread *td)
2193 {
2194         struct mbuf *control, **initial_controlp;
2195         struct proc *p;
2196         struct filedesc *fdesc;
2197         struct bintime *bt;
2198         struct cmsghdr *cm;
2199         struct cmsgcred *cmcred;
2200         struct filedescent *fde, **fdep, *fdev;
2201         struct file *fp;
2202         struct timeval *tv;
2203         struct timespec *ts;
2204         void *data;
2205         socklen_t clen, datalen;
2206         int i, j, error, *fdp, oldfds;
2207         u_int newlen;
2208
2209         UNP_LINK_UNLOCK_ASSERT();
2210
2211         p = td->td_proc;
2212         fdesc = p->p_fd;
2213         error = 0;
2214         control = *controlp;
2215         clen = control->m_len;
2216         *controlp = NULL;
2217         initial_controlp = controlp;
2218         for (cm = mtod(control, struct cmsghdr *); cm != NULL;) {
2219                 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
2220                     || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
2221                         error = EINVAL;
2222                         goto out;
2223                 }
2224                 data = CMSG_DATA(cm);
2225                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
2226
2227                 switch (cm->cmsg_type) {
2228                 /*
2229                  * Fill in credential information.
2230                  */
2231                 case SCM_CREDS:
2232                         *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
2233                             SCM_CREDS, SOL_SOCKET);
2234                         if (*controlp == NULL) {
2235                                 error = ENOBUFS;
2236                                 goto out;
2237                         }
2238                         cmcred = (struct cmsgcred *)
2239                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2240                         cmcred->cmcred_pid = p->p_pid;
2241                         cmcred->cmcred_uid = td->td_ucred->cr_ruid;
2242                         cmcred->cmcred_gid = td->td_ucred->cr_rgid;
2243                         cmcred->cmcred_euid = td->td_ucred->cr_uid;
2244                         cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
2245                             CMGROUP_MAX);
2246                         for (i = 0; i < cmcred->cmcred_ngroups; i++)
2247                                 cmcred->cmcred_groups[i] =
2248                                     td->td_ucred->cr_groups[i];
2249                         break;
2250
2251                 case SCM_RIGHTS:
2252                         oldfds = datalen / sizeof (int);
2253                         if (oldfds == 0)
2254                                 break;
2255                         /*
2256                          * Check that all the FDs passed in refer to legal
2257                          * files.  If not, reject the entire operation.
2258                          */
2259                         fdp = data;
2260                         FILEDESC_SLOCK(fdesc);
2261                         for (i = 0; i < oldfds; i++, fdp++) {
2262                                 fp = fget_locked(fdesc, *fdp);
2263                                 if (fp == NULL) {
2264                                         FILEDESC_SUNLOCK(fdesc);
2265                                         error = EBADF;
2266                                         goto out;
2267                                 }
2268                                 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
2269                                         FILEDESC_SUNLOCK(fdesc);
2270                                         error = EOPNOTSUPP;
2271                                         goto out;
2272                                 }
2273                         }
2274
2275                         /*
2276                          * Now replace the integer FDs with pointers to the
2277                          * file structure and capability rights.
2278                          */
2279                         newlen = oldfds * sizeof(fdep[0]);
2280                         *controlp = sbcreatecontrol(NULL, newlen,
2281                             SCM_RIGHTS, SOL_SOCKET);
2282                         if (*controlp == NULL) {
2283                                 FILEDESC_SUNLOCK(fdesc);
2284                                 error = E2BIG;
2285                                 goto out;
2286                         }
2287                         fdp = data;
2288                         for (i = 0; i < oldfds; i++, fdp++) {
2289                                 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) {
2290                                         fdp = data;
2291                                         for (j = 0; j < i; j++, fdp++) {
2292                                                 fdrop(fdesc->fd_ofiles[*fdp].
2293                                                     fde_file, td);
2294                                         }
2295                                         FILEDESC_SUNLOCK(fdesc);
2296                                         error = EBADF;
2297                                         goto out;
2298                                 }
2299                         }
2300                         fdp = data;
2301                         fdep = (struct filedescent **)
2302                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2303                         fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
2304                             M_WAITOK);
2305                         for (i = 0; i < oldfds; i++, fdev++, fdp++) {
2306                                 fde = &fdesc->fd_ofiles[*fdp];
2307                                 fdep[i] = fdev;
2308                                 fdep[i]->fde_file = fde->fde_file;
2309                                 filecaps_copy(&fde->fde_caps,
2310                                     &fdep[i]->fde_caps, true);
2311                                 unp_internalize_fp(fdep[i]->fde_file);
2312                         }
2313                         FILEDESC_SUNLOCK(fdesc);
2314                         break;
2315
2316                 case SCM_TIMESTAMP:
2317                         *controlp = sbcreatecontrol(NULL, sizeof(*tv),
2318                             SCM_TIMESTAMP, SOL_SOCKET);
2319                         if (*controlp == NULL) {
2320                                 error = ENOBUFS;
2321                                 goto out;
2322                         }
2323                         tv = (struct timeval *)
2324                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2325                         microtime(tv);
2326                         break;
2327
2328                 case SCM_BINTIME:
2329                         *controlp = sbcreatecontrol(NULL, sizeof(*bt),
2330                             SCM_BINTIME, SOL_SOCKET);
2331                         if (*controlp == NULL) {
2332                                 error = ENOBUFS;
2333                                 goto out;
2334                         }
2335                         bt = (struct bintime *)
2336                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2337                         bintime(bt);
2338                         break;
2339
2340                 case SCM_REALTIME:
2341                         *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2342                             SCM_REALTIME, SOL_SOCKET);
2343                         if (*controlp == NULL) {
2344                                 error = ENOBUFS;
2345                                 goto out;
2346                         }
2347                         ts = (struct timespec *)
2348                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2349                         nanotime(ts);
2350                         break;
2351
2352                 case SCM_MONOTONIC:
2353                         *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2354                             SCM_MONOTONIC, SOL_SOCKET);
2355                         if (*controlp == NULL) {
2356                                 error = ENOBUFS;
2357                                 goto out;
2358                         }
2359                         ts = (struct timespec *)
2360                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2361                         nanouptime(ts);
2362                         break;
2363
2364                 default:
2365                         error = EINVAL;
2366                         goto out;
2367                 }
2368
2369                 if (*controlp != NULL)
2370                         controlp = &(*controlp)->m_next;
2371                 if (CMSG_SPACE(datalen) < clen) {
2372                         clen -= CMSG_SPACE(datalen);
2373                         cm = (struct cmsghdr *)
2374                             ((caddr_t)cm + CMSG_SPACE(datalen));
2375                 } else {
2376                         clen = 0;
2377                         cm = NULL;
2378                 }
2379         }
2380
2381 out:
2382         if (error != 0 && initial_controlp != NULL)
2383                 unp_internalize_cleanup_rights(*initial_controlp);
2384         m_freem(control);
2385         return (error);
2386 }
2387
2388 static struct mbuf *
2389 unp_addsockcred(struct thread *td, struct mbuf *control, int mode)
2390 {
2391         struct mbuf *m, *n, *n_prev;
2392         const struct cmsghdr *cm;
2393         int ngroups, i, cmsgtype;
2394         size_t ctrlsz;
2395
2396         ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2397         if (mode & UNP_WANTCRED_ALWAYS) {
2398                 ctrlsz = SOCKCRED2SIZE(ngroups);
2399                 cmsgtype = SCM_CREDS2;
2400         } else {
2401                 ctrlsz = SOCKCREDSIZE(ngroups);
2402                 cmsgtype = SCM_CREDS;
2403         }
2404
2405         m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET);
2406         if (m == NULL)
2407                 return (control);
2408
2409         if (mode & UNP_WANTCRED_ALWAYS) {
2410                 struct sockcred2 *sc;
2411
2412                 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *));
2413                 sc->sc_version = 0;
2414                 sc->sc_pid = td->td_proc->p_pid;
2415                 sc->sc_uid = td->td_ucred->cr_ruid;
2416                 sc->sc_euid = td->td_ucred->cr_uid;
2417                 sc->sc_gid = td->td_ucred->cr_rgid;
2418                 sc->sc_egid = td->td_ucred->cr_gid;
2419                 sc->sc_ngroups = ngroups;
2420                 for (i = 0; i < sc->sc_ngroups; i++)
2421                         sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2422         } else {
2423                 struct sockcred *sc;
2424
2425                 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *));
2426                 sc->sc_uid = td->td_ucred->cr_ruid;
2427                 sc->sc_euid = td->td_ucred->cr_uid;
2428                 sc->sc_gid = td->td_ucred->cr_rgid;
2429                 sc->sc_egid = td->td_ucred->cr_gid;
2430                 sc->sc_ngroups = ngroups;
2431                 for (i = 0; i < sc->sc_ngroups; i++)
2432                         sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2433         }
2434
2435         /*
2436          * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2437          * created SCM_CREDS control message (struct sockcred) has another
2438          * format.
2439          */
2440         if (control != NULL && cmsgtype == SCM_CREDS)
2441                 for (n = control, n_prev = NULL; n != NULL;) {
2442                         cm = mtod(n, struct cmsghdr *);
2443                         if (cm->cmsg_level == SOL_SOCKET &&
2444                             cm->cmsg_type == SCM_CREDS) {
2445                                 if (n_prev == NULL)
2446                                         control = n->m_next;
2447                                 else
2448                                         n_prev->m_next = n->m_next;
2449                                 n = m_free(n);
2450                         } else {
2451                                 n_prev = n;
2452                                 n = n->m_next;
2453                         }
2454                 }
2455
2456         /* Prepend it to the head. */
2457         m->m_next = control;
2458         return (m);
2459 }
2460
2461 static struct unpcb *
2462 fptounp(struct file *fp)
2463 {
2464         struct socket *so;
2465
2466         if (fp->f_type != DTYPE_SOCKET)
2467                 return (NULL);
2468         if ((so = fp->f_data) == NULL)
2469                 return (NULL);
2470         if (so->so_proto->pr_domain != &localdomain)
2471                 return (NULL);
2472         return sotounpcb(so);
2473 }
2474
2475 static void
2476 unp_discard(struct file *fp)
2477 {
2478         struct unp_defer *dr;
2479
2480         if (unp_externalize_fp(fp)) {
2481                 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2482                 dr->ud_fp = fp;
2483                 UNP_DEFERRED_LOCK();
2484                 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2485                 UNP_DEFERRED_UNLOCK();
2486                 atomic_add_int(&unp_defers_count, 1);
2487                 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2488         } else
2489                 (void) closef(fp, (struct thread *)NULL);
2490 }
2491
2492 static void
2493 unp_process_defers(void *arg __unused, int pending)
2494 {
2495         struct unp_defer *dr;
2496         SLIST_HEAD(, unp_defer) drl;
2497         int count;
2498
2499         SLIST_INIT(&drl);
2500         for (;;) {
2501                 UNP_DEFERRED_LOCK();
2502                 if (SLIST_FIRST(&unp_defers) == NULL) {
2503                         UNP_DEFERRED_UNLOCK();
2504                         break;
2505                 }
2506                 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2507                 UNP_DEFERRED_UNLOCK();
2508                 count = 0;
2509                 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2510                         SLIST_REMOVE_HEAD(&drl, ud_link);
2511                         closef(dr->ud_fp, NULL);
2512                         free(dr, M_TEMP);
2513                         count++;
2514                 }
2515                 atomic_add_int(&unp_defers_count, -count);
2516         }
2517 }
2518
2519 static void
2520 unp_internalize_fp(struct file *fp)
2521 {
2522         struct unpcb *unp;
2523
2524         UNP_LINK_WLOCK();
2525         if ((unp = fptounp(fp)) != NULL) {
2526                 unp->unp_file = fp;
2527                 unp->unp_msgcount++;
2528         }
2529         unp_rights++;
2530         UNP_LINK_WUNLOCK();
2531 }
2532
2533 static int
2534 unp_externalize_fp(struct file *fp)
2535 {
2536         struct unpcb *unp;
2537         int ret;
2538
2539         UNP_LINK_WLOCK();
2540         if ((unp = fptounp(fp)) != NULL) {
2541                 unp->unp_msgcount--;
2542                 ret = 1;
2543         } else
2544                 ret = 0;
2545         unp_rights--;
2546         UNP_LINK_WUNLOCK();
2547         return (ret);
2548 }
2549
2550 /*
2551  * unp_defer indicates whether additional work has been defered for a future
2552  * pass through unp_gc().  It is thread local and does not require explicit
2553  * synchronization.
2554  */
2555 static int      unp_marked;
2556
2557 static void
2558 unp_remove_dead_ref(struct filedescent **fdep, int fdcount)
2559 {
2560         struct unpcb *unp;
2561         struct file *fp;
2562         int i;
2563
2564         /*
2565          * This function can only be called from the gc task.
2566          */
2567         KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0,
2568             ("%s: not on gc callout", __func__));
2569         UNP_LINK_LOCK_ASSERT();
2570
2571         for (i = 0; i < fdcount; i++) {
2572                 fp = fdep[i]->fde_file;
2573                 if ((unp = fptounp(fp)) == NULL)
2574                         continue;
2575                 if ((unp->unp_gcflag & UNPGC_DEAD) == 0)
2576                         continue;
2577                 unp->unp_gcrefs--;
2578         }
2579 }
2580
2581 static void
2582 unp_restore_undead_ref(struct filedescent **fdep, int fdcount)
2583 {
2584         struct unpcb *unp;
2585         struct file *fp;
2586         int i;
2587
2588         /*
2589          * This function can only be called from the gc task.
2590          */
2591         KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0,
2592             ("%s: not on gc callout", __func__));
2593         UNP_LINK_LOCK_ASSERT();
2594
2595         for (i = 0; i < fdcount; i++) {
2596                 fp = fdep[i]->fde_file;
2597                 if ((unp = fptounp(fp)) == NULL)
2598                         continue;
2599                 if ((unp->unp_gcflag & UNPGC_DEAD) == 0)
2600                         continue;
2601                 unp->unp_gcrefs++;
2602                 unp_marked++;
2603         }
2604 }
2605
2606 static void
2607 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int))
2608 {
2609         struct socket *so, *soa;
2610
2611         so = unp->unp_socket;
2612         SOCK_LOCK(so);
2613         if (SOLISTENING(so)) {
2614                 /*
2615                  * Mark all sockets in our accept queue.
2616                  */
2617                 TAILQ_FOREACH(soa, &so->sol_comp, so_list) {
2618                         if (sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS)
2619                                 continue;
2620                         SOCKBUF_LOCK(&soa->so_rcv);
2621                         unp_scan(soa->so_rcv.sb_mb, op);
2622                         SOCKBUF_UNLOCK(&soa->so_rcv);
2623                 }
2624         } else {
2625                 /*
2626                  * Mark all sockets we reference with RIGHTS.
2627                  */
2628                 if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2629                         SOCKBUF_LOCK(&so->so_rcv);
2630                         unp_scan(so->so_rcv.sb_mb, op);
2631                         SOCKBUF_UNLOCK(&so->so_rcv);
2632                 }
2633         }
2634         SOCK_UNLOCK(so);
2635 }
2636
2637 static int unp_recycled;
2638 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2639     "Number of unreachable sockets claimed by the garbage collector.");
2640
2641 static int unp_taskcount;
2642 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2643     "Number of times the garbage collector has run.");
2644
2645 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0,
2646     "Number of active local sockets.");
2647
2648 static void
2649 unp_gc(__unused void *arg, int pending)
2650 {
2651         struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2652                                     NULL };
2653         struct unp_head **head;
2654         struct unp_head unp_deadhead;   /* List of potentially-dead sockets. */
2655         struct file *f, **unref;
2656         struct unpcb *unp, *unptmp;
2657         int i, total, unp_unreachable;
2658
2659         LIST_INIT(&unp_deadhead);
2660         unp_taskcount++;
2661         UNP_LINK_RLOCK();
2662         /*
2663          * First determine which sockets may be in cycles.
2664          */
2665         unp_unreachable = 0;
2666
2667         for (head = heads; *head != NULL; head++)
2668                 LIST_FOREACH(unp, *head, unp_link) {
2669                         KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0,
2670                             ("%s: unp %p has unexpected gc flags 0x%x",
2671                             __func__, unp, (unsigned int)unp->unp_gcflag));
2672
2673                         f = unp->unp_file;
2674
2675                         /*
2676                          * Check for an unreachable socket potentially in a
2677                          * cycle.  It must be in a queue as indicated by
2678                          * msgcount, and this must equal the file reference
2679                          * count.  Note that when msgcount is 0 the file is
2680                          * NULL.
2681                          */
2682                         if (f != NULL && unp->unp_msgcount != 0 &&
2683                             refcount_load(&f->f_count) == unp->unp_msgcount) {
2684                                 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead);
2685                                 unp->unp_gcflag |= UNPGC_DEAD;
2686                                 unp->unp_gcrefs = unp->unp_msgcount;
2687                                 unp_unreachable++;
2688                         }
2689                 }
2690
2691         /*
2692          * Scan all sockets previously marked as potentially being in a cycle
2693          * and remove the references each socket holds on any UNPGC_DEAD
2694          * sockets in its queue.  After this step, all remaining references on
2695          * sockets marked UNPGC_DEAD should not be part of any cycle.
2696          */
2697         LIST_FOREACH(unp, &unp_deadhead, unp_dead)
2698                 unp_gc_scan(unp, unp_remove_dead_ref);
2699
2700         /*
2701          * If a socket still has a non-negative refcount, it cannot be in a
2702          * cycle.  In this case increment refcount of all children iteratively.
2703          * Stop the scan once we do a complete loop without discovering
2704          * a new reachable socket.
2705          */
2706         do {
2707                 unp_marked = 0;
2708                 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp)
2709                         if (unp->unp_gcrefs > 0) {
2710                                 unp->unp_gcflag &= ~UNPGC_DEAD;
2711                                 LIST_REMOVE(unp, unp_dead);
2712                                 KASSERT(unp_unreachable > 0,
2713                                     ("%s: unp_unreachable underflow.",
2714                                     __func__));
2715                                 unp_unreachable--;
2716                                 unp_gc_scan(unp, unp_restore_undead_ref);
2717                         }
2718         } while (unp_marked);
2719
2720         UNP_LINK_RUNLOCK();
2721
2722         if (unp_unreachable == 0)
2723                 return;
2724
2725         /*
2726          * Allocate space for a local array of dead unpcbs.
2727          * TODO: can this path be simplified by instead using the local
2728          * dead list at unp_deadhead, after taking out references
2729          * on the file object and/or unpcb and dropping the link lock?
2730          */
2731         unref = malloc(unp_unreachable * sizeof(struct file *),
2732             M_TEMP, M_WAITOK);
2733
2734         /*
2735          * Iterate looking for sockets which have been specifically marked
2736          * as unreachable and store them locally.
2737          */
2738         UNP_LINK_RLOCK();
2739         total = 0;
2740         LIST_FOREACH(unp, &unp_deadhead, unp_dead) {
2741                 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0,
2742                     ("%s: unp %p not marked UNPGC_DEAD", __func__, unp));
2743                 unp->unp_gcflag &= ~UNPGC_DEAD;
2744                 f = unp->unp_file;
2745                 if (unp->unp_msgcount == 0 || f == NULL ||
2746                     refcount_load(&f->f_count) != unp->unp_msgcount ||
2747                     !fhold(f))
2748                         continue;
2749                 unref[total++] = f;
2750                 KASSERT(total <= unp_unreachable,
2751                     ("%s: incorrect unreachable count.", __func__));
2752         }
2753         UNP_LINK_RUNLOCK();
2754
2755         /*
2756          * Now flush all sockets, free'ing rights.  This will free the
2757          * struct files associated with these sockets but leave each socket
2758          * with one remaining ref.
2759          */
2760         for (i = 0; i < total; i++) {
2761                 struct socket *so;
2762
2763                 so = unref[i]->f_data;
2764                 CURVNET_SET(so->so_vnet);
2765                 sorflush(so);
2766                 CURVNET_RESTORE();
2767         }
2768
2769         /*
2770          * And finally release the sockets so they can be reclaimed.
2771          */
2772         for (i = 0; i < total; i++)
2773                 fdrop(unref[i], NULL);
2774         unp_recycled += total;
2775         free(unref, M_TEMP);
2776 }
2777
2778 static void
2779 unp_dispose_mbuf(struct mbuf *m)
2780 {
2781
2782         if (m)
2783                 unp_scan(m, unp_freerights);
2784 }
2785
2786 /*
2787  * Synchronize against unp_gc, which can trip over data as we are freeing it.
2788  */
2789 static void
2790 unp_dispose(struct socket *so)
2791 {
2792         struct unpcb *unp;
2793
2794         unp = sotounpcb(so);
2795         UNP_LINK_WLOCK();
2796         unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2797         UNP_LINK_WUNLOCK();
2798         if (!SOLISTENING(so))
2799                 unp_dispose_mbuf(so->so_rcv.sb_mb);
2800 }
2801
2802 static void
2803 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2804 {
2805         struct mbuf *m;
2806         struct cmsghdr *cm;
2807         void *data;
2808         socklen_t clen, datalen;
2809
2810         while (m0 != NULL) {
2811                 for (m = m0; m; m = m->m_next) {
2812                         if (m->m_type != MT_CONTROL)
2813                                 continue;
2814
2815                         cm = mtod(m, struct cmsghdr *);
2816                         clen = m->m_len;
2817
2818                         while (cm != NULL) {
2819                                 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2820                                         break;
2821
2822                                 data = CMSG_DATA(cm);
2823                                 datalen = (caddr_t)cm + cm->cmsg_len
2824                                     - (caddr_t)data;
2825
2826                                 if (cm->cmsg_level == SOL_SOCKET &&
2827                                     cm->cmsg_type == SCM_RIGHTS) {
2828                                         (*op)(data, datalen /
2829                                             sizeof(struct filedescent *));
2830                                 }
2831
2832                                 if (CMSG_SPACE(datalen) < clen) {
2833                                         clen -= CMSG_SPACE(datalen);
2834                                         cm = (struct cmsghdr *)
2835                                             ((caddr_t)cm + CMSG_SPACE(datalen));
2836                                 } else {
2837                                         clen = 0;
2838                                         cm = NULL;
2839                                 }
2840                         }
2841                 }
2842                 m0 = m0->m_nextpkt;
2843         }
2844 }
2845
2846 /*
2847  * A helper function called by VFS before socket-type vnode reclamation.
2848  * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2849  * use count.
2850  */
2851 void
2852 vfs_unp_reclaim(struct vnode *vp)
2853 {
2854         struct unpcb *unp;
2855         int active;
2856         struct mtx *vplock;
2857
2858         ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2859         KASSERT(vp->v_type == VSOCK,
2860             ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2861
2862         active = 0;
2863         vplock = mtx_pool_find(mtxpool_sleep, vp);
2864         mtx_lock(vplock);
2865         VOP_UNP_CONNECT(vp, &unp);
2866         if (unp == NULL)
2867                 goto done;
2868         UNP_PCB_LOCK(unp);
2869         if (unp->unp_vnode == vp) {
2870                 VOP_UNP_DETACH(vp);
2871                 unp->unp_vnode = NULL;
2872                 active = 1;
2873         }
2874         UNP_PCB_UNLOCK(unp);
2875  done:
2876         mtx_unlock(vplock);
2877         if (active)
2878                 vunref(vp);
2879 }
2880
2881 #ifdef DDB
2882 static void
2883 db_print_indent(int indent)
2884 {
2885         int i;
2886
2887         for (i = 0; i < indent; i++)
2888                 db_printf(" ");
2889 }
2890
2891 static void
2892 db_print_unpflags(int unp_flags)
2893 {
2894         int comma;
2895
2896         comma = 0;
2897         if (unp_flags & UNP_HAVEPC) {
2898                 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2899                 comma = 1;
2900         }
2901         if (unp_flags & UNP_WANTCRED_ALWAYS) {
2902                 db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : "");
2903                 comma = 1;
2904         }
2905         if (unp_flags & UNP_WANTCRED_ONESHOT) {
2906                 db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : "");
2907                 comma = 1;
2908         }
2909         if (unp_flags & UNP_CONNWAIT) {
2910                 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2911                 comma = 1;
2912         }
2913         if (unp_flags & UNP_CONNECTING) {
2914                 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2915                 comma = 1;
2916         }
2917         if (unp_flags & UNP_BINDING) {
2918                 db_printf("%sUNP_BINDING", comma ? ", " : "");
2919                 comma = 1;
2920         }
2921 }
2922
2923 static void
2924 db_print_xucred(int indent, struct xucred *xu)
2925 {
2926         int comma, i;
2927
2928         db_print_indent(indent);
2929         db_printf("cr_version: %u   cr_uid: %u   cr_pid: %d   cr_ngroups: %d\n",
2930             xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups);
2931         db_print_indent(indent);
2932         db_printf("cr_groups: ");
2933         comma = 0;
2934         for (i = 0; i < xu->cr_ngroups; i++) {
2935                 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2936                 comma = 1;
2937         }
2938         db_printf("\n");
2939 }
2940
2941 static void
2942 db_print_unprefs(int indent, struct unp_head *uh)
2943 {
2944         struct unpcb *unp;
2945         int counter;
2946
2947         counter = 0;
2948         LIST_FOREACH(unp, uh, unp_reflink) {
2949                 if (counter % 4 == 0)
2950                         db_print_indent(indent);
2951                 db_printf("%p  ", unp);
2952                 if (counter % 4 == 3)
2953                         db_printf("\n");
2954                 counter++;
2955         }
2956         if (counter != 0 && counter % 4 != 0)
2957                 db_printf("\n");
2958 }
2959
2960 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2961 {
2962         struct unpcb *unp;
2963
2964         if (!have_addr) {
2965                 db_printf("usage: show unpcb <addr>\n");
2966                 return;
2967         }
2968         unp = (struct unpcb *)addr;
2969
2970         db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
2971             unp->unp_vnode);
2972
2973         db_printf("unp_ino: %ju   unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2974             unp->unp_conn);
2975
2976         db_printf("unp_refs:\n");
2977         db_print_unprefs(2, &unp->unp_refs);
2978
2979         /* XXXRW: Would be nice to print the full address, if any. */
2980         db_printf("unp_addr: %p\n", unp->unp_addr);
2981
2982         db_printf("unp_gencnt: %llu\n",
2983             (unsigned long long)unp->unp_gencnt);
2984
2985         db_printf("unp_flags: %x (", unp->unp_flags);
2986         db_print_unpflags(unp->unp_flags);
2987         db_printf(")\n");
2988
2989         db_printf("unp_peercred:\n");
2990         db_print_xucred(2, &unp->unp_peercred);
2991
2992         db_printf("unp_refcount: %u\n", unp->unp_refcount);
2993 }
2994 #endif