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_CAPATTACH,
 433         .pr_ctloutput =         &uipc_ctloutput,
 434         .pr_usrreqs =           &uipc_usrreqs_stream
 435 },
 436 {
 437         .pr_type =              SOCK_DGRAM,
 438         .pr_domain =            &localdomain,
 439         .pr_flags =             PR_ATOMIC|PR_ADDR|PR_RIGHTS|PR_CAPATTACH,
 440         .pr_ctloutput =         &uipc_ctloutput,
 441         .pr_usrreqs =           &uipc_usrreqs_dgram
 442 },
 443 {
 444         .pr_type =              SOCK_SEQPACKET,
 445         .pr_domain =            &localdomain,
 446
 447         /*
 448          * XXXRW: For now, PR_ADDR because soreceive will bump into them
 449          * due to our use of sbappendaddr.  A new sbappend variants is needed
 450          * that supports both atomic record writes and control data.
 451          */
 452         .pr_flags =             PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|
 453                                     PR_WANTRCVD|PR_RIGHTS|PR_CAPATTACH,
 454         .pr_ctloutput =         &uipc_ctloutput,
 455         .pr_usrreqs =           &uipc_usrreqs_seqpacket,
 456 },
 457 };
 458
 459 static struct domain localdomain = {
 460         .dom_family =           AF_LOCAL,
 461         .dom_name =             "local",
 462         .dom_init =             unp_init,
 463         .dom_externalize =      unp_externalize,
 464         .dom_dispose =          unp_dispose,
 465         .dom_protosw =          localsw,
 466         .dom_protoswNPROTOSW =  &localsw[nitems(localsw)]
 467 };
 468 DOMAIN_SET(local);
 469
 470 static void
 471 uipc_abort(struct socket *so)
 472 {
 473         struct unpcb *unp, *unp2;
 474
 475         unp = sotounpcb(so);
 476         KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
 477         UNP_PCB_UNLOCK_ASSERT(unp);
 478
 479         UNP_PCB_LOCK(unp);
 480         unp2 = unp->unp_conn;
 481         if (unp2 != NULL) {
 482                 unp_pcb_hold(unp2);
 483                 UNP_PCB_UNLOCK(unp);
 484                 unp_drop(unp2);
 485         } else
 486                 UNP_PCB_UNLOCK(unp);
 487 }
 488
 489 static int
 490 uipc_accept(struct socket *so, struct sockaddr **nam)
 491 {
 492         struct unpcb *unp, *unp2;
 493         const struct sockaddr *sa;
 494
 495         /*
 496          * Pass back name of connected socket, if it was bound and we are
 497          * still connected (our peer may have closed already!).
 498          */
 499         unp = sotounpcb(so);
 500         KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
 501
 502         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 503         UNP_PCB_LOCK(unp);
 504         unp2 = unp_pcb_lock_peer(unp);
 505         if (unp2 != NULL && unp2->unp_addr != NULL)
 506                 sa = (struct sockaddr *)unp2->unp_addr;
 507         else
 508                 sa = &sun_noname;
 509         bcopy(sa, *nam, sa->sa_len);
 510         if (unp2 != NULL)
 511                 unp_pcb_unlock_pair(unp, unp2);
 512         else
 513                 UNP_PCB_UNLOCK(unp);
 514         return (0);
 515 }
 516
 517 static int
 518 uipc_attach(struct socket *so, int proto, struct thread *td)
 519 {
 520         u_long sendspace, recvspace;
 521         struct unpcb *unp;
 522         int error;
 523         bool locked;
 524
 525         KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
 526         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
 527                 switch (so->so_type) {
 528                 case SOCK_STREAM:
 529                         sendspace = unpst_sendspace;
 530                         recvspace = unpst_recvspace;
 531                         break;
 532
 533                 case SOCK_DGRAM:
 534                         sendspace = unpdg_sendspace;
 535                         recvspace = unpdg_recvspace;
 536                         break;
 537
 538                 case SOCK_SEQPACKET:
 539                         sendspace = unpsp_sendspace;
 540                         recvspace = unpsp_recvspace;
 541                         break;
 542
 543                 default:
 544                         panic("uipc_attach");
 545                 }
 546                 error = soreserve(so, sendspace, recvspace);
 547                 if (error)
 548                         return (error);
 549         }
 550         unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
 551         if (unp == NULL)
 552                 return (ENOBUFS);
 553         LIST_INIT(&unp->unp_refs);
 554         UNP_PCB_LOCK_INIT(unp);
 555         unp->unp_socket = so;
 556         so->so_pcb = unp;
 557         refcount_init(&unp->unp_refcount, 1);
 558
 559         if ((locked = UNP_LINK_WOWNED()) == false)
 560                 UNP_LINK_WLOCK();
 561
 562         unp->unp_gencnt = ++unp_gencnt;
 563         unp->unp_ino = ++unp_ino;
 564         unp_count++;
 565         switch (so->so_type) {
 566         case SOCK_STREAM:
 567                 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
 568                 break;
 569
 570         case SOCK_DGRAM:
 571                 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
 572                 break;
 573
 574         case SOCK_SEQPACKET:
 575                 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
 576                 break;
 577
 578         default:
 579                 panic("uipc_attach");
 580         }
 581
 582         if (locked == false)
 583                 UNP_LINK_WUNLOCK();
 584
 585         return (0);
 586 }
 587
 588 static int
 589 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
 590 {
 591         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
 592         struct vattr vattr;
 593         int error, namelen;
 594         struct nameidata nd;
 595         struct unpcb *unp;
 596         struct vnode *vp;
 597         struct mount *mp;
 598         cap_rights_t rights;
 599         char *buf;
 600
 601         if (nam->sa_family != AF_UNIX)
 602                 return (EAFNOSUPPORT);
 603
 604         unp = sotounpcb(so);
 605         KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
 606
 607         if (soun->sun_len > sizeof(struct sockaddr_un))
 608                 return (EINVAL);
 609         namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
 610         if (namelen <= 0)
 611                 return (EINVAL);
 612
 613         /*
 614          * We don't allow simultaneous bind() calls on a single UNIX domain
 615          * socket, so flag in-progress operations, and return an error if an
 616          * operation is already in progress.
 617          *
 618          * Historically, we have not allowed a socket to be rebound, so this
 619          * also returns an error.  Not allowing re-binding simplifies the
 620          * implementation and avoids a great many possible failure modes.
 621          */
 622         UNP_PCB_LOCK(unp);
 623         if (unp->unp_vnode != NULL) {
 624                 UNP_PCB_UNLOCK(unp);
 625                 return (EINVAL);
 626         }
 627         if (unp->unp_flags & UNP_BINDING) {
 628                 UNP_PCB_UNLOCK(unp);
 629                 return (EALREADY);
 630         }
 631         unp->unp_flags |= UNP_BINDING;
 632         UNP_PCB_UNLOCK(unp);
 633
 634         buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
 635         bcopy(soun->sun_path, buf, namelen);
 636         buf[namelen] = 0;
 637
 638 restart:
 639         NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE,
 640             UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_BINDAT),
 641             td);
 642 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
 643         error = namei(&nd);
 644         if (error)
 645                 goto error;
 646         vp = nd.ni_vp;
 647         if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
 648                 NDFREE(&nd, NDF_ONLY_PNBUF);
 649                 if (nd.ni_dvp == vp)
 650                         vrele(nd.ni_dvp);
 651                 else
 652                         vput(nd.ni_dvp);
 653                 if (vp != NULL) {
 654                         vrele(vp);
 655                         error = EADDRINUSE;
 656                         goto error;
 657                 }
 658                 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
 659                 if (error)
 660                         goto error;
 661                 goto restart;
 662         }
 663         VATTR_NULL(&vattr);
 664         vattr.va_type = VSOCK;
 665         vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_pd->pd_cmask);
 666 #ifdef MAC
 667         error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
 668             &vattr);
 669 #endif
 670         if (error == 0)
 671                 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
 672         NDFREE(&nd, NDF_ONLY_PNBUF);
 673         if (error) {
 674                 VOP_VPUT_PAIR(nd.ni_dvp, NULL, true);
 675                 vn_finished_write(mp);
 676                 if (error == ERELOOKUP)
 677                         goto restart;
 678                 goto error;
 679         }
 680         vp = nd.ni_vp;
 681         ASSERT_VOP_ELOCKED(vp, "uipc_bind");
 682         soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
 683
 684         UNP_PCB_LOCK(unp);
 685         VOP_UNP_BIND(vp, unp);
 686         unp->unp_vnode = vp;
 687         unp->unp_addr = soun;
 688         unp->unp_flags &= ~UNP_BINDING;
 689         UNP_PCB_UNLOCK(unp);
 690         vref(vp);
 691         VOP_VPUT_PAIR(nd.ni_dvp, &vp, true);
 692         vn_finished_write(mp);
 693         free(buf, M_TEMP);
 694         return (0);
 695
 696 error:
 697         UNP_PCB_LOCK(unp);
 698         unp->unp_flags &= ~UNP_BINDING;
 699         UNP_PCB_UNLOCK(unp);
 700         free(buf, M_TEMP);
 701         return (error);
 702 }
 703
 704 static int
 705 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 706 {
 707
 708         return (uipc_bindat(AT_FDCWD, so, nam, td));
 709 }
 710
 711 static int
 712 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 713 {
 714         int error;
 715
 716         KASSERT(td == curthread, ("uipc_connect: td != curthread"));
 717         error = unp_connect(so, nam, td);
 718         return (error);
 719 }
 720
 721 static int
 722 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
 723     struct thread *td)
 724 {
 725         int error;
 726
 727         KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
 728         error = unp_connectat(fd, so, nam, td);
 729         return (error);
 730 }
 731
 732 static void
 733 uipc_close(struct socket *so)
 734 {
 735         struct unpcb *unp, *unp2;
 736         struct vnode *vp = NULL;
 737         struct mtx *vplock;
 738
 739         unp = sotounpcb(so);
 740         KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
 741
 742         vplock = NULL;
 743         if ((vp = unp->unp_vnode) != NULL) {
 744                 vplock = mtx_pool_find(mtxpool_sleep, vp);
 745                 mtx_lock(vplock);
 746         }
 747         UNP_PCB_LOCK(unp);
 748         if (vp && unp->unp_vnode == NULL) {
 749                 mtx_unlock(vplock);
 750                 vp = NULL;
 751         }
 752         if (vp != NULL) {
 753                 VOP_UNP_DETACH(vp);
 754                 unp->unp_vnode = NULL;
 755         }
 756         if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
 757                 unp_disconnect(unp, unp2);
 758         else
 759                 UNP_PCB_UNLOCK(unp);
 760         if (vp) {
 761                 mtx_unlock(vplock);
 762                 vrele(vp);
 763         }
 764 }
 765
 766 static int
 767 uipc_connect2(struct socket *so1, struct socket *so2)
 768 {
 769         struct unpcb *unp, *unp2;
 770         int error;
 771
 772         unp = so1->so_pcb;
 773         KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
 774         unp2 = so2->so_pcb;
 775         KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
 776         unp_pcb_lock_pair(unp, unp2);
 777         error = unp_connect2(so1, so2, PRU_CONNECT2);
 778         unp_pcb_unlock_pair(unp, unp2);
 779         return (error);
 780 }
 781
 782 static void
 783 uipc_detach(struct socket *so)
 784 {
 785         struct unpcb *unp, *unp2;
 786         struct mtx *vplock;
 787         struct vnode *vp;
 788         int local_unp_rights;
 789
 790         unp = sotounpcb(so);
 791         KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
 792
 793         vp = NULL;
 794         vplock = NULL;
 795
 796         SOCK_LOCK(so);
 797         if (!SOLISTENING(so)) {
 798                 /*
 799                  * Once the socket is removed from the global lists,
 800                  * uipc_ready() will not be able to locate its socket buffer, so
 801                  * clear the buffer now.  At this point internalized rights have
 802                  * already been disposed of.
 803                  */
 804                 sbrelease(&so->so_rcv, so);
 805         }
 806         SOCK_UNLOCK(so);
 807
 808         UNP_LINK_WLOCK();
 809         LIST_REMOVE(unp, unp_link);
 810         if (unp->unp_gcflag & UNPGC_DEAD)
 811                 LIST_REMOVE(unp, unp_dead);
 812         unp->unp_gencnt = ++unp_gencnt;
 813         --unp_count;
 814         UNP_LINK_WUNLOCK();
 815
 816         UNP_PCB_UNLOCK_ASSERT(unp);
 817  restart:
 818         if ((vp = unp->unp_vnode) != NULL) {
 819                 vplock = mtx_pool_find(mtxpool_sleep, vp);
 820                 mtx_lock(vplock);
 821         }
 822         UNP_PCB_LOCK(unp);
 823         if (unp->unp_vnode != vp && unp->unp_vnode != NULL) {
 824                 if (vplock)
 825                         mtx_unlock(vplock);
 826                 UNP_PCB_UNLOCK(unp);
 827                 goto restart;
 828         }
 829         if ((vp = unp->unp_vnode) != NULL) {
 830                 VOP_UNP_DETACH(vp);
 831                 unp->unp_vnode = NULL;
 832         }
 833         if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
 834                 unp_disconnect(unp, unp2);
 835         else
 836                 UNP_PCB_UNLOCK(unp);
 837
 838         UNP_REF_LIST_LOCK();
 839         while (!LIST_EMPTY(&unp->unp_refs)) {
 840                 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
 841
 842                 unp_pcb_hold(ref);
 843                 UNP_REF_LIST_UNLOCK();
 844
 845                 MPASS(ref != unp);
 846                 UNP_PCB_UNLOCK_ASSERT(ref);
 847                 unp_drop(ref);
 848                 UNP_REF_LIST_LOCK();
 849         }
 850         UNP_REF_LIST_UNLOCK();
 851
 852         UNP_PCB_LOCK(unp);
 853         local_unp_rights = unp_rights;
 854         unp->unp_socket->so_pcb = NULL;
 855         unp->unp_socket = NULL;
 856         free(unp->unp_addr, M_SONAME);
 857         unp->unp_addr = NULL;
 858         if (!unp_pcb_rele(unp))
 859                 UNP_PCB_UNLOCK(unp);
 860         if (vp) {
 861                 mtx_unlock(vplock);
 862                 vrele(vp);
 863         }
 864         if (local_unp_rights)
 865                 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
 866 }
 867
 868 static int
 869 uipc_disconnect(struct socket *so)
 870 {
 871         struct unpcb *unp, *unp2;
 872
 873         unp = sotounpcb(so);
 874         KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
 875
 876         UNP_PCB_LOCK(unp);
 877         if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
 878                 unp_disconnect(unp, unp2);
 879         else
 880                 UNP_PCB_UNLOCK(unp);
 881         return (0);
 882 }
 883
 884 static int
 885 uipc_listen(struct socket *so, int backlog, struct thread *td)
 886 {
 887         struct unpcb *unp;
 888         int error;
 889
 890         if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
 891                 return (EOPNOTSUPP);
 892
 893         unp = sotounpcb(so);
 894         KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
 895
 896         UNP_PCB_LOCK(unp);
 897         if (unp->unp_vnode == NULL) {
 898                 /* Already connected or not bound to an address. */
 899                 error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ;
 900                 UNP_PCB_UNLOCK(unp);
 901                 return (error);
 902         }
 903
 904         SOCK_LOCK(so);
 905         error = solisten_proto_check(so);
 906         if (error == 0) {
 907                 cru2xt(td, &unp->unp_peercred);
 908                 solisten_proto(so, backlog);
 909         }
 910         SOCK_UNLOCK(so);
 911         UNP_PCB_UNLOCK(unp);
 912         return (error);
 913 }
 914
 915 static int
 916 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
 917 {
 918         struct unpcb *unp, *unp2;
 919         const struct sockaddr *sa;
 920
 921         unp = sotounpcb(so);
 922         KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
 923
 924         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 925         UNP_LINK_RLOCK();
 926         /*
 927          * XXX: It seems that this test always fails even when connection is
 928          * established.  So, this else clause is added as workaround to
 929          * return PF_LOCAL sockaddr.
 930          */
 931         unp2 = unp->unp_conn;
 932         if (unp2 != NULL) {
 933                 UNP_PCB_LOCK(unp2);
 934                 if (unp2->unp_addr != NULL)
 935                         sa = (struct sockaddr *) unp2->unp_addr;
 936                 else
 937                         sa = &sun_noname;
 938                 bcopy(sa, *nam, sa->sa_len);
 939                 UNP_PCB_UNLOCK(unp2);
 940         } else {
 941                 sa = &sun_noname;
 942                 bcopy(sa, *nam, sa->sa_len);
 943         }
 944         UNP_LINK_RUNLOCK();
 945         return (0);
 946 }
 947
 948 static int
 949 uipc_rcvd(struct socket *so, int flags)
 950 {
 951         struct unpcb *unp, *unp2;
 952         struct socket *so2;
 953         u_int mbcnt, sbcc;
 954
 955         unp = sotounpcb(so);
 956         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
 957         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
 958             ("%s: socktype %d", __func__, so->so_type));
 959
 960         /*
 961          * Adjust backpressure on sender and wakeup any waiting to write.
 962          *
 963          * The unp lock is acquired to maintain the validity of the unp_conn
 964          * pointer; no lock on unp2 is required as unp2->unp_socket will be
 965          * static as long as we don't permit unp2 to disconnect from unp,
 966          * which is prevented by the lock on unp.  We cache values from
 967          * so_rcv to avoid holding the so_rcv lock over the entire
 968          * transaction on the remote so_snd.
 969          */
 970         SOCKBUF_LOCK(&so->so_rcv);
 971         mbcnt = so->so_rcv.sb_mbcnt;
 972         sbcc = sbavail(&so->so_rcv);
 973         SOCKBUF_UNLOCK(&so->so_rcv);
 974         /*
 975          * There is a benign race condition at this point.  If we're planning to
 976          * clear SB_STOP, but uipc_send is called on the connected socket at
 977          * this instant, it might add data to the sockbuf and set SB_STOP.  Then
 978          * we would erroneously clear SB_STOP below, even though the sockbuf is
 979          * full.  The race is benign because the only ill effect is to allow the
 980          * sockbuf to exceed its size limit, and the size limits are not
 981          * strictly guaranteed anyway.
 982          */
 983         UNP_PCB_LOCK(unp);
 984         unp2 = unp->unp_conn;
 985         if (unp2 == NULL) {
 986                 UNP_PCB_UNLOCK(unp);
 987                 return (0);
 988         }
 989         so2 = unp2->unp_socket;
 990         SOCKBUF_LOCK(&so2->so_snd);
 991         if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
 992                 so2->so_snd.sb_flags &= ~SB_STOP;
 993         sowwakeup_locked(so2);
 994         UNP_PCB_UNLOCK(unp);
 995         return (0);
 996 }
 997
 998 static int
 999 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
1000     struct mbuf *control, struct thread *td)
1001 {
1002         struct unpcb *unp, *unp2;
1003         struct socket *so2;
1004         u_int mbcnt, sbcc;
1005         int freed, error;
1006
1007         unp = sotounpcb(so);
1008         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
1009         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
1010             so->so_type == SOCK_SEQPACKET,
1011             ("%s: socktype %d", __func__, so->so_type));
1012
1013         freed = error = 0;
1014         if (flags & PRUS_OOB) {
1015                 error = EOPNOTSUPP;
1016                 goto release;
1017         }
1018         if (control != NULL && (error = unp_internalize(&control, td)))
1019                 goto release;
1020
1021         unp2 = NULL;
1022         switch (so->so_type) {
1023         case SOCK_DGRAM:
1024         {
1025                 const struct sockaddr *from;
1026
1027                 if (nam != NULL) {
1028                         error = unp_connect(so, nam, td);
1029                         if (error != 0)
1030                                 break;
1031                 }
1032                 UNP_PCB_LOCK(unp);
1033
1034                 /*
1035                  * Because connect() and send() are non-atomic in a sendto()
1036                  * with a target address, it's possible that the socket will
1037                  * have disconnected before the send() can run.  In that case
1038                  * return the slightly counter-intuitive but otherwise
1039                  * correct error that the socket is not connected.
1040                  */
1041                 unp2 = unp_pcb_lock_peer(unp);
1042                 if (unp2 == NULL) {
1043                         UNP_PCB_UNLOCK(unp);
1044                         error = ENOTCONN;
1045                         break;
1046                 }
1047
1048                 if (unp2->unp_flags & UNP_WANTCRED_MASK)
1049                         control = unp_addsockcred(td, control,
1050                             unp2->unp_flags);
1051                 if (unp->unp_addr != NULL)
1052                         from = (struct sockaddr *)unp->unp_addr;
1053                 else
1054                         from = &sun_noname;
1055                 so2 = unp2->unp_socket;
1056                 SOCKBUF_LOCK(&so2->so_rcv);
1057                 if (sbappendaddr_locked(&so2->so_rcv, from, m,
1058                     control)) {
1059                         sorwakeup_locked(so2);
1060                         m = NULL;
1061                         control = NULL;
1062                 } else {
1063                         SOCKBUF_UNLOCK(&so2->so_rcv);
1064                         error = ENOBUFS;
1065                 }
1066                 if (nam != NULL)
1067                         unp_disconnect(unp, unp2);
1068                 else
1069                         unp_pcb_unlock_pair(unp, unp2);
1070                 break;
1071         }
1072
1073         case SOCK_SEQPACKET:
1074         case SOCK_STREAM:
1075                 if ((so->so_state & SS_ISCONNECTED) == 0) {
1076                         if (nam != NULL) {
1077                                 error = unp_connect(so, nam, td);
1078                                 if (error != 0)
1079                                         break;
1080                         } else {
1081                                 error = ENOTCONN;
1082                                 break;
1083                         }
1084                 }
1085
1086                 UNP_PCB_LOCK(unp);
1087                 if ((unp2 = unp_pcb_lock_peer(unp)) == NULL) {
1088                         UNP_PCB_UNLOCK(unp);
1089                         error = ENOTCONN;
1090                         break;
1091                 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1092                         unp_pcb_unlock_pair(unp, unp2);
1093                         error = EPIPE;
1094                         break;
1095                 }
1096                 UNP_PCB_UNLOCK(unp);
1097                 if ((so2 = unp2->unp_socket) == NULL) {
1098                         UNP_PCB_UNLOCK(unp2);
1099                         error = ENOTCONN;
1100                         break;
1101                 }
1102                 SOCKBUF_LOCK(&so2->so_rcv);
1103                 if (unp2->unp_flags & UNP_WANTCRED_MASK) {
1104                         /*
1105                          * Credentials are passed only once on SOCK_STREAM and
1106                          * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or
1107                          * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS).
1108                          */
1109                         control = unp_addsockcred(td, control, unp2->unp_flags);
1110                         unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT;
1111                 }
1112
1113                 /*
1114                  * Send to paired receive port and wake up readers.  Don't
1115                  * check for space available in the receive buffer if we're
1116                  * attaching ancillary data; Unix domain sockets only check
1117                  * for space in the sending sockbuf, and that check is
1118                  * performed one level up the stack.  At that level we cannot
1119                  * precisely account for the amount of buffer space used
1120                  * (e.g., because control messages are not yet internalized).
1121                  */
1122                 switch (so->so_type) {
1123                 case SOCK_STREAM:
1124                         if (control != NULL) {
1125                                 sbappendcontrol_locked(&so2->so_rcv, m,
1126                                     control, flags);
1127                                 control = NULL;
1128                         } else
1129                                 sbappend_locked(&so2->so_rcv, m, flags);
1130                         break;
1131
1132                 case SOCK_SEQPACKET:
1133                         if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1134                             &sun_noname, m, control))
1135                                 control = NULL;
1136                         break;
1137                 }
1138
1139                 mbcnt = so2->so_rcv.sb_mbcnt;
1140                 sbcc = sbavail(&so2->so_rcv);
1141                 if (sbcc)
1142                         sorwakeup_locked(so2);
1143                 else
1144                         SOCKBUF_UNLOCK(&so2->so_rcv);
1145
1146                 /*
1147                  * The PCB lock on unp2 protects the SB_STOP flag.  Without it,
1148                  * it would be possible for uipc_rcvd to be called at this
1149                  * point, drain the receiving sockbuf, clear SB_STOP, and then
1150                  * we would set SB_STOP below.  That could lead to an empty
1151                  * sockbuf having SB_STOP set
1152                  */
1153                 SOCKBUF_LOCK(&so->so_snd);
1154                 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1155                         so->so_snd.sb_flags |= SB_STOP;
1156                 SOCKBUF_UNLOCK(&so->so_snd);
1157                 UNP_PCB_UNLOCK(unp2);
1158                 m = NULL;
1159                 break;
1160         }
1161
1162         /*
1163          * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1164          */
1165         if (flags & PRUS_EOF) {
1166                 UNP_PCB_LOCK(unp);
1167                 socantsendmore(so);
1168                 unp_shutdown(unp);
1169                 UNP_PCB_UNLOCK(unp);
1170         }
1171         if (control != NULL && error != 0)
1172                 unp_dispose_mbuf(control);
1173
1174 release:
1175         if (control != NULL)
1176                 m_freem(control);
1177         /*
1178          * In case of PRUS_NOTREADY, uipc_ready() is responsible
1179          * for freeing memory.
1180          */
1181         if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1182                 m_freem(m);
1183         return (error);
1184 }
1185
1186 static bool
1187 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp)
1188 {
1189         struct mbuf *mb, *n;
1190         struct sockbuf *sb;
1191
1192         SOCK_LOCK(so);
1193         if (SOLISTENING(so)) {
1194                 SOCK_UNLOCK(so);
1195                 return (false);
1196         }
1197         mb = NULL;
1198         sb = &so->so_rcv;
1199         SOCKBUF_LOCK(sb);
1200         if (sb->sb_fnrdy != NULL) {
1201                 for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) {
1202                         if (mb == m) {
1203                                 *errorp = sbready(sb, m, count);
1204                                 break;
1205                         }
1206                         mb = mb->m_next;
1207                         if (mb == NULL) {
1208                                 mb = n;
1209                                 if (mb != NULL)
1210                                         n = mb->m_nextpkt;
1211                         }
1212                 }
1213         }
1214         SOCKBUF_UNLOCK(sb);
1215         SOCK_UNLOCK(so);
1216         return (mb != NULL);
1217 }
1218
1219 static int
1220 uipc_ready(struct socket *so, struct mbuf *m, int count)
1221 {
1222         struct unpcb *unp, *unp2;
1223         struct socket *so2;
1224         int error, i;
1225
1226         unp = sotounpcb(so);
1227
1228         KASSERT(so->so_type == SOCK_STREAM,
1229             ("%s: unexpected socket type for %p", __func__, so));
1230
1231         UNP_PCB_LOCK(unp);
1232         if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) {
1233                 UNP_PCB_UNLOCK(unp);
1234                 so2 = unp2->unp_socket;
1235                 SOCKBUF_LOCK(&so2->so_rcv);
1236                 if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1237                         sorwakeup_locked(so2);
1238                 else
1239                         SOCKBUF_UNLOCK(&so2->so_rcv);
1240                 UNP_PCB_UNLOCK(unp2);
1241                 return (error);
1242         }
1243         UNP_PCB_UNLOCK(unp);
1244
1245         /*
1246          * The receiving socket has been disconnected, but may still be valid.
1247          * In this case, the now-ready mbufs are still present in its socket
1248          * buffer, so perform an exhaustive search before giving up and freeing
1249          * the mbufs.
1250          */
1251         UNP_LINK_RLOCK();
1252         LIST_FOREACH(unp, &unp_shead, unp_link) {
1253                 if (uipc_ready_scan(unp->unp_socket, m, count, &error))
1254                         break;
1255         }
1256         UNP_LINK_RUNLOCK();
1257
1258         if (unp == NULL) {
1259                 for (i = 0; i < count; i++)
1260                         m = m_free(m);
1261                 error = ECONNRESET;
1262         }
1263         return (error);
1264 }
1265
1266 static int
1267 uipc_sense(struct socket *so, struct stat *sb)
1268 {
1269         struct unpcb *unp;
1270
1271         unp = sotounpcb(so);
1272         KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1273
1274         sb->st_blksize = so->so_snd.sb_hiwat;
1275         sb->st_dev = NODEV;
1276         sb->st_ino = unp->unp_ino;
1277         return (0);
1278 }
1279
1280 static int
1281 uipc_shutdown(struct socket *so)
1282 {
1283         struct unpcb *unp;
1284
1285         unp = sotounpcb(so);
1286         KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1287
1288         UNP_PCB_LOCK(unp);
1289         socantsendmore(so);
1290         unp_shutdown(unp);
1291         UNP_PCB_UNLOCK(unp);
1292         return (0);
1293 }
1294
1295 static int
1296 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1297 {
1298         struct unpcb *unp;
1299         const struct sockaddr *sa;
1300
1301         unp = sotounpcb(so);
1302         KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1303
1304         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1305         UNP_PCB_LOCK(unp);
1306         if (unp->unp_addr != NULL)
1307                 sa = (struct sockaddr *) unp->unp_addr;
1308         else
1309                 sa = &sun_noname;
1310         bcopy(sa, *nam, sa->sa_len);
1311         UNP_PCB_UNLOCK(unp);
1312         return (0);
1313 }
1314
1315 static struct pr_usrreqs uipc_usrreqs_dgram = {
1316         .pru_abort =            uipc_abort,
1317         .pru_accept =           uipc_accept,
1318         .pru_attach =           uipc_attach,
1319         .pru_bind =             uipc_bind,
1320         .pru_bindat =           uipc_bindat,
1321         .pru_connect =          uipc_connect,
1322         .pru_connectat =        uipc_connectat,
1323         .pru_connect2 =         uipc_connect2,
1324         .pru_detach =           uipc_detach,
1325         .pru_disconnect =       uipc_disconnect,
1326         .pru_listen =           uipc_listen,
1327         .pru_peeraddr =         uipc_peeraddr,
1328         .pru_rcvd =             uipc_rcvd,
1329         .pru_send =             uipc_send,
1330         .pru_sense =            uipc_sense,
1331         .pru_shutdown =         uipc_shutdown,
1332         .pru_sockaddr =         uipc_sockaddr,
1333         .pru_soreceive =        soreceive_dgram,
1334         .pru_close =            uipc_close,
1335 };
1336
1337 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1338         .pru_abort =            uipc_abort,
1339         .pru_accept =           uipc_accept,
1340         .pru_attach =           uipc_attach,
1341         .pru_bind =             uipc_bind,
1342         .pru_bindat =           uipc_bindat,
1343         .pru_connect =          uipc_connect,
1344         .pru_connectat =        uipc_connectat,
1345         .pru_connect2 =         uipc_connect2,
1346         .pru_detach =           uipc_detach,
1347         .pru_disconnect =       uipc_disconnect,
1348         .pru_listen =           uipc_listen,
1349         .pru_peeraddr =         uipc_peeraddr,
1350         .pru_rcvd =             uipc_rcvd,
1351         .pru_send =             uipc_send,
1352         .pru_sense =            uipc_sense,
1353         .pru_shutdown =         uipc_shutdown,
1354         .pru_sockaddr =         uipc_sockaddr,
1355         .pru_soreceive =        soreceive_generic,      /* XXX: or...? */
1356         .pru_close =            uipc_close,
1357 };
1358
1359 static struct pr_usrreqs uipc_usrreqs_stream = {
1360         .pru_abort =            uipc_abort,
1361         .pru_accept =           uipc_accept,
1362         .pru_attach =           uipc_attach,
1363         .pru_bind =             uipc_bind,
1364         .pru_bindat =           uipc_bindat,
1365         .pru_connect =          uipc_connect,
1366         .pru_connectat =        uipc_connectat,
1367         .pru_connect2 =         uipc_connect2,
1368         .pru_detach =           uipc_detach,
1369         .pru_disconnect =       uipc_disconnect,
1370         .pru_listen =           uipc_listen,
1371         .pru_peeraddr =         uipc_peeraddr,
1372         .pru_rcvd =             uipc_rcvd,
1373         .pru_send =             uipc_send,
1374         .pru_ready =            uipc_ready,
1375         .pru_sense =            uipc_sense,
1376         .pru_shutdown =         uipc_shutdown,
1377         .pru_sockaddr =         uipc_sockaddr,
1378         .pru_soreceive =        soreceive_generic,
1379         .pru_close =            uipc_close,
1380 };
1381
1382 static int
1383 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1384 {
1385         struct unpcb *unp;
1386         struct xucred xu;
1387         int error, optval;
1388
1389         if (sopt->sopt_level != SOL_LOCAL)
1390                 return (EINVAL);
1391
1392         unp = sotounpcb(so);
1393         KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1394         error = 0;
1395         switch (sopt->sopt_dir) {
1396         case SOPT_GET:
1397                 switch (sopt->sopt_name) {
1398                 case LOCAL_PEERCRED:
1399                         UNP_PCB_LOCK(unp);
1400                         if (unp->unp_flags & UNP_HAVEPC)
1401                                 xu = unp->unp_peercred;
1402                         else {
1403                                 if (so->so_type == SOCK_STREAM)
1404                                         error = ENOTCONN;
1405                                 else
1406                                         error = EINVAL;
1407                         }
1408                         UNP_PCB_UNLOCK(unp);
1409                         if (error == 0)
1410                                 error = sooptcopyout(sopt, &xu, sizeof(xu));
1411                         break;
1412
1413                 case LOCAL_CREDS:
1414                         /* Unlocked read. */
1415                         optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0;
1416                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1417                         break;
1418
1419                 case LOCAL_CREDS_PERSISTENT:
1420                         /* Unlocked read. */
1421                         optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0;
1422                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1423                         break;
1424
1425                 case LOCAL_CONNWAIT:
1426                         /* Unlocked read. */
1427                         optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1428                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1429                         break;
1430
1431                 default:
1432                         error = EOPNOTSUPP;
1433                         break;
1434                 }
1435                 break;
1436
1437         case SOPT_SET:
1438                 switch (sopt->sopt_name) {
1439                 case LOCAL_CREDS:
1440                 case LOCAL_CREDS_PERSISTENT:
1441                 case LOCAL_CONNWAIT:
1442                         error = sooptcopyin(sopt, &optval, sizeof(optval),
1443                                             sizeof(optval));
1444                         if (error)
1445                                 break;
1446
1447 #define OPTSET(bit, exclusive) do {                                     \
1448         UNP_PCB_LOCK(unp);                                              \
1449         if (optval) {                                                   \
1450                 if ((unp->unp_flags & (exclusive)) != 0) {              \
1451                         UNP_PCB_UNLOCK(unp);                            \
1452                         error = EINVAL;                                 \
1453                         break;                                          \
1454                 }                                                       \
1455                 unp->unp_flags |= (bit);                                \
1456         } else                                                          \
1457                 unp->unp_flags &= ~(bit);                               \
1458         UNP_PCB_UNLOCK(unp);                                            \
1459 } while (0)
1460
1461                         switch (sopt->sopt_name) {
1462                         case LOCAL_CREDS:
1463                                 OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS);
1464                                 break;
1465
1466                         case LOCAL_CREDS_PERSISTENT:
1467                                 OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT);
1468                                 break;
1469
1470                         case LOCAL_CONNWAIT:
1471                                 OPTSET(UNP_CONNWAIT, 0);
1472                                 break;
1473
1474                         default:
1475                                 break;
1476                         }
1477                         break;
1478 #undef  OPTSET
1479                 default:
1480                         error = ENOPROTOOPT;
1481                         break;
1482                 }
1483                 break;
1484
1485         default:
1486                 error = EOPNOTSUPP;
1487                 break;
1488         }
1489         return (error);
1490 }
1491
1492 static int
1493 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1494 {
1495
1496         return (unp_connectat(AT_FDCWD, so, nam, td));
1497 }
1498
1499 static int
1500 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1501     struct thread *td)
1502 {
1503         struct mtx *vplock;
1504         struct sockaddr_un *soun;
1505         struct vnode *vp;
1506         struct socket *so2;
1507         struct unpcb *unp, *unp2, *unp3;
1508         struct nameidata nd;
1509         char buf[SOCK_MAXADDRLEN];
1510         struct sockaddr *sa;
1511         cap_rights_t rights;
1512         int error, len;
1513         bool connreq;
1514
1515         if (nam->sa_family != AF_UNIX)
1516                 return (EAFNOSUPPORT);
1517         if (nam->sa_len > sizeof(struct sockaddr_un))
1518                 return (EINVAL);
1519         len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1520         if (len <= 0)
1521                 return (EINVAL);
1522         soun = (struct sockaddr_un *)nam;
1523         bcopy(soun->sun_path, buf, len);
1524         buf[len] = 0;
1525
1526         unp = sotounpcb(so);
1527         UNP_PCB_LOCK(unp);
1528         for (;;) {
1529                 /*
1530                  * Wait for connection state to stabilize.  If a connection
1531                  * already exists, give up.  For datagram sockets, which permit
1532                  * multiple consecutive connect(2) calls, upper layers are
1533                  * responsible for disconnecting in advance of a subsequent
1534                  * connect(2), but this is not synchronized with PCB connection
1535                  * state.
1536                  *
1537                  * Also make sure that no threads are currently attempting to
1538                  * lock the peer socket, to ensure that unp_conn cannot
1539                  * transition between two valid sockets while locks are dropped.
1540                  */
1541                 if (unp->unp_conn != NULL) {
1542                         UNP_PCB_UNLOCK(unp);
1543                         return (EISCONN);
1544                 }
1545                 if ((unp->unp_flags & UNP_CONNECTING) != 0) {
1546                         UNP_PCB_UNLOCK(unp);
1547                         return (EALREADY);
1548                 }
1549                 if (unp->unp_pairbusy > 0) {
1550                         unp->unp_flags |= UNP_WAITING;
1551                         mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0);
1552                         continue;
1553                 }
1554                 break;
1555         }
1556         unp->unp_flags |= UNP_CONNECTING;
1557         UNP_PCB_UNLOCK(unp);
1558
1559         connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0;
1560         if (connreq)
1561                 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1562         else
1563                 sa = NULL;
1564         NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1565             UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT),
1566             td);
1567         error = namei(&nd);
1568         if (error)
1569                 vp = NULL;
1570         else
1571                 vp = nd.ni_vp;
1572         ASSERT_VOP_LOCKED(vp, "unp_connect");
1573         NDFREE_NOTHING(&nd);
1574         if (error)
1575                 goto bad;
1576
1577         if (vp->v_type != VSOCK) {
1578                 error = ENOTSOCK;
1579                 goto bad;
1580         }
1581 #ifdef MAC
1582         error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1583         if (error)
1584                 goto bad;
1585 #endif
1586         error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1587         if (error)
1588                 goto bad;
1589
1590         unp = sotounpcb(so);
1591         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1592
1593         vplock = mtx_pool_find(mtxpool_sleep, vp);
1594         mtx_lock(vplock);
1595         VOP_UNP_CONNECT(vp, &unp2);
1596         if (unp2 == NULL) {
1597                 error = ECONNREFUSED;
1598                 goto bad2;
1599         }
1600         so2 = unp2->unp_socket;
1601         if (so->so_type != so2->so_type) {
1602                 error = EPROTOTYPE;
1603                 goto bad2;
1604         }
1605         if (connreq) {
1606                 if (SOLISTENING(so2)) {
1607                         CURVNET_SET(so2->so_vnet);
1608                         so2 = sonewconn(so2, 0);
1609                         CURVNET_RESTORE();
1610                 } else
1611                         so2 = NULL;
1612                 if (so2 == NULL) {
1613                         error = ECONNREFUSED;
1614                         goto bad2;
1615                 }
1616                 unp3 = sotounpcb(so2);
1617                 unp_pcb_lock_pair(unp2, unp3);
1618                 if (unp2->unp_addr != NULL) {
1619                         bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1620                         unp3->unp_addr = (struct sockaddr_un *) sa;
1621                         sa = NULL;
1622                 }
1623
1624                 unp_copy_peercred(td, unp3, unp, unp2);
1625
1626                 UNP_PCB_UNLOCK(unp2);
1627                 unp2 = unp3;
1628
1629                 /*
1630                  * It is safe to block on the PCB lock here since unp2 is
1631                  * nascent and cannot be connected to any other sockets.
1632                  */
1633                 UNP_PCB_LOCK(unp);
1634 #ifdef MAC
1635                 mac_socketpeer_set_from_socket(so, so2);
1636                 mac_socketpeer_set_from_socket(so2, so);
1637 #endif
1638         } else {
1639                 unp_pcb_lock_pair(unp, unp2);
1640         }
1641         KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 &&
1642             sotounpcb(so2) == unp2,
1643             ("%s: unp2 %p so2 %p", __func__, unp2, so2));
1644         error = unp_connect2(so, so2, PRU_CONNECT);
1645         unp_pcb_unlock_pair(unp, unp2);
1646 bad2:
1647         mtx_unlock(vplock);
1648 bad:
1649         if (vp != NULL) {
1650                 vput(vp);
1651         }
1652         free(sa, M_SONAME);
1653         UNP_PCB_LOCK(unp);
1654         KASSERT((unp->unp_flags & UNP_CONNECTING) != 0,
1655             ("%s: unp %p has UNP_CONNECTING clear", __func__, unp));
1656         unp->unp_flags &= ~UNP_CONNECTING;
1657         UNP_PCB_UNLOCK(unp);
1658         return (error);
1659 }
1660
1661 /*
1662  * Set socket peer credentials at connection time.
1663  *
1664  * The client's PCB credentials are copied from its process structure.  The
1665  * server's PCB credentials are copied from the socket on which it called
1666  * listen(2).  uipc_listen cached that process's credentials at the time.
1667  */
1668 void
1669 unp_copy_peercred(struct thread *td, struct unpcb *client_unp,
1670     struct unpcb *server_unp, struct unpcb *listen_unp)
1671 {
1672         cru2xt(td, &client_unp->unp_peercred);
1673         client_unp->unp_flags |= UNP_HAVEPC;
1674
1675         memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred,
1676             sizeof(server_unp->unp_peercred));
1677         server_unp->unp_flags |= UNP_HAVEPC;
1678         client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK);
1679 }
1680
1681 static int
1682 unp_connect2(struct socket *so, struct socket *so2, int req)
1683 {
1684         struct unpcb *unp;
1685         struct unpcb *unp2;
1686
1687         unp = sotounpcb(so);
1688         KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1689         unp2 = sotounpcb(so2);
1690         KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1691
1692         UNP_PCB_LOCK_ASSERT(unp);
1693         UNP_PCB_LOCK_ASSERT(unp2);
1694         KASSERT(unp->unp_conn == NULL,
1695             ("%s: socket %p is already connected", __func__, unp));
1696
1697         if (so2->so_type != so->so_type)
1698                 return (EPROTOTYPE);
1699         unp->unp_conn = unp2;
1700         unp_pcb_hold(unp2);
1701         unp_pcb_hold(unp);
1702         switch (so->so_type) {
1703         case SOCK_DGRAM:
1704                 UNP_REF_LIST_LOCK();
1705                 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1706                 UNP_REF_LIST_UNLOCK();
1707                 soisconnected(so);
1708                 break;
1709
1710         case SOCK_STREAM:
1711         case SOCK_SEQPACKET:
1712                 KASSERT(unp2->unp_conn == NULL,
1713                     ("%s: socket %p is already connected", __func__, unp2));
1714                 unp2->unp_conn = unp;
1715                 if (req == PRU_CONNECT &&
1716                     ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1717                         soisconnecting(so);
1718                 else
1719                         soisconnected(so);
1720                 soisconnected(so2);
1721                 break;
1722
1723         default:
1724                 panic("unp_connect2");
1725         }
1726         return (0);
1727 }
1728
1729 static void
1730 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1731 {
1732         struct socket *so, *so2;
1733 #ifdef INVARIANTS
1734         struct unpcb *unptmp;
1735 #endif
1736
1737         UNP_PCB_LOCK_ASSERT(unp);
1738         UNP_PCB_LOCK_ASSERT(unp2);
1739         KASSERT(unp->unp_conn == unp2,
1740             ("%s: unpcb %p is not connected to %p", __func__, unp, unp2));
1741
1742         unp->unp_conn = NULL;
1743         so = unp->unp_socket;
1744         so2 = unp2->unp_socket;
1745         switch (unp->unp_socket->so_type) {
1746         case SOCK_DGRAM:
1747                 UNP_REF_LIST_LOCK();
1748 #ifdef INVARIANTS
1749                 LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) {
1750                         if (unptmp == unp)
1751                                 break;
1752                 }
1753                 KASSERT(unptmp != NULL,
1754                     ("%s: %p not found in reflist of %p", __func__, unp, unp2));
1755 #endif
1756                 LIST_REMOVE(unp, unp_reflink);
1757                 UNP_REF_LIST_UNLOCK();
1758                 if (so) {
1759                         SOCK_LOCK(so);
1760                         so->so_state &= ~SS_ISCONNECTED;
1761                         SOCK_UNLOCK(so);
1762                 }
1763                 break;
1764
1765         case SOCK_STREAM:
1766         case SOCK_SEQPACKET:
1767                 if (so)
1768                         soisdisconnected(so);
1769                 MPASS(unp2->unp_conn == unp);
1770                 unp2->unp_conn = NULL;
1771                 if (so2)
1772                         soisdisconnected(so2);
1773                 break;
1774         }
1775
1776         if (unp == unp2) {
1777                 unp_pcb_rele_notlast(unp);
1778                 if (!unp_pcb_rele(unp))
1779                         UNP_PCB_UNLOCK(unp);
1780         } else {
1781                 if (!unp_pcb_rele(unp))
1782                         UNP_PCB_UNLOCK(unp);
1783                 if (!unp_pcb_rele(unp2))
1784                         UNP_PCB_UNLOCK(unp2);
1785         }
1786 }
1787
1788 /*
1789  * unp_pcblist() walks the global list of struct unpcb's to generate a
1790  * pointer list, bumping the refcount on each unpcb.  It then copies them out
1791  * sequentially, validating the generation number on each to see if it has
1792  * been detached.  All of this is necessary because copyout() may sleep on
1793  * disk I/O.
1794  */
1795 static int
1796 unp_pcblist(SYSCTL_HANDLER_ARGS)
1797 {
1798         struct unpcb *unp, **unp_list;
1799         unp_gen_t gencnt;
1800         struct xunpgen *xug;
1801         struct unp_head *head;
1802         struct xunpcb *xu;
1803         u_int i;
1804         int error, n;
1805
1806         switch ((intptr_t)arg1) {
1807         case SOCK_STREAM:
1808                 head = &unp_shead;
1809                 break;
1810
1811         case SOCK_DGRAM:
1812                 head = &unp_dhead;
1813                 break;
1814
1815         case SOCK_SEQPACKET:
1816                 head = &unp_sphead;
1817                 break;
1818
1819         default:
1820                 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1821         }
1822
1823         /*
1824          * The process of preparing the PCB list is too time-consuming and
1825          * resource-intensive to repeat twice on every request.
1826          */
1827         if (req->oldptr == NULL) {
1828                 n = unp_count;
1829                 req->oldidx = 2 * (sizeof *xug)
1830                         + (n + n/8) * sizeof(struct xunpcb);
1831                 return (0);
1832         }
1833
1834         if (req->newptr != NULL)
1835                 return (EPERM);
1836
1837         /*
1838          * OK, now we're committed to doing something.
1839          */
1840         xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO);
1841         UNP_LINK_RLOCK();
1842         gencnt = unp_gencnt;
1843         n = unp_count;
1844         UNP_LINK_RUNLOCK();
1845
1846         xug->xug_len = sizeof *xug;
1847         xug->xug_count = n;
1848         xug->xug_gen = gencnt;
1849         xug->xug_sogen = so_gencnt;
1850         error = SYSCTL_OUT(req, xug, sizeof *xug);
1851         if (error) {
1852                 free(xug, M_TEMP);
1853                 return (error);
1854         }
1855
1856         unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1857
1858         UNP_LINK_RLOCK();
1859         for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1860              unp = LIST_NEXT(unp, unp_link)) {
1861                 UNP_PCB_LOCK(unp);
1862                 if (unp->unp_gencnt <= gencnt) {
1863                         if (cr_cansee(req->td->td_ucred,
1864                             unp->unp_socket->so_cred)) {
1865                                 UNP_PCB_UNLOCK(unp);
1866                                 continue;
1867                         }
1868                         unp_list[i++] = unp;
1869                         unp_pcb_hold(unp);
1870                 }
1871                 UNP_PCB_UNLOCK(unp);
1872         }
1873         UNP_LINK_RUNLOCK();
1874         n = i;                  /* In case we lost some during malloc. */
1875
1876         error = 0;
1877         xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1878         for (i = 0; i < n; i++) {
1879                 unp = unp_list[i];
1880                 UNP_PCB_LOCK(unp);
1881                 if (unp_pcb_rele(unp))
1882                         continue;
1883
1884                 if (unp->unp_gencnt <= gencnt) {
1885                         xu->xu_len = sizeof *xu;
1886                         xu->xu_unpp = (uintptr_t)unp;
1887                         /*
1888                          * XXX - need more locking here to protect against
1889                          * connect/disconnect races for SMP.
1890                          */
1891                         if (unp->unp_addr != NULL)
1892                                 bcopy(unp->unp_addr, &xu->xu_addr,
1893                                       unp->unp_addr->sun_len);
1894                         else
1895                                 bzero(&xu->xu_addr, sizeof(xu->xu_addr));
1896                         if (unp->unp_conn != NULL &&
1897                             unp->unp_conn->unp_addr != NULL)
1898                                 bcopy(unp->unp_conn->unp_addr,
1899                                       &xu->xu_caddr,
1900                                       unp->unp_conn->unp_addr->sun_len);
1901                         else
1902                                 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr));
1903                         xu->unp_vnode = (uintptr_t)unp->unp_vnode;
1904                         xu->unp_conn = (uintptr_t)unp->unp_conn;
1905                         xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs);
1906                         xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink);
1907                         xu->unp_gencnt = unp->unp_gencnt;
1908                         sotoxsocket(unp->unp_socket, &xu->xu_socket);
1909                         UNP_PCB_UNLOCK(unp);
1910                         error = SYSCTL_OUT(req, xu, sizeof *xu);
1911                 } else {
1912                         UNP_PCB_UNLOCK(unp);
1913                 }
1914         }
1915         free(xu, M_TEMP);
1916         if (!error) {
1917                 /*
1918                  * Give the user an updated idea of our state.  If the
1919                  * generation differs from what we told her before, she knows
1920                  * that something happened while we were processing this
1921                  * request, and it might be necessary to retry.
1922                  */
1923                 xug->xug_gen = unp_gencnt;
1924                 xug->xug_sogen = so_gencnt;
1925                 xug->xug_count = unp_count;
1926                 error = SYSCTL_OUT(req, xug, sizeof *xug);
1927         }
1928         free(unp_list, M_TEMP);
1929         free(xug, M_TEMP);
1930         return (error);
1931 }
1932
1933 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist,
1934     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1935     (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1936     "List of active local datagram sockets");
1937 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist,
1938     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1939     (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1940     "List of active local stream sockets");
1941 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1942     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1943     (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1944     "List of active local seqpacket sockets");
1945
1946 static void
1947 unp_shutdown(struct unpcb *unp)
1948 {
1949         struct unpcb *unp2;
1950         struct socket *so;
1951
1952         UNP_PCB_LOCK_ASSERT(unp);
1953
1954         unp2 = unp->unp_conn;
1955         if ((unp->unp_socket->so_type == SOCK_STREAM ||
1956             (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1957                 so = unp2->unp_socket;
1958                 if (so != NULL)
1959                         socantrcvmore(so);
1960         }
1961 }
1962
1963 static void
1964 unp_drop(struct unpcb *unp)
1965 {
1966         struct socket *so = unp->unp_socket;
1967         struct unpcb *unp2;
1968
1969         /*
1970          * Regardless of whether the socket's peer dropped the connection
1971          * with this socket by aborting or disconnecting, POSIX requires
1972          * that ECONNRESET is returned.
1973          */
1974
1975         UNP_PCB_LOCK(unp);
1976         if (so)
1977                 so->so_error = ECONNRESET;
1978         if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) {
1979                 /* Last reference dropped in unp_disconnect(). */
1980                 unp_pcb_rele_notlast(unp);
1981                 unp_disconnect(unp, unp2);
1982         } else if (!unp_pcb_rele(unp)) {
1983                 UNP_PCB_UNLOCK(unp);
1984         }
1985 }
1986
1987 static void
1988 unp_freerights(struct filedescent **fdep, int fdcount)
1989 {
1990         struct file *fp;
1991         int i;
1992
1993         KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1994
1995         for (i = 0; i < fdcount; i++) {
1996                 fp = fdep[i]->fde_file;
1997                 filecaps_free(&fdep[i]->fde_caps);
1998                 unp_discard(fp);
1999         }
2000         free(fdep[0], M_FILECAPS);
2001 }
2002
2003 static int
2004 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
2005 {
2006         struct thread *td = curthread;          /* XXX */
2007         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
2008         int i;
2009         int *fdp;
2010         struct filedesc *fdesc = td->td_proc->p_fd;
2011         struct filedescent **fdep;
2012         void *data;
2013         socklen_t clen = control->m_len, datalen;
2014         int error, newfds;
2015         u_int newlen;
2016
2017         UNP_LINK_UNLOCK_ASSERT();
2018
2019         error = 0;
2020         if (controlp != NULL) /* controlp == NULL => free control messages */
2021                 *controlp = NULL;
2022         while (cm != NULL) {
2023                 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
2024                         error = EINVAL;
2025                         break;
2026                 }
2027                 data = CMSG_DATA(cm);
2028                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
2029                 if (cm->cmsg_level == SOL_SOCKET
2030                     && cm->cmsg_type == SCM_RIGHTS) {
2031                         newfds = datalen / sizeof(*fdep);
2032                         if (newfds == 0)
2033                                 goto next;
2034                         fdep = data;
2035
2036                         /* If we're not outputting the descriptors free them. */
2037                         if (error || controlp == NULL) {
2038                                 unp_freerights(fdep, newfds);
2039                                 goto next;
2040                         }
2041                         FILEDESC_XLOCK(fdesc);
2042
2043                         /*
2044                          * Now change each pointer to an fd in the global
2045                          * table to an integer that is the index to the local
2046                          * fd table entry that we set up to point to the
2047                          * global one we are transferring.
2048                          */
2049                         newlen = newfds * sizeof(int);
2050                         *controlp = sbcreatecontrol(NULL, newlen,
2051                             SCM_RIGHTS, SOL_SOCKET);
2052                         if (*controlp == NULL) {
2053                                 FILEDESC_XUNLOCK(fdesc);
2054                                 error = E2BIG;
2055                                 unp_freerights(fdep, newfds);
2056                                 goto next;
2057                         }
2058
2059                         fdp = (int *)
2060                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2061                         if (fdallocn(td, 0, fdp, newfds) != 0) {
2062                                 FILEDESC_XUNLOCK(fdesc);
2063                                 error = EMSGSIZE;
2064                                 unp_freerights(fdep, newfds);
2065                                 m_freem(*controlp);
2066                                 *controlp = NULL;
2067                                 goto next;
2068                         }
2069                         for (i = 0; i < newfds; i++, fdp++) {
2070                                 _finstall(fdesc, fdep[i]->fde_file, *fdp,
2071                                     (flags & MSG_CMSG_CLOEXEC) != 0 ? O_CLOEXEC : 0,
2072                                     &fdep[i]->fde_caps);
2073                                 unp_externalize_fp(fdep[i]->fde_file);
2074                         }
2075
2076                         /*
2077                          * The new type indicates that the mbuf data refers to
2078                          * kernel resources that may need to be released before
2079                          * the mbuf is freed.
2080                          */
2081                         m_chtype(*controlp, MT_EXTCONTROL);
2082                         FILEDESC_XUNLOCK(fdesc);
2083                         free(fdep[0], M_FILECAPS);
2084                 } else {
2085                         /* We can just copy anything else across. */
2086                         if (error || controlp == NULL)
2087                                 goto next;
2088                         *controlp = sbcreatecontrol(NULL, datalen,
2089                             cm->cmsg_type, cm->cmsg_level);
2090                         if (*controlp == NULL) {
2091                                 error = ENOBUFS;
2092                                 goto next;
2093                         }
2094                         bcopy(data,
2095                             CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
2096                             datalen);
2097                 }
2098                 controlp = &(*controlp)->m_next;
2099
2100 next:
2101                 if (CMSG_SPACE(datalen) < clen) {
2102                         clen -= CMSG_SPACE(datalen);
2103                         cm = (struct cmsghdr *)
2104                             ((caddr_t)cm + CMSG_SPACE(datalen));
2105                 } else {
2106                         clen = 0;
2107                         cm = NULL;
2108                 }
2109         }
2110
2111         m_freem(control);
2112         return (error);
2113 }
2114
2115 static void
2116 unp_zone_change(void *tag)
2117 {
2118
2119         uma_zone_set_max(unp_zone, maxsockets);
2120 }
2121
2122 #ifdef INVARIANTS
2123 static void
2124 unp_zdtor(void *mem, int size __unused, void *arg __unused)
2125 {
2126         struct unpcb *unp;
2127
2128         unp = mem;
2129
2130         KASSERT(LIST_EMPTY(&unp->unp_refs),
2131             ("%s: unpcb %p has lingering refs", __func__, unp));
2132         KASSERT(unp->unp_socket == NULL,
2133             ("%s: unpcb %p has socket backpointer", __func__, unp));
2134         KASSERT(unp->unp_vnode == NULL,
2135             ("%s: unpcb %p has vnode references", __func__, unp));
2136         KASSERT(unp->unp_conn == NULL,
2137             ("%s: unpcb %p is still connected", __func__, unp));
2138         KASSERT(unp->unp_addr == NULL,
2139             ("%s: unpcb %p has leaked addr", __func__, unp));
2140 }
2141 #endif
2142
2143 static void
2144 unp_init(void)
2145 {
2146         uma_dtor dtor;
2147
2148 #ifdef VIMAGE
2149         if (!IS_DEFAULT_VNET(curvnet))
2150                 return;
2151 #endif
2152
2153 #ifdef INVARIANTS
2154         dtor = unp_zdtor;
2155 #else
2156         dtor = NULL;
2157 #endif
2158         unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor,
2159             NULL, NULL, UMA_ALIGN_CACHE, 0);
2160         uma_zone_set_max(unp_zone, maxsockets);
2161         uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
2162         EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
2163             NULL, EVENTHANDLER_PRI_ANY);
2164         LIST_INIT(&unp_dhead);
2165         LIST_INIT(&unp_shead);
2166         LIST_INIT(&unp_sphead);
2167         SLIST_INIT(&unp_defers);
2168         TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
2169         TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
2170         UNP_LINK_LOCK_INIT();
2171         UNP_DEFERRED_LOCK_INIT();
2172 }
2173
2174 static void
2175 unp_internalize_cleanup_rights(struct mbuf *control)
2176 {
2177         struct cmsghdr *cp;
2178         struct mbuf *m;
2179         void *data;
2180         socklen_t datalen;
2181
2182         for (m = control; m != NULL; m = m->m_next) {
2183                 cp = mtod(m, struct cmsghdr *);
2184                 if (cp->cmsg_level != SOL_SOCKET ||
2185                     cp->cmsg_type != SCM_RIGHTS)
2186                         continue;
2187                 data = CMSG_DATA(cp);
2188                 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data;
2189                 unp_freerights(data, datalen / sizeof(struct filedesc *));
2190         }
2191 }
2192
2193 static int
2194 unp_internalize(struct mbuf **controlp, struct thread *td)
2195 {
2196         struct mbuf *control, **initial_controlp;
2197         struct proc *p;
2198         struct filedesc *fdesc;
2199         struct bintime *bt;
2200         struct cmsghdr *cm;
2201         struct cmsgcred *cmcred;
2202         struct filedescent *fde, **fdep, *fdev;
2203         struct file *fp;
2204         struct timeval *tv;
2205         struct timespec *ts;
2206         void *data;
2207         socklen_t clen, datalen;
2208         int i, j, error, *fdp, oldfds;
2209         u_int newlen;
2210
2211         UNP_LINK_UNLOCK_ASSERT();
2212
2213         p = td->td_proc;
2214         fdesc = p->p_fd;
2215         error = 0;
2216         control = *controlp;
2217         clen = control->m_len;
2218         *controlp = NULL;
2219         initial_controlp = controlp;
2220         for (cm = mtod(control, struct cmsghdr *); cm != NULL;) {
2221                 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
2222                     || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
2223                         error = EINVAL;
2224                         goto out;
2225                 }
2226                 data = CMSG_DATA(cm);
2227                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
2228
2229                 switch (cm->cmsg_type) {
2230                 /*
2231                  * Fill in credential information.
2232                  */
2233                 case SCM_CREDS:
2234                         *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
2235                             SCM_CREDS, SOL_SOCKET);
2236                         if (*controlp == NULL) {
2237                                 error = ENOBUFS;
2238                                 goto out;
2239                         }
2240                         cmcred = (struct cmsgcred *)
2241                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2242                         cmcred->cmcred_pid = p->p_pid;
2243                         cmcred->cmcred_uid = td->td_ucred->cr_ruid;
2244                         cmcred->cmcred_gid = td->td_ucred->cr_rgid;
2245                         cmcred->cmcred_euid = td->td_ucred->cr_uid;
2246                         cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
2247                             CMGROUP_MAX);
2248                         for (i = 0; i < cmcred->cmcred_ngroups; i++)
2249                                 cmcred->cmcred_groups[i] =
2250                                     td->td_ucred->cr_groups[i];
2251                         break;
2252
2253                 case SCM_RIGHTS:
2254                         oldfds = datalen / sizeof (int);
2255                         if (oldfds == 0)
2256                                 break;
2257                         /*
2258                          * Check that all the FDs passed in refer to legal
2259                          * files.  If not, reject the entire operation.
2260                          */
2261                         fdp = data;
2262                         FILEDESC_SLOCK(fdesc);
2263                         for (i = 0; i < oldfds; i++, fdp++) {
2264                                 fp = fget_locked(fdesc, *fdp);
2265                                 if (fp == NULL) {
2266                                         FILEDESC_SUNLOCK(fdesc);
2267                                         error = EBADF;
2268                                         goto out;
2269                                 }
2270                                 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
2271                                         FILEDESC_SUNLOCK(fdesc);
2272                                         error = EOPNOTSUPP;
2273                                         goto out;
2274                                 }
2275                         }
2276
2277                         /*
2278                          * Now replace the integer FDs with pointers to the
2279                          * file structure and capability rights.
2280                          */
2281                         newlen = oldfds * sizeof(fdep[0]);
2282                         *controlp = sbcreatecontrol(NULL, newlen,
2283                             SCM_RIGHTS, SOL_SOCKET);
2284                         if (*controlp == NULL) {
2285                                 FILEDESC_SUNLOCK(fdesc);
2286                                 error = E2BIG;
2287                                 goto out;
2288                         }
2289                         fdp = data;
2290                         for (i = 0; i < oldfds; i++, fdp++) {
2291                                 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) {
2292                                         fdp = data;
2293                                         for (j = 0; j < i; j++, fdp++) {
2294                                                 fdrop(fdesc->fd_ofiles[*fdp].
2295                                                     fde_file, td);
2296                                         }
2297                                         FILEDESC_SUNLOCK(fdesc);
2298                                         error = EBADF;
2299                                         goto out;
2300                                 }
2301                         }
2302                         fdp = data;
2303                         fdep = (struct filedescent **)
2304                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2305                         fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
2306                             M_WAITOK);
2307                         for (i = 0; i < oldfds; i++, fdev++, fdp++) {
2308                                 fde = &fdesc->fd_ofiles[*fdp];
2309                                 fdep[i] = fdev;
2310                                 fdep[i]->fde_file = fde->fde_file;
2311                                 filecaps_copy(&fde->fde_caps,
2312                                     &fdep[i]->fde_caps, true);
2313                                 unp_internalize_fp(fdep[i]->fde_file);
2314                         }
2315                         FILEDESC_SUNLOCK(fdesc);
2316                         break;
2317
2318                 case SCM_TIMESTAMP:
2319                         *controlp = sbcreatecontrol(NULL, sizeof(*tv),
2320                             SCM_TIMESTAMP, SOL_SOCKET);
2321                         if (*controlp == NULL) {
2322                                 error = ENOBUFS;
2323                                 goto out;
2324                         }
2325                         tv = (struct timeval *)
2326                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2327                         microtime(tv);
2328                         break;
2329
2330                 case SCM_BINTIME:
2331                         *controlp = sbcreatecontrol(NULL, sizeof(*bt),
2332                             SCM_BINTIME, SOL_SOCKET);
2333                         if (*controlp == NULL) {
2334                                 error = ENOBUFS;
2335                                 goto out;
2336                         }
2337                         bt = (struct bintime *)
2338                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2339                         bintime(bt);
2340                         break;
2341
2342                 case SCM_REALTIME:
2343                         *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2344                             SCM_REALTIME, SOL_SOCKET);
2345                         if (*controlp == NULL) {
2346                                 error = ENOBUFS;
2347                                 goto out;
2348                         }
2349                         ts = (struct timespec *)
2350                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2351                         nanotime(ts);
2352                         break;
2353
2354                 case SCM_MONOTONIC:
2355                         *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2356                             SCM_MONOTONIC, SOL_SOCKET);
2357                         if (*controlp == NULL) {
2358                                 error = ENOBUFS;
2359                                 goto out;
2360                         }
2361                         ts = (struct timespec *)
2362                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2363                         nanouptime(ts);
2364                         break;
2365
2366                 default:
2367                         error = EINVAL;
2368                         goto out;
2369                 }
2370
2371                 if (*controlp != NULL)
2372                         controlp = &(*controlp)->m_next;
2373                 if (CMSG_SPACE(datalen) < clen) {
2374                         clen -= CMSG_SPACE(datalen);
2375                         cm = (struct cmsghdr *)
2376                             ((caddr_t)cm + CMSG_SPACE(datalen));
2377                 } else {
2378                         clen = 0;
2379                         cm = NULL;
2380                 }
2381         }
2382
2383 out:
2384         if (error != 0 && initial_controlp != NULL)
2385                 unp_internalize_cleanup_rights(*initial_controlp);
2386         m_freem(control);
2387         return (error);
2388 }
2389
2390 static struct mbuf *
2391 unp_addsockcred(struct thread *td, struct mbuf *control, int mode)
2392 {
2393         struct mbuf *m, *n, *n_prev;
2394         const struct cmsghdr *cm;
2395         int ngroups, i, cmsgtype;
2396         size_t ctrlsz;
2397
2398         ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2399         if (mode & UNP_WANTCRED_ALWAYS) {
2400                 ctrlsz = SOCKCRED2SIZE(ngroups);
2401                 cmsgtype = SCM_CREDS2;
2402         } else {
2403                 ctrlsz = SOCKCREDSIZE(ngroups);
2404                 cmsgtype = SCM_CREDS;
2405         }
2406
2407         m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET);
2408         if (m == NULL)
2409                 return (control);
2410
2411         if (mode & UNP_WANTCRED_ALWAYS) {
2412                 struct sockcred2 *sc;
2413
2414                 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *));
2415                 sc->sc_version = 0;
2416                 sc->sc_pid = td->td_proc->p_pid;
2417                 sc->sc_uid = td->td_ucred->cr_ruid;
2418                 sc->sc_euid = td->td_ucred->cr_uid;
2419                 sc->sc_gid = td->td_ucred->cr_rgid;
2420                 sc->sc_egid = td->td_ucred->cr_gid;
2421                 sc->sc_ngroups = ngroups;
2422                 for (i = 0; i < sc->sc_ngroups; i++)
2423                         sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2424         } else {
2425                 struct sockcred *sc;
2426
2427                 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *));
2428                 sc->sc_uid = td->td_ucred->cr_ruid;
2429                 sc->sc_euid = td->td_ucred->cr_uid;
2430                 sc->sc_gid = td->td_ucred->cr_rgid;
2431                 sc->sc_egid = td->td_ucred->cr_gid;
2432                 sc->sc_ngroups = ngroups;
2433                 for (i = 0; i < sc->sc_ngroups; i++)
2434                         sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2435         }
2436
2437         /*
2438          * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2439          * created SCM_CREDS control message (struct sockcred) has another
2440          * format.
2441          */
2442         if (control != NULL && cmsgtype == SCM_CREDS)
2443                 for (n = control, n_prev = NULL; n != NULL;) {
2444                         cm = mtod(n, struct cmsghdr *);
2445                         if (cm->cmsg_level == SOL_SOCKET &&
2446                             cm->cmsg_type == SCM_CREDS) {
2447                                 if (n_prev == NULL)
2448                                         control = n->m_next;
2449                                 else
2450                                         n_prev->m_next = n->m_next;
2451                                 n = m_free(n);
2452                         } else {
2453                                 n_prev = n;
2454                                 n = n->m_next;
2455                         }
2456                 }
2457
2458         /* Prepend it to the head. */
2459         m->m_next = control;
2460         return (m);
2461 }
2462
2463 static struct unpcb *
2464 fptounp(struct file *fp)
2465 {
2466         struct socket *so;
2467
2468         if (fp->f_type != DTYPE_SOCKET)
2469                 return (NULL);
2470         if ((so = fp->f_data) == NULL)
2471                 return (NULL);
2472         if (so->so_proto->pr_domain != &localdomain)
2473                 return (NULL);
2474         return sotounpcb(so);
2475 }
2476
2477 static void
2478 unp_discard(struct file *fp)
2479 {
2480         struct unp_defer *dr;
2481
2482         if (unp_externalize_fp(fp)) {
2483                 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2484                 dr->ud_fp = fp;
2485                 UNP_DEFERRED_LOCK();
2486                 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2487                 UNP_DEFERRED_UNLOCK();
2488                 atomic_add_int(&unp_defers_count, 1);
2489                 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2490         } else
2491                 closef_nothread(fp);
2492 }
2493
2494 static void
2495 unp_process_defers(void *arg __unused, int pending)
2496 {
2497         struct unp_defer *dr;
2498         SLIST_HEAD(, unp_defer) drl;
2499         int count;
2500
2501         SLIST_INIT(&drl);
2502         for (;;) {
2503                 UNP_DEFERRED_LOCK();
2504                 if (SLIST_FIRST(&unp_defers) == NULL) {
2505                         UNP_DEFERRED_UNLOCK();
2506                         break;
2507                 }
2508                 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2509                 UNP_DEFERRED_UNLOCK();
2510                 count = 0;
2511                 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2512                         SLIST_REMOVE_HEAD(&drl, ud_link);
2513                         closef_nothread(dr->ud_fp);
2514                         free(dr, M_TEMP);
2515                         count++;
2516                 }
2517                 atomic_add_int(&unp_defers_count, -count);
2518         }
2519 }
2520
2521 static void
2522 unp_internalize_fp(struct file *fp)
2523 {
2524         struct unpcb *unp;
2525
2526         UNP_LINK_WLOCK();
2527         if ((unp = fptounp(fp)) != NULL) {
2528                 unp->unp_file = fp;
2529                 unp->unp_msgcount++;
2530         }
2531         unp_rights++;
2532         UNP_LINK_WUNLOCK();
2533 }
2534
2535 static int
2536 unp_externalize_fp(struct file *fp)
2537 {
2538         struct unpcb *unp;
2539         int ret;
2540
2541         UNP_LINK_WLOCK();
2542         if ((unp = fptounp(fp)) != NULL) {
2543                 unp->unp_msgcount--;
2544                 ret = 1;
2545         } else
2546                 ret = 0;
2547         unp_rights--;
2548         UNP_LINK_WUNLOCK();
2549         return (ret);
2550 }
2551
2552 /*
2553  * unp_defer indicates whether additional work has been defered for a future
2554  * pass through unp_gc().  It is thread local and does not require explicit
2555  * synchronization.
2556  */
2557 static int      unp_marked;
2558
2559 static void
2560 unp_remove_dead_ref(struct filedescent **fdep, int fdcount)
2561 {
2562         struct unpcb *unp;
2563         struct file *fp;
2564         int i;
2565
2566         /*
2567          * This function can only be called from the gc task.
2568          */
2569         KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0,
2570             ("%s: not on gc callout", __func__));
2571         UNP_LINK_LOCK_ASSERT();
2572
2573         for (i = 0; i < fdcount; i++) {
2574                 fp = fdep[i]->fde_file;
2575                 if ((unp = fptounp(fp)) == NULL)
2576                         continue;
2577                 if ((unp->unp_gcflag & UNPGC_DEAD) == 0)
2578                         continue;
2579                 unp->unp_gcrefs--;
2580         }
2581 }
2582
2583 static void
2584 unp_restore_undead_ref(struct filedescent **fdep, int fdcount)
2585 {
2586         struct unpcb *unp;
2587         struct file *fp;
2588         int i;
2589
2590         /*
2591          * This function can only be called from the gc task.
2592          */
2593         KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0,
2594             ("%s: not on gc callout", __func__));
2595         UNP_LINK_LOCK_ASSERT();
2596
2597         for (i = 0; i < fdcount; i++) {
2598                 fp = fdep[i]->fde_file;
2599                 if ((unp = fptounp(fp)) == NULL)
2600                         continue;
2601                 if ((unp->unp_gcflag & UNPGC_DEAD) == 0)
2602                         continue;
2603                 unp->unp_gcrefs++;
2604                 unp_marked++;
2605         }
2606 }
2607
2608 static void
2609 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int))
2610 {
2611         struct socket *so, *soa;
2612
2613         so = unp->unp_socket;
2614         SOCK_LOCK(so);
2615         if (SOLISTENING(so)) {
2616                 /*
2617                  * Mark all sockets in our accept queue.
2618                  */
2619                 TAILQ_FOREACH(soa, &so->sol_comp, so_list) {
2620                         if (sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS)
2621                                 continue;
2622                         SOCKBUF_LOCK(&soa->so_rcv);
2623                         unp_scan(soa->so_rcv.sb_mb, op);
2624                         SOCKBUF_UNLOCK(&soa->so_rcv);
2625                 }
2626         } else {
2627                 /*
2628                  * Mark all sockets we reference with RIGHTS.
2629                  */
2630                 if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2631                         SOCKBUF_LOCK(&so->so_rcv);
2632                         unp_scan(so->so_rcv.sb_mb, op);
2633                         SOCKBUF_UNLOCK(&so->so_rcv);
2634                 }
2635         }
2636         SOCK_UNLOCK(so);
2637 }
2638
2639 static int unp_recycled;
2640 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2641     "Number of unreachable sockets claimed by the garbage collector.");
2642
2643 static int unp_taskcount;
2644 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2645     "Number of times the garbage collector has run.");
2646
2647 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0,
2648     "Number of active local sockets.");
2649
2650 static void
2651 unp_gc(__unused void *arg, int pending)
2652 {
2653         struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2654                                     NULL };
2655         struct unp_head **head;
2656         struct unp_head unp_deadhead;   /* List of potentially-dead sockets. */
2657         struct file *f, **unref;
2658         struct unpcb *unp, *unptmp;
2659         int i, total, unp_unreachable;
2660
2661         LIST_INIT(&unp_deadhead);
2662         unp_taskcount++;
2663         UNP_LINK_RLOCK();
2664         /*
2665          * First determine which sockets may be in cycles.
2666          */
2667         unp_unreachable = 0;
2668
2669         for (head = heads; *head != NULL; head++)
2670                 LIST_FOREACH(unp, *head, unp_link) {
2671                         KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0,
2672                             ("%s: unp %p has unexpected gc flags 0x%x",
2673                             __func__, unp, (unsigned int)unp->unp_gcflag));
2674
2675                         f = unp->unp_file;
2676
2677                         /*
2678                          * Check for an unreachable socket potentially in a
2679                          * cycle.  It must be in a queue as indicated by
2680                          * msgcount, and this must equal the file reference
2681                          * count.  Note that when msgcount is 0 the file is
2682                          * NULL.
2683                          */
2684                         if (f != NULL && unp->unp_msgcount != 0 &&
2685                             refcount_load(&f->f_count) == unp->unp_msgcount) {
2686                                 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead);
2687                                 unp->unp_gcflag |= UNPGC_DEAD;
2688                                 unp->unp_gcrefs = unp->unp_msgcount;
2689                                 unp_unreachable++;
2690                         }
2691                 }
2692
2693         /*
2694          * Scan all sockets previously marked as potentially being in a cycle
2695          * and remove the references each socket holds on any UNPGC_DEAD
2696          * sockets in its queue.  After this step, all remaining references on
2697          * sockets marked UNPGC_DEAD should not be part of any cycle.
2698          */
2699         LIST_FOREACH(unp, &unp_deadhead, unp_dead)
2700                 unp_gc_scan(unp, unp_remove_dead_ref);
2701
2702         /*
2703          * If a socket still has a non-negative refcount, it cannot be in a
2704          * cycle.  In this case increment refcount of all children iteratively.
2705          * Stop the scan once we do a complete loop without discovering
2706          * a new reachable socket.
2707          */
2708         do {
2709                 unp_marked = 0;
2710                 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp)
2711                         if (unp->unp_gcrefs > 0) {
2712                                 unp->unp_gcflag &= ~UNPGC_DEAD;
2713                                 LIST_REMOVE(unp, unp_dead);
2714                                 KASSERT(unp_unreachable > 0,
2715                                     ("%s: unp_unreachable underflow.",
2716                                     __func__));
2717                                 unp_unreachable--;
2718                                 unp_gc_scan(unp, unp_restore_undead_ref);
2719                         }
2720         } while (unp_marked);
2721
2722         UNP_LINK_RUNLOCK();
2723
2724         if (unp_unreachable == 0)
2725                 return;
2726
2727         /*
2728          * Allocate space for a local array of dead unpcbs.
2729          * TODO: can this path be simplified by instead using the local
2730          * dead list at unp_deadhead, after taking out references
2731          * on the file object and/or unpcb and dropping the link lock?
2732          */
2733         unref = malloc(unp_unreachable * sizeof(struct file *),
2734             M_TEMP, M_WAITOK);
2735
2736         /*
2737          * Iterate looking for sockets which have been specifically marked
2738          * as unreachable and store them locally.
2739          */
2740         UNP_LINK_RLOCK();
2741         total = 0;
2742         LIST_FOREACH(unp, &unp_deadhead, unp_dead) {
2743                 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0,
2744                     ("%s: unp %p not marked UNPGC_DEAD", __func__, unp));
2745                 unp->unp_gcflag &= ~UNPGC_DEAD;
2746                 f = unp->unp_file;
2747                 if (unp->unp_msgcount == 0 || f == NULL ||
2748                     refcount_load(&f->f_count) != unp->unp_msgcount ||
2749                     !fhold(f))
2750                         continue;
2751                 unref[total++] = f;
2752                 KASSERT(total <= unp_unreachable,
2753                     ("%s: incorrect unreachable count.", __func__));
2754         }
2755         UNP_LINK_RUNLOCK();
2756
2757         /*
2758          * Now flush all sockets, free'ing rights.  This will free the
2759          * struct files associated with these sockets but leave each socket
2760          * with one remaining ref.
2761          */
2762         for (i = 0; i < total; i++) {
2763                 struct socket *so;
2764
2765                 so = unref[i]->f_data;
2766                 CURVNET_SET(so->so_vnet);
2767                 sorflush(so);
2768                 CURVNET_RESTORE();
2769         }
2770
2771         /*
2772          * And finally release the sockets so they can be reclaimed.
2773          */
2774         for (i = 0; i < total; i++)
2775                 fdrop(unref[i], NULL);
2776         unp_recycled += total;
2777         free(unref, M_TEMP);
2778 }
2779
2780 static void
2781 unp_dispose_mbuf(struct mbuf *m)
2782 {
2783
2784         if (m)
2785                 unp_scan(m, unp_freerights);
2786 }
2787
2788 /*
2789  * Synchronize against unp_gc, which can trip over data as we are freeing it.
2790  */
2791 static void
2792 unp_dispose(struct socket *so)
2793 {
2794         struct unpcb *unp;
2795
2796         unp = sotounpcb(so);
2797         UNP_LINK_WLOCK();
2798         unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2799         UNP_LINK_WUNLOCK();
2800         if (!SOLISTENING(so))
2801                 unp_dispose_mbuf(so->so_rcv.sb_mb);
2802 }
2803
2804 static void
2805 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2806 {
2807         struct mbuf *m;
2808         struct cmsghdr *cm;
2809         void *data;
2810         socklen_t clen, datalen;
2811
2812         while (m0 != NULL) {
2813                 for (m = m0; m; m = m->m_next) {
2814                         if (m->m_type != MT_CONTROL)
2815                                 continue;
2816
2817                         cm = mtod(m, struct cmsghdr *);
2818                         clen = m->m_len;
2819
2820                         while (cm != NULL) {
2821                                 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2822                                         break;
2823
2824                                 data = CMSG_DATA(cm);
2825                                 datalen = (caddr_t)cm + cm->cmsg_len
2826                                     - (caddr_t)data;
2827
2828                                 if (cm->cmsg_level == SOL_SOCKET &&
2829                                     cm->cmsg_type == SCM_RIGHTS) {
2830                                         (*op)(data, datalen /
2831                                             sizeof(struct filedescent *));
2832                                 }
2833
2834                                 if (CMSG_SPACE(datalen) < clen) {
2835                                         clen -= CMSG_SPACE(datalen);
2836                                         cm = (struct cmsghdr *)
2837                                             ((caddr_t)cm + CMSG_SPACE(datalen));
2838                                 } else {
2839                                         clen = 0;
2840                                         cm = NULL;
2841                                 }
2842                         }
2843                 }
2844                 m0 = m0->m_nextpkt;
2845         }
2846 }
2847
2848 /*
2849  * A helper function called by VFS before socket-type vnode reclamation.
2850  * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2851  * use count.
2852  */
2853 void
2854 vfs_unp_reclaim(struct vnode *vp)
2855 {
2856         struct unpcb *unp;
2857         int active;
2858         struct mtx *vplock;
2859
2860         ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2861         KASSERT(vp->v_type == VSOCK,
2862             ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2863
2864         active = 0;
2865         vplock = mtx_pool_find(mtxpool_sleep, vp);
2866         mtx_lock(vplock);
2867         VOP_UNP_CONNECT(vp, &unp);
2868         if (unp == NULL)
2869                 goto done;
2870         UNP_PCB_LOCK(unp);
2871         if (unp->unp_vnode == vp) {
2872                 VOP_UNP_DETACH(vp);
2873                 unp->unp_vnode = NULL;
2874                 active = 1;
2875         }
2876         UNP_PCB_UNLOCK(unp);
2877  done:
2878         mtx_unlock(vplock);
2879         if (active)
2880                 vunref(vp);
2881 }
2882
2883 #ifdef DDB
2884 static void
2885 db_print_indent(int indent)
2886 {
2887         int i;
2888
2889         for (i = 0; i < indent; i++)
2890                 db_printf(" ");
2891 }
2892
2893 static void
2894 db_print_unpflags(int unp_flags)
2895 {
2896         int comma;
2897
2898         comma = 0;
2899         if (unp_flags & UNP_HAVEPC) {
2900                 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2901                 comma = 1;
2902         }
2903         if (unp_flags & UNP_WANTCRED_ALWAYS) {
2904                 db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : "");
2905                 comma = 1;
2906         }
2907         if (unp_flags & UNP_WANTCRED_ONESHOT) {
2908                 db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : "");
2909                 comma = 1;
2910         }
2911         if (unp_flags & UNP_CONNWAIT) {
2912                 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2913                 comma = 1;
2914         }
2915         if (unp_flags & UNP_CONNECTING) {
2916                 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2917                 comma = 1;
2918         }
2919         if (unp_flags & UNP_BINDING) {
2920                 db_printf("%sUNP_BINDING", comma ? ", " : "");
2921                 comma = 1;
2922         }
2923 }
2924
2925 static void
2926 db_print_xucred(int indent, struct xucred *xu)
2927 {
2928         int comma, i;
2929
2930         db_print_indent(indent);
2931         db_printf("cr_version: %u   cr_uid: %u   cr_pid: %d   cr_ngroups: %d\n",
2932             xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups);
2933         db_print_indent(indent);
2934         db_printf("cr_groups: ");
2935         comma = 0;
2936         for (i = 0; i < xu->cr_ngroups; i++) {
2937                 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2938                 comma = 1;
2939         }
2940         db_printf("\n");
2941 }
2942
2943 static void
2944 db_print_unprefs(int indent, struct unp_head *uh)
2945 {
2946         struct unpcb *unp;
2947         int counter;
2948
2949         counter = 0;
2950         LIST_FOREACH(unp, uh, unp_reflink) {
2951                 if (counter % 4 == 0)
2952                         db_print_indent(indent);
2953                 db_printf("%p  ", unp);
2954                 if (counter % 4 == 3)
2955                         db_printf("\n");
2956                 counter++;
2957         }
2958         if (counter != 0 && counter % 4 != 0)
2959                 db_printf("\n");
2960 }
2961
2962 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2963 {
2964         struct unpcb *unp;
2965
2966         if (!have_addr) {
2967                 db_printf("usage: show unpcb <addr>\n");
2968                 return;
2969         }
2970         unp = (struct unpcb *)addr;
2971
2972         db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
2973             unp->unp_vnode);
2974
2975         db_printf("unp_ino: %ju   unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2976             unp->unp_conn);
2977
2978         db_printf("unp_refs:\n");
2979         db_print_unprefs(2, &unp->unp_refs);
2980
2981         /* XXXRW: Would be nice to print the full address, if any. */
2982         db_printf("unp_addr: %p\n", unp->unp_addr);
2983
2984         db_printf("unp_gencnt: %llu\n",
2985             (unsigned long long)unp->unp_gencnt);
2986
2987         db_printf("unp_flags: %x (", unp->unp_flags);
2988         db_print_unpflags(unp->unp_flags);
2989         db_printf(")\n");
2990
2991         db_printf("unp_peercred:\n");
2992         db_print_xucred(2, &unp->unp_peercred);
2993
2994         db_printf("unp_refcount: %u\n", unp->unp_refcount);
2995 }
2996 #endif