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