sys/kern/uipc_usrreq.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-3-Clause
   3  *
   4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
   5  *      The Regents of the University of California. All Rights Reserved.
   6  * Copyright (c) 2004-2009 Robert N. M. Watson All Rights Reserved.
   7  * Copyright (c) 2018 Matthew Macy
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in the
  16  *    documentation and/or other materials provided with the distribution.
  17  * 3. Neither the name of the University nor the names of its contributors
  18  *    may be used to endorse or promote products derived from this software
  19  *    without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  *
  33  *      From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
  34  */
  35
  36 /*
  37  * UNIX Domain (Local) Sockets
  38  *
  39  * This is an implementation of UNIX (local) domain sockets.  Each socket has
  40  * an associated struct unpcb (UNIX protocol control block).  Stream sockets
  41  * may be connected to 0 or 1 other socket.  Datagram sockets may be
  42  * connected to 0, 1, or many other sockets.  Sockets may be created and
  43  * connected in pairs (socketpair(2)), or bound/connected to using the file
  44  * system name space.  For most purposes, only the receive socket buffer is
  45  * used, as sending on one socket delivers directly to the receive socket
  46  * buffer of a second socket.
  47  *
  48  * The implementation is substantially complicated by the fact that
  49  * "ancillary data", such as file descriptors or credentials, may be passed
  50  * across UNIX domain sockets.  The potential for passing UNIX domain sockets
  51  * over other UNIX domain sockets requires the implementation of a simple
  52  * garbage collector to find and tear down cycles of disconnected sockets.
  53  *
  54  * TODO:
  55  *      RDM
  56  *      rethink name space problems
  57  *      need a proper out-of-band
  58  */
  59
  60 #include <sys/cdefs.h>
  61 __FBSDID("$FreeBSD$");
  62
  63 #include "opt_ddb.h"
  64
  65 #include <sys/param.h>
  66 #include <sys/capsicum.h>
  67 #include <sys/domain.h>
  68 #include <sys/eventhandler.h>
  69 #include <sys/fcntl.h>
  70 #include <sys/file.h>
  71 #include <sys/filedesc.h>
  72 #include <sys/kernel.h>
  73 #include <sys/lock.h>
  74 #include <sys/malloc.h>
  75 #include <sys/mbuf.h>
  76 #include <sys/mount.h>
  77 #include <sys/mutex.h>
  78 #include <sys/namei.h>
  79 #include <sys/proc.h>
  80 #include <sys/protosw.h>
  81 #include <sys/queue.h>
  82 #include <sys/resourcevar.h>
  83 #include <sys/rwlock.h>
  84 #include <sys/socket.h>
  85 #include <sys/socketvar.h>
  86 #include <sys/signalvar.h>
  87 #include <sys/stat.h>
  88 #include <sys/sx.h>
  89 #include <sys/sysctl.h>
  90 #include <sys/systm.h>
  91 #include <sys/taskqueue.h>
  92 #include <sys/un.h>
  93 #include <sys/unpcb.h>
  94 #include <sys/vnode.h>
  95
  96 #include <net/vnet.h>
  97
  98 #ifdef DDB
  99 #include <ddb/ddb.h>
 100 #endif
 101
 102 #include <security/mac/mac_framework.h>
 103
 104 #include <vm/uma.h>
 105
 106 MALLOC_DECLARE(M_FILECAPS);
 107
 108 /*
 109  * See unpcb.h for the locking key.
 110  */
 111
 112 static uma_zone_t       unp_zone;
 113 static unp_gen_t        unp_gencnt;     /* (l) */
 114 static u_int            unp_count;      /* (l) Count of local sockets. */
 115 static ino_t            unp_ino;        /* Prototype for fake inode numbers. */
 116 static int              unp_rights;     /* (g) File descriptors in flight. */
 117 static struct unp_head  unp_shead;      /* (l) List of stream sockets. */
 118 static struct unp_head  unp_dhead;      /* (l) List of datagram sockets. */
 119 static struct unp_head  unp_sphead;     /* (l) List of seqpacket sockets. */
 120
 121 struct unp_defer {
 122         SLIST_ENTRY(unp_defer) ud_link;
 123         struct file *ud_fp;
 124 };
 125 static SLIST_HEAD(, unp_defer) unp_defers;
 126 static int unp_defers_count;
 127
 128 static const struct sockaddr    sun_noname = { sizeof(sun_noname), AF_LOCAL };
 129
 130 /*
 131  * Garbage collection of cyclic file descriptor/socket references occurs
 132  * asynchronously in a taskqueue context in order to avoid recursion and
 133  * reentrance in the UNIX domain socket, file descriptor, and socket layer
 134  * code.  See unp_gc() for a full description.
 135  */
 136 static struct timeout_task unp_gc_task;
 137
 138 /*
 139  * The close of unix domain sockets attached as SCM_RIGHTS is
 140  * postponed to the taskqueue, to avoid arbitrary recursion depth.
 141  * The attached sockets might have another sockets attached.
 142  */
 143 static struct task      unp_defer_task;
 144
 145 /*
 146  * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
 147  * stream sockets, although the total for sender and receiver is actually
 148  * only PIPSIZ.
 149  *
 150  * Datagram sockets really use the sendspace as the maximum datagram size,
 151  * and don't really want to reserve the sendspace.  Their recvspace should be
 152  * large enough for at least one max-size datagram plus address.
 153  */
 154 #ifndef PIPSIZ
 155 #define PIPSIZ  8192
 156 #endif
 157 static u_long   unpst_sendspace = PIPSIZ;
 158 static u_long   unpst_recvspace = PIPSIZ;
 159 static u_long   unpdg_sendspace = 2*1024;       /* really max datagram size */
 160 static u_long   unpdg_recvspace = 4*1024;
 161 static u_long   unpsp_sendspace = PIPSIZ;       /* really max datagram size */
 162 static u_long   unpsp_recvspace = PIPSIZ;
 163
 164 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 165     "Local domain");
 166 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream,
 167     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 168     "SOCK_STREAM");
 169 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram,
 170     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 171     "SOCK_DGRAM");
 172 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket,
 173     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 174     "SOCK_SEQPACKET");
 175
 176 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
 177            &unpst_sendspace, 0, "Default stream send space.");
 178 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
 179            &unpst_recvspace, 0, "Default stream receive space.");
 180 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
 181            &unpdg_sendspace, 0, "Default datagram send space.");
 182 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
 183            &unpdg_recvspace, 0, "Default datagram receive space.");
 184 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
 185            &unpsp_sendspace, 0, "Default seqpacket send space.");
 186 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
 187            &unpsp_recvspace, 0, "Default seqpacket receive space.");
 188 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
 189     "File descriptors in flight.");
 190 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
 191     &unp_defers_count, 0,
 192     "File descriptors deferred to taskqueue for close.");
 193
 194 /*
 195  * Locking and synchronization:
 196  *
 197  * Several types of locks exist in the local domain socket implementation:
 198  * - a global linkage lock
 199  * - a global connection list lock
 200  * - the mtxpool lock
 201  * - per-unpcb mutexes
 202  *
 203  * The linkage lock protects the global socket lists, the generation number
 204  * counter and garbage collector state.
 205  *
 206  * The connection list lock protects the list of referring sockets in a datagram
 207  * socket PCB.  This lock is also overloaded to protect a global list of
 208  * sockets whose buffers contain socket references in the form of SCM_RIGHTS
 209  * messages.  To avoid recursion, such references are released by a dedicated
 210  * thread.
 211  *
 212  * The mtxpool lock protects the vnode from being modified while referenced.
 213  * Lock ordering rules require that it be acquired before any PCB locks.
 214  *
 215  * The unpcb lock (unp_mtx) protects the most commonly referenced fields in the
 216  * unpcb.  This includes the unp_conn field, which either links two connected
 217  * PCBs together (for connected socket types) or points at the destination
 218  * socket (for connectionless socket types).  The operations of creating or
 219  * destroying a connection therefore involve locking multiple PCBs.  To avoid
 220  * lock order reversals, in some cases this involves dropping a PCB lock and
 221  * using a reference counter to maintain liveness.
 222  *
 223  * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
 224  * allocated in pru_attach() and freed in pru_detach().  The validity of that
 225  * pointer is an invariant, so no lock is required to dereference the so_pcb
 226  * pointer if a valid socket reference is held by the caller.  In practice,
 227  * this is always true during operations performed on a socket.  Each unpcb
 228  * has a back-pointer to its socket, unp_socket, which will be stable under
 229  * the same circumstances.
 230  *
 231  * This pointer may only be safely dereferenced as long as a valid reference
 232  * to the unpcb is held.  Typically, this reference will be from the socket,
 233  * or from another unpcb when the referring unpcb's lock is held (in order
 234  * that the reference not be invalidated during use).  For example, to follow
 235  * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee
 236  * that detach is not run clearing unp_socket.
 237  *
 238  * Blocking with UNIX domain sockets is a tricky issue: unlike most network
 239  * protocols, bind() is a non-atomic operation, and connect() requires
 240  * potential sleeping in the protocol, due to potentially waiting on local or
 241  * distributed file systems.  We try to separate "lookup" operations, which
 242  * may sleep, and the IPC operations themselves, which typically can occur
 243  * with relative atomicity as locks can be held over the entire operation.
 244  *
 245  * Another tricky issue is simultaneous multi-threaded or multi-process
 246  * access to a single UNIX domain socket.  These are handled by the flags
 247  * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
 248  * binding, both of which involve dropping UNIX domain socket locks in order
 249  * to perform namei() and other file system operations.
 250  */
 251 static struct rwlock    unp_link_rwlock;
 252 static struct mtx       unp_defers_lock;
 253
 254 #define UNP_LINK_LOCK_INIT()            rw_init(&unp_link_rwlock,       \
 255                                             "unp_link_rwlock")
 256
 257 #define UNP_LINK_LOCK_ASSERT()          rw_assert(&unp_link_rwlock,     \
 258                                             RA_LOCKED)
 259 #define UNP_LINK_UNLOCK_ASSERT()        rw_assert(&unp_link_rwlock,     \
 260                                             RA_UNLOCKED)
 261
 262 #define UNP_LINK_RLOCK()                rw_rlock(&unp_link_rwlock)
 263 #define UNP_LINK_RUNLOCK()              rw_runlock(&unp_link_rwlock)
 264 #define UNP_LINK_WLOCK()                rw_wlock(&unp_link_rwlock)
 265 #define UNP_LINK_WUNLOCK()              rw_wunlock(&unp_link_rwlock)
 266 #define UNP_LINK_WLOCK_ASSERT()         rw_assert(&unp_link_rwlock,     \
 267                                             RA_WLOCKED)
 268 #define UNP_LINK_WOWNED()               rw_wowned(&unp_link_rwlock)
 269
 270 #define UNP_DEFERRED_LOCK_INIT()        mtx_init(&unp_defers_lock, \
 271                                             "unp_defer", NULL, MTX_DEF)
 272 #define UNP_DEFERRED_LOCK()             mtx_lock(&unp_defers_lock)
 273 #define UNP_DEFERRED_UNLOCK()           mtx_unlock(&unp_defers_lock)
 274
 275 #define UNP_REF_LIST_LOCK()             UNP_DEFERRED_LOCK();
 276 #define UNP_REF_LIST_UNLOCK()           UNP_DEFERRED_UNLOCK();
 277
 278 #define UNP_PCB_LOCK_INIT(unp)          mtx_init(&(unp)->unp_mtx,       \
 279                                             "unp", "unp",       \
 280                                             MTX_DUPOK|MTX_DEF)
 281 #define UNP_PCB_LOCK_DESTROY(unp)       mtx_destroy(&(unp)->unp_mtx)
 282 #define UNP_PCB_LOCKPTR(unp)            (&(unp)->unp_mtx)
 283 #define UNP_PCB_LOCK(unp)               mtx_lock(&(unp)->unp_mtx)
 284 #define UNP_PCB_TRYLOCK(unp)            mtx_trylock(&(unp)->unp_mtx)
 285 #define UNP_PCB_UNLOCK(unp)             mtx_unlock(&(unp)->unp_mtx)
 286 #define UNP_PCB_OWNED(unp)              mtx_owned(&(unp)->unp_mtx)
 287 #define UNP_PCB_LOCK_ASSERT(unp)        mtx_assert(&(unp)->unp_mtx, MA_OWNED)
 288 #define UNP_PCB_UNLOCK_ASSERT(unp)      mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED)
 289
 290 static int      uipc_connect2(struct socket *, struct socket *);
 291 static int      uipc_ctloutput(struct socket *, struct sockopt *);
 292 static int      unp_connect(struct socket *, struct sockaddr *,
 293                     struct thread *);
 294 static int      unp_connectat(int, struct socket *, struct sockaddr *,
 295                     struct thread *);
 296 static int      unp_connect2(struct socket *so, struct socket *so2, int);
 297 static void     unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
 298 static void     unp_dispose(struct socket *so);
 299 static void     unp_dispose_mbuf(struct mbuf *);
 300 static void     unp_shutdown(struct unpcb *);
 301 static void     unp_drop(struct unpcb *);
 302 static void     unp_gc(__unused void *, int);
 303 static void     unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
 304 static void     unp_discard(struct file *);
 305 static void     unp_freerights(struct filedescent **, int);
 306 static void     unp_init(void);
 307 static int      unp_internalize(struct mbuf **, struct thread *);
 308 static void     unp_internalize_fp(struct file *);
 309 static int      unp_externalize(struct mbuf *, struct mbuf **, int);
 310 static int      unp_externalize_fp(struct file *);
 311 static struct mbuf      *unp_addsockcred(struct thread *, struct mbuf *, int);
 312 static void     unp_process_defers(void * __unused, int);
 313
 314 static void
 315 unp_pcb_hold(struct unpcb *unp)
 316 {
 317         u_int old __unused;
 318
 319         old = refcount_acquire(&unp->unp_refcount);
 320         KASSERT(old > 0, ("%s: unpcb %p has no references", __func__, unp));
 321 }
 322
 323 static __result_use_check bool
 324 unp_pcb_rele(struct unpcb *unp)
 325 {
 326         bool ret;
 327
 328         UNP_PCB_LOCK_ASSERT(unp);
 329
 330         if ((ret = refcount_release(&unp->unp_refcount))) {
 331                 UNP_PCB_UNLOCK(unp);
 332                 UNP_PCB_LOCK_DESTROY(unp);
 333                 uma_zfree(unp_zone, unp);
 334         }
 335         return (ret);
 336 }
 337
 338 static void
 339 unp_pcb_rele_notlast(struct unpcb *unp)
 340 {
 341         bool ret __unused;
 342
 343         ret = refcount_release(&unp->unp_refcount);
 344         KASSERT(!ret, ("%s: unpcb %p has no references", __func__, unp));
 345 }
 346
 347 static void
 348 unp_pcb_lock_pair(struct unpcb *unp, struct unpcb *unp2)
 349 {
 350         UNP_PCB_UNLOCK_ASSERT(unp);
 351         UNP_PCB_UNLOCK_ASSERT(unp2);
 352
 353         if (unp == unp2) {
 354                 UNP_PCB_LOCK(unp);
 355         } else if ((uintptr_t)unp2 > (uintptr_t)unp) {
 356                 UNP_PCB_LOCK(unp);
 357                 UNP_PCB_LOCK(unp2);
 358         } else {
 359                 UNP_PCB_LOCK(unp2);
 360                 UNP_PCB_LOCK(unp);
 361         }
 362 }
 363
 364 static void
 365 unp_pcb_unlock_pair(struct unpcb *unp, struct unpcb *unp2)
 366 {
 367         UNP_PCB_UNLOCK(unp);
 368         if (unp != unp2)
 369                 UNP_PCB_UNLOCK(unp2);
 370 }
 371
 372 /*
 373  * Try to lock the connected peer of an already locked socket.  In some cases
 374  * this requires that we unlock the current socket.  The pairbusy counter is
 375  * used to block concurrent connection attempts while the lock is dropped.  The
 376  * caller must be careful to revalidate PCB state.
 377  */
 378 static struct unpcb *
 379 unp_pcb_lock_peer(struct unpcb *unp)
 380 {
 381         struct unpcb *unp2;
 382
 383         UNP_PCB_LOCK_ASSERT(unp);
 384         unp2 = unp->unp_conn;
 385         if (unp2 == NULL)
 386                 return (NULL);
 387         if (__predict_false(unp == unp2))
 388                 return (unp);
 389
 390         UNP_PCB_UNLOCK_ASSERT(unp2);
 391
 392         if (__predict_true(UNP_PCB_TRYLOCK(unp2)))
 393                 return (unp2);
 394         if ((uintptr_t)unp2 > (uintptr_t)unp) {
 395                 UNP_PCB_LOCK(unp2);
 396                 return (unp2);
 397         }
 398         unp->unp_pairbusy++;
 399         unp_pcb_hold(unp2);
 400         UNP_PCB_UNLOCK(unp);
 401
 402         UNP_PCB_LOCK(unp2);
 403         UNP_PCB_LOCK(unp);
 404         KASSERT(unp->unp_conn == unp2 || unp->unp_conn == NULL,
 405             ("%s: socket %p was reconnected", __func__, unp));
 406         if (--unp->unp_pairbusy == 0 && (unp->unp_flags & UNP_WAITING) != 0) {
 407                 unp->unp_flags &= ~UNP_WAITING;
 408                 wakeup(unp);
 409         }
 410         if (unp_pcb_rele(unp2)) {
 411                 /* unp2 is unlocked. */
 412                 return (NULL);
 413         }
 414         if (unp->unp_conn == NULL) {
 415                 UNP_PCB_UNLOCK(unp2);
 416                 return (NULL);
 417         }
 418         return (unp2);
 419 }
 420
 421 /*
 422  * Definitions of protocols supported in the LOCAL domain.
 423  */
 424 static struct domain localdomain;
 425 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
 426 static struct pr_usrreqs uipc_usrreqs_seqpacket;
 427 static struct protosw localsw[] = {
 428 {
 429         .pr_type =              SOCK_STREAM,
 430         .pr_domain =            &localdomain,
 431         .pr_flags =             PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
 432         .pr_ctloutput =         &uipc_ctloutput,
 433         .pr_usrreqs =           &uipc_usrreqs_stream
 434 },
 435 {
 436         .pr_type =              SOCK_DGRAM,
 437         .pr_domain =            &localdomain,
 438         .pr_flags =             PR_ATOMIC|PR_ADDR|PR_RIGHTS,
 439         .pr_ctloutput =         &uipc_ctloutput,
 440         .pr_usrreqs =           &uipc_usrreqs_dgram
 441 },
 442 {
 443         .pr_type =              SOCK_SEQPACKET,
 444         .pr_domain =            &localdomain,
 445
 446         /*
 447          * XXXRW: For now, PR_ADDR because soreceive will bump into them
 448          * due to our use of sbappendaddr.  A new sbappend variants is needed
 449          * that supports both atomic record writes and control data.
 450          */
 451         .pr_flags =             PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
 452                                     PR_RIGHTS,
 453         .pr_ctloutput =         &uipc_ctloutput,
 454         .pr_usrreqs =           &uipc_usrreqs_seqpacket,
 455 },
 456 };
 457
 458 static struct domain localdomain = {
 459         .dom_family =           AF_LOCAL,
 460         .dom_name =             "local",
 461         .dom_init =             unp_init,
 462         .dom_externalize =      unp_externalize,
 463         .dom_dispose =          unp_dispose,
 464         .dom_protosw =          localsw,
 465         .dom_protoswNPROTOSW =  &localsw[nitems(localsw)]
 466 };
 467 DOMAIN_SET(local);
 468
 469 static void
 470 uipc_abort(struct socket *so)
 471 {
 472         struct unpcb *unp, *unp2;
 473
 474         unp = sotounpcb(so);
 475         KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
 476         UNP_PCB_UNLOCK_ASSERT(unp);
 477
 478         UNP_PCB_LOCK(unp);
 479         unp2 = unp->unp_conn;
 480         if (unp2 != NULL) {
 481                 unp_pcb_hold(unp2);
 482                 UNP_PCB_UNLOCK(unp);
 483                 unp_drop(unp2);
 484         } else
 485                 UNP_PCB_UNLOCK(unp);
 486 }
 487
 488 static int
 489 uipc_accept(struct socket *so, struct sockaddr **nam)
 490 {
 491         struct unpcb *unp, *unp2;
 492         const struct sockaddr *sa;
 493
 494         /*
 495          * Pass back name of connected socket, if it was bound and we are
 496          * still connected (our peer may have closed already!).
 497          */
 498         unp = sotounpcb(so);
 499         KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
 500
 501         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 502         UNP_PCB_LOCK(unp);
 503         unp2 = unp_pcb_lock_peer(unp);
 504         if (unp2 != NULL && unp2->unp_addr != NULL)
 505                 sa = (struct sockaddr *)unp2->unp_addr;
 506         else
 507                 sa = &sun_noname;
 508         bcopy(sa, *nam, sa->sa_len);
 509         if (unp2 != NULL)
 510                 unp_pcb_unlock_pair(unp, unp2);
 511         else
 512                 UNP_PCB_UNLOCK(unp);
 513         return (0);
 514 }
 515
 516 static int
 517 uipc_attach(struct socket *so, int proto, struct thread *td)
 518 {
 519         u_long sendspace, recvspace;
 520         struct unpcb *unp;
 521         int error;
 522         bool locked;
 523
 524         KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
 525         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
 526                 switch (so->so_type) {
 527                 case SOCK_STREAM:
 528                         sendspace = unpst_sendspace;
 529                         recvspace = unpst_recvspace;
 530                         break;
 531
 532                 case SOCK_DGRAM:
 533                         sendspace = unpdg_sendspace;
 534                         recvspace = unpdg_recvspace;
 535                         break;
 536
 537                 case SOCK_SEQPACKET:
 538                         sendspace = unpsp_sendspace;
 539                         recvspace = unpsp_recvspace;
 540                         break;
 541
 542                 default:
 543                         panic("uipc_attach");
 544                 }
 545                 error = soreserve(so, sendspace, recvspace);
 546                 if (error)
 547                         return (error);
 548         }
 549         unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
 550         if (unp == NULL)
 551                 return (ENOBUFS);
 552         LIST_INIT(&unp->unp_refs);
 553         UNP_PCB_LOCK_INIT(unp);
 554         unp->unp_socket = so;
 555         so->so_pcb = unp;
 556         refcount_init(&unp->unp_refcount, 1);
 557
 558         if ((locked = UNP_LINK_WOWNED()) == false)
 559                 UNP_LINK_WLOCK();
 560
 561         unp->unp_gencnt = ++unp_gencnt;
 562         unp->unp_ino = ++unp_ino;
 563         unp_count++;
 564         switch (so->so_type) {
 565         case SOCK_STREAM:
 566                 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
 567                 break;
 568
 569         case SOCK_DGRAM:
 570                 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
 571                 break;
 572
 573         case SOCK_SEQPACKET:
 574                 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
 575                 break;
 576
 577         default:
 578                 panic("uipc_attach");
 579         }
 580
 581         if (locked == false)
 582                 UNP_LINK_WUNLOCK();
 583
 584         return (0);
 585 }
 586
 587 static int
 588 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
 589 {
 590         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
 591         struct vattr vattr;
 592         int error, namelen;
 593         struct nameidata nd;
 594         struct unpcb *unp;
 595         struct vnode *vp;
 596         struct mount *mp;
 597         cap_rights_t rights;
 598         char *buf;
 599
 600         if (nam->sa_family != AF_UNIX)
 601                 return (EAFNOSUPPORT);
 602
 603         unp = sotounpcb(so);
 604         KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
 605
 606         if (soun->sun_len > sizeof(struct sockaddr_un))
 607                 return (EINVAL);
 608         namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
 609         if (namelen <= 0)
 610                 return (EINVAL);
 611
 612         /*
 613          * We don't allow simultaneous bind() calls on a single UNIX domain
 614          * socket, so flag in-progress operations, and return an error if an
 615          * operation is already in progress.
 616          *
 617          * Historically, we have not allowed a socket to be rebound, so this
 618          * also returns an error.  Not allowing re-binding simplifies the
 619          * implementation and avoids a great many possible failure modes.
 620          */
 621         UNP_PCB_LOCK(unp);
 622         if (unp->unp_vnode != NULL) {
 623                 UNP_PCB_UNLOCK(unp);
 624                 return (EINVAL);
 625         }
 626         if (unp->unp_flags & UNP_BINDING) {
 627                 UNP_PCB_UNLOCK(unp);
 628                 return (EALREADY);
 629         }
 630         unp->unp_flags |= UNP_BINDING;
 631         UNP_PCB_UNLOCK(unp);
 632
 633         buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
 634         bcopy(soun->sun_path, buf, namelen);
 635         buf[namelen] = 0;
 636
 637 restart:
 638         NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE,
 639             UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_BINDAT),
 640             td);
 641 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
 642         error = namei(&nd);
 643         if (error)
 644                 goto error;
 645         vp = nd.ni_vp;
 646         if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
 647                 NDFREE(&nd, NDF_ONLY_PNBUF);
 648                 if (nd.ni_dvp == vp)
 649                         vrele(nd.ni_dvp);
 650                 else
 651                         vput(nd.ni_dvp);
 652                 if (vp != NULL) {
 653                         vrele(vp);
 654                         error = EADDRINUSE;
 655                         goto error;
 656                 }
 657                 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
 658                 if (error)
 659                         goto error;
 660                 goto restart;
 661         }
 662         VATTR_NULL(&vattr);
 663         vattr.va_type = VSOCK;
 664         vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_pd->pd_cmask);
 665 #ifdef MAC
 666         error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
 667             &vattr);
 668 #endif
 669         if (error == 0)
 670                 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
 671         NDFREE(&nd, NDF_ONLY_PNBUF);
 672         if (error) {
 673                 VOP_VPUT_PAIR(nd.ni_dvp, NULL, true);
 674                 vn_finished_write(mp);
 675                 if (error == ERELOOKUP)
 676                         goto restart;
 677                 goto error;
 678         }
 679         vp = nd.ni_vp;
 680         ASSERT_VOP_ELOCKED(vp, "uipc_bind");
 681         soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
 682
 683         UNP_PCB_LOCK(unp);
 684         VOP_UNP_BIND(vp, unp);
 685         unp->unp_vnode = vp;
 686         unp->unp_addr = soun;
 687         unp->unp_flags &= ~UNP_BINDING;
 688         UNP_PCB_UNLOCK(unp);
 689         vref(vp);
 690         VOP_VPUT_PAIR(nd.ni_dvp, &vp, true);
 691         vn_finished_write(mp);
 692         free(buf, M_TEMP);
 693         return (0);
 694
 695 error:
 696         UNP_PCB_LOCK(unp);
 697         unp->unp_flags &= ~UNP_BINDING;
 698         UNP_PCB_UNLOCK(unp);
 699         free(buf, M_TEMP);
 700         return (error);
 701 }
 702
 703 static int
 704 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 705 {
 706
 707         return (uipc_bindat(AT_FDCWD, so, nam, td));
 708 }
 709
 710 static int
 711 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 712 {
 713         int error;
 714
 715         KASSERT(td == curthread, ("uipc_connect: td != curthread"));
 716         error = unp_connect(so, nam, td);
 717         return (error);
 718 }
 719
 720 static int
 721 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
 722     struct thread *td)
 723 {
 724         int error;
 725
 726         KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
 727         error = unp_connectat(fd, so, nam, td);
 728         return (error);
 729 }
 730
 731 static void
 732 uipc_close(struct socket *so)
 733 {
 734         struct unpcb *unp, *unp2;
 735         struct vnode *vp = NULL;
 736         struct mtx *vplock;
 737
 738         unp = sotounpcb(so);
 739         KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
 740
 741         vplock = NULL;
 742         if ((vp = unp->unp_vnode) != NULL) {
 743                 vplock = mtx_pool_find(mtxpool_sleep, vp);
 744                 mtx_lock(vplock);
 745         }
 746         UNP_PCB_LOCK(unp);
 747         if (vp && unp->unp_vnode == NULL) {
 748                 mtx_unlock(vplock);
 749                 vp = NULL;
 750         }
 751         if (vp != NULL) {
 752                 VOP_UNP_DETACH(vp);
 753                 unp->unp_vnode = NULL;
 754         }
 755         if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
 756                 unp_disconnect(unp, unp2);
 757         else
 758                 UNP_PCB_UNLOCK(unp);
 759         if (vp) {
 760                 mtx_unlock(vplock);
 761                 vrele(vp);
 762         }
 763 }
 764
 765 static int
 766 uipc_connect2(struct socket *so1, struct socket *so2)
 767 {
 768         struct unpcb *unp, *unp2;
 769         int error;
 770
 771         unp = so1->so_pcb;
 772         KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
 773         unp2 = so2->so_pcb;
 774         KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
 775         unp_pcb_lock_pair(unp, unp2);
 776         error = unp_connect2(so1, so2, PRU_CONNECT2);
 777         unp_pcb_unlock_pair(unp, unp2);
 778         return (error);
 779 }
 780
 781 static void
 782 uipc_detach(struct socket *so)
 783 {
 784         struct unpcb *unp, *unp2;
 785         struct mtx *vplock;
 786         struct vnode *vp;
 787         int local_unp_rights;
 788
 789         unp = sotounpcb(so);
 790         KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
 791
 792         vp = NULL;
 793         vplock = NULL;
 794
 795         SOCK_LOCK(so);
 796         if (!SOLISTENING(so)) {
 797                 /*
 798                  * Once the socket is removed from the global lists,
 799                  * uipc_ready() will not be able to locate its socket buffer, so
 800                  * clear the buffer now.  At this point internalized rights have
 801                  * already been disposed of.
 802                  */
 803                 sbrelease(&so->so_rcv, so);
 804         }
 805         SOCK_UNLOCK(so);
 806
 807         UNP_LINK_WLOCK();
 808         LIST_REMOVE(unp, unp_link);
 809         if (unp->unp_gcflag & UNPGC_DEAD)
 810                 LIST_REMOVE(unp, unp_dead);
 811         unp->unp_gencnt = ++unp_gencnt;
 812         --unp_count;
 813         UNP_LINK_WUNLOCK();
 814
 815         UNP_PCB_UNLOCK_ASSERT(unp);
 816  restart:
 817         if ((vp = unp->unp_vnode) != NULL) {
 818                 vplock = mtx_pool_find(mtxpool_sleep, vp);
 819                 mtx_lock(vplock);
 820         }
 821         UNP_PCB_LOCK(unp);
 822         if (unp->unp_vnode != vp && unp->unp_vnode != NULL) {
 823                 if (vplock)
 824                         mtx_unlock(vplock);
 825                 UNP_PCB_UNLOCK(unp);
 826                 goto restart;
 827         }
 828         if ((vp = unp->unp_vnode) != NULL) {
 829                 VOP_UNP_DETACH(vp);
 830                 unp->unp_vnode = NULL;
 831         }
 832         if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
 833                 unp_disconnect(unp, unp2);
 834         else
 835                 UNP_PCB_UNLOCK(unp);
 836
 837         UNP_REF_LIST_LOCK();
 838         while (!LIST_EMPTY(&unp->unp_refs)) {
 839                 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
 840
 841                 unp_pcb_hold(ref);
 842                 UNP_REF_LIST_UNLOCK();
 843
 844                 MPASS(ref != unp);
 845                 UNP_PCB_UNLOCK_ASSERT(ref);
 846                 unp_drop(ref);
 847                 UNP_REF_LIST_LOCK();
 848         }
 849         UNP_REF_LIST_UNLOCK();
 850
 851         UNP_PCB_LOCK(unp);
 852         local_unp_rights = unp_rights;
 853         unp->unp_socket->so_pcb = NULL;
 854         unp->unp_socket = NULL;
 855         free(unp->unp_addr, M_SONAME);
 856         unp->unp_addr = NULL;
 857         if (!unp_pcb_rele(unp))
 858                 UNP_PCB_UNLOCK(unp);
 859         if (vp) {
 860                 mtx_unlock(vplock);
 861                 vrele(vp);
 862         }
 863         if (local_unp_rights)
 864                 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
 865 }
 866
 867 static int
 868 uipc_disconnect(struct socket *so)
 869 {
 870         struct unpcb *unp, *unp2;
 871
 872         unp = sotounpcb(so);
 873         KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
 874
 875         UNP_PCB_LOCK(unp);
 876         if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
 877                 unp_disconnect(unp, unp2);
 878         else
 879                 UNP_PCB_UNLOCK(unp);
 880         return (0);
 881 }
 882
 883 static int
 884 uipc_listen(struct socket *so, int backlog, struct thread *td)
 885 {
 886         struct unpcb *unp;
 887         int error;
 888
 889         if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
 890                 return (EOPNOTSUPP);
 891
 892         unp = sotounpcb(so);
 893         KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
 894
 895         UNP_PCB_LOCK(unp);
 896         if (unp->unp_vnode == NULL) {
 897                 /* Already connected or not bound to an address. */
 898                 error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ;
 899                 UNP_PCB_UNLOCK(unp);
 900                 return (error);
 901         }
 902
 903         SOCK_LOCK(so);
 904         error = solisten_proto_check(so);
 905         if (error == 0) {
 906                 cru2xt(td, &unp->unp_peercred);
 907                 solisten_proto(so, backlog);
 908         }
 909         SOCK_UNLOCK(so);
 910         UNP_PCB_UNLOCK(unp);
 911         return (error);
 912 }
 913
 914 static int
 915 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
 916 {
 917         struct unpcb *unp, *unp2;
 918         const struct sockaddr *sa;
 919
 920         unp = sotounpcb(so);
 921         KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
 922
 923         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 924         UNP_LINK_RLOCK();
 925         /*
 926          * XXX: It seems that this test always fails even when connection is
 927          * established.  So, this else clause is added as workaround to
 928          * return PF_LOCAL sockaddr.
 929          */
 930         unp2 = unp->unp_conn;
 931         if (unp2 != NULL) {
 932                 UNP_PCB_LOCK(unp2);
 933                 if (unp2->unp_addr != NULL)
 934                         sa = (struct sockaddr *) unp2->unp_addr;
 935                 else
 936                         sa = &sun_noname;
 937                 bcopy(sa, *nam, sa->sa_len);
 938                 UNP_PCB_UNLOCK(unp2);
 939         } else {
 940                 sa = &sun_noname;
 941                 bcopy(sa, *nam, sa->sa_len);
 942         }
 943         UNP_LINK_RUNLOCK();
 944         return (0);
 945 }
 946
 947 static int
 948 uipc_rcvd(struct socket *so, int flags)
 949 {
 950         struct unpcb *unp, *unp2;
 951         struct socket *so2;
 952         u_int mbcnt, sbcc;
 953
 954         unp = sotounpcb(so);
 955         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
 956         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
 957             ("%s: socktype %d", __func__, so->so_type));
 958
 959         /*
 960          * Adjust backpressure on sender and wakeup any waiting to write.
 961          *
 962          * The unp lock is acquired to maintain the validity of the unp_conn
 963          * pointer; no lock on unp2 is required as unp2->unp_socket will be
 964          * static as long as we don't permit unp2 to disconnect from unp,
 965          * which is prevented by the lock on unp.  We cache values from
 966          * so_rcv to avoid holding the so_rcv lock over the entire
 967          * transaction on the remote so_snd.
 968          */
 969         SOCKBUF_LOCK(&so->so_rcv);
 970         mbcnt = so->so_rcv.sb_mbcnt;
 971         sbcc = sbavail(&so->so_rcv);
 972         SOCKBUF_UNLOCK(&so->so_rcv);
 973         /*
 974          * There is a benign race condition at this point.  If we're planning to
 975          * clear SB_STOP, but uipc_send is called on the connected socket at
 976          * this instant, it might add data to the sockbuf and set SB_STOP.  Then
 977          * we would erroneously clear SB_STOP below, even though the sockbuf is
 978          * full.  The race is benign because the only ill effect is to allow the
 979          * sockbuf to exceed its size limit, and the size limits are not
 980          * strictly guaranteed anyway.
 981          */
 982         UNP_PCB_LOCK(unp);
 983         unp2 = unp->unp_conn;
 984         if (unp2 == NULL) {
 985                 UNP_PCB_UNLOCK(unp);
 986                 return (0);
 987         }
 988         so2 = unp2->unp_socket;
 989         SOCKBUF_LOCK(&so2->so_snd);
 990         if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
 991                 so2->so_snd.sb_flags &= ~SB_STOP;
 992         sowwakeup_locked(so2);
 993         UNP_PCB_UNLOCK(unp);
 994         return (0);
 995 }
 996
 997 static int
 998 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
 999     struct mbuf *control, struct thread *td)
1000 {
1001         struct unpcb *unp, *unp2;
1002         struct socket *so2;
1003         u_int mbcnt, sbcc;
1004         int freed, error;
1005
1006         unp = sotounpcb(so);
1007         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
1008         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
1009             so->so_type == SOCK_SEQPACKET,
1010             ("%s: socktype %d", __func__, so->so_type));
1011
1012         freed = error = 0;
1013         if (flags & PRUS_OOB) {
1014                 error = EOPNOTSUPP;
1015                 goto release;
1016         }
1017         if (control != NULL && (error = unp_internalize(&control, td)))
1018                 goto release;
1019
1020         unp2 = NULL;
1021         switch (so->so_type) {
1022         case SOCK_DGRAM:
1023         {
1024                 const struct sockaddr *from;
1025
1026                 if (nam != NULL) {
1027                         error = unp_connect(so, nam, td);
1028                         if (error != 0)
1029                                 break;
1030                 }
1031                 UNP_PCB_LOCK(unp);
1032
1033                 /*
1034                  * Because connect() and send() are non-atomic in a sendto()
1035                  * with a target address, it's possible that the socket will
1036                  * have disconnected before the send() can run.  In that case
1037                  * return the slightly counter-intuitive but otherwise
1038                  * correct error that the socket is not connected.
1039                  */
1040                 unp2 = unp_pcb_lock_peer(unp);
1041                 if (unp2 == NULL) {
1042                         UNP_PCB_UNLOCK(unp);
1043                         error = ENOTCONN;
1044                         break;
1045                 }
1046
1047                 if (unp2->unp_flags & UNP_WANTCRED_MASK)
1048                         control = unp_addsockcred(td, control,
1049                             unp2->unp_flags);
1050                 if (unp->unp_addr != NULL)
1051                         from = (struct sockaddr *)unp->unp_addr;
1052                 else
1053                         from = &sun_noname;
1054                 so2 = unp2->unp_socket;
1055                 SOCKBUF_LOCK(&so2->so_rcv);
1056                 if (sbappendaddr_locked(&so2->so_rcv, from, m,
1057                     control)) {
1058                         sorwakeup_locked(so2);
1059                         m = NULL;
1060                         control = NULL;
1061                 } else {
1062                         SOCKBUF_UNLOCK(&so2->so_rcv);
1063                         error = ENOBUFS;
1064                 }
1065                 if (nam != NULL)
1066                         unp_disconnect(unp, unp2);
1067                 else
1068                         unp_pcb_unlock_pair(unp, unp2);
1069                 break;
1070         }
1071
1072         case SOCK_SEQPACKET:
1073         case SOCK_STREAM:
1074                 if ((so->so_state & SS_ISCONNECTED) == 0) {
1075                         if (nam != NULL) {
1076                                 error = unp_connect(so, nam, td);
1077                                 if (error != 0)
1078                                         break;
1079                         } else {
1080                                 error = ENOTCONN;
1081                                 break;
1082                         }
1083                 }
1084
1085                 UNP_PCB_LOCK(unp);
1086                 if ((unp2 = unp_pcb_lock_peer(unp)) == NULL) {
1087                         UNP_PCB_UNLOCK(unp);
1088                         error = ENOTCONN;
1089                         break;
1090                 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1091                         unp_pcb_unlock_pair(unp, unp2);
1092                         error = EPIPE;
1093                         break;
1094                 }
1095                 UNP_PCB_UNLOCK(unp);
1096                 if ((so2 = unp2->unp_socket) == NULL) {
1097                         UNP_PCB_UNLOCK(unp2);
1098                         error = ENOTCONN;
1099                         break;
1100                 }
1101                 SOCKBUF_LOCK(&so2->so_rcv);
1102                 if (unp2->unp_flags & UNP_WANTCRED_MASK) {
1103                         /*
1104                          * Credentials are passed only once on SOCK_STREAM and
1105                          * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or
1106                          * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS).
1107                          */
1108                         control = unp_addsockcred(td, control, unp2->unp_flags);
1109                         unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT;
1110                 }
1111
1112                 /*
1113                  * Send to paired receive port and wake up readers.  Don't
1114                  * check for space available in the receive buffer if we're
1115                  * attaching ancillary data; Unix domain sockets only check
1116                  * for space in the sending sockbuf, and that check is
1117                  * performed one level up the stack.  At that level we cannot
1118                  * precisely account for the amount of buffer space used
1119                  * (e.g., because control messages are not yet internalized).
1120                  */
1121                 switch (so->so_type) {
1122                 case SOCK_STREAM:
1123                         if (control != NULL) {
1124                                 sbappendcontrol_locked(&so2->so_rcv, m,
1125                                     control, flags);
1126                                 control = NULL;
1127                         } else
1128                                 sbappend_locked(&so2->so_rcv, m, flags);
1129                         break;
1130
1131                 case SOCK_SEQPACKET:
1132                         if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1133                             &sun_noname, m, control))
1134                                 control = NULL;
1135                         break;
1136                 }
1137
1138                 mbcnt = so2->so_rcv.sb_mbcnt;
1139                 sbcc = sbavail(&so2->so_rcv);
1140                 if (sbcc)
1141                         sorwakeup_locked(so2);
1142                 else
1143                         SOCKBUF_UNLOCK(&so2->so_rcv);
1144
1145                 /*
1146                  * The PCB lock on unp2 protects the SB_STOP flag.  Without it,
1147                  * it would be possible for uipc_rcvd to be called at this
1148                  * point, drain the receiving sockbuf, clear SB_STOP, and then
1149                  * we would set SB_STOP below.  That could lead to an empty
1150                  * sockbuf having SB_STOP set
1151                  */
1152                 SOCKBUF_LOCK(&so->so_snd);
1153                 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1154                         so->so_snd.sb_flags |= SB_STOP;
1155                 SOCKBUF_UNLOCK(&so->so_snd);
1156                 UNP_PCB_UNLOCK(unp2);
1157                 m = NULL;
1158                 break;
1159         }
1160
1161         /*
1162          * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1163          */
1164         if (flags & PRUS_EOF) {
1165                 UNP_PCB_LOCK(unp);
1166                 socantsendmore(so);
1167                 unp_shutdown(unp);
1168                 UNP_PCB_UNLOCK(unp);
1169         }
1170         if (control != NULL && error != 0)
1171                 unp_dispose_mbuf(control);
1172
1173 release:
1174         if (control != NULL)
1175                 m_freem(control);
1176         /*
1177          * In case of PRUS_NOTREADY, uipc_ready() is responsible
1178          * for freeing memory.
1179          */
1180         if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1181                 m_freem(m);
1182         return (error);
1183 }
1184
1185 static bool
1186 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp)
1187 {
1188         struct mbuf *mb, *n;
1189         struct sockbuf *sb;
1190
1191         SOCK_LOCK(so);
1192         if (SOLISTENING(so)) {
1193                 SOCK_UNLOCK(so);
1194                 return (false);
1195         }
1196         mb = NULL;
1197         sb = &so->so_rcv;
1198         SOCKBUF_LOCK(sb);
1199         if (sb->sb_fnrdy != NULL) {
1200                 for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) {
1201                         if (mb == m) {
1202                                 *errorp = sbready(sb, m, count);
1203                                 break;
1204                         }
1205                         mb = mb->m_next;
1206                         if (mb == NULL) {
1207                                 mb = n;
1208                                 if (mb != NULL)
1209                                         n = mb->m_nextpkt;
1210                         }
1211                 }
1212         }
1213         SOCKBUF_UNLOCK(sb);
1214         SOCK_UNLOCK(so);
1215         return (mb != NULL);
1216 }
1217
1218 static int
1219 uipc_ready(struct socket *so, struct mbuf *m, int count)
1220 {
1221         struct unpcb *unp, *unp2;
1222         struct socket *so2;
1223         int error, i;
1224
1225         unp = sotounpcb(so);
1226
1227         KASSERT(so->so_type == SOCK_STREAM,
1228             ("%s: unexpected socket type for %p", __func__, so));
1229
1230         UNP_PCB_LOCK(unp);
1231         if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) {
1232                 UNP_PCB_UNLOCK(unp);
1233                 so2 = unp2->unp_socket;
1234                 SOCKBUF_LOCK(&so2->so_rcv);
1235                 if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1236                         sorwakeup_locked(so2);
1237                 else
1238                         SOCKBUF_UNLOCK(&so2->so_rcv);
1239                 UNP_PCB_UNLOCK(unp2);
1240                 return (error);
1241         }
1242         UNP_PCB_UNLOCK(unp);
1243
1244         /*
1245          * The receiving socket has been disconnected, but may still be valid.
1246          * In this case, the now-ready mbufs are still present in its socket
1247          * buffer, so perform an exhaustive search before giving up and freeing
1248          * the mbufs.
1249          */
1250         UNP_LINK_RLOCK();
1251         LIST_FOREACH(unp, &unp_shead, unp_link) {
1252                 if (uipc_ready_scan(unp->unp_socket, m, count, &error))
1253                         break;
1254         }
1255         UNP_LINK_RUNLOCK();
1256
1257         if (unp == NULL) {
1258                 for (i = 0; i < count; i++)
1259                         m = m_free(m);
1260                 error = ECONNRESET;
1261         }
1262         return (error);
1263 }
1264
1265 static int
1266 uipc_sense(struct socket *so, struct stat *sb)
1267 {
1268         struct unpcb *unp;
1269
1270         unp = sotounpcb(so);
1271         KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1272
1273         sb->st_blksize = so->so_snd.sb_hiwat;
1274         sb->st_dev = NODEV;
1275         sb->st_ino = unp->unp_ino;
1276         return (0);
1277 }
1278
1279 static int
1280 uipc_shutdown(struct socket *so)
1281 {
1282         struct unpcb *unp;
1283
1284         unp = sotounpcb(so);
1285         KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1286
1287         UNP_PCB_LOCK(unp);
1288         socantsendmore(so);
1289         unp_shutdown(unp);
1290         UNP_PCB_UNLOCK(unp);
1291         return (0);
1292 }
1293
1294 static int
1295 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1296 {
1297         struct unpcb *unp;
1298         const struct sockaddr *sa;
1299
1300         unp = sotounpcb(so);
1301         KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1302
1303         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1304         UNP_PCB_LOCK(unp);
1305         if (unp->unp_addr != NULL)
1306                 sa = (struct sockaddr *) unp->unp_addr;
1307         else
1308                 sa = &sun_noname;
1309         bcopy(sa, *nam, sa->sa_len);
1310         UNP_PCB_UNLOCK(unp);
1311         return (0);
1312 }
1313
1314 static struct pr_usrreqs uipc_usrreqs_dgram = {
1315         .pru_abort =            uipc_abort,
1316         .pru_accept =           uipc_accept,
1317         .pru_attach =           uipc_attach,
1318         .pru_bind =             uipc_bind,
1319         .pru_bindat =           uipc_bindat,
1320         .pru_connect =          uipc_connect,
1321         .pru_connectat =        uipc_connectat,
1322         .pru_connect2 =         uipc_connect2,
1323         .pru_detach =           uipc_detach,
1324         .pru_disconnect =       uipc_disconnect,
1325         .pru_listen =           uipc_listen,
1326         .pru_peeraddr =         uipc_peeraddr,
1327         .pru_rcvd =             uipc_rcvd,
1328         .pru_send =             uipc_send,
1329         .pru_sense =            uipc_sense,
1330         .pru_shutdown =         uipc_shutdown,
1331         .pru_sockaddr =         uipc_sockaddr,
1332         .pru_soreceive =        soreceive_dgram,
1333         .pru_close =            uipc_close,
1334 };
1335
1336 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1337         .pru_abort =            uipc_abort,
1338         .pru_accept =           uipc_accept,
1339         .pru_attach =           uipc_attach,
1340         .pru_bind =             uipc_bind,
1341         .pru_bindat =           uipc_bindat,
1342         .pru_connect =          uipc_connect,
1343         .pru_connectat =        uipc_connectat,
1344         .pru_connect2 =         uipc_connect2,
1345         .pru_detach =           uipc_detach,
1346         .pru_disconnect =       uipc_disconnect,
1347         .pru_listen =           uipc_listen,
1348         .pru_peeraddr =         uipc_peeraddr,
1349         .pru_rcvd =             uipc_rcvd,
1350         .pru_send =             uipc_send,
1351         .pru_sense =            uipc_sense,
1352         .pru_shutdown =         uipc_shutdown,
1353         .pru_sockaddr =         uipc_sockaddr,
1354         .pru_soreceive =        soreceive_generic,      /* XXX: or...? */
1355         .pru_close =            uipc_close,
1356 };
1357
1358 static struct pr_usrreqs uipc_usrreqs_stream = {
1359         .pru_abort =            uipc_abort,
1360         .pru_accept =           uipc_accept,
1361         .pru_attach =           uipc_attach,
1362         .pru_bind =             uipc_bind,
1363         .pru_bindat =           uipc_bindat,
1364         .pru_connect =          uipc_connect,
1365         .pru_connectat =        uipc_connectat,
1366         .pru_connect2 =         uipc_connect2,
1367         .pru_detach =           uipc_detach,
1368         .pru_disconnect =       uipc_disconnect,
1369         .pru_listen =           uipc_listen,
1370         .pru_peeraddr =         uipc_peeraddr,
1371         .pru_rcvd =             uipc_rcvd,
1372         .pru_send =             uipc_send,
1373         .pru_ready =            uipc_ready,
1374         .pru_sense =            uipc_sense,
1375         .pru_shutdown =         uipc_shutdown,
1376         .pru_sockaddr =         uipc_sockaddr,
1377         .pru_soreceive =        soreceive_generic,
1378         .pru_close =            uipc_close,
1379 };
1380
1381 static int
1382 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1383 {
1384         struct unpcb *unp;
1385         struct xucred xu;
1386         int error, optval;
1387
1388         if (sopt->sopt_level != SOL_LOCAL)
1389                 return (EINVAL);
1390
1391         unp = sotounpcb(so);
1392         KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1393         error = 0;
1394         switch (sopt->sopt_dir) {
1395         case SOPT_GET:
1396                 switch (sopt->sopt_name) {
1397                 case LOCAL_PEERCRED:
1398                         UNP_PCB_LOCK(unp);
1399                         if (unp->unp_flags & UNP_HAVEPC)
1400                                 xu = unp->unp_peercred;
1401                         else {
1402                                 if (so->so_type == SOCK_STREAM)
1403                                         error = ENOTCONN;
1404                                 else
1405                                         error = EINVAL;
1406                         }
1407                         UNP_PCB_UNLOCK(unp);
1408                         if (error == 0)
1409                                 error = sooptcopyout(sopt, &xu, sizeof(xu));
1410                         break;
1411
1412                 case LOCAL_CREDS:
1413                         /* Unlocked read. */
1414                         optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0;
1415                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1416                         break;
1417
1418                 case LOCAL_CREDS_PERSISTENT:
1419                         /* Unlocked read. */
1420                         optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0;
1421                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1422                         break;
1423
1424                 case LOCAL_CONNWAIT:
1425                         /* Unlocked read. */
1426                         optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1427                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1428                         break;
1429
1430                 default:
1431                         error = EOPNOTSUPP;
1432                         break;
1433                 }
1434                 break;
1435
1436         case SOPT_SET:
1437                 switch (sopt->sopt_name) {
1438                 case LOCAL_CREDS:
1439                 case LOCAL_CREDS_PERSISTENT:
1440                 case LOCAL_CONNWAIT:
1441                         error = sooptcopyin(sopt, &optval, sizeof(optval),
1442                                             sizeof(optval));
1443                         if (error)
1444                                 break;
1445
1446 #define OPTSET(bit, exclusive) do {                                     \
1447         UNP_PCB_LOCK(unp);                                              \
1448         if (optval) {                                                   \
1449                 if ((unp->unp_flags & (exclusive)) != 0) {              \
1450                         UNP_PCB_UNLOCK(unp);                            \
1451                         error = EINVAL;                                 \
1452                         break;                                          \
1453                 }                                                       \
1454                 unp->unp_flags |= (bit);                                \
1455         } else                                                          \
1456                 unp->unp_flags &= ~(bit);                               \
1457         UNP_PCB_UNLOCK(unp);                                            \
1458 } while (0)
1459
1460                         switch (sopt->sopt_name) {
1461                         case LOCAL_CREDS:
1462                                 OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS);
1463                                 break;
1464
1465                         case LOCAL_CREDS_PERSISTENT:
1466                                 OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT);
1467                                 break;
1468
1469                         case LOCAL_CONNWAIT:
1470                                 OPTSET(UNP_CONNWAIT, 0);
1471                                 break;
1472
1473                         default:
1474                                 break;
1475                         }
1476                         break;
1477 #undef  OPTSET
1478                 default:
1479                         error = ENOPROTOOPT;
1480                         break;
1481                 }
1482                 break;
1483
1484         default:
1485                 error = EOPNOTSUPP;
1486                 break;
1487         }
1488         return (error);
1489 }
1490
1491 static int
1492 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1493 {
1494
1495         return (unp_connectat(AT_FDCWD, so, nam, td));
1496 }
1497
1498 static int
1499 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1500     struct thread *td)
1501 {
1502         struct mtx *vplock;
1503         struct sockaddr_un *soun;
1504         struct vnode *vp;
1505         struct socket *so2;
1506         struct unpcb *unp, *unp2, *unp3;
1507         struct nameidata nd;
1508         char buf[SOCK_MAXADDRLEN];
1509         struct sockaddr *sa;
1510         cap_rights_t rights;
1511         int error, len;
1512         bool connreq;
1513
1514         if (nam->sa_family != AF_UNIX)
1515                 return (EAFNOSUPPORT);
1516         if (nam->sa_len > sizeof(struct sockaddr_un))
1517                 return (EINVAL);
1518         len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1519         if (len <= 0)
1520                 return (EINVAL);
1521         soun = (struct sockaddr_un *)nam;
1522         bcopy(soun->sun_path, buf, len);
1523         buf[len] = 0;
1524
1525         unp = sotounpcb(so);
1526         UNP_PCB_LOCK(unp);
1527         for (;;) {
1528                 /*
1529                  * Wait for connection state to stabilize.  If a connection
1530                  * already exists, give up.  For datagram sockets, which permit
1531                  * multiple consecutive connect(2) calls, upper layers are
1532                  * responsible for disconnecting in advance of a subsequent
1533                  * connect(2), but this is not synchronized with PCB connection
1534                  * state.
1535                  *
1536                  * Also make sure that no threads are currently attempting to
1537                  * lock the peer socket, to ensure that unp_conn cannot
1538                  * transition between two valid sockets while locks are dropped.
1539                  */
1540                 if (unp->unp_conn != NULL) {
1541                         UNP_PCB_UNLOCK(unp);
1542                         return (EISCONN);
1543                 }
1544                 if ((unp->unp_flags & UNP_CONNECTING) != 0) {
1545                         UNP_PCB_UNLOCK(unp);
1546                         return (EALREADY);
1547                 }
1548                 if (unp->unp_pairbusy > 0) {
1549                         unp->unp_flags |= UNP_WAITING;
1550                         mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0);
1551                         continue;
1552                 }
1553                 break;
1554         }
1555         unp->unp_flags |= UNP_CONNECTING;
1556         UNP_PCB_UNLOCK(unp);
1557
1558         connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0;
1559         if (connreq)
1560                 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1561         else
1562                 sa = NULL;
1563         NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1564             UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT),
1565             td);
1566         error = namei(&nd);
1567         if (error)
1568                 vp = NULL;
1569         else
1570                 vp = nd.ni_vp;
1571         ASSERT_VOP_LOCKED(vp, "unp_connect");
1572         NDFREE_NOTHING(&nd);
1573         if (error)
1574                 goto bad;
1575
1576         if (vp->v_type != VSOCK) {
1577                 error = ENOTSOCK;
1578                 goto bad;
1579         }
1580 #ifdef MAC
1581         error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1582         if (error)
1583                 goto bad;
1584 #endif
1585         error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1586         if (error)
1587                 goto bad;
1588
1589         unp = sotounpcb(so);
1590         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1591
1592         vplock = mtx_pool_find(mtxpool_sleep, vp);
1593         mtx_lock(vplock);
1594         VOP_UNP_CONNECT(vp, &unp2);
1595         if (unp2 == NULL) {
1596                 error = ECONNREFUSED;
1597                 goto bad2;
1598         }
1599         so2 = unp2->unp_socket;
1600         if (so->so_type != so2->so_type) {
1601                 error = EPROTOTYPE;
1602                 goto bad2;
1603         }
1604         if (connreq) {
1605                 if (SOLISTENING(so2)) {
1606                         CURVNET_SET(so2->so_vnet);
1607                         so2 = sonewconn(so2, 0);
1608                         CURVNET_RESTORE();
1609                 } else
1610                         so2 = NULL;
1611                 if (so2 == NULL) {
1612                         error = ECONNREFUSED;
1613                         goto bad2;
1614                 }
1615                 unp3 = sotounpcb(so2);
1616                 unp_pcb_lock_pair(unp2, unp3);
1617                 if (unp2->unp_addr != NULL) {
1618                         bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1619                         unp3->unp_addr = (struct sockaddr_un *) sa;
1620                         sa = NULL;
1621                 }
1622
1623                 unp_copy_peercred(td, unp3, unp, unp2);
1624
1625                 UNP_PCB_UNLOCK(unp2);
1626                 unp2 = unp3;
1627
1628                 /*
1629                  * It is safe to block on the PCB lock here since unp2 is
1630                  * nascent and cannot be connected to any other sockets.
1631                  */
1632                 UNP_PCB_LOCK(unp);
1633 #ifdef MAC
1634                 mac_socketpeer_set_from_socket(so, so2);
1635                 mac_socketpeer_set_from_socket(so2, so);
1636 #endif
1637         } else {
1638                 unp_pcb_lock_pair(unp, unp2);
1639         }
1640         KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 &&
1641             sotounpcb(so2) == unp2,
1642             ("%s: unp2 %p so2 %p", __func__, unp2, so2));
1643         error = unp_connect2(so, so2, PRU_CONNECT);
1644         unp_pcb_unlock_pair(unp, unp2);
1645 bad2:
1646         mtx_unlock(vplock);
1647 bad:
1648         if (vp != NULL) {
1649                 vput(vp);
1650         }
1651         free(sa, M_SONAME);
1652         UNP_PCB_LOCK(unp);
1653         KASSERT((unp->unp_flags & UNP_CONNECTING) != 0,
1654             ("%s: unp %p has UNP_CONNECTING clear", __func__, unp));
1655         unp->unp_flags &= ~UNP_CONNECTING;
1656         UNP_PCB_UNLOCK(unp);
1657         return (error);
1658 }
1659
1660 /*
1661  * Set socket peer credentials at connection time.
1662  *
1663  * The client's PCB credentials are copied from its process structure.  The
1664  * server's PCB credentials are copied from the socket on which it called
1665  * listen(2).  uipc_listen cached that process's credentials at the time.
1666  */
1667 void
1668 unp_copy_peercred(struct thread *td, struct unpcb *client_unp,
1669     struct unpcb *server_unp, struct unpcb *listen_unp)
1670 {
1671         cru2xt(td, &client_unp->unp_peercred);
1672         client_unp->unp_flags |= UNP_HAVEPC;
1673
1674         memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred,
1675             sizeof(server_unp->unp_peercred));
1676         server_unp->unp_flags |= UNP_HAVEPC;
1677         client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK);
1678 }
1679
1680 static int
1681 unp_connect2(struct socket *so, struct socket *so2, int req)
1682 {
1683         struct unpcb *unp;
1684         struct unpcb *unp2;
1685
1686         unp = sotounpcb(so);
1687         KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1688         unp2 = sotounpcb(so2);
1689         KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1690
1691         UNP_PCB_LOCK_ASSERT(unp);
1692         UNP_PCB_LOCK_ASSERT(unp2);
1693         KASSERT(unp->unp_conn == NULL,
1694             ("%s: socket %p is already connected", __func__, unp));
1695
1696         if (so2->so_type != so->so_type)
1697                 return (EPROTOTYPE);
1698         unp->unp_conn = unp2;
1699         unp_pcb_hold(unp2);
1700         unp_pcb_hold(unp);
1701         switch (so->so_type) {
1702         case SOCK_DGRAM:
1703                 UNP_REF_LIST_LOCK();
1704                 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1705                 UNP_REF_LIST_UNLOCK();
1706                 soisconnected(so);
1707                 break;
1708
1709         case SOCK_STREAM:
1710         case SOCK_SEQPACKET:
1711                 KASSERT(unp2->unp_conn == NULL,
1712                     ("%s: socket %p is already connected", __func__, unp2));
1713                 unp2->unp_conn = unp;
1714                 if (req == PRU_CONNECT &&
1715                     ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1716                         soisconnecting(so);
1717                 else
1718                         soisconnected(so);
1719                 soisconnected(so2);
1720                 break;
1721
1722         default:
1723                 panic("unp_connect2");
1724         }
1725         return (0);
1726 }
1727
1728 static void
1729 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1730 {
1731         struct socket *so, *so2;
1732 #ifdef INVARIANTS
1733         struct unpcb *unptmp;
1734 #endif
1735
1736         UNP_PCB_LOCK_ASSERT(unp);
1737         UNP_PCB_LOCK_ASSERT(unp2);
1738         KASSERT(unp->unp_conn == unp2,
1739             ("%s: unpcb %p is not connected to %p", __func__, unp, unp2));
1740
1741         unp->unp_conn = NULL;
1742         so = unp->unp_socket;
1743         so2 = unp2->unp_socket;
1744         switch (unp->unp_socket->so_type) {
1745         case SOCK_DGRAM:
1746                 UNP_REF_LIST_LOCK();
1747 #ifdef INVARIANTS
1748                 LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) {
1749                         if (unptmp == unp)
1750                                 break;
1751                 }
1752                 KASSERT(unptmp != NULL,
1753                     ("%s: %p not found in reflist of %p", __func__, unp, unp2));
1754 #endif
1755                 LIST_REMOVE(unp, unp_reflink);
1756                 UNP_REF_LIST_UNLOCK();
1757                 if (so) {
1758                         SOCK_LOCK(so);
1759                         so->so_state &= ~SS_ISCONNECTED;
1760                         SOCK_UNLOCK(so);
1761                 }
1762                 break;
1763
1764         case SOCK_STREAM:
1765         case SOCK_SEQPACKET:
1766                 if (so)
1767                         soisdisconnected(so);
1768                 MPASS(unp2->unp_conn == unp);
1769                 unp2->unp_conn = NULL;
1770                 if (so2)
1771                         soisdisconnected(so2);
1772                 break;
1773         }
1774
1775         if (unp == unp2) {
1776                 unp_pcb_rele_notlast(unp);
1777                 if (!unp_pcb_rele(unp))
1778                         UNP_PCB_UNLOCK(unp);
1779         } else {
1780                 if (!unp_pcb_rele(unp))
1781                         UNP_PCB_UNLOCK(unp);
1782                 if (!unp_pcb_rele(unp2))
1783                         UNP_PCB_UNLOCK(unp2);
1784         }
1785 }
1786
1787 /*
1788  * unp_pcblist() walks the global list of struct unpcb's to generate a
1789  * pointer list, bumping the refcount on each unpcb.  It then copies them out
1790  * sequentially, validating the generation number on each to see if it has
1791  * been detached.  All of this is necessary because copyout() may sleep on
1792  * disk I/O.
1793  */
1794 static int
1795 unp_pcblist(SYSCTL_HANDLER_ARGS)
1796 {
1797         struct unpcb *unp, **unp_list;
1798         unp_gen_t gencnt;
1799         struct xunpgen *xug;
1800         struct unp_head *head;
1801         struct xunpcb *xu;
1802         u_int i;
1803         int error, n;
1804
1805         switch ((intptr_t)arg1) {
1806         case SOCK_STREAM:
1807                 head = &unp_shead;
1808                 break;
1809
1810         case SOCK_DGRAM:
1811                 head = &unp_dhead;
1812                 break;
1813
1814         case SOCK_SEQPACKET:
1815                 head = &unp_sphead;
1816                 break;
1817
1818         default:
1819                 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1820         }
1821
1822         /*
1823          * The process of preparing the PCB list is too time-consuming and
1824          * resource-intensive to repeat twice on every request.
1825          */
1826         if (req->oldptr == NULL) {
1827                 n = unp_count;
1828                 req->oldidx = 2 * (sizeof *xug)
1829                         + (n + n/8) * sizeof(struct xunpcb);
1830                 return (0);
1831         }
1832
1833         if (req->newptr != NULL)
1834                 return (EPERM);
1835
1836         /*
1837          * OK, now we're committed to doing something.
1838          */
1839         xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO);
1840         UNP_LINK_RLOCK();
1841         gencnt = unp_gencnt;
1842         n = unp_count;
1843         UNP_LINK_RUNLOCK();
1844
1845         xug->xug_len = sizeof *xug;
1846         xug->xug_count = n;
1847         xug->xug_gen = gencnt;
1848         xug->xug_sogen = so_gencnt;
1849         error = SYSCTL_OUT(req, xug, sizeof *xug);
1850         if (error) {
1851                 free(xug, M_TEMP);
1852                 return (error);
1853         }
1854
1855         unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1856
1857         UNP_LINK_RLOCK();
1858         for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1859              unp = LIST_NEXT(unp, unp_link)) {
1860                 UNP_PCB_LOCK(unp);
1861                 if (unp->unp_gencnt <= gencnt) {
1862                         if (cr_cansee(req->td->td_ucred,
1863                             unp->unp_socket->so_cred)) {
1864                                 UNP_PCB_UNLOCK(unp);
1865                                 continue;
1866                         }
1867                         unp_list[i++] = unp;
1868                         unp_pcb_hold(unp);
1869                 }
1870                 UNP_PCB_UNLOCK(unp);
1871         }
1872         UNP_LINK_RUNLOCK();
1873         n = i;                  /* In case we lost some during malloc. */
1874
1875         error = 0;
1876         xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1877         for (i = 0; i < n; i++) {
1878                 unp = unp_list[i];
1879                 UNP_PCB_LOCK(unp);
1880                 if (unp_pcb_rele(unp))
1881                         continue;
1882
1883                 if (unp->unp_gencnt <= gencnt) {
1884                         xu->xu_len = sizeof *xu;
1885                         xu->xu_unpp = (uintptr_t)unp;
1886                         /*
1887                          * XXX - need more locking here to protect against
1888                          * connect/disconnect races for SMP.
1889                          */
1890                         if (unp->unp_addr != NULL)
1891                                 bcopy(unp->unp_addr, &xu->xu_addr,
1892                                       unp->unp_addr->sun_len);
1893                         else
1894                                 bzero(&xu->xu_addr, sizeof(xu->xu_addr));
1895                         if (unp->unp_conn != NULL &&
1896                             unp->unp_conn->unp_addr != NULL)
1897                                 bcopy(unp->unp_conn->unp_addr,
1898                                       &xu->xu_caddr,
1899                                       unp->unp_conn->unp_addr->sun_len);
1900                         else
1901                                 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr));
1902                         xu->unp_vnode = (uintptr_t)unp->unp_vnode;
1903                         xu->unp_conn = (uintptr_t)unp->unp_conn;
1904                         xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs);
1905                         xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink);
1906                         xu->unp_gencnt = unp->unp_gencnt;
1907                         sotoxsocket(unp->unp_socket, &xu->xu_socket);
1908                         UNP_PCB_UNLOCK(unp);
1909                         error = SYSCTL_OUT(req, xu, sizeof *xu);
1910                 } else {
1911                         UNP_PCB_UNLOCK(unp);
1912                 }
1913         }
1914         free(xu, M_TEMP);
1915         if (!error) {
1916                 /*
1917                  * Give the user an updated idea of our state.  If the
1918                  * generation differs from what we told her before, she knows
1919                  * that something happened while we were processing this
1920                  * request, and it might be necessary to retry.
1921                  */
1922                 xug->xug_gen = unp_gencnt;
1923                 xug->xug_sogen = so_gencnt;
1924                 xug->xug_count = unp_count;
1925                 error = SYSCTL_OUT(req, xug, sizeof *xug);
1926         }
1927         free(unp_list, M_TEMP);
1928         free(xug, M_TEMP);
1929         return (error);
1930 }
1931
1932 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist,
1933     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1934     (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1935     "List of active local datagram sockets");
1936 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist,
1937     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1938     (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1939     "List of active local stream sockets");
1940 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1941     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1942     (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1943     "List of active local seqpacket sockets");
1944
1945 static void
1946 unp_shutdown(struct unpcb *unp)
1947 {
1948         struct unpcb *unp2;
1949         struct socket *so;
1950
1951         UNP_PCB_LOCK_ASSERT(unp);
1952
1953         unp2 = unp->unp_conn;
1954         if ((unp->unp_socket->so_type == SOCK_STREAM ||
1955             (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1956                 so = unp2->unp_socket;
1957                 if (so != NULL)
1958                         socantrcvmore(so);
1959         }
1960 }
1961
1962 static void
1963 unp_drop(struct unpcb *unp)
1964 {
1965         struct socket *so = unp->unp_socket;
1966         struct unpcb *unp2;
1967
1968         /*
1969          * Regardless of whether the socket's peer dropped the connection
1970          * with this socket by aborting or disconnecting, POSIX requires
1971          * that ECONNRESET is returned.
1972          */
1973
1974         UNP_PCB_LOCK(unp);
1975         if (so)
1976                 so->so_error = ECONNRESET;
1977         if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) {
1978                 /* Last reference dropped in unp_disconnect(). */
1979                 unp_pcb_rele_notlast(unp);
1980                 unp_disconnect(unp, unp2);
1981         } else if (!unp_pcb_rele(unp)) {
1982                 UNP_PCB_UNLOCK(unp);
1983         }
1984 }
1985
1986 static void
1987 unp_freerights(struct filedescent **fdep, int fdcount)
1988 {
1989         struct file *fp;
1990         int i;
1991
1992         KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1993
1994         for (i = 0; i < fdcount; i++) {
1995                 fp = fdep[i]->fde_file;
1996                 filecaps_free(&fdep[i]->fde_caps);
1997                 unp_discard(fp);
1998         }
1999         free(fdep[0], M_FILECAPS);
2000 }
2001
2002 static int
2003 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
2004 {
2005         struct thread *td = curthread;          /* XXX */
2006         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
2007         int i;
2008         int *fdp;
2009         struct filedesc *fdesc = td->td_proc->p_fd;
2010         struct filedescent **fdep;
2011         void *data;
2012         socklen_t clen = control->m_len, datalen;
2013         int error, newfds;
2014         u_int newlen;
2015
2016         UNP_LINK_UNLOCK_ASSERT();
2017
2018         error = 0;
2019         if (controlp != NULL) /* controlp == NULL => free control messages */
2020                 *controlp = NULL;
2021         while (cm != NULL) {
2022                 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
2023                         error = EINVAL;
2024                         break;
2025                 }
2026                 data = CMSG_DATA(cm);
2027                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
2028                 if (cm->cmsg_level == SOL_SOCKET
2029                     && cm->cmsg_type == SCM_RIGHTS) {
2030                         newfds = datalen / sizeof(*fdep);
2031                         if (newfds == 0)
2032                                 goto next;
2033                         fdep = data;
2034
2035                         /* If we're not outputting the descriptors free them. */
2036                         if (error || controlp == NULL) {
2037                                 unp_freerights(fdep, newfds);
2038                                 goto next;
2039                         }
2040                         FILEDESC_XLOCK(fdesc);
2041
2042                         /*
2043                          * Now change each pointer to an fd in the global
2044                          * table to an integer that is the index to the local
2045                          * fd table entry that we set up to point to the
2046                          * global one we are transferring.
2047                          */
2048                         newlen = newfds * sizeof(int);
2049                         *controlp = sbcreatecontrol(NULL, newlen,
2050                             SCM_RIGHTS, SOL_SOCKET);
2051                         if (*controlp == NULL) {
2052                                 FILEDESC_XUNLOCK(fdesc);
2053                                 error = E2BIG;
2054                                 unp_freerights(fdep, newfds);
2055                                 goto next;
2056                         }
2057
2058                         fdp = (int *)
2059                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2060                         if (fdallocn(td, 0, fdp, newfds) != 0) {
2061                                 FILEDESC_XUNLOCK(fdesc);
2062                                 error = EMSGSIZE;
2063                                 unp_freerights(fdep, newfds);
2064                                 m_freem(*controlp);
2065                                 *controlp = NULL;
2066                                 goto next;
2067                         }
2068                         for (i = 0; i < newfds; i++, fdp++) {
2069                                 _finstall(fdesc, fdep[i]->fde_file, *fdp,
2070                                     (flags & MSG_CMSG_CLOEXEC) != 0 ? O_CLOEXEC : 0,
2071                                     &fdep[i]->fde_caps);
2072                                 unp_externalize_fp(fdep[i]->fde_file);
2073                         }
2074
2075                         /*
2076                          * The new type indicates that the mbuf data refers to
2077                          * kernel resources that may need to be released before
2078                          * the mbuf is freed.
2079                          */
2080                         m_chtype(*controlp, MT_EXTCONTROL);
2081                         FILEDESC_XUNLOCK(fdesc);
2082                         free(fdep[0], M_FILECAPS);
2083                 } else {
2084                         /* We can just copy anything else across. */
2085                         if (error || controlp == NULL)
2086                                 goto next;
2087                         *controlp = sbcreatecontrol(NULL, datalen,
2088                             cm->cmsg_type, cm->cmsg_level);
2089                         if (*controlp == NULL) {
2090                                 error = ENOBUFS;
2091                                 goto next;
2092                         }
2093                         bcopy(data,
2094                             CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
2095                             datalen);
2096                 }
2097                 controlp = &(*controlp)->m_next;
2098
2099 next:
2100                 if (CMSG_SPACE(datalen) < clen) {
2101                         clen -= CMSG_SPACE(datalen);
2102                         cm = (struct cmsghdr *)
2103                             ((caddr_t)cm + CMSG_SPACE(datalen));
2104                 } else {
2105                         clen = 0;
2106                         cm = NULL;
2107                 }
2108         }
2109
2110         m_freem(control);
2111         return (error);
2112 }
2113
2114 static void
2115 unp_zone_change(void *tag)
2116 {
2117
2118         uma_zone_set_max(unp_zone, maxsockets);
2119 }
2120
2121 #ifdef INVARIANTS
2122 static void
2123 unp_zdtor(void *mem, int size __unused, void *arg __unused)
2124 {
2125         struct unpcb *unp;
2126
2127         unp = mem;
2128
2129         KASSERT(LIST_EMPTY(&unp->unp_refs),
2130             ("%s: unpcb %p has lingering refs", __func__, unp));
2131         KASSERT(unp->unp_socket == NULL,
2132             ("%s: unpcb %p has socket backpointer", __func__, unp));
2133         KASSERT(unp->unp_vnode == NULL,
2134             ("%s: unpcb %p has vnode references", __func__, unp));
2135         KASSERT(unp->unp_conn == NULL,
2136             ("%s: unpcb %p is still connected", __func__, unp));
2137         KASSERT(unp->unp_addr == NULL,
2138             ("%s: unpcb %p has leaked addr", __func__, unp));
2139 }
2140 #endif
2141
2142 static void
2143 unp_init(void)
2144 {
2145         uma_dtor dtor;
2146
2147 #ifdef VIMAGE
2148         if (!IS_DEFAULT_VNET(curvnet))
2149                 return;
2150 #endif
2151
2152 #ifdef INVARIANTS
2153         dtor = unp_zdtor;
2154 #else
2155         dtor = NULL;
2156 #endif
2157         unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor,
2158             NULL, NULL, UMA_ALIGN_CACHE, 0);
2159         uma_zone_set_max(unp_zone, maxsockets);
2160         uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
2161         EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
2162             NULL, EVENTHANDLER_PRI_ANY);
2163         LIST_INIT(&unp_dhead);
2164         LIST_INIT(&unp_shead);
2165         LIST_INIT(&unp_sphead);
2166         SLIST_INIT(&unp_defers);
2167         TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
2168         TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
2169         UNP_LINK_LOCK_INIT();
2170         UNP_DEFERRED_LOCK_INIT();
2171 }
2172
2173 static void
2174 unp_internalize_cleanup_rights(struct mbuf *control)
2175 {
2176         struct cmsghdr *cp;
2177         struct mbuf *m;
2178         void *data;
2179         socklen_t datalen;
2180
2181         for (m = control; m != NULL; m = m->m_next) {
2182                 cp = mtod(m, struct cmsghdr *);
2183                 if (cp->cmsg_level != SOL_SOCKET ||
2184                     cp->cmsg_type != SCM_RIGHTS)
2185                         continue;
2186                 data = CMSG_DATA(cp);
2187                 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data;
2188                 unp_freerights(data, datalen / sizeof(struct filedesc *));
2189         }
2190 }
2191
2192 static int
2193 unp_internalize(struct mbuf **controlp, struct thread *td)
2194 {
2195         struct mbuf *control, **initial_controlp;
2196         struct proc *p;
2197         struct filedesc *fdesc;
2198         struct bintime *bt;
2199         struct cmsghdr *cm;
2200         struct cmsgcred *cmcred;
2201         struct filedescent *fde, **fdep, *fdev;
2202         struct file *fp;
2203         struct timeval *tv;
2204         struct timespec *ts;
2205         void *data;
2206         socklen_t clen, datalen;
2207         int i, j, error, *fdp, oldfds;
2208         u_int newlen;
2209
2210         UNP_LINK_UNLOCK_ASSERT();
2211
2212         p = td->td_proc;
2213         fdesc = p->p_fd;
2214         error = 0;
2215         control = *controlp;
2216         clen = control->m_len;
2217         *controlp = NULL;
2218         initial_controlp = controlp;
2219         for (cm = mtod(control, struct cmsghdr *); cm != NULL;) {
2220                 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
2221                     || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
2222                         error = EINVAL;
2223                         goto out;
2224                 }
2225                 data = CMSG_DATA(cm);
2226                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
2227
2228                 switch (cm->cmsg_type) {
2229                 /*
2230                  * Fill in credential information.
2231                  */
2232                 case SCM_CREDS:
2233                         *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
2234                             SCM_CREDS, SOL_SOCKET);
2235                         if (*controlp == NULL) {
2236                                 error = ENOBUFS;
2237                                 goto out;
2238                         }
2239                         cmcred = (struct cmsgcred *)
2240                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2241                         cmcred->cmcred_pid = p->p_pid;
2242                         cmcred->cmcred_uid = td->td_ucred->cr_ruid;
2243                         cmcred->cmcred_gid = td->td_ucred->cr_rgid;
2244                         cmcred->cmcred_euid = td->td_ucred->cr_uid;
2245                         cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
2246                             CMGROUP_MAX);
2247                         for (i = 0; i < cmcred->cmcred_ngroups; i++)
2248                                 cmcred->cmcred_groups[i] =
2249                                     td->td_ucred->cr_groups[i];
2250                         break;
2251
2252                 case SCM_RIGHTS:
2253                         oldfds = datalen / sizeof (int);
2254                         if (oldfds == 0)
2255                                 break;
2256                         /*
2257                          * Check that all the FDs passed in refer to legal
2258                          * files.  If not, reject the entire operation.
2259                          */
2260                         fdp = data;
2261                         FILEDESC_SLOCK(fdesc);
2262                         for (i = 0; i < oldfds; i++, fdp++) {
2263                                 fp = fget_locked(fdesc, *fdp);
2264                                 if (fp == NULL) {
2265                                         FILEDESC_SUNLOCK(fdesc);
2266                                         error = EBADF;
2267                                         goto out;
2268                                 }
2269                                 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
2270                                         FILEDESC_SUNLOCK(fdesc);
2271                                         error = EOPNOTSUPP;
2272                                         goto out;
2273                                 }
2274                         }
2275
2276                         /*
2277                          * Now replace the integer FDs with pointers to the
2278                          * file structure and capability rights.
2279                          */
2280                         newlen = oldfds * sizeof(fdep[0]);
2281                         *controlp = sbcreatecontrol(NULL, newlen,
2282                             SCM_RIGHTS, SOL_SOCKET);
2283                         if (*controlp == NULL) {
2284                                 FILEDESC_SUNLOCK(fdesc);
2285                                 error = E2BIG;
2286                                 goto out;
2287                         }
2288                         fdp = data;
2289                         for (i = 0; i < oldfds; i++, fdp++) {
2290                                 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) {
2291                                         fdp = data;
2292                                         for (j = 0; j < i; j++, fdp++) {
2293                                                 fdrop(fdesc->fd_ofiles[*fdp].
2294                                                     fde_file, td);
2295                                         }
2296                                         FILEDESC_SUNLOCK(fdesc);
2297                                         error = EBADF;
2298                                         goto out;
2299                                 }
2300                         }
2301                         fdp = data;
2302                         fdep = (struct filedescent **)
2303                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2304                         fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
2305                             M_WAITOK);
2306                         for (i = 0; i < oldfds; i++, fdev++, fdp++) {
2307                                 fde = &fdesc->fd_ofiles[*fdp];
2308                                 fdep[i] = fdev;
2309                                 fdep[i]->fde_file = fde->fde_file;
2310                                 filecaps_copy(&fde->fde_caps,
2311                                     &fdep[i]->fde_caps, true);
2312                                 unp_internalize_fp(fdep[i]->fde_file);
2313                         }
2314                         FILEDESC_SUNLOCK(fdesc);
2315                         break;
2316
2317                 case SCM_TIMESTAMP:
2318                         *controlp = sbcreatecontrol(NULL, sizeof(*tv),
2319                             SCM_TIMESTAMP, SOL_SOCKET);
2320                         if (*controlp == NULL) {
2321                                 error = ENOBUFS;
2322                                 goto out;
2323                         }
2324                         tv = (struct timeval *)
2325                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2326                         microtime(tv);
2327                         break;
2328
2329                 case SCM_BINTIME:
2330                         *controlp = sbcreatecontrol(NULL, sizeof(*bt),
2331                             SCM_BINTIME, SOL_SOCKET);
2332                         if (*controlp == NULL) {
2333                                 error = ENOBUFS;
2334                                 goto out;
2335                         }
2336                         bt = (struct bintime *)
2337                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2338                         bintime(bt);
2339                         break;
2340
2341                 case SCM_REALTIME:
2342                         *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2343                             SCM_REALTIME, SOL_SOCKET);
2344                         if (*controlp == NULL) {
2345                                 error = ENOBUFS;
2346                                 goto out;
2347                         }
2348                         ts = (struct timespec *)
2349                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2350                         nanotime(ts);
2351                         break;
2352
2353                 case SCM_MONOTONIC:
2354                         *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2355                             SCM_MONOTONIC, SOL_SOCKET);
2356                         if (*controlp == NULL) {
2357                                 error = ENOBUFS;
2358                                 goto out;
2359                         }
2360                         ts = (struct timespec *)
2361                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2362                         nanouptime(ts);
2363                         break;
2364
2365                 default:
2366                         error = EINVAL;
2367                         goto out;
2368                 }
2369
2370                 if (*controlp != NULL)
2371                         controlp = &(*controlp)->m_next;
2372                 if (CMSG_SPACE(datalen) < clen) {
2373                         clen -= CMSG_SPACE(datalen);
2374                         cm = (struct cmsghdr *)
2375                             ((caddr_t)cm + CMSG_SPACE(datalen));
2376                 } else {
2377                         clen = 0;
2378                         cm = NULL;
2379                 }
2380         }
2381
2382 out:
2383         if (error != 0 && initial_controlp != NULL)
2384                 unp_internalize_cleanup_rights(*initial_controlp);
2385         m_freem(control);
2386         return (error);
2387 }
2388
2389 static struct mbuf *
2390 unp_addsockcred(struct thread *td, struct mbuf *control, int mode)
2391 {
2392         struct mbuf *m, *n, *n_prev;
2393         const struct cmsghdr *cm;
2394         int ngroups, i, cmsgtype;
2395         size_t ctrlsz;
2396
2397         ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2398         if (mode & UNP_WANTCRED_ALWAYS) {
2399                 ctrlsz = SOCKCRED2SIZE(ngroups);
2400                 cmsgtype = SCM_CREDS2;
2401         } else {
2402                 ctrlsz = SOCKCREDSIZE(ngroups);
2403                 cmsgtype = SCM_CREDS;
2404         }
2405
2406         m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET);
2407         if (m == NULL)
2408                 return (control);
2409
2410         if (mode & UNP_WANTCRED_ALWAYS) {
2411                 struct sockcred2 *sc;
2412
2413                 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *));
2414                 sc->sc_version = 0;
2415                 sc->sc_pid = td->td_proc->p_pid;
2416                 sc->sc_uid = td->td_ucred->cr_ruid;
2417                 sc->sc_euid = td->td_ucred->cr_uid;
2418                 sc->sc_gid = td->td_ucred->cr_rgid;
2419                 sc->sc_egid = td->td_ucred->cr_gid;
2420                 sc->sc_ngroups = ngroups;
2421                 for (i = 0; i < sc->sc_ngroups; i++)
2422                         sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2423         } else {
2424                 struct sockcred *sc;
2425
2426                 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *));
2427                 sc->sc_uid = td->td_ucred->cr_ruid;
2428                 sc->sc_euid = td->td_ucred->cr_uid;
2429                 sc->sc_gid = td->td_ucred->cr_rgid;
2430                 sc->sc_egid = td->td_ucred->cr_gid;
2431                 sc->sc_ngroups = ngroups;
2432                 for (i = 0; i < sc->sc_ngroups; i++)
2433                         sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2434         }
2435
2436         /*
2437          * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2438          * created SCM_CREDS control message (struct sockcred) has another
2439          * format.
2440          */
2441         if (control != NULL && cmsgtype == SCM_CREDS)
2442                 for (n = control, n_prev = NULL; n != NULL;) {
2443                         cm = mtod(n, struct cmsghdr *);
2444                         if (cm->cmsg_level == SOL_SOCKET &&
2445                             cm->cmsg_type == SCM_CREDS) {
2446                                 if (n_prev == NULL)
2447                                         control = n->m_next;
2448                                 else
2449                                         n_prev->m_next = n->m_next;
2450                                 n = m_free(n);
2451                         } else {
2452                                 n_prev = n;
2453                                 n = n->m_next;
2454                         }
2455                 }
2456
2457         /* Prepend it to the head. */
2458         m->m_next = control;
2459         return (m);
2460 }
2461
2462 static struct unpcb *
2463 fptounp(struct file *fp)
2464 {
2465         struct socket *so;
2466
2467         if (fp->f_type != DTYPE_SOCKET)
2468                 return (NULL);
2469         if ((so = fp->f_data) == NULL)
2470                 return (NULL);
2471         if (so->so_proto->pr_domain != &localdomain)
2472                 return (NULL);
2473         return sotounpcb(so);
2474 }
2475
2476 static void
2477 unp_discard(struct file *fp)
2478 {
2479         struct unp_defer *dr;
2480
2481         if (unp_externalize_fp(fp)) {
2482                 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2483                 dr->ud_fp = fp;
2484                 UNP_DEFERRED_LOCK();
2485                 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2486                 UNP_DEFERRED_UNLOCK();
2487                 atomic_add_int(&unp_defers_count, 1);
2488                 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2489         } else
2490                 closef_nothread(fp);
2491 }
2492
2493 static void
2494 unp_process_defers(void *arg __unused, int pending)
2495 {
2496         struct unp_defer *dr;
2497         SLIST_HEAD(, unp_defer) drl;
2498         int count;
2499
2500         SLIST_INIT(&drl);
2501         for (;;) {
2502                 UNP_DEFERRED_LOCK();
2503                 if (SLIST_FIRST(&unp_defers) == NULL) {
2504                         UNP_DEFERRED_UNLOCK();
2505                         break;
2506                 }
2507                 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2508                 UNP_DEFERRED_UNLOCK();
2509                 count = 0;
2510                 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2511                         SLIST_REMOVE_HEAD(&drl, ud_link);
2512                         closef_nothread(dr->ud_fp);
2513                         free(dr, M_TEMP);
2514                         count++;
2515                 }
2516                 atomic_add_int(&unp_defers_count, -count);
2517         }
2518 }
2519
2520 static void
2521 unp_internalize_fp(struct file *fp)
2522 {
2523         struct unpcb *unp;
2524
2525         UNP_LINK_WLOCK();
2526         if ((unp = fptounp(fp)) != NULL) {
2527                 unp->unp_file = fp;
2528                 unp->unp_msgcount++;
2529         }
2530         unp_rights++;
2531         UNP_LINK_WUNLOCK();
2532 }
2533
2534 static int
2535 unp_externalize_fp(struct file *fp)
2536 {
2537         struct unpcb *unp;
2538         int ret;
2539
2540         UNP_LINK_WLOCK();
2541         if ((unp = fptounp(fp)) != NULL) {
2542                 unp->unp_msgcount--;
2543                 ret = 1;
2544         } else
2545                 ret = 0;
2546         unp_rights--;
2547         UNP_LINK_WUNLOCK();
2548         return (ret);
2549 }
2550
2551 /*
2552  * unp_defer indicates whether additional work has been defered for a future
2553  * pass through unp_gc().  It is thread local and does not require explicit
2554  * synchronization.
2555  */
2556 static int      unp_marked;
2557
2558 static void
2559 unp_remove_dead_ref(struct filedescent **fdep, int fdcount)
2560 {
2561         struct unpcb *unp;
2562         struct file *fp;
2563         int i;
2564
2565         /*
2566          * This function can only be called from the gc task.
2567          */
2568         KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0,
2569             ("%s: not on gc callout", __func__));
2570         UNP_LINK_LOCK_ASSERT();
2571
2572         for (i = 0; i < fdcount; i++) {
2573                 fp = fdep[i]->fde_file;
2574                 if ((unp = fptounp(fp)) == NULL)
2575                         continue;
2576                 if ((unp->unp_gcflag & UNPGC_DEAD) == 0)
2577                         continue;
2578                 unp->unp_gcrefs--;
2579         }
2580 }
2581
2582 static void
2583 unp_restore_undead_ref(struct filedescent **fdep, int fdcount)
2584 {
2585         struct unpcb *unp;
2586         struct file *fp;
2587         int i;
2588
2589         /*
2590          * This function can only be called from the gc task.
2591          */
2592         KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0,
2593             ("%s: not on gc callout", __func__));
2594         UNP_LINK_LOCK_ASSERT();
2595
2596         for (i = 0; i < fdcount; i++) {
2597                 fp = fdep[i]->fde_file;
2598                 if ((unp = fptounp(fp)) == NULL)
2599                         continue;
2600                 if ((unp->unp_gcflag & UNPGC_DEAD) == 0)
2601                         continue;
2602                 unp->unp_gcrefs++;
2603                 unp_marked++;
2604         }
2605 }
2606
2607 static void
2608 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int))
2609 {
2610         struct socket *so, *soa;
2611
2612         so = unp->unp_socket;
2613         SOCK_LOCK(so);
2614         if (SOLISTENING(so)) {
2615                 /*
2616                  * Mark all sockets in our accept queue.
2617                  */
2618                 TAILQ_FOREACH(soa, &so->sol_comp, so_list) {
2619                         if (sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS)
2620                                 continue;
2621                         SOCKBUF_LOCK(&soa->so_rcv);
2622                         unp_scan(soa->so_rcv.sb_mb, op);
2623                         SOCKBUF_UNLOCK(&soa->so_rcv);
2624                 }
2625         } else {
2626                 /*
2627                  * Mark all sockets we reference with RIGHTS.
2628                  */
2629                 if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2630                         SOCKBUF_LOCK(&so->so_rcv);
2631                         unp_scan(so->so_rcv.sb_mb, op);
2632                         SOCKBUF_UNLOCK(&so->so_rcv);
2633                 }
2634         }
2635         SOCK_UNLOCK(so);
2636 }
2637
2638 static int unp_recycled;
2639 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2640     "Number of unreachable sockets claimed by the garbage collector.");
2641
2642 static int unp_taskcount;
2643 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2644     "Number of times the garbage collector has run.");
2645
2646 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0,
2647     "Number of active local sockets.");
2648
2649 static void
2650 unp_gc(__unused void *arg, int pending)
2651 {
2652         struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2653                                     NULL };
2654         struct unp_head **head;
2655         struct unp_head unp_deadhead;   /* List of potentially-dead sockets. */
2656         struct file *f, **unref;
2657         struct unpcb *unp, *unptmp;
2658         int i, total, unp_unreachable;
2659
2660         LIST_INIT(&unp_deadhead);
2661         unp_taskcount++;
2662         UNP_LINK_RLOCK();
2663         /*
2664          * First determine which sockets may be in cycles.
2665          */
2666         unp_unreachable = 0;
2667
2668         for (head = heads; *head != NULL; head++)
2669                 LIST_FOREACH(unp, *head, unp_link) {
2670                         KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0,
2671                             ("%s: unp %p has unexpected gc flags 0x%x",
2672                             __func__, unp, (unsigned int)unp->unp_gcflag));
2673
2674                         f = unp->unp_file;
2675
2676                         /*
2677                          * Check for an unreachable socket potentially in a
2678                          * cycle.  It must be in a queue as indicated by
2679                          * msgcount, and this must equal the file reference
2680                          * count.  Note that when msgcount is 0 the file is
2681                          * NULL.
2682                          */
2683                         if (f != NULL && unp->unp_msgcount != 0 &&
2684                             refcount_load(&f->f_count) == unp->unp_msgcount) {
2685                                 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead);
2686                                 unp->unp_gcflag |= UNPGC_DEAD;
2687                                 unp->unp_gcrefs = unp->unp_msgcount;
2688                                 unp_unreachable++;
2689                         }
2690                 }
2691
2692         /*
2693          * Scan all sockets previously marked as potentially being in a cycle
2694          * and remove the references each socket holds on any UNPGC_DEAD
2695          * sockets in its queue.  After this step, all remaining references on
2696          * sockets marked UNPGC_DEAD should not be part of any cycle.
2697          */
2698         LIST_FOREACH(unp, &unp_deadhead, unp_dead)
2699                 unp_gc_scan(unp, unp_remove_dead_ref);
2700
2701         /*
2702          * If a socket still has a non-negative refcount, it cannot be in a
2703          * cycle.  In this case increment refcount of all children iteratively.
2704          * Stop the scan once we do a complete loop without discovering
2705          * a new reachable socket.
2706          */
2707         do {
2708                 unp_marked = 0;
2709                 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp)
2710                         if (unp->unp_gcrefs > 0) {
2711                                 unp->unp_gcflag &= ~UNPGC_DEAD;
2712                                 LIST_REMOVE(unp, unp_dead);
2713                                 KASSERT(unp_unreachable > 0,
2714                                     ("%s: unp_unreachable underflow.",
2715                                     __func__));
2716                                 unp_unreachable--;
2717                                 unp_gc_scan(unp, unp_restore_undead_ref);
2718                         }
2719         } while (unp_marked);
2720
2721         UNP_LINK_RUNLOCK();
2722
2723         if (unp_unreachable == 0)
2724                 return;
2725
2726         /*
2727          * Allocate space for a local array of dead unpcbs.
2728          * TODO: can this path be simplified by instead using the local
2729          * dead list at unp_deadhead, after taking out references
2730          * on the file object and/or unpcb and dropping the link lock?
2731          */
2732         unref = malloc(unp_unreachable * sizeof(struct file *),
2733             M_TEMP, M_WAITOK);
2734
2735         /*
2736          * Iterate looking for sockets which have been specifically marked
2737          * as unreachable and store them locally.
2738          */
2739         UNP_LINK_RLOCK();
2740         total = 0;
2741         LIST_FOREACH(unp, &unp_deadhead, unp_dead) {
2742                 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0,
2743                     ("%s: unp %p not marked UNPGC_DEAD", __func__, unp));
2744                 unp->unp_gcflag &= ~UNPGC_DEAD;
2745                 f = unp->unp_file;
2746                 if (unp->unp_msgcount == 0 || f == NULL ||
2747                     refcount_load(&f->f_count) != unp->unp_msgcount ||
2748                     !fhold(f))
2749                         continue;
2750                 unref[total++] = f;
2751                 KASSERT(total <= unp_unreachable,
2752                     ("%s: incorrect unreachable count.", __func__));
2753         }
2754         UNP_LINK_RUNLOCK();
2755
2756         /*
2757          * Now flush all sockets, free'ing rights.  This will free the
2758          * struct files associated with these sockets but leave each socket
2759          * with one remaining ref.
2760          */
2761         for (i = 0; i < total; i++) {
2762                 struct socket *so;
2763
2764                 so = unref[i]->f_data;
2765                 CURVNET_SET(so->so_vnet);
2766                 sorflush(so);
2767                 CURVNET_RESTORE();
2768         }
2769
2770         /*
2771          * And finally release the sockets so they can be reclaimed.
2772          */
2773         for (i = 0; i < total; i++)
2774                 fdrop(unref[i], NULL);
2775         unp_recycled += total;
2776         free(unref, M_TEMP);
2777 }
2778
2779 static void
2780 unp_dispose_mbuf(struct mbuf *m)
2781 {
2782
2783         if (m)
2784                 unp_scan(m, unp_freerights);
2785 }
2786
2787 /*
2788  * Synchronize against unp_gc, which can trip over data as we are freeing it.
2789  */
2790 static void
2791 unp_dispose(struct socket *so)
2792 {
2793         struct unpcb *unp;
2794
2795         unp = sotounpcb(so);
2796         UNP_LINK_WLOCK();
2797         unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2798         UNP_LINK_WUNLOCK();
2799         if (!SOLISTENING(so))
2800                 unp_dispose_mbuf(so->so_rcv.sb_mb);
2801 }
2802
2803 static void
2804 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2805 {
2806         struct mbuf *m;
2807         struct cmsghdr *cm;
2808         void *data;
2809         socklen_t clen, datalen;
2810
2811         while (m0 != NULL) {
2812                 for (m = m0; m; m = m->m_next) {
2813                         if (m->m_type != MT_CONTROL)
2814                                 continue;
2815
2816                         cm = mtod(m, struct cmsghdr *);
2817                         clen = m->m_len;
2818
2819                         while (cm != NULL) {
2820                                 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2821                                         break;
2822
2823                                 data = CMSG_DATA(cm);
2824                                 datalen = (caddr_t)cm + cm->cmsg_len
2825                                     - (caddr_t)data;
2826
2827                                 if (cm->cmsg_level == SOL_SOCKET &&
2828                                     cm->cmsg_type == SCM_RIGHTS) {
2829                                         (*op)(data, datalen /
2830                                             sizeof(struct filedescent *));
2831                                 }
2832
2833                                 if (CMSG_SPACE(datalen) < clen) {
2834                                         clen -= CMSG_SPACE(datalen);
2835                                         cm = (struct cmsghdr *)
2836                                             ((caddr_t)cm + CMSG_SPACE(datalen));
2837                                 } else {
2838                                         clen = 0;
2839                                         cm = NULL;
2840                                 }
2841                         }
2842                 }
2843                 m0 = m0->m_nextpkt;
2844         }
2845 }
2846
2847 /*
2848  * A helper function called by VFS before socket-type vnode reclamation.
2849  * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2850  * use count.
2851  */
2852 void
2853 vfs_unp_reclaim(struct vnode *vp)
2854 {
2855         struct unpcb *unp;
2856         int active;
2857         struct mtx *vplock;
2858
2859         ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2860         KASSERT(vp->v_type == VSOCK,
2861             ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2862
2863         active = 0;
2864         vplock = mtx_pool_find(mtxpool_sleep, vp);
2865         mtx_lock(vplock);
2866         VOP_UNP_CONNECT(vp, &unp);
2867         if (unp == NULL)
2868                 goto done;
2869         UNP_PCB_LOCK(unp);
2870         if (unp->unp_vnode == vp) {
2871                 VOP_UNP_DETACH(vp);
2872                 unp->unp_vnode = NULL;
2873                 active = 1;
2874         }
2875         UNP_PCB_UNLOCK(unp);
2876  done:
2877         mtx_unlock(vplock);
2878         if (active)
2879                 vunref(vp);
2880 }
2881
2882 #ifdef DDB
2883 static void
2884 db_print_indent(int indent)
2885 {
2886         int i;
2887
2888         for (i = 0; i < indent; i++)
2889                 db_printf(" ");
2890 }
2891
2892 static void
2893 db_print_unpflags(int unp_flags)
2894 {
2895         int comma;
2896
2897         comma = 0;
2898         if (unp_flags & UNP_HAVEPC) {
2899                 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2900                 comma = 1;
2901         }
2902         if (unp_flags & UNP_WANTCRED_ALWAYS) {
2903                 db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : "");
2904                 comma = 1;
2905         }
2906         if (unp_flags & UNP_WANTCRED_ONESHOT) {
2907                 db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : "");
2908                 comma = 1;
2909         }
2910         if (unp_flags & UNP_CONNWAIT) {
2911                 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2912                 comma = 1;
2913         }
2914         if (unp_flags & UNP_CONNECTING) {
2915                 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2916                 comma = 1;
2917         }
2918         if (unp_flags & UNP_BINDING) {
2919                 db_printf("%sUNP_BINDING", comma ? ", " : "");
2920                 comma = 1;
2921         }
2922 }
2923
2924 static void
2925 db_print_xucred(int indent, struct xucred *xu)
2926 {
2927         int comma, i;
2928
2929         db_print_indent(indent);
2930         db_printf("cr_version: %u   cr_uid: %u   cr_pid: %d   cr_ngroups: %d\n",
2931             xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups);
2932         db_print_indent(indent);
2933         db_printf("cr_groups: ");
2934         comma = 0;
2935         for (i = 0; i < xu->cr_ngroups; i++) {
2936                 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2937                 comma = 1;
2938         }
2939         db_printf("\n");
2940 }
2941
2942 static void
2943 db_print_unprefs(int indent, struct unp_head *uh)
2944 {
2945         struct unpcb *unp;
2946         int counter;
2947
2948         counter = 0;
2949         LIST_FOREACH(unp, uh, unp_reflink) {
2950                 if (counter % 4 == 0)
2951                         db_print_indent(indent);
2952                 db_printf("%p  ", unp);
2953                 if (counter % 4 == 3)
2954                         db_printf("\n");
2955                 counter++;
2956         }
2957         if (counter != 0 && counter % 4 != 0)
2958                 db_printf("\n");
2959 }
2960
2961 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2962 {
2963         struct unpcb *unp;
2964
2965         if (!have_addr) {
2966                 db_printf("usage: show unpcb <addr>\n");
2967                 return;
2968         }
2969         unp = (struct unpcb *)addr;
2970
2971         db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
2972             unp->unp_vnode);
2973
2974         db_printf("unp_ino: %ju   unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2975             unp->unp_conn);
2976
2977         db_printf("unp_refs:\n");
2978         db_print_unprefs(2, &unp->unp_refs);
2979
2980         /* XXXRW: Would be nice to print the full address, if any. */
2981         db_printf("unp_addr: %p\n", unp->unp_addr);
2982
2983         db_printf("unp_gencnt: %llu\n",
2984             (unsigned long long)unp->unp_gencnt);
2985
2986         db_printf("unp_flags: %x (", unp->unp_flags);
2987         db_print_unpflags(unp->unp_flags);
2988         db_printf(")\n");
2989
2990         db_printf("unp_peercred:\n");
2991         db_print_xucred(2, &unp->unp_peercred);
2992
2993         db_printf("unp_refcount: %u\n", unp->unp_refcount);
2994 }
2995 #endif