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