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/fcntl.h>
  69 #include <sys/malloc.h>         /* XXX must be before <sys/file.h> */
  70 #include <sys/eventhandler.h>
  71 #include <sys/file.h>
  72 #include <sys/filedesc.h>
  73 #include <sys/kernel.h>
  74 #include <sys/lock.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  * Locking key:
 110  * (l)  Locked using list lock
 111  * (g)  Locked using linkage lock
 112  */
 113
 114 static uma_zone_t       unp_zone;
 115 static unp_gen_t        unp_gencnt;     /* (l) */
 116 static u_int            unp_count;      /* (l) Count of local sockets. */
 117 static ino_t            unp_ino;        /* Prototype for fake inode numbers. */
 118 static int              unp_rights;     /* (g) File descriptors in flight. */
 119 static struct unp_head  unp_shead;      /* (l) List of stream sockets. */
 120 static struct unp_head  unp_dhead;      /* (l) List of datagram sockets. */
 121 static struct unp_head  unp_sphead;     /* (l) List of seqpacket sockets. */
 122
 123 struct unp_defer {
 124         SLIST_ENTRY(unp_defer) ud_link;
 125         struct file *ud_fp;
 126 };
 127 static SLIST_HEAD(, unp_defer) unp_defers;
 128 static int unp_defers_count;
 129
 130 static const struct sockaddr    sun_noname = { sizeof(sun_noname), AF_LOCAL };
 131
 132 /*
 133  * Garbage collection of cyclic file descriptor/socket references occurs
 134  * asynchronously in a taskqueue context in order to avoid recursion and
 135  * reentrance in the UNIX domain socket, file descriptor, and socket layer
 136  * code.  See unp_gc() for a full description.
 137  */
 138 static struct timeout_task unp_gc_task;
 139
 140 /*
 141  * The close of unix domain sockets attached as SCM_RIGHTS is
 142  * postponed to the taskqueue, to avoid arbitrary recursion depth.
 143  * The attached sockets might have another sockets attached.
 144  */
 145 static struct task      unp_defer_task;
 146
 147 /*
 148  * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
 149  * stream sockets, although the total for sender and receiver is actually
 150  * only PIPSIZ.
 151  *
 152  * Datagram sockets really use the sendspace as the maximum datagram size,
 153  * and don't really want to reserve the sendspace.  Their recvspace should be
 154  * large enough for at least one max-size datagram plus address.
 155  */
 156 #ifndef PIPSIZ
 157 #define PIPSIZ  8192
 158 #endif
 159 static u_long   unpst_sendspace = PIPSIZ;
 160 static u_long   unpst_recvspace = PIPSIZ;
 161 static u_long   unpdg_sendspace = 2*1024;       /* really max datagram size */
 162 static u_long   unpdg_recvspace = 4*1024;
 163 static u_long   unpsp_sendspace = PIPSIZ;       /* really max datagram size */
 164 static u_long   unpsp_recvspace = PIPSIZ;
 165
 166 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
 167 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0,
 168     "SOCK_STREAM");
 169 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
 170 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0,
 171     "SOCK_SEQPACKET");
 172
 173 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
 174            &unpst_sendspace, 0, "Default stream send space.");
 175 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
 176            &unpst_recvspace, 0, "Default stream receive space.");
 177 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
 178            &unpdg_sendspace, 0, "Default datagram send space.");
 179 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
 180            &unpdg_recvspace, 0, "Default datagram receive space.");
 181 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
 182            &unpsp_sendspace, 0, "Default seqpacket send space.");
 183 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
 184            &unpsp_recvspace, 0, "Default seqpacket receive space.");
 185 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
 186     "File descriptors in flight.");
 187 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
 188     &unp_defers_count, 0,
 189     "File descriptors deferred to taskqueue for close.");
 190
 191 /*
 192  * Locking and synchronization:
 193  *
 194  * Three types of locks exist in the local domain socket implementation: a
 195  * a global linkage rwlock, the mtxpool lock, and per-unpcb mutexes.
 196  * The linkage lock protects the socket count, global generation number,
 197  * and stream/datagram global lists.
 198  *
 199  * The mtxpool lock protects the vnode from being modified while referenced.
 200  * Lock ordering requires that it be acquired before any unpcb locks.
 201  *
 202  * The unpcb lock (unp_mtx) protects all fields in the unpcb. Of particular
 203  * note is that this includes the unp_conn field. So long as the unpcb lock
 204  * is held the reference to the unpcb pointed to by unp_conn is valid. If we
 205  * require that the unpcb pointed to by unp_conn remain live in cases where
 206  * we need to drop the unp_mtx as when we need to acquire the lock for a
 207  * second unpcb the caller must first acquire an additional reference on the
 208  * second unpcb and then revalidate any state (typically check that unp_conn
 209  * is non-NULL) upon requiring the initial unpcb lock. The lock ordering
 210  * between unpcbs is the conventional ascending address order. Two helper
 211  * routines exist for this:
 212  *
 213  *   - unp_pcb_lock2(unp, unp2) - which just acquires the two locks in the
 214  *     safe ordering.
 215  *
 216  *   - unp_pcb_owned_lock2(unp, unp2, freed) - the lock for unp is held
 217  *     when called. If unp is unlocked and unp2 is subsequently freed
 218  *     freed will be set to 1.
 219  *
 220  * The helper routines for references are:
 221  *
 222  *   - unp_pcb_hold(unp): Can be called any time we currently hold a valid
 223  *     reference to unp.
 224  *
 225  *    - unp_pcb_rele(unp): The caller must hold the unp lock. If we are
 226  *      releasing the last reference, detach must have been called thus
 227  *      unp->unp_socket be NULL.
 228  *
 229  * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
 230  * allocated in pru_attach() and freed in pru_detach().  The validity of that
 231  * pointer is an invariant, so no lock is required to dereference the so_pcb
 232  * pointer if a valid socket reference is held by the caller.  In practice,
 233  * this is always true during operations performed on a socket.  Each unpcb
 234  * has a back-pointer to its socket, unp_socket, which will be stable under
 235  * the same circumstances.
 236  *
 237  * This pointer may only be safely dereferenced as long as a valid reference
 238  * to the unpcb is held.  Typically, this reference will be from the socket,
 239  * or from another unpcb when the referring unpcb's lock is held (in order
 240  * that the reference not be invalidated during use).  For example, to follow
 241  * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee
 242  * that detach is not run clearing unp_socket.
 243  *
 244  * Blocking with UNIX domain sockets is a tricky issue: unlike most network
 245  * protocols, bind() is a non-atomic operation, and connect() requires
 246  * potential sleeping in the protocol, due to potentially waiting on local or
 247  * distributed file systems.  We try to separate "lookup" operations, which
 248  * may sleep, and the IPC operations themselves, which typically can occur
 249  * with relative atomicity as locks can be held over the entire operation.
 250  *
 251  * Another tricky issue is simultaneous multi-threaded or multi-process
 252  * access to a single UNIX domain socket.  These are handled by the flags
 253  * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
 254  * binding, both of which involve dropping UNIX domain socket locks in order
 255  * to perform namei() and other file system operations.
 256  */
 257 static struct rwlock    unp_link_rwlock;
 258 static struct mtx       unp_defers_lock;
 259
 260 #define UNP_LINK_LOCK_INIT()            rw_init(&unp_link_rwlock,       \
 261                                             "unp_link_rwlock")
 262
 263 #define UNP_LINK_LOCK_ASSERT()  rw_assert(&unp_link_rwlock,     \
 264                                             RA_LOCKED)
 265 #define UNP_LINK_UNLOCK_ASSERT()        rw_assert(&unp_link_rwlock,     \
 266                                             RA_UNLOCKED)
 267
 268 #define UNP_LINK_RLOCK()                rw_rlock(&unp_link_rwlock)
 269 #define UNP_LINK_RUNLOCK()              rw_runlock(&unp_link_rwlock)
 270 #define UNP_LINK_WLOCK()                rw_wlock(&unp_link_rwlock)
 271 #define UNP_LINK_WUNLOCK()              rw_wunlock(&unp_link_rwlock)
 272 #define UNP_LINK_WLOCK_ASSERT()         rw_assert(&unp_link_rwlock,     \
 273                                             RA_WLOCKED)
 274 #define UNP_LINK_WOWNED()               rw_wowned(&unp_link_rwlock)
 275
 276 #define UNP_DEFERRED_LOCK_INIT()        mtx_init(&unp_defers_lock, \
 277                                             "unp_defer", NULL, MTX_DEF)
 278 #define UNP_DEFERRED_LOCK()             mtx_lock(&unp_defers_lock)
 279 #define UNP_DEFERRED_UNLOCK()           mtx_unlock(&unp_defers_lock)
 280
 281 #define UNP_REF_LIST_LOCK()             UNP_DEFERRED_LOCK();
 282 #define UNP_REF_LIST_UNLOCK()           UNP_DEFERRED_UNLOCK();
 283
 284 #define UNP_PCB_LOCK_INIT(unp)          mtx_init(&(unp)->unp_mtx,       \
 285                                             "unp", "unp",       \
 286                                             MTX_DUPOK|MTX_DEF)
 287 #define UNP_PCB_LOCK_DESTROY(unp)       mtx_destroy(&(unp)->unp_mtx)
 288 #define UNP_PCB_LOCK(unp)               mtx_lock(&(unp)->unp_mtx)
 289 #define UNP_PCB_TRYLOCK(unp)            mtx_trylock(&(unp)->unp_mtx)
 290 #define UNP_PCB_UNLOCK(unp)             mtx_unlock(&(unp)->unp_mtx)
 291 #define UNP_PCB_OWNED(unp)              mtx_owned(&(unp)->unp_mtx)
 292 #define UNP_PCB_LOCK_ASSERT(unp)        mtx_assert(&(unp)->unp_mtx, MA_OWNED)
 293 #define UNP_PCB_UNLOCK_ASSERT(unp)      mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED)
 294
 295 static int      uipc_connect2(struct socket *, struct socket *);
 296 static int      uipc_ctloutput(struct socket *, struct sockopt *);
 297 static int      unp_connect(struct socket *, struct sockaddr *,
 298                     struct thread *);
 299 static int      unp_connectat(int, struct socket *, struct sockaddr *,
 300                     struct thread *);
 301 static int      unp_connect2(struct socket *so, struct socket *so2, int);
 302 static void     unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
 303 static void     unp_dispose(struct socket *so);
 304 static void     unp_dispose_mbuf(struct mbuf *);
 305 static void     unp_shutdown(struct unpcb *);
 306 static void     unp_drop(struct unpcb *);
 307 static void     unp_gc(__unused void *, int);
 308 static void     unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
 309 static void     unp_discard(struct file *);
 310 static void     unp_freerights(struct filedescent **, int);
 311 static void     unp_init(void);
 312 static int      unp_internalize(struct mbuf **, struct thread *);
 313 static void     unp_internalize_fp(struct file *);
 314 static int      unp_externalize(struct mbuf *, struct mbuf **, int);
 315 static int      unp_externalize_fp(struct file *);
 316 static struct mbuf      *unp_addsockcred(struct thread *, struct mbuf *);
 317 static void     unp_process_defers(void * __unused, int);
 318
 319
 320 static void
 321 unp_pcb_hold(struct unpcb *unp)
 322 {
 323         MPASS(unp->unp_refcount);
 324         refcount_acquire(&unp->unp_refcount);
 325 }
 326
 327 static int
 328 unp_pcb_rele(struct unpcb *unp)
 329 {
 330         int freed;
 331
 332         UNP_PCB_LOCK_ASSERT(unp);
 333         MPASS(unp->unp_refcount);
 334         if ((freed = refcount_release(&unp->unp_refcount))) {
 335                 /* we got here with having detached? */
 336                 MPASS(unp->unp_socket == NULL);
 337                 UNP_PCB_UNLOCK(unp);
 338                 UNP_PCB_LOCK_DESTROY(unp);
 339                 uma_zfree(unp_zone, unp);
 340         }
 341         return (freed);
 342 }
 343
 344 static void
 345 unp_pcb_lock2(struct unpcb *unp, struct unpcb *unp2)
 346 {
 347         MPASS(unp != unp2);
 348         UNP_PCB_UNLOCK_ASSERT(unp);
 349         UNP_PCB_UNLOCK_ASSERT(unp2);
 350         if ((uintptr_t)unp2 > (uintptr_t)unp) {
 351                 UNP_PCB_LOCK(unp);
 352                 UNP_PCB_LOCK(unp2);
 353         } else {
 354                 UNP_PCB_LOCK(unp2);
 355                 UNP_PCB_LOCK(unp);
 356         }
 357 }
 358
 359 static __noinline void
 360 unp_pcb_owned_lock2_slowpath(struct unpcb *unp, struct unpcb **unp2p,
 361     int *freed)
 362 {
 363         struct unpcb *unp2;
 364
 365         unp2 = *unp2p;
 366         unp_pcb_hold(unp2);
 367         UNP_PCB_UNLOCK(unp);
 368         UNP_PCB_LOCK(unp2);
 369         UNP_PCB_LOCK(unp);
 370         *freed = unp_pcb_rele(unp2);
 371         if (*freed)
 372                 *unp2p = NULL;
 373 }
 374
 375 #define unp_pcb_owned_lock2(unp, unp2, freed) do {                      \
 376         freed = 0;                                                      \
 377         UNP_PCB_LOCK_ASSERT(unp);                                       \
 378         UNP_PCB_UNLOCK_ASSERT(unp2);                                    \
 379         MPASS((unp) != (unp2));                                         \
 380         if (__predict_true(UNP_PCB_TRYLOCK(unp2)))                      \
 381                 break;                                                  \
 382         else if ((uintptr_t)(unp2) > (uintptr_t)(unp))                  \
 383                 UNP_PCB_LOCK(unp2);                                     \
 384         else                                                            \
 385                 unp_pcb_owned_lock2_slowpath((unp), &(unp2), &freed);   \
 386 } while (0)
 387
 388
 389 /*
 390  * Definitions of protocols supported in the LOCAL domain.
 391  */
 392 static struct domain localdomain;
 393 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
 394 static struct pr_usrreqs uipc_usrreqs_seqpacket;
 395 static struct protosw localsw[] = {
 396 {
 397         .pr_type =              SOCK_STREAM,
 398         .pr_domain =            &localdomain,
 399         .pr_flags =             PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
 400         .pr_ctloutput =         &uipc_ctloutput,
 401         .pr_usrreqs =           &uipc_usrreqs_stream
 402 },
 403 {
 404         .pr_type =              SOCK_DGRAM,
 405         .pr_domain =            &localdomain,
 406         .pr_flags =             PR_ATOMIC|PR_ADDR|PR_RIGHTS,
 407         .pr_ctloutput =         &uipc_ctloutput,
 408         .pr_usrreqs =           &uipc_usrreqs_dgram
 409 },
 410 {
 411         .pr_type =              SOCK_SEQPACKET,
 412         .pr_domain =            &localdomain,
 413
 414         /*
 415          * XXXRW: For now, PR_ADDR because soreceive will bump into them
 416          * due to our use of sbappendaddr.  A new sbappend variants is needed
 417          * that supports both atomic record writes and control data.
 418          */
 419         .pr_flags =             PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
 420                                     PR_RIGHTS,
 421         .pr_ctloutput =         &uipc_ctloutput,
 422         .pr_usrreqs =           &uipc_usrreqs_seqpacket,
 423 },
 424 };
 425
 426 static struct domain localdomain = {
 427         .dom_family =           AF_LOCAL,
 428         .dom_name =             "local",
 429         .dom_init =             unp_init,
 430         .dom_externalize =      unp_externalize,
 431         .dom_dispose =          unp_dispose,
 432         .dom_protosw =          localsw,
 433         .dom_protoswNPROTOSW =  &localsw[nitems(localsw)]
 434 };
 435 DOMAIN_SET(local);
 436
 437 static void
 438 uipc_abort(struct socket *so)
 439 {
 440         struct unpcb *unp, *unp2;
 441
 442         unp = sotounpcb(so);
 443         KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
 444         UNP_PCB_UNLOCK_ASSERT(unp);
 445
 446         UNP_PCB_LOCK(unp);
 447         unp2 = unp->unp_conn;
 448         if (unp2 != NULL) {
 449                 unp_pcb_hold(unp2);
 450                 UNP_PCB_UNLOCK(unp);
 451                 unp_drop(unp2);
 452         } else
 453                 UNP_PCB_UNLOCK(unp);
 454 }
 455
 456 static int
 457 uipc_accept(struct socket *so, struct sockaddr **nam)
 458 {
 459         struct unpcb *unp, *unp2;
 460         const struct sockaddr *sa;
 461
 462         /*
 463          * Pass back name of connected socket, if it was bound and we are
 464          * still connected (our peer may have closed already!).
 465          */
 466         unp = sotounpcb(so);
 467         KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
 468
 469         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 470         UNP_LINK_RLOCK();
 471         unp2 = unp->unp_conn;
 472         if (unp2 != NULL && unp2->unp_addr != NULL) {
 473                 UNP_PCB_LOCK(unp2);
 474                 sa = (struct sockaddr *) unp2->unp_addr;
 475                 bcopy(sa, *nam, sa->sa_len);
 476                 UNP_PCB_UNLOCK(unp2);
 477         } else {
 478                 sa = &sun_noname;
 479                 bcopy(sa, *nam, sa->sa_len);
 480         }
 481         UNP_LINK_RUNLOCK();
 482         return (0);
 483 }
 484
 485 static int
 486 uipc_attach(struct socket *so, int proto, struct thread *td)
 487 {
 488         u_long sendspace, recvspace;
 489         struct unpcb *unp;
 490         int error;
 491         bool locked;
 492
 493         KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
 494         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
 495                 switch (so->so_type) {
 496                 case SOCK_STREAM:
 497                         sendspace = unpst_sendspace;
 498                         recvspace = unpst_recvspace;
 499                         break;
 500
 501                 case SOCK_DGRAM:
 502                         sendspace = unpdg_sendspace;
 503                         recvspace = unpdg_recvspace;
 504                         break;
 505
 506                 case SOCK_SEQPACKET:
 507                         sendspace = unpsp_sendspace;
 508                         recvspace = unpsp_recvspace;
 509                         break;
 510
 511                 default:
 512                         panic("uipc_attach");
 513                 }
 514                 error = soreserve(so, sendspace, recvspace);
 515                 if (error)
 516                         return (error);
 517         }
 518         unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
 519         if (unp == NULL)
 520                 return (ENOBUFS);
 521         LIST_INIT(&unp->unp_refs);
 522         UNP_PCB_LOCK_INIT(unp);
 523         unp->unp_socket = so;
 524         so->so_pcb = unp;
 525         unp->unp_refcount = 1;
 526         if (so->so_listen != NULL)
 527                 unp->unp_flags |= UNP_NASCENT;
 528
 529         if ((locked = UNP_LINK_WOWNED()) == false)
 530                 UNP_LINK_WLOCK();
 531
 532         unp->unp_gencnt = ++unp_gencnt;
 533         unp->unp_ino = ++unp_ino;
 534         unp_count++;
 535         switch (so->so_type) {
 536         case SOCK_STREAM:
 537                 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
 538                 break;
 539
 540         case SOCK_DGRAM:
 541                 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
 542                 break;
 543
 544         case SOCK_SEQPACKET:
 545                 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
 546                 break;
 547
 548         default:
 549                 panic("uipc_attach");
 550         }
 551
 552         if (locked == false)
 553                 UNP_LINK_WUNLOCK();
 554
 555         return (0);
 556 }
 557
 558 static int
 559 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
 560 {
 561         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
 562         struct vattr vattr;
 563         int error, namelen;
 564         struct nameidata nd;
 565         struct unpcb *unp;
 566         struct vnode *vp;
 567         struct mount *mp;
 568         cap_rights_t rights;
 569         char *buf;
 570
 571         if (nam->sa_family != AF_UNIX)
 572                 return (EAFNOSUPPORT);
 573
 574         unp = sotounpcb(so);
 575         KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
 576
 577         if (soun->sun_len > sizeof(struct sockaddr_un))
 578                 return (EINVAL);
 579         namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
 580         if (namelen <= 0)
 581                 return (EINVAL);
 582
 583         /*
 584          * We don't allow simultaneous bind() calls on a single UNIX domain
 585          * socket, so flag in-progress operations, and return an error if an
 586          * operation is already in progress.
 587          *
 588          * Historically, we have not allowed a socket to be rebound, so this
 589          * also returns an error.  Not allowing re-binding simplifies the
 590          * implementation and avoids a great many possible failure modes.
 591          */
 592         UNP_PCB_LOCK(unp);
 593         if (unp->unp_vnode != NULL) {
 594                 UNP_PCB_UNLOCK(unp);
 595                 return (EINVAL);
 596         }
 597         if (unp->unp_flags & UNP_BINDING) {
 598                 UNP_PCB_UNLOCK(unp);
 599                 return (EALREADY);
 600         }
 601         unp->unp_flags |= UNP_BINDING;
 602         UNP_PCB_UNLOCK(unp);
 603
 604         buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
 605         bcopy(soun->sun_path, buf, namelen);
 606         buf[namelen] = 0;
 607
 608 restart:
 609         NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE,
 610             UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td);
 611 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
 612         error = namei(&nd);
 613         if (error)
 614                 goto error;
 615         vp = nd.ni_vp;
 616         if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
 617                 NDFREE(&nd, NDF_ONLY_PNBUF);
 618                 if (nd.ni_dvp == vp)
 619                         vrele(nd.ni_dvp);
 620                 else
 621                         vput(nd.ni_dvp);
 622                 if (vp != NULL) {
 623                         vrele(vp);
 624                         error = EADDRINUSE;
 625                         goto error;
 626                 }
 627                 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
 628                 if (error)
 629                         goto error;
 630                 goto restart;
 631         }
 632         VATTR_NULL(&vattr);
 633         vattr.va_type = VSOCK;
 634         vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
 635 #ifdef MAC
 636         error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
 637             &vattr);
 638 #endif
 639         if (error == 0)
 640                 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
 641         NDFREE(&nd, NDF_ONLY_PNBUF);
 642         vput(nd.ni_dvp);
 643         if (error) {
 644                 vn_finished_write(mp);
 645                 goto error;
 646         }
 647         vp = nd.ni_vp;
 648         ASSERT_VOP_ELOCKED(vp, "uipc_bind");
 649         soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
 650
 651         UNP_PCB_LOCK(unp);
 652         VOP_UNP_BIND(vp, unp);
 653         unp->unp_vnode = vp;
 654         unp->unp_addr = soun;
 655         unp->unp_flags &= ~UNP_BINDING;
 656         UNP_PCB_UNLOCK(unp);
 657         VOP_UNLOCK(vp, 0);
 658         vn_finished_write(mp);
 659         free(buf, M_TEMP);
 660         return (0);
 661
 662 error:
 663         UNP_PCB_LOCK(unp);
 664         unp->unp_flags &= ~UNP_BINDING;
 665         UNP_PCB_UNLOCK(unp);
 666         free(buf, M_TEMP);
 667         return (error);
 668 }
 669
 670 static int
 671 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 672 {
 673
 674         return (uipc_bindat(AT_FDCWD, so, nam, td));
 675 }
 676
 677 static int
 678 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 679 {
 680         int error;
 681
 682         KASSERT(td == curthread, ("uipc_connect: td != curthread"));
 683         error = unp_connect(so, nam, td);
 684         return (error);
 685 }
 686
 687 static int
 688 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
 689     struct thread *td)
 690 {
 691         int error;
 692
 693         KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
 694         error = unp_connectat(fd, so, nam, td);
 695         return (error);
 696 }
 697
 698 static void
 699 uipc_close(struct socket *so)
 700 {
 701         struct unpcb *unp, *unp2;
 702         struct vnode *vp = NULL;
 703         struct mtx *vplock;
 704         int freed;
 705         unp = sotounpcb(so);
 706         KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
 707
 708
 709         vplock = NULL;
 710         if ((vp = unp->unp_vnode) != NULL) {
 711                 vplock = mtx_pool_find(mtxpool_sleep, vp);
 712                 mtx_lock(vplock);
 713         }
 714         UNP_PCB_LOCK(unp);
 715         if (vp && unp->unp_vnode == NULL) {
 716                 mtx_unlock(vplock);
 717                 vp = NULL;
 718         }
 719         if (vp != NULL) {
 720                 VOP_UNP_DETACH(vp);
 721                 unp->unp_vnode = NULL;
 722         }
 723         unp2 = unp->unp_conn;
 724         unp_pcb_hold(unp);
 725         if (__predict_false(unp == unp2)) {
 726                 unp_disconnect(unp, unp2);
 727         } else if (unp2 != NULL) {
 728                 unp_pcb_hold(unp2);
 729                 unp_pcb_owned_lock2(unp, unp2, freed);
 730                 unp_disconnect(unp, unp2);
 731                 if (unp_pcb_rele(unp2) == 0)
 732                         UNP_PCB_UNLOCK(unp2);
 733         }
 734         if (unp_pcb_rele(unp) == 0)
 735                 UNP_PCB_UNLOCK(unp);
 736         if (vp) {
 737                 mtx_unlock(vplock);
 738                 vrele(vp);
 739         }
 740 }
 741
 742 static int
 743 uipc_connect2(struct socket *so1, struct socket *so2)
 744 {
 745         struct unpcb *unp, *unp2;
 746         int error;
 747
 748         unp = so1->so_pcb;
 749         KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
 750         unp2 = so2->so_pcb;
 751         KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
 752         if (unp != unp2)
 753                 unp_pcb_lock2(unp, unp2);
 754         else
 755                 UNP_PCB_LOCK(unp);
 756         error = unp_connect2(so1, so2, PRU_CONNECT2);
 757         if (unp != unp2)
 758                 UNP_PCB_UNLOCK(unp2);
 759         UNP_PCB_UNLOCK(unp);
 760         return (error);
 761 }
 762
 763 static void
 764 uipc_detach(struct socket *so)
 765 {
 766         struct unpcb *unp, *unp2;
 767         struct mtx *vplock;
 768         struct sockaddr_un *saved_unp_addr;
 769         struct vnode *vp;
 770         int freeunp, local_unp_rights;
 771
 772         unp = sotounpcb(so);
 773         KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
 774
 775         vp = NULL;
 776         vplock = NULL;
 777         local_unp_rights = 0;
 778
 779         UNP_LINK_WLOCK();
 780         LIST_REMOVE(unp, unp_link);
 781         unp->unp_gencnt = ++unp_gencnt;
 782         --unp_count;
 783         UNP_LINK_WUNLOCK();
 784
 785         UNP_PCB_UNLOCK_ASSERT(unp);
 786  restart:
 787         if ((vp = unp->unp_vnode) != NULL) {
 788                 vplock = mtx_pool_find(mtxpool_sleep, vp);
 789                 mtx_lock(vplock);
 790         }
 791         UNP_PCB_LOCK(unp);
 792         if (unp->unp_vnode != vp &&
 793                 unp->unp_vnode != NULL) {
 794                 if (vplock)
 795                         mtx_unlock(vplock);
 796                 UNP_PCB_UNLOCK(unp);
 797                 goto restart;
 798         }
 799         if ((unp->unp_flags & UNP_NASCENT) != 0) {
 800                 goto teardown;
 801         }
 802         if ((vp = unp->unp_vnode) != NULL) {
 803                 VOP_UNP_DETACH(vp);
 804                 unp->unp_vnode = NULL;
 805         }
 806         if (__predict_false(unp == unp->unp_conn)) {
 807                 unp_disconnect(unp, unp);
 808                 unp2 = NULL;
 809                 goto connect_self;
 810         }
 811         if ((unp2 = unp->unp_conn) != NULL) {
 812                 unp_pcb_owned_lock2(unp, unp2, freeunp);
 813                 if (freeunp)
 814                         unp2 = NULL;
 815         }
 816         unp_pcb_hold(unp);
 817         if (unp2 != NULL) {
 818                 unp_pcb_hold(unp2);
 819                 unp_disconnect(unp, unp2);
 820                 if (unp_pcb_rele(unp2) == 0)
 821                         UNP_PCB_UNLOCK(unp2);
 822         }
 823  connect_self:
 824         UNP_PCB_UNLOCK(unp);
 825         UNP_REF_LIST_LOCK();
 826         while (!LIST_EMPTY(&unp->unp_refs)) {
 827                 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
 828
 829                 unp_pcb_hold(ref);
 830                 UNP_REF_LIST_UNLOCK();
 831
 832                 MPASS(ref != unp);
 833                 UNP_PCB_UNLOCK_ASSERT(ref);
 834                 unp_drop(ref);
 835                 UNP_REF_LIST_LOCK();
 836         }
 837
 838         UNP_REF_LIST_UNLOCK();
 839         UNP_PCB_LOCK(unp);
 840         freeunp = unp_pcb_rele(unp);
 841         MPASS(freeunp == 0);
 842         local_unp_rights = unp_rights;
 843 teardown:
 844         unp->unp_socket->so_pcb = NULL;
 845         saved_unp_addr = unp->unp_addr;
 846         unp->unp_addr = NULL;
 847         unp->unp_socket = NULL;
 848         freeunp = unp_pcb_rele(unp);
 849         if (saved_unp_addr != NULL)
 850                 free(saved_unp_addr, M_SONAME);
 851         if (!freeunp)
 852                 UNP_PCB_UNLOCK(unp);
 853         if (vp) {
 854                 mtx_unlock(vplock);
 855                 vrele(vp);
 856         }
 857         if (local_unp_rights)
 858                 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
 859 }
 860
 861 static int
 862 uipc_disconnect(struct socket *so)
 863 {
 864         struct unpcb *unp, *unp2;
 865         int freed;
 866
 867         unp = sotounpcb(so);
 868         KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
 869
 870         UNP_PCB_LOCK(unp);
 871         if ((unp2 = unp->unp_conn) == NULL) {
 872                 UNP_PCB_UNLOCK(unp);
 873                 return (0);
 874         }
 875         if (__predict_true(unp != unp2)) {
 876                 unp_pcb_owned_lock2(unp, unp2, freed);
 877                 if (__predict_false(freed)) {
 878                         UNP_PCB_UNLOCK(unp);
 879                         return (0);
 880                 }
 881                 unp_pcb_hold(unp2);
 882         }
 883         unp_pcb_hold(unp);
 884         unp_disconnect(unp, unp2);
 885         if (unp_pcb_rele(unp) == 0)
 886                 UNP_PCB_UNLOCK(unp);
 887         if ((unp != unp2) && unp_pcb_rele(unp2) == 0)
 888                 UNP_PCB_UNLOCK(unp2);
 889         return (0);
 890 }
 891
 892 static int
 893 uipc_listen(struct socket *so, int backlog, struct thread *td)
 894 {
 895         struct unpcb *unp;
 896         int error;
 897
 898         if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
 899                 return (EOPNOTSUPP);
 900
 901         unp = sotounpcb(so);
 902         KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
 903
 904         UNP_PCB_LOCK(unp);
 905         if (unp->unp_vnode == NULL) {
 906                 /* Already connected or not bound to an address. */
 907                 error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ;
 908                 UNP_PCB_UNLOCK(unp);
 909                 return (error);
 910         }
 911
 912         SOCK_LOCK(so);
 913         error = solisten_proto_check(so);
 914         if (error == 0) {
 915                 cru2x(td->td_ucred, &unp->unp_peercred);
 916                 solisten_proto(so, backlog);
 917         }
 918         SOCK_UNLOCK(so);
 919         UNP_PCB_UNLOCK(unp);
 920         return (error);
 921 }
 922
 923 static int
 924 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
 925 {
 926         struct unpcb *unp, *unp2;
 927         const struct sockaddr *sa;
 928
 929         unp = sotounpcb(so);
 930         KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
 931
 932         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 933         UNP_LINK_RLOCK();
 934         /*
 935          * XXX: It seems that this test always fails even when connection is
 936          * established.  So, this else clause is added as workaround to
 937          * return PF_LOCAL sockaddr.
 938          */
 939         unp2 = unp->unp_conn;
 940         if (unp2 != NULL) {
 941                 UNP_PCB_LOCK(unp2);
 942                 if (unp2->unp_addr != NULL)
 943                         sa = (struct sockaddr *) unp2->unp_addr;
 944                 else
 945                         sa = &sun_noname;
 946                 bcopy(sa, *nam, sa->sa_len);
 947                 UNP_PCB_UNLOCK(unp2);
 948         } else {
 949                 sa = &sun_noname;
 950                 bcopy(sa, *nam, sa->sa_len);
 951         }
 952         UNP_LINK_RUNLOCK();
 953         return (0);
 954 }
 955
 956 static int
 957 uipc_rcvd(struct socket *so, int flags)
 958 {
 959         struct unpcb *unp, *unp2;
 960         struct socket *so2;
 961         u_int mbcnt, sbcc;
 962
 963         unp = sotounpcb(so);
 964         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
 965         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
 966             ("%s: socktype %d", __func__, so->so_type));
 967
 968         /*
 969          * Adjust backpressure on sender and wakeup any waiting to write.
 970          *
 971          * The unp lock is acquired to maintain the validity of the unp_conn
 972          * pointer; no lock on unp2 is required as unp2->unp_socket will be
 973          * static as long as we don't permit unp2 to disconnect from unp,
 974          * which is prevented by the lock on unp.  We cache values from
 975          * so_rcv to avoid holding the so_rcv lock over the entire
 976          * transaction on the remote so_snd.
 977          */
 978         SOCKBUF_LOCK(&so->so_rcv);
 979         mbcnt = so->so_rcv.sb_mbcnt;
 980         sbcc = sbavail(&so->so_rcv);
 981         SOCKBUF_UNLOCK(&so->so_rcv);
 982         /*
 983          * There is a benign race condition at this point.  If we're planning to
 984          * clear SB_STOP, but uipc_send is called on the connected socket at
 985          * this instant, it might add data to the sockbuf and set SB_STOP.  Then
 986          * we would erroneously clear SB_STOP below, even though the sockbuf is
 987          * full.  The race is benign because the only ill effect is to allow the
 988          * sockbuf to exceed its size limit, and the size limits are not
 989          * strictly guaranteed anyway.
 990          */
 991         UNP_PCB_LOCK(unp);
 992         unp2 = unp->unp_conn;
 993         if (unp2 == NULL) {
 994                 UNP_PCB_UNLOCK(unp);
 995                 return (0);
 996         }
 997         so2 = unp2->unp_socket;
 998         SOCKBUF_LOCK(&so2->so_snd);
 999         if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
1000                 so2->so_snd.sb_flags &= ~SB_STOP;
1001         sowwakeup_locked(so2);
1002         UNP_PCB_UNLOCK(unp);
1003         return (0);
1004 }
1005
1006 static int
1007 connect_internal(struct socket *so, struct sockaddr *nam, struct thread *td)
1008 {
1009         int error;
1010         struct unpcb *unp;
1011
1012         unp = so->so_pcb;
1013         if (unp->unp_conn != NULL)
1014                 return (EISCONN);
1015         error = unp_connect(so, nam, td);
1016         if (error)
1017                 return (error);
1018         UNP_PCB_LOCK(unp);
1019         if (unp->unp_conn == NULL) {
1020                 UNP_PCB_UNLOCK(unp);
1021                 if (error == 0)
1022                         error = ENOTCONN;
1023         }
1024         return (error);
1025 }
1026
1027
1028 static int
1029 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
1030     struct mbuf *control, struct thread *td)
1031 {
1032         struct unpcb *unp, *unp2;
1033         struct socket *so2;
1034         u_int mbcnt, sbcc;
1035         int freed, error;
1036
1037         unp = sotounpcb(so);
1038         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
1039         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
1040             so->so_type == SOCK_SEQPACKET,
1041             ("%s: socktype %d", __func__, so->so_type));
1042
1043         freed = error = 0;
1044         if (flags & PRUS_OOB) {
1045                 error = EOPNOTSUPP;
1046                 goto release;
1047         }
1048         if (control != NULL && (error = unp_internalize(&control, td)))
1049                 goto release;
1050
1051         unp2 = NULL;
1052         switch (so->so_type) {
1053         case SOCK_DGRAM:
1054         {
1055                 const struct sockaddr *from;
1056
1057                 if (nam != NULL) {
1058                         /*
1059                          * We return with UNP_PCB_LOCK_HELD so we know that
1060                          * the reference is live if the pointer is valid.
1061                          */
1062                         if ((error = connect_internal(so, nam, td)))
1063                                 break;
1064                         MPASS(unp->unp_conn != NULL);
1065                         unp2 = unp->unp_conn;
1066                 } else  {
1067                         UNP_PCB_LOCK(unp);
1068
1069                         /*
1070                          * Because connect() and send() are non-atomic in a sendto()
1071                          * with a target address, it's possible that the socket will
1072                          * have disconnected before the send() can run.  In that case
1073                          * return the slightly counter-intuitive but otherwise
1074                          * correct error that the socket is not connected.
1075                          */
1076                         if ((unp2 = unp->unp_conn)  == NULL) {
1077                                 UNP_PCB_UNLOCK(unp);
1078                                 error = ENOTCONN;
1079                                 break;
1080                         }
1081                 }
1082                 if (__predict_false(unp == unp2)) {
1083                         if (unp->unp_socket == NULL) {
1084                                 error = ENOTCONN;
1085                                 break;
1086                         }
1087                         goto connect_self;
1088                 }
1089                 unp_pcb_owned_lock2(unp, unp2, freed);
1090                 if (__predict_false(freed)) {
1091                         UNP_PCB_UNLOCK(unp);
1092                         error = ENOTCONN;
1093                         break;
1094                 }
1095                 /*
1096                  * The socket referencing unp2 may have been closed
1097                  * or unp may have been disconnected if the unp lock
1098                  * was dropped to acquire unp2.
1099                  */
1100                 if (__predict_false(unp->unp_conn == NULL) ||
1101                         unp2->unp_socket == NULL) {
1102                         UNP_PCB_UNLOCK(unp);
1103                         if (unp_pcb_rele(unp2) == 0)
1104                                 UNP_PCB_UNLOCK(unp2);
1105                         error = ENOTCONN;
1106                         break;
1107                 }
1108         connect_self:
1109                 if (unp2->unp_flags & UNP_WANTCRED)
1110                         control = unp_addsockcred(td, control);
1111                 if (unp->unp_addr != NULL)
1112                         from = (struct sockaddr *)unp->unp_addr;
1113                 else
1114                         from = &sun_noname;
1115                 so2 = unp2->unp_socket;
1116                 SOCKBUF_LOCK(&so2->so_rcv);
1117                 if (sbappendaddr_locked(&so2->so_rcv, from, m,
1118                     control)) {
1119                         sorwakeup_locked(so2);
1120                         m = NULL;
1121                         control = NULL;
1122                 } else {
1123                         SOCKBUF_UNLOCK(&so2->so_rcv);
1124                         error = ENOBUFS;
1125                 }
1126                 if (nam != NULL)
1127                         unp_disconnect(unp, unp2);
1128                 if (__predict_true(unp != unp2))
1129                         UNP_PCB_UNLOCK(unp2);
1130                 UNP_PCB_UNLOCK(unp);
1131                 break;
1132         }
1133
1134         case SOCK_SEQPACKET:
1135         case SOCK_STREAM:
1136                 if ((so->so_state & SS_ISCONNECTED) == 0) {
1137                         if (nam != NULL) {
1138                                 if ((error = connect_internal(so, nam, td)))
1139                                         break;
1140                         } else  {
1141                                 error = ENOTCONN;
1142                                 break;
1143                         }
1144                 } else if ((unp2 = unp->unp_conn) == NULL) {
1145                         error = ENOTCONN;
1146                         break;
1147                 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1148                         error = EPIPE;
1149                         break;
1150                 } else {
1151                         UNP_PCB_LOCK(unp);
1152                         if ((unp2 = unp->unp_conn) == NULL) {
1153                                 UNP_PCB_UNLOCK(unp);
1154                                 error = ENOTCONN;
1155                                 break;
1156                         }
1157                 }
1158                 unp_pcb_owned_lock2(unp, unp2, freed);
1159                 UNP_PCB_UNLOCK(unp);
1160                 if (__predict_false(freed)) {
1161                         error = ENOTCONN;
1162                         break;
1163                 }
1164                 if ((so2 = unp2->unp_socket) == NULL) {
1165                         UNP_PCB_UNLOCK(unp2);
1166                         error = ENOTCONN;
1167                         break;
1168                 }
1169                 SOCKBUF_LOCK(&so2->so_rcv);
1170                 if (unp2->unp_flags & UNP_WANTCRED) {
1171                         /*
1172                          * Credentials are passed only once on SOCK_STREAM
1173                          * and SOCK_SEQPACKET.
1174                          */
1175                         unp2->unp_flags &= ~UNP_WANTCRED;
1176                         control = unp_addsockcred(td, control);
1177                 }
1178
1179                 /*
1180                  * Send to paired receive port and wake up readers.  Don't
1181                  * check for space available in the receive buffer if we're
1182                  * attaching ancillary data; Unix domain sockets only check
1183                  * for space in the sending sockbuf, and that check is
1184                  * performed one level up the stack.  At that level we cannot
1185                  * precisely account for the amount of buffer space used
1186                  * (e.g., because control messages are not yet internalized).
1187                  */
1188                 switch (so->so_type) {
1189                 case SOCK_STREAM:
1190                         if (control != NULL) {
1191                                 sbappendcontrol_locked(&so2->so_rcv, m,
1192                                     control);
1193                                 control = NULL;
1194                         } else
1195                                 sbappend_locked(&so2->so_rcv, m, flags);
1196                         break;
1197
1198                 case SOCK_SEQPACKET: {
1199                         const struct sockaddr *from;
1200
1201                         from = &sun_noname;
1202                         if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1203                             from, m, control))
1204                                 control = NULL;
1205                         break;
1206                         }
1207                 }
1208
1209                 mbcnt = so2->so_rcv.sb_mbcnt;
1210                 sbcc = sbavail(&so2->so_rcv);
1211                 if (sbcc)
1212                         sorwakeup_locked(so2);
1213                 else
1214                         SOCKBUF_UNLOCK(&so2->so_rcv);
1215
1216                 /*
1217                  * The PCB lock on unp2 protects the SB_STOP flag.  Without it,
1218                  * it would be possible for uipc_rcvd to be called at this
1219                  * point, drain the receiving sockbuf, clear SB_STOP, and then
1220                  * we would set SB_STOP below.  That could lead to an empty
1221                  * sockbuf having SB_STOP set
1222                  */
1223                 SOCKBUF_LOCK(&so->so_snd);
1224                 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1225                         so->so_snd.sb_flags |= SB_STOP;
1226                 SOCKBUF_UNLOCK(&so->so_snd);
1227                 UNP_PCB_UNLOCK(unp2);
1228                 m = NULL;
1229                 break;
1230         }
1231
1232         /*
1233          * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1234          */
1235         if (flags & PRUS_EOF) {
1236                 UNP_PCB_LOCK(unp);
1237                 socantsendmore(so);
1238                 unp_shutdown(unp);
1239                 UNP_PCB_UNLOCK(unp);
1240         }
1241         if (control != NULL && error != 0)
1242                 unp_dispose_mbuf(control);
1243
1244 release:
1245         if (control != NULL)
1246                 m_freem(control);
1247         /*
1248          * In case of PRUS_NOTREADY, uipc_ready() is responsible
1249          * for freeing memory.
1250          */
1251         if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1252                 m_freem(m);
1253         return (error);
1254 }
1255
1256 static int
1257 uipc_ready(struct socket *so, struct mbuf *m, int count)
1258 {
1259         struct unpcb *unp, *unp2;
1260         struct socket *so2;
1261         int error;
1262
1263         unp = sotounpcb(so);
1264
1265         UNP_PCB_LOCK(unp);
1266         if ((unp2 = unp->unp_conn) == NULL) {
1267                 UNP_PCB_UNLOCK(unp);
1268                 goto error;
1269         }
1270         if (unp != unp2) {
1271                 if (UNP_PCB_TRYLOCK(unp2) == 0) {
1272                         unp_pcb_hold(unp2);
1273                         UNP_PCB_UNLOCK(unp);
1274                         UNP_PCB_LOCK(unp2);
1275                         if (unp_pcb_rele(unp2))
1276                                 goto error;
1277                 } else
1278                         UNP_PCB_UNLOCK(unp);
1279         }
1280         so2 = unp2->unp_socket;
1281
1282         SOCKBUF_LOCK(&so2->so_rcv);
1283         if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1284                 sorwakeup_locked(so2);
1285         else
1286                 SOCKBUF_UNLOCK(&so2->so_rcv);
1287
1288         UNP_PCB_UNLOCK(unp2);
1289
1290         return (error);
1291  error:
1292         for (int i = 0; i < count; i++)
1293                 m = m_free(m);
1294         return (ECONNRESET);
1295 }
1296
1297 static int
1298 uipc_sense(struct socket *so, struct stat *sb)
1299 {
1300         struct unpcb *unp;
1301
1302         unp = sotounpcb(so);
1303         KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1304
1305         sb->st_blksize = so->so_snd.sb_hiwat;
1306         sb->st_dev = NODEV;
1307         sb->st_ino = unp->unp_ino;
1308         return (0);
1309 }
1310
1311 static int
1312 uipc_shutdown(struct socket *so)
1313 {
1314         struct unpcb *unp;
1315
1316         unp = sotounpcb(so);
1317         KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1318
1319         UNP_PCB_LOCK(unp);
1320         socantsendmore(so);
1321         unp_shutdown(unp);
1322         UNP_PCB_UNLOCK(unp);
1323         return (0);
1324 }
1325
1326 static int
1327 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1328 {
1329         struct unpcb *unp;
1330         const struct sockaddr *sa;
1331
1332         unp = sotounpcb(so);
1333         KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1334
1335         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1336         UNP_PCB_LOCK(unp);
1337         if (unp->unp_addr != NULL)
1338                 sa = (struct sockaddr *) unp->unp_addr;
1339         else
1340                 sa = &sun_noname;
1341         bcopy(sa, *nam, sa->sa_len);
1342         UNP_PCB_UNLOCK(unp);
1343         return (0);
1344 }
1345
1346 static struct pr_usrreqs uipc_usrreqs_dgram = {
1347         .pru_abort =            uipc_abort,
1348         .pru_accept =           uipc_accept,
1349         .pru_attach =           uipc_attach,
1350         .pru_bind =             uipc_bind,
1351         .pru_bindat =           uipc_bindat,
1352         .pru_connect =          uipc_connect,
1353         .pru_connectat =        uipc_connectat,
1354         .pru_connect2 =         uipc_connect2,
1355         .pru_detach =           uipc_detach,
1356         .pru_disconnect =       uipc_disconnect,
1357         .pru_listen =           uipc_listen,
1358         .pru_peeraddr =         uipc_peeraddr,
1359         .pru_rcvd =             uipc_rcvd,
1360         .pru_send =             uipc_send,
1361         .pru_sense =            uipc_sense,
1362         .pru_shutdown =         uipc_shutdown,
1363         .pru_sockaddr =         uipc_sockaddr,
1364         .pru_soreceive =        soreceive_dgram,
1365         .pru_close =            uipc_close,
1366 };
1367
1368 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1369         .pru_abort =            uipc_abort,
1370         .pru_accept =           uipc_accept,
1371         .pru_attach =           uipc_attach,
1372         .pru_bind =             uipc_bind,
1373         .pru_bindat =           uipc_bindat,
1374         .pru_connect =          uipc_connect,
1375         .pru_connectat =        uipc_connectat,
1376         .pru_connect2 =         uipc_connect2,
1377         .pru_detach =           uipc_detach,
1378         .pru_disconnect =       uipc_disconnect,
1379         .pru_listen =           uipc_listen,
1380         .pru_peeraddr =         uipc_peeraddr,
1381         .pru_rcvd =             uipc_rcvd,
1382         .pru_send =             uipc_send,
1383         .pru_sense =            uipc_sense,
1384         .pru_shutdown =         uipc_shutdown,
1385         .pru_sockaddr =         uipc_sockaddr,
1386         .pru_soreceive =        soreceive_generic,      /* XXX: or...? */
1387         .pru_close =            uipc_close,
1388 };
1389
1390 static struct pr_usrreqs uipc_usrreqs_stream = {
1391         .pru_abort =            uipc_abort,
1392         .pru_accept =           uipc_accept,
1393         .pru_attach =           uipc_attach,
1394         .pru_bind =             uipc_bind,
1395         .pru_bindat =           uipc_bindat,
1396         .pru_connect =          uipc_connect,
1397         .pru_connectat =        uipc_connectat,
1398         .pru_connect2 =         uipc_connect2,
1399         .pru_detach =           uipc_detach,
1400         .pru_disconnect =       uipc_disconnect,
1401         .pru_listen =           uipc_listen,
1402         .pru_peeraddr =         uipc_peeraddr,
1403         .pru_rcvd =             uipc_rcvd,
1404         .pru_send =             uipc_send,
1405         .pru_ready =            uipc_ready,
1406         .pru_sense =            uipc_sense,
1407         .pru_shutdown =         uipc_shutdown,
1408         .pru_sockaddr =         uipc_sockaddr,
1409         .pru_soreceive =        soreceive_generic,
1410         .pru_close =            uipc_close,
1411 };
1412
1413 static int
1414 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1415 {
1416         struct unpcb *unp;
1417         struct xucred xu;
1418         int error, optval;
1419
1420         if (sopt->sopt_level != 0)
1421                 return (EINVAL);
1422
1423         unp = sotounpcb(so);
1424         KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1425         error = 0;
1426         switch (sopt->sopt_dir) {
1427         case SOPT_GET:
1428                 switch (sopt->sopt_name) {
1429                 case LOCAL_PEERCRED:
1430                         UNP_PCB_LOCK(unp);
1431                         if (unp->unp_flags & UNP_HAVEPC)
1432                                 xu = unp->unp_peercred;
1433                         else {
1434                                 if (so->so_type == SOCK_STREAM)
1435                                         error = ENOTCONN;
1436                                 else
1437                                         error = EINVAL;
1438                         }
1439                         UNP_PCB_UNLOCK(unp);
1440                         if (error == 0)
1441                                 error = sooptcopyout(sopt, &xu, sizeof(xu));
1442                         break;
1443
1444                 case LOCAL_CREDS:
1445                         /* Unlocked read. */
1446                         optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1447                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1448                         break;
1449
1450                 case LOCAL_CONNWAIT:
1451                         /* Unlocked read. */
1452                         optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1453                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1454                         break;
1455
1456                 default:
1457                         error = EOPNOTSUPP;
1458                         break;
1459                 }
1460                 break;
1461
1462         case SOPT_SET:
1463                 switch (sopt->sopt_name) {
1464                 case LOCAL_CREDS:
1465                 case LOCAL_CONNWAIT:
1466                         error = sooptcopyin(sopt, &optval, sizeof(optval),
1467                                             sizeof(optval));
1468                         if (error)
1469                                 break;
1470
1471 #define OPTSET(bit) do {                                                \
1472         UNP_PCB_LOCK(unp);                                              \
1473         if (optval)                                                     \
1474                 unp->unp_flags |= bit;                                  \
1475         else                                                            \
1476                 unp->unp_flags &= ~bit;                                 \
1477         UNP_PCB_UNLOCK(unp);                                            \
1478 } while (0)
1479
1480                         switch (sopt->sopt_name) {
1481                         case LOCAL_CREDS:
1482                                 OPTSET(UNP_WANTCRED);
1483                                 break;
1484
1485                         case LOCAL_CONNWAIT:
1486                                 OPTSET(UNP_CONNWAIT);
1487                                 break;
1488
1489                         default:
1490                                 break;
1491                         }
1492                         break;
1493 #undef  OPTSET
1494                 default:
1495                         error = ENOPROTOOPT;
1496                         break;
1497                 }
1498                 break;
1499
1500         default:
1501                 error = EOPNOTSUPP;
1502                 break;
1503         }
1504         return (error);
1505 }
1506
1507 static int
1508 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1509 {
1510
1511         return (unp_connectat(AT_FDCWD, so, nam, td));
1512 }
1513
1514 static int
1515 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1516     struct thread *td)
1517 {
1518         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1519         struct vnode *vp;
1520         struct socket *so2;
1521         struct unpcb *unp, *unp2, *unp3;
1522         struct nameidata nd;
1523         char buf[SOCK_MAXADDRLEN];
1524         struct sockaddr *sa;
1525         cap_rights_t rights;
1526         int error, len, freed;
1527         struct mtx *vplock;
1528
1529         if (nam->sa_family != AF_UNIX)
1530                 return (EAFNOSUPPORT);
1531         if (nam->sa_len > sizeof(struct sockaddr_un))
1532                 return (EINVAL);
1533         len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1534         if (len <= 0)
1535                 return (EINVAL);
1536         bcopy(soun->sun_path, buf, len);
1537         buf[len] = 0;
1538
1539         unp = sotounpcb(so);
1540         UNP_PCB_LOCK(unp);
1541         if (unp->unp_flags & UNP_CONNECTING) {
1542                 UNP_PCB_UNLOCK(unp);
1543                 return (EALREADY);
1544         }
1545         unp->unp_flags |= UNP_CONNECTING;
1546         UNP_PCB_UNLOCK(unp);
1547
1548         sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1549         NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1550             UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1551         error = namei(&nd);
1552         if (error)
1553                 vp = NULL;
1554         else
1555                 vp = nd.ni_vp;
1556         ASSERT_VOP_LOCKED(vp, "unp_connect");
1557         NDFREE(&nd, NDF_ONLY_PNBUF);
1558         if (error)
1559                 goto bad;
1560
1561         if (vp->v_type != VSOCK) {
1562                 error = ENOTSOCK;
1563                 goto bad;
1564         }
1565 #ifdef MAC
1566         error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1567         if (error)
1568                 goto bad;
1569 #endif
1570         error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1571         if (error)
1572                 goto bad;
1573
1574         unp = sotounpcb(so);
1575         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1576
1577         vplock = mtx_pool_find(mtxpool_sleep, vp);
1578         mtx_lock(vplock);
1579         VOP_UNP_CONNECT(vp, &unp2);
1580         if (unp2 == NULL) {
1581                 error = ECONNREFUSED;
1582                 goto bad2;
1583         }
1584         so2 = unp2->unp_socket;
1585         if (so->so_type != so2->so_type) {
1586                 error = EPROTOTYPE;
1587                 goto bad2;
1588         }
1589         if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1590                 if (so2->so_options & SO_ACCEPTCONN) {
1591                         CURVNET_SET(so2->so_vnet);
1592                         so2 = sonewconn(so2, 0);
1593                         CURVNET_RESTORE();
1594                 } else
1595                         so2 = NULL;
1596                 if (so2 == NULL) {
1597                         error = ECONNREFUSED;
1598                         goto bad2;
1599                 }
1600                 unp3 = sotounpcb(so2);
1601                 unp_pcb_lock2(unp2, unp3);
1602                 if (unp2->unp_addr != NULL) {
1603                         bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1604                         unp3->unp_addr = (struct sockaddr_un *) sa;
1605                         sa = NULL;
1606                 }
1607
1608                 unp_copy_peercred(td, unp3, unp, unp2);
1609
1610                 UNP_PCB_UNLOCK(unp2);
1611                 unp2 = unp3;
1612                 unp_pcb_owned_lock2(unp2, unp, freed);
1613                 if (__predict_false(freed)) {
1614                         UNP_PCB_UNLOCK(unp2);
1615                         error = ECONNREFUSED;
1616                         goto bad2;
1617                 }
1618 #ifdef MAC
1619                 mac_socketpeer_set_from_socket(so, so2);
1620                 mac_socketpeer_set_from_socket(so2, so);
1621 #endif
1622         } else {
1623                 if (unp == unp2)
1624                         UNP_PCB_LOCK(unp);
1625                 else
1626                         unp_pcb_lock2(unp, unp2);
1627         }
1628         KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 &&
1629             sotounpcb(so2) == unp2,
1630             ("%s: unp2 %p so2 %p", __func__, unp2, so2));
1631         error = unp_connect2(so, so2, PRU_CONNECT);
1632         if (unp != unp2)
1633                 UNP_PCB_UNLOCK(unp2);
1634         UNP_PCB_UNLOCK(unp);
1635 bad2:
1636         mtx_unlock(vplock);
1637 bad:
1638         if (vp != NULL) {
1639                 vput(vp);
1640         }
1641         free(sa, M_SONAME);
1642         UNP_PCB_LOCK(unp);
1643         unp->unp_flags &= ~UNP_CONNECTING;
1644         UNP_PCB_UNLOCK(unp);
1645         return (error);
1646 }
1647
1648 /*
1649  * Set socket peer credentials at connection time.
1650  *
1651  * The client's PCB credentials are copied from its process structure.  The
1652  * server's PCB credentials are copied from the socket on which it called
1653  * listen(2).  uipc_listen cached that process's credentials at the time.
1654  */
1655 void
1656 unp_copy_peercred(struct thread *td, struct unpcb *client_unp,
1657     struct unpcb *server_unp, struct unpcb *listen_unp)
1658 {
1659         cru2x(td->td_ucred, &client_unp->unp_peercred);
1660         client_unp->unp_flags |= UNP_HAVEPC;
1661
1662         memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred,
1663             sizeof(server_unp->unp_peercred));
1664         server_unp->unp_flags |= UNP_HAVEPC;
1665         if (listen_unp->unp_flags & UNP_WANTCRED)
1666                 client_unp->unp_flags |= UNP_WANTCRED;
1667 }
1668
1669 static int
1670 unp_connect2(struct socket *so, struct socket *so2, int req)
1671 {
1672         struct unpcb *unp;
1673         struct unpcb *unp2;
1674
1675         unp = sotounpcb(so);
1676         KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1677         unp2 = sotounpcb(so2);
1678         KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1679
1680         UNP_PCB_LOCK_ASSERT(unp);
1681         UNP_PCB_LOCK_ASSERT(unp2);
1682
1683         if (so2->so_type != so->so_type)
1684                 return (EPROTOTYPE);
1685         unp2->unp_flags &= ~UNP_NASCENT;
1686         unp->unp_conn = unp2;
1687         unp_pcb_hold(unp2);
1688         unp_pcb_hold(unp);
1689         switch (so->so_type) {
1690         case SOCK_DGRAM:
1691                 UNP_REF_LIST_LOCK();
1692                 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1693                 UNP_REF_LIST_UNLOCK();
1694                 soisconnected(so);
1695                 break;
1696
1697         case SOCK_STREAM:
1698         case SOCK_SEQPACKET:
1699                 unp2->unp_conn = unp;
1700                 if (req == PRU_CONNECT &&
1701                     ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1702                         soisconnecting(so);
1703                 else
1704                         soisconnected(so);
1705                 soisconnected(so2);
1706                 break;
1707
1708         default:
1709                 panic("unp_connect2");
1710         }
1711         return (0);
1712 }
1713
1714 static void
1715 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1716 {
1717         struct socket *so, *so2;
1718         int freed __unused;
1719
1720         KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1721
1722         UNP_PCB_LOCK_ASSERT(unp);
1723         UNP_PCB_LOCK_ASSERT(unp2);
1724
1725         if (unp->unp_conn == NULL && unp2->unp_conn == NULL)
1726                 return;
1727
1728         MPASS(unp->unp_conn == unp2);
1729         unp->unp_conn = NULL;
1730         so = unp->unp_socket;
1731         so2 = unp2->unp_socket;
1732         switch (unp->unp_socket->so_type) {
1733         case SOCK_DGRAM:
1734                 UNP_REF_LIST_LOCK();
1735                 LIST_REMOVE(unp, unp_reflink);
1736                 UNP_REF_LIST_UNLOCK();
1737                 if (so) {
1738                         SOCK_LOCK(so);
1739                         so->so_state &= ~SS_ISCONNECTED;
1740                         SOCK_UNLOCK(so);
1741                 }
1742                 break;
1743
1744         case SOCK_STREAM:
1745         case SOCK_SEQPACKET:
1746                 if (so)
1747                         soisdisconnected(so);
1748                 MPASS(unp2->unp_conn == unp);
1749                 unp2->unp_conn = NULL;
1750                 if (so2)
1751                         soisdisconnected(so2);
1752                 break;
1753         }
1754         freed = unp_pcb_rele(unp);
1755         MPASS(freed == 0);
1756         freed = unp_pcb_rele(unp2);
1757         MPASS(freed == 0);
1758 }
1759
1760 /*
1761  * unp_pcblist() walks the global list of struct unpcb's to generate a
1762  * pointer list, bumping the refcount on each unpcb.  It then copies them out
1763  * sequentially, validating the generation number on each to see if it has
1764  * been detached.  All of this is necessary because copyout() may sleep on
1765  * disk I/O.
1766  */
1767 static int
1768 unp_pcblist(SYSCTL_HANDLER_ARGS)
1769 {
1770         struct unpcb *unp, **unp_list;
1771         unp_gen_t gencnt;
1772         struct xunpgen *xug;
1773         struct unp_head *head;
1774         struct xunpcb *xu;
1775         u_int i;
1776         int error, freeunp, n;
1777
1778         switch ((intptr_t)arg1) {
1779         case SOCK_STREAM:
1780                 head = &unp_shead;
1781                 break;
1782
1783         case SOCK_DGRAM:
1784                 head = &unp_dhead;
1785                 break;
1786
1787         case SOCK_SEQPACKET:
1788                 head = &unp_sphead;
1789                 break;
1790
1791         default:
1792                 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1793         }
1794
1795         /*
1796          * The process of preparing the PCB list is too time-consuming and
1797          * resource-intensive to repeat twice on every request.
1798          */
1799         if (req->oldptr == NULL) {
1800                 n = unp_count;
1801                 req->oldidx = 2 * (sizeof *xug)
1802                         + (n + n/8) * sizeof(struct xunpcb);
1803                 return (0);
1804         }
1805
1806         if (req->newptr != NULL)
1807                 return (EPERM);
1808
1809         /*
1810          * OK, now we're committed to doing something.
1811          */
1812         xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO);
1813         UNP_LINK_RLOCK();
1814         gencnt = unp_gencnt;
1815         n = unp_count;
1816         UNP_LINK_RUNLOCK();
1817
1818         xug->xug_len = sizeof *xug;
1819         xug->xug_count = n;
1820         xug->xug_gen = gencnt;
1821         xug->xug_sogen = so_gencnt;
1822         error = SYSCTL_OUT(req, xug, sizeof *xug);
1823         if (error) {
1824                 free(xug, M_TEMP);
1825                 return (error);
1826         }
1827
1828         unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1829
1830         UNP_LINK_RLOCK();
1831         for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1832              unp = LIST_NEXT(unp, unp_link)) {
1833                 UNP_PCB_LOCK(unp);
1834                 if (unp->unp_gencnt <= gencnt) {
1835                         if (cr_cansee(req->td->td_ucred,
1836                             unp->unp_socket->so_cred)) {
1837                                 UNP_PCB_UNLOCK(unp);
1838                                 continue;
1839                         }
1840                         unp_list[i++] = unp;
1841                         unp_pcb_hold(unp);
1842                 }
1843                 UNP_PCB_UNLOCK(unp);
1844         }
1845         UNP_LINK_RUNLOCK();
1846         n = i;                  /* In case we lost some during malloc. */
1847
1848         error = 0;
1849         xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1850         for (i = 0; i < n; i++) {
1851                 unp = unp_list[i];
1852                 UNP_PCB_LOCK(unp);
1853                 freeunp = unp_pcb_rele(unp);
1854
1855                 if (freeunp == 0 && unp->unp_gencnt <= gencnt) {
1856                         xu->xu_len = sizeof *xu;
1857                         xu->xu_unpp = (uintptr_t)unp;
1858                         /*
1859                          * XXX - need more locking here to protect against
1860                          * connect/disconnect races for SMP.
1861                          */
1862                         if (unp->unp_addr != NULL)
1863                                 bcopy(unp->unp_addr, &xu->xu_addr,
1864                                       unp->unp_addr->sun_len);
1865                         else
1866                                 bzero(&xu->xu_addr, sizeof(xu->xu_addr));
1867                         if (unp->unp_conn != NULL &&
1868                             unp->unp_conn->unp_addr != NULL)
1869                                 bcopy(unp->unp_conn->unp_addr,
1870                                       &xu->xu_caddr,
1871                                       unp->unp_conn->unp_addr->sun_len);
1872                         else
1873                                 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr));
1874                         xu->unp_vnode = (uintptr_t)unp->unp_vnode;
1875                         xu->unp_conn = (uintptr_t)unp->unp_conn;
1876                         xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs);
1877                         xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink);
1878                         xu->unp_gencnt = unp->unp_gencnt;
1879                         sotoxsocket(unp->unp_socket, &xu->xu_socket);
1880                         UNP_PCB_UNLOCK(unp);
1881                         error = SYSCTL_OUT(req, xu, sizeof *xu);
1882                 } else  if (freeunp == 0)
1883                         UNP_PCB_UNLOCK(unp);
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         int freed;
1937
1938         /*
1939          * Regardless of whether the socket's peer dropped the connection
1940          * with this socket by aborting or disconnecting, POSIX requires
1941          * that ECONNRESET is returned.
1942          */
1943         /* acquire a reference so that unp isn't freed from underneath us */
1944
1945         UNP_PCB_LOCK(unp);
1946         if (so)
1947                 so->so_error = ECONNRESET;
1948         unp2 = unp->unp_conn;
1949         if (unp2 == unp) {
1950                 unp_disconnect(unp, unp2);
1951         } else if (unp2 != NULL) {
1952                 unp_pcb_hold(unp2);
1953                 unp_pcb_owned_lock2(unp, unp2, freed);
1954                 unp_disconnect(unp, unp2);
1955                 if (unp_pcb_rele(unp2) == 0)
1956                         UNP_PCB_UNLOCK(unp2);
1957         }
1958         if (unp_pcb_rele(unp) == 0)
1959                 UNP_PCB_UNLOCK(unp);
1960 }
1961
1962 static void
1963 unp_freerights(struct filedescent **fdep, int fdcount)
1964 {
1965         struct file *fp;
1966         int i;
1967
1968         KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1969
1970         for (i = 0; i < fdcount; i++) {
1971                 fp = fdep[i]->fde_file;
1972                 filecaps_free(&fdep[i]->fde_caps);
1973                 unp_discard(fp);
1974         }
1975         free(fdep[0], M_FILECAPS);
1976 }
1977
1978 static int
1979 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1980 {
1981         struct thread *td = curthread;          /* XXX */
1982         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1983         int i;
1984         int *fdp;
1985         struct filedesc *fdesc = td->td_proc->p_fd;
1986         struct filedescent **fdep;
1987         void *data;
1988         socklen_t clen = control->m_len, datalen;
1989         int error, newfds;
1990         u_int newlen;
1991
1992         UNP_LINK_UNLOCK_ASSERT();
1993
1994         error = 0;
1995         if (controlp != NULL) /* controlp == NULL => free control messages */
1996                 *controlp = NULL;
1997         while (cm != NULL) {
1998                 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1999                         error = EINVAL;
2000                         break;
2001                 }
2002                 data = CMSG_DATA(cm);
2003                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
2004                 if (cm->cmsg_level == SOL_SOCKET
2005                     && cm->cmsg_type == SCM_RIGHTS) {
2006                         newfds = datalen / sizeof(*fdep);
2007                         if (newfds == 0)
2008                                 goto next;
2009                         fdep = data;
2010
2011                         /* If we're not outputting the descriptors free them. */
2012                         if (error || controlp == NULL) {
2013                                 unp_freerights(fdep, newfds);
2014                                 goto next;
2015                         }
2016                         FILEDESC_XLOCK(fdesc);
2017
2018                         /*
2019                          * Now change each pointer to an fd in the global
2020                          * table to an integer that is the index to the local
2021                          * fd table entry that we set up to point to the
2022                          * global one we are transferring.
2023                          */
2024                         newlen = newfds * sizeof(int);
2025                         *controlp = sbcreatecontrol(NULL, newlen,
2026                             SCM_RIGHTS, SOL_SOCKET);
2027                         if (*controlp == NULL) {
2028                                 FILEDESC_XUNLOCK(fdesc);
2029                                 error = E2BIG;
2030                                 unp_freerights(fdep, newfds);
2031                                 goto next;
2032                         }
2033
2034                         fdp = (int *)
2035                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2036                         if (fdallocn(td, 0, fdp, newfds) != 0) {
2037                                 FILEDESC_XUNLOCK(fdesc);
2038                                 error = EMSGSIZE;
2039                                 unp_freerights(fdep, newfds);
2040                                 m_freem(*controlp);
2041                                 *controlp = NULL;
2042                                 goto next;
2043                         }
2044                         for (i = 0; i < newfds; i++, fdp++) {
2045                                 _finstall(fdesc, fdep[i]->fde_file, *fdp,
2046                                     (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0,
2047                                     &fdep[i]->fde_caps);
2048                                 unp_externalize_fp(fdep[i]->fde_file);
2049                         }
2050
2051                         /*
2052                          * The new type indicates that the mbuf data refers to
2053                          * kernel resources that may need to be released before
2054                          * the mbuf is freed.
2055                          */
2056                         m_chtype(*controlp, MT_EXTCONTROL);
2057                         FILEDESC_XUNLOCK(fdesc);
2058                         free(fdep[0], M_FILECAPS);
2059                 } else {
2060                         /* We can just copy anything else across. */
2061                         if (error || controlp == NULL)
2062                                 goto next;
2063                         *controlp = sbcreatecontrol(NULL, datalen,
2064                             cm->cmsg_type, cm->cmsg_level);
2065                         if (*controlp == NULL) {
2066                                 error = ENOBUFS;
2067                                 goto next;
2068                         }
2069                         bcopy(data,
2070                             CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
2071                             datalen);
2072                 }
2073                 controlp = &(*controlp)->m_next;
2074
2075 next:
2076                 if (CMSG_SPACE(datalen) < clen) {
2077                         clen -= CMSG_SPACE(datalen);
2078                         cm = (struct cmsghdr *)
2079                             ((caddr_t)cm + CMSG_SPACE(datalen));
2080                 } else {
2081                         clen = 0;
2082                         cm = NULL;
2083                 }
2084         }
2085
2086         m_freem(control);
2087         return (error);
2088 }
2089
2090 static void
2091 unp_zone_change(void *tag)
2092 {
2093
2094         uma_zone_set_max(unp_zone, maxsockets);
2095 }
2096
2097 static void
2098 unp_init(void)
2099 {
2100
2101 #ifdef VIMAGE
2102         if (!IS_DEFAULT_VNET(curvnet))
2103                 return;
2104 #endif
2105         unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
2106             NULL, NULL, UMA_ALIGN_CACHE, 0);
2107         if (unp_zone == NULL)
2108                 panic("unp_init");
2109         uma_zone_set_max(unp_zone, maxsockets);
2110         uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
2111         EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
2112             NULL, EVENTHANDLER_PRI_ANY);
2113         LIST_INIT(&unp_dhead);
2114         LIST_INIT(&unp_shead);
2115         LIST_INIT(&unp_sphead);
2116         SLIST_INIT(&unp_defers);
2117         TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
2118         TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
2119         UNP_LINK_LOCK_INIT();
2120         UNP_DEFERRED_LOCK_INIT();
2121 }
2122
2123 static void
2124 unp_internalize_cleanup_rights(struct mbuf *control)
2125 {
2126         struct cmsghdr *cp;
2127         struct mbuf *m;
2128         void *data;
2129         socklen_t datalen;
2130
2131         for (m = control; m != NULL; m = m->m_next) {
2132                 cp = mtod(m, struct cmsghdr *);
2133                 if (cp->cmsg_level != SOL_SOCKET ||
2134                     cp->cmsg_type != SCM_RIGHTS)
2135                         continue;
2136                 data = CMSG_DATA(cp);
2137                 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data;
2138                 unp_freerights(data, datalen / sizeof(struct filedesc *));
2139         }
2140 }
2141
2142 static int
2143 unp_internalize(struct mbuf **controlp, struct thread *td)
2144 {
2145         struct mbuf *control, **initial_controlp;
2146         struct proc *p;
2147         struct filedesc *fdesc;
2148         struct bintime *bt;
2149         struct cmsghdr *cm;
2150         struct cmsgcred *cmcred;
2151         struct filedescent *fde, **fdep, *fdev;
2152         struct file *fp;
2153         struct timeval *tv;
2154         struct timespec *ts;
2155         void *data;
2156         socklen_t clen, datalen;
2157         int i, j, error, *fdp, oldfds;
2158         u_int newlen;
2159
2160         UNP_LINK_UNLOCK_ASSERT();
2161
2162         p = td->td_proc;
2163         fdesc = p->p_fd;
2164         error = 0;
2165         control = *controlp;
2166         clen = control->m_len;
2167         *controlp = NULL;
2168         initial_controlp = controlp;
2169         for (cm = mtod(control, struct cmsghdr *); cm != NULL;) {
2170                 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
2171                     || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
2172                         error = EINVAL;
2173                         goto out;
2174                 }
2175                 data = CMSG_DATA(cm);
2176                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
2177
2178                 switch (cm->cmsg_type) {
2179                 /*
2180                  * Fill in credential information.
2181                  */
2182                 case SCM_CREDS:
2183                         *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
2184                             SCM_CREDS, SOL_SOCKET);
2185                         if (*controlp == NULL) {
2186                                 error = ENOBUFS;
2187                                 goto out;
2188                         }
2189                         cmcred = (struct cmsgcred *)
2190                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2191                         cmcred->cmcred_pid = p->p_pid;
2192                         cmcred->cmcred_uid = td->td_ucred->cr_ruid;
2193                         cmcred->cmcred_gid = td->td_ucred->cr_rgid;
2194                         cmcred->cmcred_euid = td->td_ucred->cr_uid;
2195                         cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
2196                             CMGROUP_MAX);
2197                         for (i = 0; i < cmcred->cmcred_ngroups; i++)
2198                                 cmcred->cmcred_groups[i] =
2199                                     td->td_ucred->cr_groups[i];
2200                         break;
2201
2202                 case SCM_RIGHTS:
2203                         oldfds = datalen / sizeof (int);
2204                         if (oldfds == 0)
2205                                 break;
2206                         /*
2207                          * Check that all the FDs passed in refer to legal
2208                          * files.  If not, reject the entire operation.
2209                          */
2210                         fdp = data;
2211                         FILEDESC_SLOCK(fdesc);
2212                         for (i = 0; i < oldfds; i++, fdp++) {
2213                                 fp = fget_locked(fdesc, *fdp);
2214                                 if (fp == NULL) {
2215                                         FILEDESC_SUNLOCK(fdesc);
2216                                         error = EBADF;
2217                                         goto out;
2218                                 }
2219                                 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
2220                                         FILEDESC_SUNLOCK(fdesc);
2221                                         error = EOPNOTSUPP;
2222                                         goto out;
2223                                 }
2224
2225                         }
2226
2227                         /*
2228                          * Now replace the integer FDs with pointers to the
2229                          * file structure and capability rights.
2230                          */
2231                         newlen = oldfds * sizeof(fdep[0]);
2232                         *controlp = sbcreatecontrol(NULL, newlen,
2233                             SCM_RIGHTS, SOL_SOCKET);
2234                         if (*controlp == NULL) {
2235                                 FILEDESC_SUNLOCK(fdesc);
2236                                 error = E2BIG;
2237                                 goto out;
2238                         }
2239                         fdp = data;
2240                         for (i = 0; i < oldfds; i++, fdp++) {
2241                                 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) {
2242                                         fdp = data;
2243                                         for (j = 0; j < i; j++, fdp++) {
2244                                                 fdrop(fdesc->fd_ofiles[*fdp].
2245                                                     fde_file, td);
2246                                         }
2247                                         FILEDESC_SUNLOCK(fdesc);
2248                                         error = EBADF;
2249                                         goto out;
2250                                 }
2251                         }
2252                         fdp = data;
2253                         fdep = (struct filedescent **)
2254                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2255                         fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
2256                             M_WAITOK);
2257                         for (i = 0; i < oldfds; i++, fdev++, fdp++) {
2258                                 fde = &fdesc->fd_ofiles[*fdp];
2259                                 fdep[i] = fdev;
2260                                 fdep[i]->fde_file = fde->fde_file;
2261                                 filecaps_copy(&fde->fde_caps,
2262                                     &fdep[i]->fde_caps, true);
2263                                 unp_internalize_fp(fdep[i]->fde_file);
2264                         }
2265                         FILEDESC_SUNLOCK(fdesc);
2266                         break;
2267
2268                 case SCM_TIMESTAMP:
2269                         *controlp = sbcreatecontrol(NULL, sizeof(*tv),
2270                             SCM_TIMESTAMP, SOL_SOCKET);
2271                         if (*controlp == NULL) {
2272                                 error = ENOBUFS;
2273                                 goto out;
2274                         }
2275                         tv = (struct timeval *)
2276                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2277                         microtime(tv);
2278                         break;
2279
2280                 case SCM_BINTIME:
2281                         *controlp = sbcreatecontrol(NULL, sizeof(*bt),
2282                             SCM_BINTIME, SOL_SOCKET);
2283                         if (*controlp == NULL) {
2284                                 error = ENOBUFS;
2285                                 goto out;
2286                         }
2287                         bt = (struct bintime *)
2288                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2289                         bintime(bt);
2290                         break;
2291
2292                 case SCM_REALTIME:
2293                         *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2294                             SCM_REALTIME, SOL_SOCKET);
2295                         if (*controlp == NULL) {
2296                                 error = ENOBUFS;
2297                                 goto out;
2298                         }
2299                         ts = (struct timespec *)
2300                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2301                         nanotime(ts);
2302                         break;
2303
2304                 case SCM_MONOTONIC:
2305                         *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2306                             SCM_MONOTONIC, SOL_SOCKET);
2307                         if (*controlp == NULL) {
2308                                 error = ENOBUFS;
2309                                 goto out;
2310                         }
2311                         ts = (struct timespec *)
2312                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2313                         nanouptime(ts);
2314                         break;
2315
2316                 default:
2317                         error = EINVAL;
2318                         goto out;
2319                 }
2320
2321                 if (*controlp != NULL)
2322                         controlp = &(*controlp)->m_next;
2323                 if (CMSG_SPACE(datalen) < clen) {
2324                         clen -= CMSG_SPACE(datalen);
2325                         cm = (struct cmsghdr *)
2326                             ((caddr_t)cm + CMSG_SPACE(datalen));
2327                 } else {
2328                         clen = 0;
2329                         cm = NULL;
2330                 }
2331         }
2332
2333 out:
2334         if (error != 0 && initial_controlp != NULL)
2335                 unp_internalize_cleanup_rights(*initial_controlp);
2336         m_freem(control);
2337         return (error);
2338 }
2339
2340 static struct mbuf *
2341 unp_addsockcred(struct thread *td, struct mbuf *control)
2342 {
2343         struct mbuf *m, *n, *n_prev;
2344         struct sockcred *sc;
2345         const struct cmsghdr *cm;
2346         int ngroups;
2347         int i;
2348
2349         ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2350         m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2351         if (m == NULL)
2352                 return (control);
2353
2354         sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2355         sc->sc_uid = td->td_ucred->cr_ruid;
2356         sc->sc_euid = td->td_ucred->cr_uid;
2357         sc->sc_gid = td->td_ucred->cr_rgid;
2358         sc->sc_egid = td->td_ucred->cr_gid;
2359         sc->sc_ngroups = ngroups;
2360         for (i = 0; i < sc->sc_ngroups; i++)
2361                 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2362
2363         /*
2364          * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2365          * created SCM_CREDS control message (struct sockcred) has another
2366          * format.
2367          */
2368         if (control != NULL)
2369                 for (n = control, n_prev = NULL; n != NULL;) {
2370                         cm = mtod(n, struct cmsghdr *);
2371                         if (cm->cmsg_level == SOL_SOCKET &&
2372                             cm->cmsg_type == SCM_CREDS) {
2373                                 if (n_prev == NULL)
2374                                         control = n->m_next;
2375                                 else
2376                                         n_prev->m_next = n->m_next;
2377                                 n = m_free(n);
2378                         } else {
2379                                 n_prev = n;
2380                                 n = n->m_next;
2381                         }
2382                 }
2383
2384         /* Prepend it to the head. */
2385         m->m_next = control;
2386         return (m);
2387 }
2388
2389 static struct unpcb *
2390 fptounp(struct file *fp)
2391 {
2392         struct socket *so;
2393
2394         if (fp->f_type != DTYPE_SOCKET)
2395                 return (NULL);
2396         if ((so = fp->f_data) == NULL)
2397                 return (NULL);
2398         if (so->so_proto->pr_domain != &localdomain)
2399                 return (NULL);
2400         return sotounpcb(so);
2401 }
2402
2403 static void
2404 unp_discard(struct file *fp)
2405 {
2406         struct unp_defer *dr;
2407
2408         if (unp_externalize_fp(fp)) {
2409                 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2410                 dr->ud_fp = fp;
2411                 UNP_DEFERRED_LOCK();
2412                 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2413                 UNP_DEFERRED_UNLOCK();
2414                 atomic_add_int(&unp_defers_count, 1);
2415                 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2416         } else
2417                 (void) closef(fp, (struct thread *)NULL);
2418 }
2419
2420 static void
2421 unp_process_defers(void *arg __unused, int pending)
2422 {
2423         struct unp_defer *dr;
2424         SLIST_HEAD(, unp_defer) drl;
2425         int count;
2426
2427         SLIST_INIT(&drl);
2428         for (;;) {
2429                 UNP_DEFERRED_LOCK();
2430                 if (SLIST_FIRST(&unp_defers) == NULL) {
2431                         UNP_DEFERRED_UNLOCK();
2432                         break;
2433                 }
2434                 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2435                 UNP_DEFERRED_UNLOCK();
2436                 count = 0;
2437                 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2438                         SLIST_REMOVE_HEAD(&drl, ud_link);
2439                         closef(dr->ud_fp, NULL);
2440                         free(dr, M_TEMP);
2441                         count++;
2442                 }
2443                 atomic_add_int(&unp_defers_count, -count);
2444         }
2445 }
2446
2447 static void
2448 unp_internalize_fp(struct file *fp)
2449 {
2450         struct unpcb *unp;
2451
2452         UNP_LINK_WLOCK();
2453         if ((unp = fptounp(fp)) != NULL) {
2454                 unp->unp_file = fp;
2455                 unp->unp_msgcount++;
2456         }
2457         unp_rights++;
2458         UNP_LINK_WUNLOCK();
2459 }
2460
2461 static int
2462 unp_externalize_fp(struct file *fp)
2463 {
2464         struct unpcb *unp;
2465         int ret;
2466
2467         UNP_LINK_WLOCK();
2468         if ((unp = fptounp(fp)) != NULL) {
2469                 unp->unp_msgcount--;
2470                 ret = 1;
2471         } else
2472                 ret = 0;
2473         unp_rights--;
2474         UNP_LINK_WUNLOCK();
2475         return (ret);
2476 }
2477
2478 /*
2479  * unp_defer indicates whether additional work has been defered for a future
2480  * pass through unp_gc().  It is thread local and does not require explicit
2481  * synchronization.
2482  */
2483 static int      unp_marked;
2484 static int      unp_unreachable;
2485
2486 static void
2487 unp_accessable(struct filedescent **fdep, int fdcount)
2488 {
2489         struct unpcb *unp;
2490         struct file *fp;
2491         int i;
2492
2493         for (i = 0; i < fdcount; i++) {
2494                 fp = fdep[i]->fde_file;
2495                 if ((unp = fptounp(fp)) == NULL)
2496                         continue;
2497                 if (unp->unp_gcflag & UNPGC_REF)
2498                         continue;
2499                 unp->unp_gcflag &= ~UNPGC_DEAD;
2500                 unp->unp_gcflag |= UNPGC_REF;
2501                 unp_marked++;
2502         }
2503 }
2504
2505 static void
2506 unp_gc_process(struct unpcb *unp)
2507 {
2508         struct socket *so, *soa;
2509         struct file *fp;
2510
2511         /* Already processed. */
2512         if (unp->unp_gcflag & UNPGC_SCANNED)
2513                 return;
2514         fp = unp->unp_file;
2515
2516         /*
2517          * Check for a socket potentially in a cycle.  It must be in a
2518          * queue as indicated by msgcount, and this must equal the file
2519          * reference count.  Note that when msgcount is 0 the file is NULL.
2520          */
2521         if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2522             unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2523                 unp->unp_gcflag |= UNPGC_DEAD;
2524                 unp_unreachable++;
2525                 return;
2526         }
2527
2528         so = unp->unp_socket;
2529         SOCK_LOCK(so);
2530         if (SOLISTENING(so)) {
2531                 /*
2532                  * Mark all sockets in our accept queue.
2533                  */
2534                 TAILQ_FOREACH(soa, &so->sol_comp, so_list) {
2535                         if (sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS)
2536                                 continue;
2537                         SOCKBUF_LOCK(&soa->so_rcv);
2538                         unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2539                         SOCKBUF_UNLOCK(&soa->so_rcv);
2540                 }
2541         } else {
2542                 /*
2543                  * Mark all sockets we reference with RIGHTS.
2544                  */
2545                 if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2546                         SOCKBUF_LOCK(&so->so_rcv);
2547                         unp_scan(so->so_rcv.sb_mb, unp_accessable);
2548                         SOCKBUF_UNLOCK(&so->so_rcv);
2549                 }
2550         }
2551         SOCK_UNLOCK(so);
2552         unp->unp_gcflag |= UNPGC_SCANNED;
2553 }
2554
2555 static int unp_recycled;
2556 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2557     "Number of unreachable sockets claimed by the garbage collector.");
2558
2559 static int unp_taskcount;
2560 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2561     "Number of times the garbage collector has run.");
2562
2563 static void
2564 unp_gc(__unused void *arg, int pending)
2565 {
2566         struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2567                                     NULL };
2568         struct unp_head **head;
2569         struct file *f, **unref;
2570         struct unpcb *unp;
2571         int i, total;
2572
2573         unp_taskcount++;
2574         UNP_LINK_RLOCK();
2575         /*
2576          * First clear all gc flags from previous runs, apart from
2577          * UNPGC_IGNORE_RIGHTS.
2578          */
2579         for (head = heads; *head != NULL; head++)
2580                 LIST_FOREACH(unp, *head, unp_link)
2581                         unp->unp_gcflag =
2582                             (unp->unp_gcflag & UNPGC_IGNORE_RIGHTS);
2583
2584         /*
2585          * Scan marking all reachable sockets with UNPGC_REF.  Once a socket
2586          * is reachable all of the sockets it references are reachable.
2587          * Stop the scan once we do a complete loop without discovering
2588          * a new reachable socket.
2589          */
2590         do {
2591                 unp_unreachable = 0;
2592                 unp_marked = 0;
2593                 for (head = heads; *head != NULL; head++)
2594                         LIST_FOREACH(unp, *head, unp_link)
2595                                 unp_gc_process(unp);
2596         } while (unp_marked);
2597         UNP_LINK_RUNLOCK();
2598         if (unp_unreachable == 0)
2599                 return;
2600
2601         /*
2602          * Allocate space for a local list of dead unpcbs.
2603          */
2604         unref = malloc(unp_unreachable * sizeof(struct file *),
2605             M_TEMP, M_WAITOK);
2606
2607         /*
2608          * Iterate looking for sockets which have been specifically marked
2609          * as as unreachable and store them locally.
2610          */
2611         UNP_LINK_RLOCK();
2612         for (total = 0, head = heads; *head != NULL; head++)
2613                 LIST_FOREACH(unp, *head, unp_link)
2614                         if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2615                                 f = unp->unp_file;
2616                                 if (unp->unp_msgcount == 0 || f == NULL ||
2617                                     f->f_count != unp->unp_msgcount ||
2618                                     !fhold(f))
2619                                         continue;
2620                                 unref[total++] = f;
2621                                 KASSERT(total <= unp_unreachable,
2622                                     ("unp_gc: incorrect unreachable count."));
2623                         }
2624         UNP_LINK_RUNLOCK();
2625
2626         /*
2627          * Now flush all sockets, free'ing rights.  This will free the
2628          * struct files associated with these sockets but leave each socket
2629          * with one remaining ref.
2630          */
2631         for (i = 0; i < total; i++) {
2632                 struct socket *so;
2633
2634                 so = unref[i]->f_data;
2635                 CURVNET_SET(so->so_vnet);
2636                 sorflush(so);
2637                 CURVNET_RESTORE();
2638         }
2639
2640         /*
2641          * And finally release the sockets so they can be reclaimed.
2642          */
2643         for (i = 0; i < total; i++)
2644                 fdrop(unref[i], NULL);
2645         unp_recycled += total;
2646         free(unref, M_TEMP);
2647 }
2648
2649 static void
2650 unp_dispose_mbuf(struct mbuf *m)
2651 {
2652
2653         if (m)
2654                 unp_scan(m, unp_freerights);
2655 }
2656
2657 /*
2658  * Synchronize against unp_gc, which can trip over data as we are freeing it.
2659  */
2660 static void
2661 unp_dispose(struct socket *so)
2662 {
2663         struct unpcb *unp;
2664
2665         unp = sotounpcb(so);
2666         UNP_LINK_WLOCK();
2667         unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2668         UNP_LINK_WUNLOCK();
2669         if (!SOLISTENING(so))
2670                 unp_dispose_mbuf(so->so_rcv.sb_mb);
2671 }
2672
2673 static void
2674 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2675 {
2676         struct mbuf *m;
2677         struct cmsghdr *cm;
2678         void *data;
2679         socklen_t clen, datalen;
2680
2681         while (m0 != NULL) {
2682                 for (m = m0; m; m = m->m_next) {
2683                         if (m->m_type != MT_CONTROL)
2684                                 continue;
2685
2686                         cm = mtod(m, struct cmsghdr *);
2687                         clen = m->m_len;
2688
2689                         while (cm != NULL) {
2690                                 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2691                                         break;
2692
2693                                 data = CMSG_DATA(cm);
2694                                 datalen = (caddr_t)cm + cm->cmsg_len
2695                                     - (caddr_t)data;
2696
2697                                 if (cm->cmsg_level == SOL_SOCKET &&
2698                                     cm->cmsg_type == SCM_RIGHTS) {
2699                                         (*op)(data, datalen /
2700                                             sizeof(struct filedescent *));
2701                                 }
2702
2703                                 if (CMSG_SPACE(datalen) < clen) {
2704                                         clen -= CMSG_SPACE(datalen);
2705                                         cm = (struct cmsghdr *)
2706                                             ((caddr_t)cm + CMSG_SPACE(datalen));
2707                                 } else {
2708                                         clen = 0;
2709                                         cm = NULL;
2710                                 }
2711                         }
2712                 }
2713                 m0 = m0->m_nextpkt;
2714         }
2715 }
2716
2717 /*
2718  * A helper function called by VFS before socket-type vnode reclamation.
2719  * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2720  * use count.
2721  */
2722 void
2723 vfs_unp_reclaim(struct vnode *vp)
2724 {
2725         struct unpcb *unp;
2726         int active;
2727         struct mtx *vplock;
2728
2729         ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2730         KASSERT(vp->v_type == VSOCK,
2731             ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2732
2733         active = 0;
2734         vplock = mtx_pool_find(mtxpool_sleep, vp);
2735         mtx_lock(vplock);
2736         VOP_UNP_CONNECT(vp, &unp);
2737         if (unp == NULL)
2738                 goto done;
2739         UNP_PCB_LOCK(unp);
2740         if (unp->unp_vnode == vp) {
2741                 VOP_UNP_DETACH(vp);
2742                 unp->unp_vnode = NULL;
2743                 active = 1;
2744         }
2745         UNP_PCB_UNLOCK(unp);
2746  done:
2747         mtx_unlock(vplock);
2748         if (active)
2749                 vunref(vp);
2750 }
2751
2752 #ifdef DDB
2753 static void
2754 db_print_indent(int indent)
2755 {
2756         int i;
2757
2758         for (i = 0; i < indent; i++)
2759                 db_printf(" ");
2760 }
2761
2762 static void
2763 db_print_unpflags(int unp_flags)
2764 {
2765         int comma;
2766
2767         comma = 0;
2768         if (unp_flags & UNP_HAVEPC) {
2769                 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2770                 comma = 1;
2771         }
2772         if (unp_flags & UNP_WANTCRED) {
2773                 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2774                 comma = 1;
2775         }
2776         if (unp_flags & UNP_CONNWAIT) {
2777                 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2778                 comma = 1;
2779         }
2780         if (unp_flags & UNP_CONNECTING) {
2781                 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2782                 comma = 1;
2783         }
2784         if (unp_flags & UNP_BINDING) {
2785                 db_printf("%sUNP_BINDING", comma ? ", " : "");
2786                 comma = 1;
2787         }
2788 }
2789
2790 static void
2791 db_print_xucred(int indent, struct xucred *xu)
2792 {
2793         int comma, i;
2794
2795         db_print_indent(indent);
2796         db_printf("cr_version: %u   cr_uid: %u   cr_ngroups: %d\n",
2797             xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2798         db_print_indent(indent);
2799         db_printf("cr_groups: ");
2800         comma = 0;
2801         for (i = 0; i < xu->cr_ngroups; i++) {
2802                 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2803                 comma = 1;
2804         }
2805         db_printf("\n");
2806 }
2807
2808 static void
2809 db_print_unprefs(int indent, struct unp_head *uh)
2810 {
2811         struct unpcb *unp;
2812         int counter;
2813
2814         counter = 0;
2815         LIST_FOREACH(unp, uh, unp_reflink) {
2816                 if (counter % 4 == 0)
2817                         db_print_indent(indent);
2818                 db_printf("%p  ", unp);
2819                 if (counter % 4 == 3)
2820                         db_printf("\n");
2821                 counter++;
2822         }
2823         if (counter != 0 && counter % 4 != 0)
2824                 db_printf("\n");
2825 }
2826
2827 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2828 {
2829         struct unpcb *unp;
2830
2831         if (!have_addr) {
2832                 db_printf("usage: show unpcb <addr>\n");
2833                 return;
2834         }
2835         unp = (struct unpcb *)addr;
2836
2837         db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
2838             unp->unp_vnode);
2839
2840         db_printf("unp_ino: %ju   unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2841             unp->unp_conn);
2842
2843         db_printf("unp_refs:\n");
2844         db_print_unprefs(2, &unp->unp_refs);
2845
2846         /* XXXRW: Would be nice to print the full address, if any. */
2847         db_printf("unp_addr: %p\n", unp->unp_addr);
2848
2849         db_printf("unp_gencnt: %llu\n",
2850             (unsigned long long)unp->unp_gencnt);
2851
2852         db_printf("unp_flags: %x (", unp->unp_flags);
2853         db_print_unpflags(unp->unp_flags);
2854         db_printf(")\n");
2855
2856         db_printf("unp_peercred:\n");
2857         db_print_xucred(2, &unp->unp_peercred);
2858
2859         db_printf("unp_refcount: %u\n", unp->unp_refcount);
2860 }
2861 #endif