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, int *freed)
 361
 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                 }                                                                                                                       \
 387 } while (0)
 388
 389
 390 /*
 391  * Definitions of protocols supported in the LOCAL domain.
 392  */
 393 static struct domain localdomain;
 394 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
 395 static struct pr_usrreqs uipc_usrreqs_seqpacket;
 396 static struct protosw localsw[] = {
 397 {
 398         .pr_type =              SOCK_STREAM,
 399         .pr_domain =            &localdomain,
 400         .pr_flags =             PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
 401         .pr_ctloutput =         &uipc_ctloutput,
 402         .pr_usrreqs =           &uipc_usrreqs_stream
 403 },
 404 {
 405         .pr_type =              SOCK_DGRAM,
 406         .pr_domain =            &localdomain,
 407         .pr_flags =             PR_ATOMIC|PR_ADDR|PR_RIGHTS,
 408         .pr_ctloutput =         &uipc_ctloutput,
 409         .pr_usrreqs =           &uipc_usrreqs_dgram
 410 },
 411 {
 412         .pr_type =              SOCK_SEQPACKET,
 413         .pr_domain =            &localdomain,
 414
 415         /*
 416          * XXXRW: For now, PR_ADDR because soreceive will bump into them
 417          * due to our use of sbappendaddr.  A new sbappend variants is needed
 418          * that supports both atomic record writes and control data.
 419          */
 420         .pr_flags =             PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
 421                                     PR_RIGHTS,
 422         .pr_ctloutput =         &uipc_ctloutput,
 423         .pr_usrreqs =           &uipc_usrreqs_seqpacket,
 424 },
 425 };
 426
 427 static struct domain localdomain = {
 428         .dom_family =           AF_LOCAL,
 429         .dom_name =             "local",
 430         .dom_init =             unp_init,
 431         .dom_externalize =      unp_externalize,
 432         .dom_dispose =          unp_dispose,
 433         .dom_protosw =          localsw,
 434         .dom_protoswNPROTOSW =  &localsw[nitems(localsw)]
 435 };
 436 DOMAIN_SET(local);
 437
 438 static void
 439 uipc_abort(struct socket *so)
 440 {
 441         struct unpcb *unp, *unp2;
 442
 443         unp = sotounpcb(so);
 444         KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
 445         UNP_PCB_UNLOCK_ASSERT(unp);
 446
 447         UNP_PCB_LOCK(unp);
 448         unp2 = unp->unp_conn;
 449         if (unp2 != NULL) {
 450                 unp_pcb_hold(unp2);
 451                 UNP_PCB_UNLOCK(unp);
 452                 unp_drop(unp2);
 453         } else
 454                 UNP_PCB_UNLOCK(unp);
 455 }
 456
 457 static int
 458 uipc_accept(struct socket *so, struct sockaddr **nam)
 459 {
 460         struct unpcb *unp, *unp2;
 461         const struct sockaddr *sa;
 462
 463         /*
 464          * Pass back name of connected socket, if it was bound and we are
 465          * still connected (our peer may have closed already!).
 466          */
 467         unp = sotounpcb(so);
 468         KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
 469
 470         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 471         UNP_LINK_RLOCK();
 472         unp2 = unp->unp_conn;
 473         if (unp2 != NULL && unp2->unp_addr != NULL) {
 474                 UNP_PCB_LOCK(unp2);
 475                 sa = (struct sockaddr *) unp2->unp_addr;
 476                 bcopy(sa, *nam, sa->sa_len);
 477                 UNP_PCB_UNLOCK(unp2);
 478         } else {
 479                 sa = &sun_noname;
 480                 bcopy(sa, *nam, sa->sa_len);
 481         }
 482         UNP_LINK_RUNLOCK();
 483         return (0);
 484 }
 485
 486 static int
 487 uipc_attach(struct socket *so, int proto, struct thread *td)
 488 {
 489         u_long sendspace, recvspace;
 490         struct unpcb *unp;
 491         int error;
 492         bool locked;
 493
 494         KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
 495         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
 496                 switch (so->so_type) {
 497                 case SOCK_STREAM:
 498                         sendspace = unpst_sendspace;
 499                         recvspace = unpst_recvspace;
 500                         break;
 501
 502                 case SOCK_DGRAM:
 503                         sendspace = unpdg_sendspace;
 504                         recvspace = unpdg_recvspace;
 505                         break;
 506
 507                 case SOCK_SEQPACKET:
 508                         sendspace = unpsp_sendspace;
 509                         recvspace = unpsp_recvspace;
 510                         break;
 511
 512                 default:
 513                         panic("uipc_attach");
 514                 }
 515                 error = soreserve(so, sendspace, recvspace);
 516                 if (error)
 517                         return (error);
 518         }
 519         unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
 520         if (unp == NULL)
 521                 return (ENOBUFS);
 522         LIST_INIT(&unp->unp_refs);
 523         UNP_PCB_LOCK_INIT(unp);
 524         unp->unp_socket = so;
 525         so->so_pcb = unp;
 526         unp->unp_refcount = 1;
 527         if (so->so_listen != NULL)
 528                 unp->unp_flags |= UNP_NASCENT;
 529
 530         if ((locked = UNP_LINK_WOWNED()) == false)
 531                 UNP_LINK_WLOCK();
 532
 533         unp->unp_gencnt = ++unp_gencnt;
 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 (unp == unp2) {
 876                 if (unp_pcb_rele(unp) == 0)
 877                         UNP_PCB_UNLOCK(unp);
 878         }
 879         unp_pcb_owned_lock2(unp, unp2, freed);
 880         if (__predict_false(freed)) {
 881                 UNP_PCB_UNLOCK(unp);
 882                 return (0);
 883         }
 884         unp_pcb_hold(unp2);
 885         unp_pcb_hold(unp);
 886         unp_disconnect(unp, unp2);
 887         if (unp_pcb_rele(unp) == 0)
 888                 UNP_PCB_UNLOCK(unp);
 889         if (unp_pcb_rele(unp2) == 0)
 890                 UNP_PCB_UNLOCK(unp2);
 891         return (0);
 892 }
 893
 894 static int
 895 uipc_listen(struct socket *so, int backlog, struct thread *td)
 896 {
 897         struct unpcb *unp;
 898         int error;
 899
 900         if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
 901                 return (EOPNOTSUPP);
 902
 903         unp = sotounpcb(so);
 904         KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
 905
 906         UNP_PCB_LOCK(unp);
 907         if (unp->unp_vnode == NULL) {
 908                 /* Already connected or not bound to an address. */
 909                 error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ;
 910                 UNP_PCB_UNLOCK(unp);
 911                 return (error);
 912         }
 913
 914         SOCK_LOCK(so);
 915         error = solisten_proto_check(so);
 916         if (error == 0) {
 917                 cru2x(td->td_ucred, &unp->unp_peercred);
 918                 solisten_proto(so, backlog);
 919         }
 920         SOCK_UNLOCK(so);
 921         UNP_PCB_UNLOCK(unp);
 922         return (error);
 923 }
 924
 925 static int
 926 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
 927 {
 928         struct unpcb *unp, *unp2;
 929         const struct sockaddr *sa;
 930
 931         unp = sotounpcb(so);
 932         KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
 933
 934         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 935         UNP_LINK_RLOCK();
 936         /*
 937          * XXX: It seems that this test always fails even when connection is
 938          * established.  So, this else clause is added as workaround to
 939          * return PF_LOCAL sockaddr.
 940          */
 941         unp2 = unp->unp_conn;
 942         if (unp2 != NULL) {
 943                 UNP_PCB_LOCK(unp2);
 944                 if (unp2->unp_addr != NULL)
 945                         sa = (struct sockaddr *) unp2->unp_addr;
 946                 else
 947                         sa = &sun_noname;
 948                 bcopy(sa, *nam, sa->sa_len);
 949                 UNP_PCB_UNLOCK(unp2);
 950         } else {
 951                 sa = &sun_noname;
 952                 bcopy(sa, *nam, sa->sa_len);
 953         }
 954         UNP_LINK_RUNLOCK();
 955         return (0);
 956 }
 957
 958 static int
 959 uipc_rcvd(struct socket *so, int flags)
 960 {
 961         struct unpcb *unp, *unp2;
 962         struct socket *so2;
 963         u_int mbcnt, sbcc;
 964
 965         unp = sotounpcb(so);
 966         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
 967         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
 968             ("%s: socktype %d", __func__, so->so_type));
 969
 970         /*
 971          * Adjust backpressure on sender and wakeup any waiting to write.
 972          *
 973          * The unp lock is acquired to maintain the validity of the unp_conn
 974          * pointer; no lock on unp2 is required as unp2->unp_socket will be
 975          * static as long as we don't permit unp2 to disconnect from unp,
 976          * which is prevented by the lock on unp.  We cache values from
 977          * so_rcv to avoid holding the so_rcv lock over the entire
 978          * transaction on the remote so_snd.
 979          */
 980         SOCKBUF_LOCK(&so->so_rcv);
 981         mbcnt = so->so_rcv.sb_mbcnt;
 982         sbcc = sbavail(&so->so_rcv);
 983         SOCKBUF_UNLOCK(&so->so_rcv);
 984         /*
 985          * There is a benign race condition at this point.  If we're planning to
 986          * clear SB_STOP, but uipc_send is called on the connected socket at
 987          * this instant, it might add data to the sockbuf and set SB_STOP.  Then
 988          * we would erroneously clear SB_STOP below, even though the sockbuf is
 989          * full.  The race is benign because the only ill effect is to allow the
 990          * sockbuf to exceed its size limit, and the size limits are not
 991          * strictly guaranteed anyway.
 992          */
 993         UNP_PCB_LOCK(unp);
 994         unp2 = unp->unp_conn;
 995         if (unp2 == NULL) {
 996                 UNP_PCB_UNLOCK(unp);
 997                 return (0);
 998         }
 999         so2 = unp2->unp_socket;
1000         SOCKBUF_LOCK(&so2->so_snd);
1001         if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
1002                 so2->so_snd.sb_flags &= ~SB_STOP;
1003         sowwakeup_locked(so2);
1004         UNP_PCB_UNLOCK(unp);
1005         return (0);
1006 }
1007
1008 static int
1009 connect_internal(struct socket *so, struct sockaddr *nam, struct thread *td)
1010 {
1011         int error;
1012         struct unpcb *unp;
1013
1014         unp = so->so_pcb;
1015         if (unp->unp_conn != NULL)
1016                 return (EISCONN);
1017         error = unp_connect(so, nam, td);
1018         if (error)
1019                 return (error);
1020         UNP_PCB_LOCK(unp);
1021         if (unp->unp_conn == NULL) {
1022                 UNP_PCB_UNLOCK(unp);
1023                 if (error == 0)
1024                         error = ENOTCONN;
1025         }
1026         return (error);
1027 }
1028
1029
1030 static int
1031 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
1032     struct mbuf *control, struct thread *td)
1033 {
1034         struct unpcb *unp, *unp2;
1035         struct socket *so2;
1036         u_int mbcnt, sbcc;
1037         int freed, error;
1038
1039         unp = sotounpcb(so);
1040         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
1041         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
1042             so->so_type == SOCK_SEQPACKET,
1043             ("%s: socktype %d", __func__, so->so_type));
1044
1045         freed = error = 0;
1046         if (flags & PRUS_OOB) {
1047                 error = EOPNOTSUPP;
1048                 goto release;
1049         }
1050         if (control != NULL && (error = unp_internalize(&control, td)))
1051                 goto release;
1052
1053         unp2 = NULL;
1054         switch (so->so_type) {
1055         case SOCK_DGRAM:
1056         {
1057                 const struct sockaddr *from;
1058
1059                 if (nam != NULL) {
1060                         /*
1061                          * We return with UNP_PCB_LOCK_HELD so we know that
1062                          * the reference is live if the pointer is valid.
1063                          */
1064                         if ((error = connect_internal(so, nam, td)))
1065                                 break;
1066                         MPASS(unp->unp_conn != NULL);
1067                         unp2 = unp->unp_conn;
1068                 } else  {
1069                         UNP_PCB_LOCK(unp);
1070
1071                         /*
1072                          * Because connect() and send() are non-atomic in a sendto()
1073                          * with a target address, it's possible that the socket will
1074                          * have disconnected before the send() can run.  In that case
1075                          * return the slightly counter-intuitive but otherwise
1076                          * correct error that the socket is not connected.
1077                          */
1078                         if ((unp2 = unp->unp_conn)  == NULL) {
1079                                 UNP_PCB_UNLOCK(unp);
1080                                 error = ENOTCONN;
1081                                 break;
1082                         }
1083                 }
1084                 if (__predict_false(unp == unp2)) {
1085                         if (unp->unp_socket == NULL) {
1086                                 error = ENOTCONN;
1087                                 break;
1088                         }
1089                         goto connect_self;
1090                 }
1091                 unp_pcb_owned_lock2(unp, unp2, freed);
1092                 if (__predict_false(freed)) {
1093                         UNP_PCB_UNLOCK(unp);
1094                         error = ENOTCONN;
1095                         break;
1096                 }
1097                 /*
1098                  * The socket referencing unp2 may have been closed
1099                  * or unp may have been disconnected if the unp lock
1100                  * was dropped to acquire unp2.
1101                  */
1102                 if (__predict_false(unp->unp_conn == NULL) ||
1103                         unp2->unp_socket == NULL) {
1104                         UNP_PCB_UNLOCK(unp);
1105                         if (unp_pcb_rele(unp2) == 0)
1106                                 UNP_PCB_UNLOCK(unp2);
1107                         error = ENOTCONN;
1108                         break;
1109                 }
1110         connect_self:
1111                 if (unp2->unp_flags & UNP_WANTCRED)
1112                         control = unp_addsockcred(td, control);
1113                 if (unp->unp_addr != NULL)
1114                         from = (struct sockaddr *)unp->unp_addr;
1115                 else
1116                         from = &sun_noname;
1117                 so2 = unp2->unp_socket;
1118                 SOCKBUF_LOCK(&so2->so_rcv);
1119                 if (sbappendaddr_locked(&so2->so_rcv, from, m,
1120                     control)) {
1121                         sorwakeup_locked(so2);
1122                         m = NULL;
1123                         control = NULL;
1124                 } else {
1125                         SOCKBUF_UNLOCK(&so2->so_rcv);
1126                         error = ENOBUFS;
1127                 }
1128                 if (nam != NULL)
1129                         unp_disconnect(unp, unp2);
1130                 if (__predict_true(unp != unp2))
1131                         UNP_PCB_UNLOCK(unp2);
1132                 UNP_PCB_UNLOCK(unp);
1133                 break;
1134         }
1135
1136         case SOCK_SEQPACKET:
1137         case SOCK_STREAM:
1138                 if ((so->so_state & SS_ISCONNECTED) == 0) {
1139                         if (nam != NULL) {
1140                                 if ((error = connect_internal(so, nam, td)))
1141                                         break;
1142                         } else  {
1143                                 error = ENOTCONN;
1144                                 break;
1145                         }
1146                 } else if ((unp2 = unp->unp_conn) == NULL) {
1147                         error = ENOTCONN;
1148                         break;
1149                 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1150                         error = EPIPE;
1151                         break;
1152                 } else {
1153                         UNP_PCB_LOCK(unp);
1154                         if ((unp2 = unp->unp_conn) == NULL) {
1155                                 UNP_PCB_UNLOCK(unp);
1156                                 error = ENOTCONN;
1157                                 break;
1158                         }
1159                 }
1160                 unp_pcb_owned_lock2(unp, unp2, freed);
1161                 UNP_PCB_UNLOCK(unp);
1162                 if (__predict_false(freed)) {
1163                         error = ENOTCONN;
1164                         break;
1165                 }
1166                 if ((so2 = unp2->unp_socket) == NULL) {
1167                         UNP_PCB_UNLOCK(unp2);
1168                         error = ENOTCONN;
1169                         break;
1170                 }
1171                 SOCKBUF_LOCK(&so2->so_rcv);
1172                 if (unp2->unp_flags & UNP_WANTCRED) {
1173                         /*
1174                          * Credentials are passed only once on SOCK_STREAM
1175                          * and SOCK_SEQPACKET.
1176                          */
1177                         unp2->unp_flags &= ~UNP_WANTCRED;
1178                         control = unp_addsockcred(td, control);
1179                 }
1180                 /*
1181                  * Send to paired receive port, and then reduce send buffer
1182                  * hiwater marks to maintain backpressure.  Wake up readers.
1183                  */
1184                 switch (so->so_type) {
1185                 case SOCK_STREAM:
1186                         if (control != NULL) {
1187                                 if (sbappendcontrol_locked(&so2->so_rcv, m,
1188                                     control))
1189                                         control = NULL;
1190                         } else
1191                                 sbappend_locked(&so2->so_rcv, m, flags);
1192                         break;
1193
1194                 case SOCK_SEQPACKET: {
1195                         const struct sockaddr *from;
1196
1197                         from = &sun_noname;
1198                         /*
1199                          * Don't check for space available in so2->so_rcv.
1200                          * Unix domain sockets only check for space in the
1201                          * sending sockbuf, and that check is performed one
1202                          * level up the stack.
1203                          */
1204                         if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1205                                 from, m, control))
1206                                 control = NULL;
1207                         break;
1208                         }
1209                 }
1210
1211                 mbcnt = so2->so_rcv.sb_mbcnt;
1212                 sbcc = sbavail(&so2->so_rcv);
1213                 if (sbcc)
1214                         sorwakeup_locked(so2);
1215                 else
1216                         SOCKBUF_UNLOCK(&so2->so_rcv);
1217
1218                 /*
1219                  * The PCB lock on unp2 protects the SB_STOP flag.  Without it,
1220                  * it would be possible for uipc_rcvd to be called at this
1221                  * point, drain the receiving sockbuf, clear SB_STOP, and then
1222                  * we would set SB_STOP below.  That could lead to an empty
1223                  * sockbuf having SB_STOP set
1224                  */
1225                 SOCKBUF_LOCK(&so->so_snd);
1226                 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1227                         so->so_snd.sb_flags |= SB_STOP;
1228                 SOCKBUF_UNLOCK(&so->so_snd);
1229                 UNP_PCB_UNLOCK(unp2);
1230                 m = NULL;
1231                 break;
1232         }
1233
1234         /*
1235          * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1236          */
1237         if (flags & PRUS_EOF) {
1238                 UNP_PCB_LOCK(unp);
1239                 socantsendmore(so);
1240                 unp_shutdown(unp);
1241                 UNP_PCB_UNLOCK(unp);
1242         }
1243         if (control != NULL && error != 0)
1244                 unp_dispose_mbuf(control);
1245
1246 release:
1247         if (control != NULL)
1248                 m_freem(control);
1249         /*
1250          * In case of PRUS_NOTREADY, uipc_ready() is responsible
1251          * for freeing memory.
1252          */
1253         if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1254                 m_freem(m);
1255         return (error);
1256 }
1257
1258 static int
1259 uipc_ready(struct socket *so, struct mbuf *m, int count)
1260 {
1261         struct unpcb *unp, *unp2;
1262         struct socket *so2;
1263         int error;
1264
1265         unp = sotounpcb(so);
1266
1267         UNP_LINK_RLOCK();
1268         if ((unp2 = unp->unp_conn) == NULL) {
1269                 UNP_LINK_RUNLOCK();
1270                 for (int i = 0; i < count; i++)
1271                         m = m_free(m);
1272                 return (ECONNRESET);
1273         }
1274         UNP_PCB_LOCK(unp2);
1275         so2 = unp2->unp_socket;
1276
1277         SOCKBUF_LOCK(&so2->so_rcv);
1278         if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1279                 sorwakeup_locked(so2);
1280         else
1281                 SOCKBUF_UNLOCK(&so2->so_rcv);
1282
1283         UNP_PCB_UNLOCK(unp2);
1284         UNP_LINK_RUNLOCK();
1285
1286         return (error);
1287 }
1288
1289 static int
1290 uipc_sense(struct socket *so, struct stat *sb)
1291 {
1292         struct unpcb *unp;
1293
1294         unp = sotounpcb(so);
1295         KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1296
1297         sb->st_blksize = so->so_snd.sb_hiwat;
1298         UNP_PCB_LOCK(unp);
1299         sb->st_dev = NODEV;
1300         if (unp->unp_ino == 0)
1301                 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1302         sb->st_ino = unp->unp_ino;
1303         UNP_PCB_UNLOCK(unp);
1304         return (0);
1305 }
1306
1307 static int
1308 uipc_shutdown(struct socket *so)
1309 {
1310         struct unpcb *unp;
1311
1312         unp = sotounpcb(so);
1313         KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1314
1315         UNP_PCB_LOCK(unp);
1316         socantsendmore(so);
1317         unp_shutdown(unp);
1318         UNP_PCB_UNLOCK(unp);
1319         return (0);
1320 }
1321
1322 static int
1323 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1324 {
1325         struct unpcb *unp;
1326         const struct sockaddr *sa;
1327
1328         unp = sotounpcb(so);
1329         KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1330
1331         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1332         UNP_PCB_LOCK(unp);
1333         if (unp->unp_addr != NULL)
1334                 sa = (struct sockaddr *) unp->unp_addr;
1335         else
1336                 sa = &sun_noname;
1337         bcopy(sa, *nam, sa->sa_len);
1338         UNP_PCB_UNLOCK(unp);
1339         return (0);
1340 }
1341
1342 static struct pr_usrreqs uipc_usrreqs_dgram = {
1343         .pru_abort =            uipc_abort,
1344         .pru_accept =           uipc_accept,
1345         .pru_attach =           uipc_attach,
1346         .pru_bind =             uipc_bind,
1347         .pru_bindat =           uipc_bindat,
1348         .pru_connect =          uipc_connect,
1349         .pru_connectat =        uipc_connectat,
1350         .pru_connect2 =         uipc_connect2,
1351         .pru_detach =           uipc_detach,
1352         .pru_disconnect =       uipc_disconnect,
1353         .pru_listen =           uipc_listen,
1354         .pru_peeraddr =         uipc_peeraddr,
1355         .pru_rcvd =             uipc_rcvd,
1356         .pru_send =             uipc_send,
1357         .pru_sense =            uipc_sense,
1358         .pru_shutdown =         uipc_shutdown,
1359         .pru_sockaddr =         uipc_sockaddr,
1360         .pru_soreceive =        soreceive_dgram,
1361         .pru_close =            uipc_close,
1362 };
1363
1364 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1365         .pru_abort =            uipc_abort,
1366         .pru_accept =           uipc_accept,
1367         .pru_attach =           uipc_attach,
1368         .pru_bind =             uipc_bind,
1369         .pru_bindat =           uipc_bindat,
1370         .pru_connect =          uipc_connect,
1371         .pru_connectat =        uipc_connectat,
1372         .pru_connect2 =         uipc_connect2,
1373         .pru_detach =           uipc_detach,
1374         .pru_disconnect =       uipc_disconnect,
1375         .pru_listen =           uipc_listen,
1376         .pru_peeraddr =         uipc_peeraddr,
1377         .pru_rcvd =             uipc_rcvd,
1378         .pru_send =             uipc_send,
1379         .pru_sense =            uipc_sense,
1380         .pru_shutdown =         uipc_shutdown,
1381         .pru_sockaddr =         uipc_sockaddr,
1382         .pru_soreceive =        soreceive_generic,      /* XXX: or...? */
1383         .pru_close =            uipc_close,
1384 };
1385
1386 static struct pr_usrreqs uipc_usrreqs_stream = {
1387         .pru_abort =            uipc_abort,
1388         .pru_accept =           uipc_accept,
1389         .pru_attach =           uipc_attach,
1390         .pru_bind =             uipc_bind,
1391         .pru_bindat =           uipc_bindat,
1392         .pru_connect =          uipc_connect,
1393         .pru_connectat =        uipc_connectat,
1394         .pru_connect2 =         uipc_connect2,
1395         .pru_detach =           uipc_detach,
1396         .pru_disconnect =       uipc_disconnect,
1397         .pru_listen =           uipc_listen,
1398         .pru_peeraddr =         uipc_peeraddr,
1399         .pru_rcvd =             uipc_rcvd,
1400         .pru_send =             uipc_send,
1401         .pru_ready =            uipc_ready,
1402         .pru_sense =            uipc_sense,
1403         .pru_shutdown =         uipc_shutdown,
1404         .pru_sockaddr =         uipc_sockaddr,
1405         .pru_soreceive =        soreceive_generic,
1406         .pru_close =            uipc_close,
1407 };
1408
1409 static int
1410 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1411 {
1412         struct unpcb *unp;
1413         struct xucred xu;
1414         int error, optval;
1415
1416         if (sopt->sopt_level != 0)
1417                 return (EINVAL);
1418
1419         unp = sotounpcb(so);
1420         KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1421         error = 0;
1422         switch (sopt->sopt_dir) {
1423         case SOPT_GET:
1424                 switch (sopt->sopt_name) {
1425                 case LOCAL_PEERCRED:
1426                         UNP_PCB_LOCK(unp);
1427                         if (unp->unp_flags & UNP_HAVEPC)
1428                                 xu = unp->unp_peercred;
1429                         else {
1430                                 if (so->so_type == SOCK_STREAM)
1431                                         error = ENOTCONN;
1432                                 else
1433                                         error = EINVAL;
1434                         }
1435                         UNP_PCB_UNLOCK(unp);
1436                         if (error == 0)
1437                                 error = sooptcopyout(sopt, &xu, sizeof(xu));
1438                         break;
1439
1440                 case LOCAL_CREDS:
1441                         /* Unlocked read. */
1442                         optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1443                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1444                         break;
1445
1446                 case LOCAL_CONNWAIT:
1447                         /* Unlocked read. */
1448                         optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1449                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1450                         break;
1451
1452                 default:
1453                         error = EOPNOTSUPP;
1454                         break;
1455                 }
1456                 break;
1457
1458         case SOPT_SET:
1459                 switch (sopt->sopt_name) {
1460                 case LOCAL_CREDS:
1461                 case LOCAL_CONNWAIT:
1462                         error = sooptcopyin(sopt, &optval, sizeof(optval),
1463                                             sizeof(optval));
1464                         if (error)
1465                                 break;
1466
1467 #define OPTSET(bit) do {                                                \
1468         UNP_PCB_LOCK(unp);                                              \
1469         if (optval)                                                     \
1470                 unp->unp_flags |= bit;                                  \
1471         else                                                            \
1472                 unp->unp_flags &= ~bit;                                 \
1473         UNP_PCB_UNLOCK(unp);                                            \
1474 } while (0)
1475
1476                         switch (sopt->sopt_name) {
1477                         case LOCAL_CREDS:
1478                                 OPTSET(UNP_WANTCRED);
1479                                 break;
1480
1481                         case LOCAL_CONNWAIT:
1482                                 OPTSET(UNP_CONNWAIT);
1483                                 break;
1484
1485                         default:
1486                                 break;
1487                         }
1488                         break;
1489 #undef  OPTSET
1490                 default:
1491                         error = ENOPROTOOPT;
1492                         break;
1493                 }
1494                 break;
1495
1496         default:
1497                 error = EOPNOTSUPP;
1498                 break;
1499         }
1500         return (error);
1501 }
1502
1503 static int
1504 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1505 {
1506
1507         return (unp_connectat(AT_FDCWD, so, nam, td));
1508 }
1509
1510 static int
1511 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1512     struct thread *td)
1513 {
1514         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1515         struct vnode *vp;
1516         struct socket *so2;
1517         struct unpcb *unp, *unp2, *unp3;
1518         struct nameidata nd;
1519         char buf[SOCK_MAXADDRLEN];
1520         struct sockaddr *sa;
1521         cap_rights_t rights;
1522         int error, len, freed;
1523         struct mtx *vplock;
1524
1525         if (nam->sa_family != AF_UNIX)
1526                 return (EAFNOSUPPORT);
1527         if (nam->sa_len > sizeof(struct sockaddr_un))
1528                 return (EINVAL);
1529         len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1530         if (len <= 0)
1531                 return (EINVAL);
1532         bcopy(soun->sun_path, buf, len);
1533         buf[len] = 0;
1534
1535         unp = sotounpcb(so);
1536         UNP_PCB_LOCK(unp);
1537         if (unp->unp_flags & UNP_CONNECTING) {
1538                 UNP_PCB_UNLOCK(unp);
1539                 return (EALREADY);
1540         }
1541         unp->unp_flags |= UNP_CONNECTING;
1542         UNP_PCB_UNLOCK(unp);
1543
1544         sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1545         NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1546             UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1547         error = namei(&nd);
1548         if (error)
1549                 vp = NULL;
1550         else
1551                 vp = nd.ni_vp;
1552         ASSERT_VOP_LOCKED(vp, "unp_connect");
1553         NDFREE(&nd, NDF_ONLY_PNBUF);
1554         if (error)
1555                 goto bad;
1556
1557         if (vp->v_type != VSOCK) {
1558                 error = ENOTSOCK;
1559                 goto bad;
1560         }
1561 #ifdef MAC
1562         error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1563         if (error)
1564                 goto bad;
1565 #endif
1566         error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1567         if (error)
1568                 goto bad;
1569
1570         unp = sotounpcb(so);
1571         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1572
1573         vplock = mtx_pool_find(mtxpool_sleep, vp);
1574         mtx_lock(vplock);
1575         VOP_UNP_CONNECT(vp, &unp2);
1576         if (unp2 == NULL) {
1577                 error = ECONNREFUSED;
1578                 goto bad2;
1579         }
1580         so2 = unp2->unp_socket;
1581         if (so->so_type != so2->so_type) {
1582                 error = EPROTOTYPE;
1583                 goto bad2;
1584         }
1585         if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1586                 if (so2->so_options & SO_ACCEPTCONN) {
1587                         CURVNET_SET(so2->so_vnet);
1588                         so2 = sonewconn(so2, 0);
1589                         CURVNET_RESTORE();
1590                 } else
1591                         so2 = NULL;
1592                 if (so2 == NULL) {
1593                         error = ECONNREFUSED;
1594                         goto bad2;
1595                 }
1596                 unp3 = sotounpcb(so2);
1597                 unp_pcb_lock2(unp2, unp3);
1598                 if (unp2->unp_addr != NULL) {
1599                         bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1600                         unp3->unp_addr = (struct sockaddr_un *) sa;
1601                         sa = NULL;
1602                 }
1603
1604                 /*
1605                  * The connector's (client's) credentials are copied from its
1606                  * process structure at the time of connect() (which is now).
1607                  */
1608                 cru2x(td->td_ucred, &unp3->unp_peercred);
1609                 unp3->unp_flags |= UNP_HAVEPC;
1610
1611                 /*
1612                  * The receiver's (server's) credentials are copied from the
1613                  * unp_peercred member of socket on which the former called
1614                  * listen(); uipc_listen() cached that process's credentials
1615                  * at that time so we can use them now.
1616                  */
1617                 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1618                     sizeof(unp->unp_peercred));
1619                 unp->unp_flags |= UNP_HAVEPC;
1620                 if (unp2->unp_flags & UNP_WANTCRED)
1621                         unp3->unp_flags |= UNP_WANTCRED;
1622                 UNP_PCB_UNLOCK(unp2);
1623                 unp2 = unp3;
1624                 unp_pcb_owned_lock2(unp2, unp, freed);
1625                 if (__predict_false(freed)) {
1626                         UNP_PCB_UNLOCK(unp2);
1627                         error = ECONNREFUSED;
1628                         goto bad2;
1629                 }
1630 #ifdef MAC
1631                 mac_socketpeer_set_from_socket(so, so2);
1632                 mac_socketpeer_set_from_socket(so2, so);
1633 #endif
1634         } else {
1635                 if (unp == unp2)
1636                         UNP_PCB_LOCK(unp);
1637                 else
1638                         unp_pcb_lock2(unp, unp2);
1639         }
1640         KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 &&
1641             sotounpcb(so2) == unp2,
1642             ("%s: unp2 %p so2 %p", __func__, unp2, so2));
1643         error = unp_connect2(so, so2, PRU_CONNECT);
1644         if (unp != unp2)
1645                 UNP_PCB_UNLOCK(unp2);
1646         UNP_PCB_UNLOCK(unp);
1647 bad2:
1648         mtx_unlock(vplock);
1649 bad:
1650         if (vp != NULL) {
1651                 vput(vp);
1652         }
1653         free(sa, M_SONAME);
1654         UNP_PCB_LOCK(unp);
1655         unp->unp_flags &= ~UNP_CONNECTING;
1656         UNP_PCB_UNLOCK(unp);
1657         return (error);
1658 }
1659
1660 static int
1661 unp_connect2(struct socket *so, struct socket *so2, int req)
1662 {
1663         struct unpcb *unp;
1664         struct unpcb *unp2;
1665
1666         unp = sotounpcb(so);
1667         KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1668         unp2 = sotounpcb(so2);
1669         KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1670
1671         UNP_PCB_LOCK_ASSERT(unp);
1672         UNP_PCB_LOCK_ASSERT(unp2);
1673
1674         if (so2->so_type != so->so_type)
1675                 return (EPROTOTYPE);
1676         unp2->unp_flags &= ~UNP_NASCENT;
1677         unp->unp_conn = unp2;
1678         unp_pcb_hold(unp2);
1679         unp_pcb_hold(unp);
1680         switch (so->so_type) {
1681         case SOCK_DGRAM:
1682                 UNP_REF_LIST_LOCK();
1683                 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1684                 UNP_REF_LIST_UNLOCK();
1685                 soisconnected(so);
1686                 break;
1687
1688         case SOCK_STREAM:
1689         case SOCK_SEQPACKET:
1690                 unp2->unp_conn = unp;
1691                 if (req == PRU_CONNECT &&
1692                     ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1693                         soisconnecting(so);
1694                 else
1695                         soisconnected(so);
1696                 soisconnected(so2);
1697                 break;
1698
1699         default:
1700                 panic("unp_connect2");
1701         }
1702         return (0);
1703 }
1704
1705 static void
1706 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1707 {
1708         struct socket *so, *so2;
1709         int freed __unused;
1710
1711         KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1712
1713         UNP_PCB_LOCK_ASSERT(unp);
1714         UNP_PCB_LOCK_ASSERT(unp2);
1715
1716         if (unp->unp_conn == NULL && unp2->unp_conn == NULL)
1717                 return;
1718
1719         MPASS(unp->unp_conn == unp2);
1720         unp->unp_conn = NULL;
1721         so = unp->unp_socket;
1722         so2 = unp2->unp_socket;
1723         switch (unp->unp_socket->so_type) {
1724         case SOCK_DGRAM:
1725                 UNP_REF_LIST_LOCK();
1726                 LIST_REMOVE(unp, unp_reflink);
1727                 UNP_REF_LIST_UNLOCK();
1728                 if (so) {
1729                         SOCK_LOCK(so);
1730                         so->so_state &= ~SS_ISCONNECTED;
1731                         SOCK_UNLOCK(so);
1732                 }
1733                 break;
1734
1735         case SOCK_STREAM:
1736         case SOCK_SEQPACKET:
1737                 if (so)
1738                         soisdisconnected(so);
1739                 MPASS(unp2->unp_conn == unp);
1740                 unp2->unp_conn = NULL;
1741                 if (so2)
1742                         soisdisconnected(so2);
1743                 break;
1744         }
1745         freed = unp_pcb_rele(unp);
1746         MPASS(freed == 0);
1747         freed = unp_pcb_rele(unp2);
1748         MPASS(freed == 0);
1749 }
1750
1751 /*
1752  * unp_pcblist() walks the global list of struct unpcb's to generate a
1753  * pointer list, bumping the refcount on each unpcb.  It then copies them out
1754  * sequentially, validating the generation number on each to see if it has
1755  * been detached.  All of this is necessary because copyout() may sleep on
1756  * disk I/O.
1757  */
1758 static int
1759 unp_pcblist(SYSCTL_HANDLER_ARGS)
1760 {
1761         struct unpcb *unp, **unp_list;
1762         unp_gen_t gencnt;
1763         struct xunpgen *xug;
1764         struct unp_head *head;
1765         struct xunpcb *xu;
1766         u_int i;
1767         int error, freeunp, n;
1768
1769         switch ((intptr_t)arg1) {
1770         case SOCK_STREAM:
1771                 head = &unp_shead;
1772                 break;
1773
1774         case SOCK_DGRAM:
1775                 head = &unp_dhead;
1776                 break;
1777
1778         case SOCK_SEQPACKET:
1779                 head = &unp_sphead;
1780                 break;
1781
1782         default:
1783                 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1784         }
1785
1786         /*
1787          * The process of preparing the PCB list is too time-consuming and
1788          * resource-intensive to repeat twice on every request.
1789          */
1790         if (req->oldptr == NULL) {
1791                 n = unp_count;
1792                 req->oldidx = 2 * (sizeof *xug)
1793                         + (n + n/8) * sizeof(struct xunpcb);
1794                 return (0);
1795         }
1796
1797         if (req->newptr != NULL)
1798                 return (EPERM);
1799
1800         /*
1801          * OK, now we're committed to doing something.
1802          */
1803         xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1804         UNP_LINK_RLOCK();
1805         gencnt = unp_gencnt;
1806         n = unp_count;
1807         UNP_LINK_RUNLOCK();
1808
1809         xug->xug_len = sizeof *xug;
1810         xug->xug_count = n;
1811         xug->xug_gen = gencnt;
1812         xug->xug_sogen = so_gencnt;
1813         error = SYSCTL_OUT(req, xug, sizeof *xug);
1814         if (error) {
1815                 free(xug, M_TEMP);
1816                 return (error);
1817         }
1818
1819         unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1820
1821         UNP_LINK_RLOCK();
1822         for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1823              unp = LIST_NEXT(unp, unp_link)) {
1824                 UNP_PCB_LOCK(unp);
1825                 if (unp->unp_gencnt <= gencnt) {
1826                         if (cr_cansee(req->td->td_ucred,
1827                             unp->unp_socket->so_cred)) {
1828                                 UNP_PCB_UNLOCK(unp);
1829                                 continue;
1830                         }
1831                         unp_list[i++] = unp;
1832                         unp_pcb_hold(unp);
1833                 }
1834                 UNP_PCB_UNLOCK(unp);
1835         }
1836         UNP_LINK_RUNLOCK();
1837         n = i;                  /* In case we lost some during malloc. */
1838
1839         error = 0;
1840         xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1841         for (i = 0; i < n; i++) {
1842                 unp = unp_list[i];
1843                 UNP_PCB_LOCK(unp);
1844                 freeunp = unp_pcb_rele(unp);
1845
1846                 if (freeunp == 0 && unp->unp_gencnt <= gencnt) {
1847                         xu->xu_len = sizeof *xu;
1848                         xu->xu_unpp = unp;
1849                         /*
1850                          * XXX - need more locking here to protect against
1851                          * connect/disconnect races for SMP.
1852                          */
1853                         if (unp->unp_addr != NULL)
1854                                 bcopy(unp->unp_addr, &xu->xu_addr,
1855                                       unp->unp_addr->sun_len);
1856                         else
1857                                 bzero(&xu->xu_addr, sizeof(xu->xu_addr));
1858                         if (unp->unp_conn != NULL &&
1859                             unp->unp_conn->unp_addr != NULL)
1860                                 bcopy(unp->unp_conn->unp_addr,
1861                                       &xu->xu_caddr,
1862                                       unp->unp_conn->unp_addr->sun_len);
1863                         else
1864                                 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr));
1865                         xu->unp_vnode = unp->unp_vnode;
1866                         xu->unp_conn = unp->unp_conn;
1867                         xu->xu_firstref = LIST_FIRST(&unp->unp_refs);
1868                         xu->xu_nextref = LIST_NEXT(unp, unp_reflink);
1869                         xu->unp_gencnt = unp->unp_gencnt;
1870                         sotoxsocket(unp->unp_socket, &xu->xu_socket);
1871                         UNP_PCB_UNLOCK(unp);
1872                         error = SYSCTL_OUT(req, xu, sizeof *xu);
1873                 } else  if (freeunp == 0)
1874                         UNP_PCB_UNLOCK(unp);
1875         }
1876         free(xu, M_TEMP);
1877         if (!error) {
1878                 /*
1879                  * Give the user an updated idea of our state.  If the
1880                  * generation differs from what we told her before, she knows
1881                  * that something happened while we were processing this
1882                  * request, and it might be necessary to retry.
1883                  */
1884                 xug->xug_gen = unp_gencnt;
1885                 xug->xug_sogen = so_gencnt;
1886                 xug->xug_count = unp_count;
1887                 error = SYSCTL_OUT(req, xug, sizeof *xug);
1888         }
1889         free(unp_list, M_TEMP);
1890         free(xug, M_TEMP);
1891         return (error);
1892 }
1893
1894 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1895     (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1896     "List of active local datagram sockets");
1897 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1898     (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1899     "List of active local stream sockets");
1900 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1901     CTLTYPE_OPAQUE | CTLFLAG_RD,
1902     (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1903     "List of active local seqpacket sockets");
1904
1905 static void
1906 unp_shutdown(struct unpcb *unp)
1907 {
1908         struct unpcb *unp2;
1909         struct socket *so;
1910
1911         UNP_PCB_LOCK_ASSERT(unp);
1912
1913         unp2 = unp->unp_conn;
1914         if ((unp->unp_socket->so_type == SOCK_STREAM ||
1915             (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1916                 so = unp2->unp_socket;
1917                 if (so != NULL)
1918                         socantrcvmore(so);
1919         }
1920 }
1921
1922 static void
1923 unp_drop(struct unpcb *unp)
1924 {
1925         struct socket *so = unp->unp_socket;
1926         struct unpcb *unp2;
1927         int freed;
1928
1929         /*
1930          * Regardless of whether the socket's peer dropped the connection
1931          * with this socket by aborting or disconnecting, POSIX requires
1932          * that ECONNRESET is returned.
1933          */
1934         /* acquire a reference so that unp isn't freed from underneath us */
1935
1936         UNP_PCB_LOCK(unp);
1937         if (so)
1938                 so->so_error = ECONNRESET;
1939         unp2 = unp->unp_conn;
1940         if (unp2 == unp) {
1941                 unp_disconnect(unp, unp2);
1942         } else if (unp2 != NULL) {
1943                 unp_pcb_hold(unp2);
1944                 unp_pcb_owned_lock2(unp, unp2, freed);
1945                 unp_disconnect(unp, unp2);
1946                 if (unp_pcb_rele(unp2) == 0)
1947                         UNP_PCB_UNLOCK(unp2);
1948         }
1949         if (unp_pcb_rele(unp) == 0)
1950                 UNP_PCB_UNLOCK(unp);
1951 }
1952
1953 static void
1954 unp_freerights(struct filedescent **fdep, int fdcount)
1955 {
1956         struct file *fp;
1957         int i;
1958
1959         KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1960
1961         for (i = 0; i < fdcount; i++) {
1962                 fp = fdep[i]->fde_file;
1963                 filecaps_free(&fdep[i]->fde_caps);
1964                 unp_discard(fp);
1965         }
1966         free(fdep[0], M_FILECAPS);
1967 }
1968
1969 static int
1970 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1971 {
1972         struct thread *td = curthread;          /* XXX */
1973         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1974         int i;
1975         int *fdp;
1976         struct filedesc *fdesc = td->td_proc->p_fd;
1977         struct filedescent **fdep;
1978         void *data;
1979         socklen_t clen = control->m_len, datalen;
1980         int error, newfds;
1981         u_int newlen;
1982
1983         UNP_LINK_UNLOCK_ASSERT();
1984
1985         error = 0;
1986         if (controlp != NULL) /* controlp == NULL => free control messages */
1987                 *controlp = NULL;
1988         while (cm != NULL) {
1989                 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1990                         error = EINVAL;
1991                         break;
1992                 }
1993                 data = CMSG_DATA(cm);
1994                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1995                 if (cm->cmsg_level == SOL_SOCKET
1996                     && cm->cmsg_type == SCM_RIGHTS) {
1997                         newfds = datalen / sizeof(*fdep);
1998                         if (newfds == 0)
1999                                 goto next;
2000                         fdep = data;
2001
2002                         /* If we're not outputting the descriptors free them. */
2003                         if (error || controlp == NULL) {
2004                                 unp_freerights(fdep, newfds);
2005                                 goto next;
2006                         }
2007                         FILEDESC_XLOCK(fdesc);
2008
2009                         /*
2010                          * Now change each pointer to an fd in the global
2011                          * table to an integer that is the index to the local
2012                          * fd table entry that we set up to point to the
2013                          * global one we are transferring.
2014                          */
2015                         newlen = newfds * sizeof(int);
2016                         *controlp = sbcreatecontrol(NULL, newlen,
2017                             SCM_RIGHTS, SOL_SOCKET);
2018                         if (*controlp == NULL) {
2019                                 FILEDESC_XUNLOCK(fdesc);
2020                                 error = E2BIG;
2021                                 unp_freerights(fdep, newfds);
2022                                 goto next;
2023                         }
2024
2025                         fdp = (int *)
2026                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2027                         if (fdallocn(td, 0, fdp, newfds) != 0) {
2028                                 FILEDESC_XUNLOCK(fdesc);
2029                                 error = EMSGSIZE;
2030                                 unp_freerights(fdep, newfds);
2031                                 m_freem(*controlp);
2032                                 *controlp = NULL;
2033                                 goto next;
2034                         }
2035                         for (i = 0; i < newfds; i++, fdp++) {
2036                                 _finstall(fdesc, fdep[i]->fde_file, *fdp,
2037                                     (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0,
2038                                     &fdep[i]->fde_caps);
2039                                 unp_externalize_fp(fdep[i]->fde_file);
2040                         }
2041                         FILEDESC_XUNLOCK(fdesc);
2042                         free(fdep[0], M_FILECAPS);
2043                 } else {
2044                         /* We can just copy anything else across. */
2045                         if (error || controlp == NULL)
2046                                 goto next;
2047                         *controlp = sbcreatecontrol(NULL, datalen,
2048                             cm->cmsg_type, cm->cmsg_level);
2049                         if (*controlp == NULL) {
2050                                 error = ENOBUFS;
2051                                 goto next;
2052                         }
2053                         bcopy(data,
2054                             CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
2055                             datalen);
2056                 }
2057                 controlp = &(*controlp)->m_next;
2058
2059 next:
2060                 if (CMSG_SPACE(datalen) < clen) {
2061                         clen -= CMSG_SPACE(datalen);
2062                         cm = (struct cmsghdr *)
2063                             ((caddr_t)cm + CMSG_SPACE(datalen));
2064                 } else {
2065                         clen = 0;
2066                         cm = NULL;
2067                 }
2068         }
2069
2070         m_freem(control);
2071         return (error);
2072 }
2073
2074 static void
2075 unp_zone_change(void *tag)
2076 {
2077
2078         uma_zone_set_max(unp_zone, maxsockets);
2079 }
2080
2081 static void
2082 unp_init(void)
2083 {
2084
2085 #ifdef VIMAGE
2086         if (!IS_DEFAULT_VNET(curvnet))
2087                 return;
2088 #endif
2089         unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
2090             NULL, NULL, UMA_ALIGN_CACHE, 0);
2091         if (unp_zone == NULL)
2092                 panic("unp_init");
2093         uma_zone_set_max(unp_zone, maxsockets);
2094         uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
2095         EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
2096             NULL, EVENTHANDLER_PRI_ANY);
2097         LIST_INIT(&unp_dhead);
2098         LIST_INIT(&unp_shead);
2099         LIST_INIT(&unp_sphead);
2100         SLIST_INIT(&unp_defers);
2101         TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
2102         TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
2103         UNP_LINK_LOCK_INIT();
2104         UNP_DEFERRED_LOCK_INIT();
2105 }
2106
2107 static int
2108 unp_internalize(struct mbuf **controlp, struct thread *td)
2109 {
2110         struct mbuf *control = *controlp;
2111         struct proc *p = td->td_proc;
2112         struct filedesc *fdesc = p->p_fd;
2113         struct bintime *bt;
2114         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
2115         struct cmsgcred *cmcred;
2116         struct filedescent *fde, **fdep, *fdev;
2117         struct file *fp;
2118         struct timeval *tv;
2119         struct timespec *ts;
2120         int i, *fdp;
2121         void *data;
2122         socklen_t clen = control->m_len, datalen;
2123         int error, oldfds;
2124         u_int newlen;
2125
2126         UNP_LINK_UNLOCK_ASSERT();
2127
2128         error = 0;
2129         *controlp = NULL;
2130         while (cm != NULL) {
2131                 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
2132                     || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
2133                         error = EINVAL;
2134                         goto out;
2135                 }
2136                 data = CMSG_DATA(cm);
2137                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
2138
2139                 switch (cm->cmsg_type) {
2140                 /*
2141                  * Fill in credential information.
2142                  */
2143                 case SCM_CREDS:
2144                         *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
2145                             SCM_CREDS, SOL_SOCKET);
2146                         if (*controlp == NULL) {
2147                                 error = ENOBUFS;
2148                                 goto out;
2149                         }
2150                         cmcred = (struct cmsgcred *)
2151                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2152                         cmcred->cmcred_pid = p->p_pid;
2153                         cmcred->cmcred_uid = td->td_ucred->cr_ruid;
2154                         cmcred->cmcred_gid = td->td_ucred->cr_rgid;
2155                         cmcred->cmcred_euid = td->td_ucred->cr_uid;
2156                         cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
2157                             CMGROUP_MAX);
2158                         for (i = 0; i < cmcred->cmcred_ngroups; i++)
2159                                 cmcred->cmcred_groups[i] =
2160                                     td->td_ucred->cr_groups[i];
2161                         break;
2162
2163                 case SCM_RIGHTS:
2164                         oldfds = datalen / sizeof (int);
2165                         if (oldfds == 0)
2166                                 break;
2167                         /*
2168                          * Check that all the FDs passed in refer to legal
2169                          * files.  If not, reject the entire operation.
2170                          */
2171                         fdp = data;
2172                         FILEDESC_SLOCK(fdesc);
2173                         for (i = 0; i < oldfds; i++, fdp++) {
2174                                 fp = fget_locked(fdesc, *fdp);
2175                                 if (fp == NULL) {
2176                                         FILEDESC_SUNLOCK(fdesc);
2177                                         error = EBADF;
2178                                         goto out;
2179                                 }
2180                                 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
2181                                         FILEDESC_SUNLOCK(fdesc);
2182                                         error = EOPNOTSUPP;
2183                                         goto out;
2184                                 }
2185
2186                         }
2187
2188                         /*
2189                          * Now replace the integer FDs with pointers to the
2190                          * file structure and capability rights.
2191                          */
2192                         newlen = oldfds * sizeof(fdep[0]);
2193                         *controlp = sbcreatecontrol(NULL, newlen,
2194                             SCM_RIGHTS, SOL_SOCKET);
2195                         if (*controlp == NULL) {
2196                                 FILEDESC_SUNLOCK(fdesc);
2197                                 error = E2BIG;
2198                                 goto out;
2199                         }
2200                         fdp = data;
2201                         fdep = (struct filedescent **)
2202                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2203                         fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
2204                             M_WAITOK);
2205                         for (i = 0; i < oldfds; i++, fdev++, fdp++) {
2206                                 fde = &fdesc->fd_ofiles[*fdp];
2207                                 fdep[i] = fdev;
2208                                 fdep[i]->fde_file = fde->fde_file;
2209                                 filecaps_copy(&fde->fde_caps,
2210                                     &fdep[i]->fde_caps, true);
2211                                 unp_internalize_fp(fdep[i]->fde_file);
2212                         }
2213                         FILEDESC_SUNLOCK(fdesc);
2214                         break;
2215
2216                 case SCM_TIMESTAMP:
2217                         *controlp = sbcreatecontrol(NULL, sizeof(*tv),
2218                             SCM_TIMESTAMP, SOL_SOCKET);
2219                         if (*controlp == NULL) {
2220                                 error = ENOBUFS;
2221                                 goto out;
2222                         }
2223                         tv = (struct timeval *)
2224                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2225                         microtime(tv);
2226                         break;
2227
2228                 case SCM_BINTIME:
2229                         *controlp = sbcreatecontrol(NULL, sizeof(*bt),
2230                             SCM_BINTIME, SOL_SOCKET);
2231                         if (*controlp == NULL) {
2232                                 error = ENOBUFS;
2233                                 goto out;
2234                         }
2235                         bt = (struct bintime *)
2236                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2237                         bintime(bt);
2238                         break;
2239
2240                 case SCM_REALTIME:
2241                         *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2242                             SCM_REALTIME, SOL_SOCKET);
2243                         if (*controlp == NULL) {
2244                                 error = ENOBUFS;
2245                                 goto out;
2246                         }
2247                         ts = (struct timespec *)
2248                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2249                         nanotime(ts);
2250                         break;
2251
2252                 case SCM_MONOTONIC:
2253                         *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2254                             SCM_MONOTONIC, SOL_SOCKET);
2255                         if (*controlp == NULL) {
2256                                 error = ENOBUFS;
2257                                 goto out;
2258                         }
2259                         ts = (struct timespec *)
2260                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2261                         nanouptime(ts);
2262                         break;
2263
2264                 default:
2265                         error = EINVAL;
2266                         goto out;
2267                 }
2268
2269                 controlp = &(*controlp)->m_next;
2270                 if (CMSG_SPACE(datalen) < clen) {
2271                         clen -= CMSG_SPACE(datalen);
2272                         cm = (struct cmsghdr *)
2273                             ((caddr_t)cm + CMSG_SPACE(datalen));
2274                 } else {
2275                         clen = 0;
2276                         cm = NULL;
2277                 }
2278         }
2279
2280 out:
2281         m_freem(control);
2282         return (error);
2283 }
2284
2285 static struct mbuf *
2286 unp_addsockcred(struct thread *td, struct mbuf *control)
2287 {
2288         struct mbuf *m, *n, *n_prev;
2289         struct sockcred *sc;
2290         const struct cmsghdr *cm;
2291         int ngroups;
2292         int i;
2293
2294         ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2295         m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2296         if (m == NULL)
2297                 return (control);
2298
2299         sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2300         sc->sc_uid = td->td_ucred->cr_ruid;
2301         sc->sc_euid = td->td_ucred->cr_uid;
2302         sc->sc_gid = td->td_ucred->cr_rgid;
2303         sc->sc_egid = td->td_ucred->cr_gid;
2304         sc->sc_ngroups = ngroups;
2305         for (i = 0; i < sc->sc_ngroups; i++)
2306                 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2307
2308         /*
2309          * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2310          * created SCM_CREDS control message (struct sockcred) has another
2311          * format.
2312          */
2313         if (control != NULL)
2314                 for (n = control, n_prev = NULL; n != NULL;) {
2315                         cm = mtod(n, struct cmsghdr *);
2316                         if (cm->cmsg_level == SOL_SOCKET &&
2317                             cm->cmsg_type == SCM_CREDS) {
2318                                 if (n_prev == NULL)
2319                                         control = n->m_next;
2320                                 else
2321                                         n_prev->m_next = n->m_next;
2322                                 n = m_free(n);
2323                         } else {
2324                                 n_prev = n;
2325                                 n = n->m_next;
2326                         }
2327                 }
2328
2329         /* Prepend it to the head. */
2330         m->m_next = control;
2331         return (m);
2332 }
2333
2334 static struct unpcb *
2335 fptounp(struct file *fp)
2336 {
2337         struct socket *so;
2338
2339         if (fp->f_type != DTYPE_SOCKET)
2340                 return (NULL);
2341         if ((so = fp->f_data) == NULL)
2342                 return (NULL);
2343         if (so->so_proto->pr_domain != &localdomain)
2344                 return (NULL);
2345         return sotounpcb(so);
2346 }
2347
2348 static void
2349 unp_discard(struct file *fp)
2350 {
2351         struct unp_defer *dr;
2352
2353         if (unp_externalize_fp(fp)) {
2354                 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2355                 dr->ud_fp = fp;
2356                 UNP_DEFERRED_LOCK();
2357                 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2358                 UNP_DEFERRED_UNLOCK();
2359                 atomic_add_int(&unp_defers_count, 1);
2360                 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2361         } else
2362                 (void) closef(fp, (struct thread *)NULL);
2363 }
2364
2365 static void
2366 unp_process_defers(void *arg __unused, int pending)
2367 {
2368         struct unp_defer *dr;
2369         SLIST_HEAD(, unp_defer) drl;
2370         int count;
2371
2372         SLIST_INIT(&drl);
2373         for (;;) {
2374                 UNP_DEFERRED_LOCK();
2375                 if (SLIST_FIRST(&unp_defers) == NULL) {
2376                         UNP_DEFERRED_UNLOCK();
2377                         break;
2378                 }
2379                 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2380                 UNP_DEFERRED_UNLOCK();
2381                 count = 0;
2382                 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2383                         SLIST_REMOVE_HEAD(&drl, ud_link);
2384                         closef(dr->ud_fp, NULL);
2385                         free(dr, M_TEMP);
2386                         count++;
2387                 }
2388                 atomic_add_int(&unp_defers_count, -count);
2389         }
2390 }
2391
2392 static void
2393 unp_internalize_fp(struct file *fp)
2394 {
2395         struct unpcb *unp;
2396
2397         UNP_LINK_WLOCK();
2398         if ((unp = fptounp(fp)) != NULL) {
2399                 unp->unp_file = fp;
2400                 unp->unp_msgcount++;
2401         }
2402         fhold(fp);
2403         unp_rights++;
2404         UNP_LINK_WUNLOCK();
2405 }
2406
2407 static int
2408 unp_externalize_fp(struct file *fp)
2409 {
2410         struct unpcb *unp;
2411         int ret;
2412
2413         UNP_LINK_WLOCK();
2414         if ((unp = fptounp(fp)) != NULL) {
2415                 unp->unp_msgcount--;
2416                 ret = 1;
2417         } else
2418                 ret = 0;
2419         unp_rights--;
2420         UNP_LINK_WUNLOCK();
2421         return (ret);
2422 }
2423
2424 /*
2425  * unp_defer indicates whether additional work has been defered for a future
2426  * pass through unp_gc().  It is thread local and does not require explicit
2427  * synchronization.
2428  */
2429 static int      unp_marked;
2430 static int      unp_unreachable;
2431
2432 static void
2433 unp_accessable(struct filedescent **fdep, int fdcount)
2434 {
2435         struct unpcb *unp;
2436         struct file *fp;
2437         int i;
2438
2439         for (i = 0; i < fdcount; i++) {
2440                 fp = fdep[i]->fde_file;
2441                 if ((unp = fptounp(fp)) == NULL)
2442                         continue;
2443                 if (unp->unp_gcflag & UNPGC_REF)
2444                         continue;
2445                 unp->unp_gcflag &= ~UNPGC_DEAD;
2446                 unp->unp_gcflag |= UNPGC_REF;
2447                 unp_marked++;
2448         }
2449 }
2450
2451 static void
2452 unp_gc_process(struct unpcb *unp)
2453 {
2454         struct socket *so, *soa;
2455         struct file *fp;
2456
2457         /* Already processed. */
2458         if (unp->unp_gcflag & UNPGC_SCANNED)
2459                 return;
2460         fp = unp->unp_file;
2461
2462         /*
2463          * Check for a socket potentially in a cycle.  It must be in a
2464          * queue as indicated by msgcount, and this must equal the file
2465          * reference count.  Note that when msgcount is 0 the file is NULL.
2466          */
2467         if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2468             unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2469                 unp->unp_gcflag |= UNPGC_DEAD;
2470                 unp_unreachable++;
2471                 return;
2472         }
2473
2474         so = unp->unp_socket;
2475         SOCK_LOCK(so);
2476         if (SOLISTENING(so)) {
2477                 /*
2478                  * Mark all sockets in our accept queue.
2479                  */
2480                 TAILQ_FOREACH(soa, &so->sol_comp, so_list) {
2481                         if (sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS)
2482                                 continue;
2483                         SOCKBUF_LOCK(&soa->so_rcv);
2484                         unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2485                         SOCKBUF_UNLOCK(&soa->so_rcv);
2486                 }
2487         } else {
2488                 /*
2489                  * Mark all sockets we reference with RIGHTS.
2490                  */
2491                 if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2492                         SOCKBUF_LOCK(&so->so_rcv);
2493                         unp_scan(so->so_rcv.sb_mb, unp_accessable);
2494                         SOCKBUF_UNLOCK(&so->so_rcv);
2495                 }
2496         }
2497         SOCK_UNLOCK(so);
2498         unp->unp_gcflag |= UNPGC_SCANNED;
2499 }
2500
2501 static int unp_recycled;
2502 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2503     "Number of unreachable sockets claimed by the garbage collector.");
2504
2505 static int unp_taskcount;
2506 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2507     "Number of times the garbage collector has run.");
2508
2509 static void
2510 unp_gc(__unused void *arg, int pending)
2511 {
2512         struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2513                                     NULL };
2514         struct unp_head **head;
2515         struct file *f, **unref;
2516         struct unpcb *unp;
2517         int i, total;
2518
2519         unp_taskcount++;
2520         UNP_LINK_RLOCK();
2521         /*
2522          * First clear all gc flags from previous runs, apart from
2523          * UNPGC_IGNORE_RIGHTS.
2524          */
2525         for (head = heads; *head != NULL; head++)
2526                 LIST_FOREACH(unp, *head, unp_link)
2527                         unp->unp_gcflag =
2528                             (unp->unp_gcflag & UNPGC_IGNORE_RIGHTS);
2529
2530         /*
2531          * Scan marking all reachable sockets with UNPGC_REF.  Once a socket
2532          * is reachable all of the sockets it references are reachable.
2533          * Stop the scan once we do a complete loop without discovering
2534          * a new reachable socket.
2535          */
2536         do {
2537                 unp_unreachable = 0;
2538                 unp_marked = 0;
2539                 for (head = heads; *head != NULL; head++)
2540                         LIST_FOREACH(unp, *head, unp_link)
2541                                 unp_gc_process(unp);
2542         } while (unp_marked);
2543         UNP_LINK_RUNLOCK();
2544         if (unp_unreachable == 0)
2545                 return;
2546
2547         /*
2548          * Allocate space for a local list of dead unpcbs.
2549          */
2550         unref = malloc(unp_unreachable * sizeof(struct file *),
2551             M_TEMP, M_WAITOK);
2552
2553         /*
2554          * Iterate looking for sockets which have been specifically marked
2555          * as as unreachable and store them locally.
2556          */
2557         UNP_LINK_RLOCK();
2558         for (total = 0, head = heads; *head != NULL; head++)
2559                 LIST_FOREACH(unp, *head, unp_link)
2560                         if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2561                                 f = unp->unp_file;
2562                                 if (unp->unp_msgcount == 0 || f == NULL ||
2563                                     f->f_count != unp->unp_msgcount)
2564                                         continue;
2565                                 unref[total++] = f;
2566                                 fhold(f);
2567                                 KASSERT(total <= unp_unreachable,
2568                                     ("unp_gc: incorrect unreachable count."));
2569                         }
2570         UNP_LINK_RUNLOCK();
2571
2572         /*
2573          * Now flush all sockets, free'ing rights.  This will free the
2574          * struct files associated with these sockets but leave each socket
2575          * with one remaining ref.
2576          */
2577         for (i = 0; i < total; i++) {
2578                 struct socket *so;
2579
2580                 so = unref[i]->f_data;
2581                 CURVNET_SET(so->so_vnet);
2582                 sorflush(so);
2583                 CURVNET_RESTORE();
2584         }
2585
2586         /*
2587          * And finally release the sockets so they can be reclaimed.
2588          */
2589         for (i = 0; i < total; i++)
2590                 fdrop(unref[i], NULL);
2591         unp_recycled += total;
2592         free(unref, M_TEMP);
2593 }
2594
2595 static void
2596 unp_dispose_mbuf(struct mbuf *m)
2597 {
2598
2599         if (m)
2600                 unp_scan(m, unp_freerights);
2601 }
2602
2603 /*
2604  * Synchronize against unp_gc, which can trip over data as we are freeing it.
2605  */
2606 static void
2607 unp_dispose(struct socket *so)
2608 {
2609         struct unpcb *unp;
2610
2611         unp = sotounpcb(so);
2612         UNP_LINK_WLOCK();
2613         unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2614         UNP_LINK_WUNLOCK();
2615         if (!SOLISTENING(so))
2616                 unp_dispose_mbuf(so->so_rcv.sb_mb);
2617 }
2618
2619 static void
2620 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2621 {
2622         struct mbuf *m;
2623         struct cmsghdr *cm;
2624         void *data;
2625         socklen_t clen, datalen;
2626
2627         while (m0 != NULL) {
2628                 for (m = m0; m; m = m->m_next) {
2629                         if (m->m_type != MT_CONTROL)
2630                                 continue;
2631
2632                         cm = mtod(m, struct cmsghdr *);
2633                         clen = m->m_len;
2634
2635                         while (cm != NULL) {
2636                                 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2637                                         break;
2638
2639                                 data = CMSG_DATA(cm);
2640                                 datalen = (caddr_t)cm + cm->cmsg_len
2641                                     - (caddr_t)data;
2642
2643                                 if (cm->cmsg_level == SOL_SOCKET &&
2644                                     cm->cmsg_type == SCM_RIGHTS) {
2645                                         (*op)(data, datalen /
2646                                             sizeof(struct filedescent *));
2647                                 }
2648
2649                                 if (CMSG_SPACE(datalen) < clen) {
2650                                         clen -= CMSG_SPACE(datalen);
2651                                         cm = (struct cmsghdr *)
2652                                             ((caddr_t)cm + CMSG_SPACE(datalen));
2653                                 } else {
2654                                         clen = 0;
2655                                         cm = NULL;
2656                                 }
2657                         }
2658                 }
2659                 m0 = m0->m_nextpkt;
2660         }
2661 }
2662
2663 /*
2664  * A helper function called by VFS before socket-type vnode reclamation.
2665  * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2666  * use count.
2667  */
2668 void
2669 vfs_unp_reclaim(struct vnode *vp)
2670 {
2671         struct unpcb *unp;
2672         int active;
2673         struct mtx *vplock;
2674
2675         ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2676         KASSERT(vp->v_type == VSOCK,
2677             ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2678
2679         active = 0;
2680         vplock = mtx_pool_find(mtxpool_sleep, vp);
2681         mtx_lock(vplock);
2682         VOP_UNP_CONNECT(vp, &unp);
2683         if (unp == NULL)
2684                 goto done;
2685         UNP_PCB_LOCK(unp);
2686         if (unp->unp_vnode == vp) {
2687                 VOP_UNP_DETACH(vp);
2688                 unp->unp_vnode = NULL;
2689                 active = 1;
2690         }
2691         UNP_PCB_UNLOCK(unp);
2692  done:
2693         mtx_unlock(vplock);
2694         if (active)
2695                 vunref(vp);
2696 }
2697
2698 #ifdef DDB
2699 static void
2700 db_print_indent(int indent)
2701 {
2702         int i;
2703
2704         for (i = 0; i < indent; i++)
2705                 db_printf(" ");
2706 }
2707
2708 static void
2709 db_print_unpflags(int unp_flags)
2710 {
2711         int comma;
2712
2713         comma = 0;
2714         if (unp_flags & UNP_HAVEPC) {
2715                 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2716                 comma = 1;
2717         }
2718         if (unp_flags & UNP_WANTCRED) {
2719                 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2720                 comma = 1;
2721         }
2722         if (unp_flags & UNP_CONNWAIT) {
2723                 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2724                 comma = 1;
2725         }
2726         if (unp_flags & UNP_CONNECTING) {
2727                 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2728                 comma = 1;
2729         }
2730         if (unp_flags & UNP_BINDING) {
2731                 db_printf("%sUNP_BINDING", comma ? ", " : "");
2732                 comma = 1;
2733         }
2734 }
2735
2736 static void
2737 db_print_xucred(int indent, struct xucred *xu)
2738 {
2739         int comma, i;
2740
2741         db_print_indent(indent);
2742         db_printf("cr_version: %u   cr_uid: %u   cr_ngroups: %d\n",
2743             xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2744         db_print_indent(indent);
2745         db_printf("cr_groups: ");
2746         comma = 0;
2747         for (i = 0; i < xu->cr_ngroups; i++) {
2748                 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2749                 comma = 1;
2750         }
2751         db_printf("\n");
2752 }
2753
2754 static void
2755 db_print_unprefs(int indent, struct unp_head *uh)
2756 {
2757         struct unpcb *unp;
2758         int counter;
2759
2760         counter = 0;
2761         LIST_FOREACH(unp, uh, unp_reflink) {
2762                 if (counter % 4 == 0)
2763                         db_print_indent(indent);
2764                 db_printf("%p  ", unp);
2765                 if (counter % 4 == 3)
2766                         db_printf("\n");
2767                 counter++;
2768         }
2769         if (counter != 0 && counter % 4 != 0)
2770                 db_printf("\n");
2771 }
2772
2773 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2774 {
2775         struct unpcb *unp;
2776
2777         if (!have_addr) {
2778                 db_printf("usage: show unpcb <addr>\n");
2779                 return;
2780         }
2781         unp = (struct unpcb *)addr;
2782
2783         db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
2784             unp->unp_vnode);
2785
2786         db_printf("unp_ino: %ju   unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2787             unp->unp_conn);
2788
2789         db_printf("unp_refs:\n");
2790         db_print_unprefs(2, &unp->unp_refs);
2791
2792         /* XXXRW: Would be nice to print the full address, if any. */
2793         db_printf("unp_addr: %p\n", unp->unp_addr);
2794
2795         db_printf("unp_gencnt: %llu\n",
2796             (unsigned long long)unp->unp_gencnt);
2797
2798         db_printf("unp_flags: %x (", unp->unp_flags);
2799         db_print_unpflags(unp->unp_flags);
2800         db_printf(")\n");
2801
2802         db_printf("unp_peercred:\n");
2803         db_print_xucred(2, &unp->unp_peercred);
2804
2805         db_printf("unp_refcount: %u\n", unp->unp_refcount);
2806 }
2807 #endif