sys/kern/uipc_usrreq.c

   1 /*-
   2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
   3  *      The Regents of the University of California.
   4  * Copyright (c) 2004-2009 Robert N. M. Watson
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer.
  12  * 2. Redistributions in binary form must reproduce the above copyright
  13  *    notice, this list of conditions and the following disclaimer in the
  14  *    documentation and/or other materials provided with the distribution.
  15  * 4. Neither the name of the University nor the names of its contributors
  16  *    may be used to endorse or promote products derived from this software
  17  *    without specific prior written permission.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  29  * SUCH DAMAGE.
  30  *
  31  *      From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
  32  */
  33
  34 /*
  35  * UNIX Domain (Local) Sockets
  36  *
  37  * This is an implementation of UNIX (local) domain sockets.  Each socket has
  38  * an associated struct unpcb (UNIX protocol control block).  Stream sockets
  39  * may be connected to 0 or 1 other socket.  Datagram sockets may be
  40  * connected to 0, 1, or many other sockets.  Sockets may be created and
  41  * connected in pairs (socketpair(2)), or bound/connected to using the file
  42  * system name space.  For most purposes, only the receive socket buffer is
  43  * used, as sending on one socket delivers directly to the receive socket
  44  * buffer of a second socket.
  45  *
  46  * The implementation is substantially complicated by the fact that
  47  * "ancillary data", such as file descriptors or credentials, may be passed
  48  * across UNIX domain sockets.  The potential for passing UNIX domain sockets
  49  * over other UNIX domain sockets requires the implementation of a simple
  50  * garbage collector to find and tear down cycles of disconnected sockets.
  51  *
  52  * TODO:
  53  *      RDM
  54  *      rethink name space problems
  55  *      need a proper out-of-band
  56  */
  57
  58 #include <sys/cdefs.h>
  59 __FBSDID("$FreeBSD$");
  60
  61 #include "opt_ddb.h"
  62
  63 #include <sys/param.h>
  64 #include <sys/capsicum.h>
  65 #include <sys/domain.h>
  66 #include <sys/fcntl.h>
  67 #include <sys/malloc.h>         /* XXX must be before <sys/file.h> */
  68 #include <sys/eventhandler.h>
  69 #include <sys/file.h>
  70 #include <sys/filedesc.h>
  71 #include <sys/kernel.h>
  72 #include <sys/lock.h>
  73 #include <sys/mbuf.h>
  74 #include <sys/mount.h>
  75 #include <sys/mutex.h>
  76 #include <sys/namei.h>
  77 #include <sys/proc.h>
  78 #include <sys/protosw.h>
  79 #include <sys/queue.h>
  80 #include <sys/resourcevar.h>
  81 #include <sys/rwlock.h>
  82 #include <sys/socket.h>
  83 #include <sys/socketvar.h>
  84 #include <sys/signalvar.h>
  85 #include <sys/stat.h>
  86 #include <sys/sx.h>
  87 #include <sys/sysctl.h>
  88 #include <sys/systm.h>
  89 #include <sys/taskqueue.h>
  90 #include <sys/un.h>
  91 #include <sys/unpcb.h>
  92 #include <sys/vnode.h>
  93
  94 #include <net/vnet.h>
  95
  96 #ifdef DDB
  97 #include <ddb/ddb.h>
  98 #endif
  99
 100 #include <security/mac/mac_framework.h>
 101
 102 #include <vm/uma.h>
 103
 104 MALLOC_DECLARE(M_FILECAPS);
 105
 106 /*
 107  * Locking key:
 108  * (l)  Locked using list lock
 109  * (g)  Locked using linkage lock
 110  */
 111
 112 static uma_zone_t       unp_zone;
 113 static unp_gen_t        unp_gencnt;     /* (l) */
 114 static u_int            unp_count;      /* (l) Count of local sockets. */
 115 static ino_t            unp_ino;        /* Prototype for fake inode numbers. */
 116 static int              unp_rights;     /* (g) File descriptors in flight. */
 117 static struct unp_head  unp_shead;      /* (l) List of stream sockets. */
 118 static struct unp_head  unp_dhead;      /* (l) List of datagram sockets. */
 119 static struct unp_head  unp_sphead;     /* (l) List of seqpacket sockets. */
 120
 121 struct unp_defer {
 122         SLIST_ENTRY(unp_defer) ud_link;
 123         struct file *ud_fp;
 124 };
 125 static SLIST_HEAD(, unp_defer) unp_defers;
 126 static int unp_defers_count;
 127
 128 static const struct sockaddr    sun_noname = { sizeof(sun_noname), AF_LOCAL };
 129
 130 /*
 131  * Garbage collection of cyclic file descriptor/socket references occurs
 132  * asynchronously in a taskqueue context in order to avoid recursion and
 133  * reentrance in the UNIX domain socket, file descriptor, and socket layer
 134  * code.  See unp_gc() for a full description.
 135  */
 136 static struct timeout_task unp_gc_task;
 137
 138 /*
 139  * The close of unix domain sockets attached as SCM_RIGHTS is
 140  * postponed to the taskqueue, to avoid arbitrary recursion depth.
 141  * The attached sockets might have another sockets attached.
 142  */
 143 static struct task      unp_defer_task;
 144
 145 /*
 146  * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
 147  * stream sockets, although the total for sender and receiver is actually
 148  * only PIPSIZ.
 149  *
 150  * Datagram sockets really use the sendspace as the maximum datagram size,
 151  * and don't really want to reserve the sendspace.  Their recvspace should be
 152  * large enough for at least one max-size datagram plus address.
 153  */
 154 #ifndef PIPSIZ
 155 #define PIPSIZ  8192
 156 #endif
 157 static u_long   unpst_sendspace = PIPSIZ;
 158 static u_long   unpst_recvspace = PIPSIZ;
 159 static u_long   unpdg_sendspace = 2*1024;       /* really max datagram size */
 160 static u_long   unpdg_recvspace = 4*1024;
 161 static u_long   unpsp_sendspace = PIPSIZ;       /* really max datagram size */
 162 static u_long   unpsp_recvspace = PIPSIZ;
 163
 164 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
 165 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0,
 166     "SOCK_STREAM");
 167 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
 168 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0,
 169     "SOCK_SEQPACKET");
 170
 171 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
 172            &unpst_sendspace, 0, "Default stream send space.");
 173 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
 174            &unpst_recvspace, 0, "Default stream receive space.");
 175 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
 176            &unpdg_sendspace, 0, "Default datagram send space.");
 177 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
 178            &unpdg_recvspace, 0, "Default datagram receive space.");
 179 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
 180            &unpsp_sendspace, 0, "Default seqpacket send space.");
 181 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
 182            &unpsp_recvspace, 0, "Default seqpacket receive space.");
 183 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
 184     "File descriptors in flight.");
 185 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
 186     &unp_defers_count, 0,
 187     "File descriptors deferred to taskqueue for close.");
 188
 189 /*
 190  * Locking and synchronization:
 191  *
 192  * Three types of locks exit in the local domain socket implementation: a
 193  * global list mutex, a global linkage rwlock, and per-unpcb mutexes.  Of the
 194  * global locks, the list lock protects the socket count, global generation
 195  * number, and stream/datagram global lists.  The linkage lock protects the
 196  * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be
 197  * held exclusively over the acquisition of multiple unpcb locks to prevent
 198  * deadlock.
 199  *
 200  * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
 201  * allocated in pru_attach() and freed in pru_detach().  The validity of that
 202  * pointer is an invariant, so no lock is required to dereference the so_pcb
 203  * pointer if a valid socket reference is held by the caller.  In practice,
 204  * this is always true during operations performed on a socket.  Each unpcb
 205  * has a back-pointer to its socket, unp_socket, which will be stable under
 206  * the same circumstances.
 207  *
 208  * This pointer may only be safely dereferenced as long as a valid reference
 209  * to the unpcb is held.  Typically, this reference will be from the socket,
 210  * or from another unpcb when the referring unpcb's lock is held (in order
 211  * that the reference not be invalidated during use).  For example, to follow
 212  * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
 213  * as unp_socket remains valid as long as the reference to unp_conn is valid.
 214  *
 215  * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx.  Individual
 216  * atomic reads without the lock may be performed "lockless", but more
 217  * complex reads and read-modify-writes require the mutex to be held.  No
 218  * lock order is defined between unpcb locks -- multiple unpcb locks may be
 219  * acquired at the same time only when holding the linkage rwlock
 220  * exclusively, which prevents deadlocks.
 221  *
 222  * Blocking with UNIX domain sockets is a tricky issue: unlike most network
 223  * protocols, bind() is a non-atomic operation, and connect() requires
 224  * potential sleeping in the protocol, due to potentially waiting on local or
 225  * distributed file systems.  We try to separate "lookup" operations, which
 226  * may sleep, and the IPC operations themselves, which typically can occur
 227  * with relative atomicity as locks can be held over the entire operation.
 228  *
 229  * Another tricky issue is simultaneous multi-threaded or multi-process
 230  * access to a single UNIX domain socket.  These are handled by the flags
 231  * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
 232  * binding, both of which involve dropping UNIX domain socket locks in order
 233  * to perform namei() and other file system operations.
 234  */
 235 static struct rwlock    unp_link_rwlock;
 236 static struct mtx       unp_list_lock;
 237 static struct mtx       unp_defers_lock;
 238
 239 #define UNP_LINK_LOCK_INIT()            rw_init(&unp_link_rwlock,       \
 240                                             "unp_link_rwlock")
 241
 242 #define UNP_LINK_LOCK_ASSERT()  rw_assert(&unp_link_rwlock,     \
 243                                             RA_LOCKED)
 244 #define UNP_LINK_UNLOCK_ASSERT()        rw_assert(&unp_link_rwlock,     \
 245                                             RA_UNLOCKED)
 246
 247 #define UNP_LINK_RLOCK()                rw_rlock(&unp_link_rwlock)
 248 #define UNP_LINK_RUNLOCK()              rw_runlock(&unp_link_rwlock)
 249 #define UNP_LINK_WLOCK()                rw_wlock(&unp_link_rwlock)
 250 #define UNP_LINK_WUNLOCK()              rw_wunlock(&unp_link_rwlock)
 251 #define UNP_LINK_WLOCK_ASSERT()         rw_assert(&unp_link_rwlock,     \
 252                                             RA_WLOCKED)
 253
 254 #define UNP_LIST_LOCK_INIT()            mtx_init(&unp_list_lock,        \
 255                                             "unp_list_lock", NULL, MTX_DEF)
 256 #define UNP_LIST_LOCK()                 mtx_lock(&unp_list_lock)
 257 #define UNP_LIST_UNLOCK()               mtx_unlock(&unp_list_lock)
 258
 259 #define UNP_DEFERRED_LOCK_INIT()        mtx_init(&unp_defers_lock, \
 260                                             "unp_defer", NULL, MTX_DEF)
 261 #define UNP_DEFERRED_LOCK()             mtx_lock(&unp_defers_lock)
 262 #define UNP_DEFERRED_UNLOCK()           mtx_unlock(&unp_defers_lock)
 263
 264 #define UNP_PCB_LOCK_INIT(unp)          mtx_init(&(unp)->unp_mtx,       \
 265                                             "unp_mtx", "unp_mtx",       \
 266                                             MTX_DUPOK|MTX_DEF|MTX_RECURSE)
 267 #define UNP_PCB_LOCK_DESTROY(unp)       mtx_destroy(&(unp)->unp_mtx)
 268 #define UNP_PCB_LOCK(unp)               mtx_lock(&(unp)->unp_mtx)
 269 #define UNP_PCB_UNLOCK(unp)             mtx_unlock(&(unp)->unp_mtx)
 270 #define UNP_PCB_LOCK_ASSERT(unp)        mtx_assert(&(unp)->unp_mtx, MA_OWNED)
 271
 272 static int      uipc_connect2(struct socket *, struct socket *);
 273 static int      uipc_ctloutput(struct socket *, struct sockopt *);
 274 static int      unp_connect(struct socket *, struct sockaddr *,
 275                     struct thread *);
 276 static int      unp_connectat(int, struct socket *, struct sockaddr *,
 277                     struct thread *);
 278 static int      unp_connect2(struct socket *so, struct socket *so2, int);
 279 static void     unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
 280 static void     unp_dispose(struct mbuf *);
 281 static void     unp_dispose_so(struct socket *so);
 282 static void     unp_shutdown(struct unpcb *);
 283 static void     unp_drop(struct unpcb *);
 284 static void     unp_gc(__unused void *, int);
 285 static void     unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
 286 static void     unp_discard(struct file *);
 287 static void     unp_freerights(struct filedescent **, int);
 288 static void     unp_init(void);
 289 static int      unp_internalize(struct mbuf **, struct thread *);
 290 static void     unp_internalize_fp(struct file *);
 291 static int      unp_externalize(struct mbuf *, struct mbuf **, int);
 292 static int      unp_externalize_fp(struct file *);
 293 static struct mbuf      *unp_addsockcred(struct thread *, struct mbuf *);
 294 static void     unp_process_defers(void * __unused, int);
 295
 296 /*
 297  * Definitions of protocols supported in the LOCAL domain.
 298  */
 299 static struct domain localdomain;
 300 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
 301 static struct pr_usrreqs uipc_usrreqs_seqpacket;
 302 static struct protosw localsw[] = {
 303 {
 304         .pr_type =              SOCK_STREAM,
 305         .pr_domain =            &localdomain,
 306         .pr_flags =             PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
 307         .pr_ctloutput =         &uipc_ctloutput,
 308         .pr_usrreqs =           &uipc_usrreqs_stream
 309 },
 310 {
 311         .pr_type =              SOCK_DGRAM,
 312         .pr_domain =            &localdomain,
 313         .pr_flags =             PR_ATOMIC|PR_ADDR|PR_RIGHTS,
 314         .pr_ctloutput =         &uipc_ctloutput,
 315         .pr_usrreqs =           &uipc_usrreqs_dgram
 316 },
 317 {
 318         .pr_type =              SOCK_SEQPACKET,
 319         .pr_domain =            &localdomain,
 320
 321         /*
 322          * XXXRW: For now, PR_ADDR because soreceive will bump into them
 323          * due to our use of sbappendaddr.  A new sbappend variants is needed
 324          * that supports both atomic record writes and control data.
 325          */
 326         .pr_flags =             PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
 327                                     PR_RIGHTS,
 328         .pr_ctloutput =         &uipc_ctloutput,
 329         .pr_usrreqs =           &uipc_usrreqs_seqpacket,
 330 },
 331 };
 332
 333 static struct domain localdomain = {
 334         .dom_family =           AF_LOCAL,
 335         .dom_name =             "local",
 336         .dom_init =             unp_init,
 337         .dom_externalize =      unp_externalize,
 338         .dom_dispose =          unp_dispose_so,
 339         .dom_protosw =          localsw,
 340         .dom_protoswNPROTOSW =  &localsw[nitems(localsw)]
 341 };
 342 DOMAIN_SET(local);
 343
 344 static void
 345 uipc_abort(struct socket *so)
 346 {
 347         struct unpcb *unp, *unp2;
 348
 349         unp = sotounpcb(so);
 350         KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
 351
 352         UNP_LINK_WLOCK();
 353         UNP_PCB_LOCK(unp);
 354         unp2 = unp->unp_conn;
 355         if (unp2 != NULL) {
 356                 UNP_PCB_LOCK(unp2);
 357                 unp_drop(unp2);
 358                 UNP_PCB_UNLOCK(unp2);
 359         }
 360         UNP_PCB_UNLOCK(unp);
 361         UNP_LINK_WUNLOCK();
 362 }
 363
 364 static int
 365 uipc_accept(struct socket *so, struct sockaddr **nam)
 366 {
 367         struct unpcb *unp, *unp2;
 368         const struct sockaddr *sa;
 369
 370         /*
 371          * Pass back name of connected socket, if it was bound and we are
 372          * still connected (our peer may have closed already!).
 373          */
 374         unp = sotounpcb(so);
 375         KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
 376
 377         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 378         UNP_LINK_RLOCK();
 379         unp2 = unp->unp_conn;
 380         if (unp2 != NULL && unp2->unp_addr != NULL) {
 381                 UNP_PCB_LOCK(unp2);
 382                 sa = (struct sockaddr *) unp2->unp_addr;
 383                 bcopy(sa, *nam, sa->sa_len);
 384                 UNP_PCB_UNLOCK(unp2);
 385         } else {
 386                 sa = &sun_noname;
 387                 bcopy(sa, *nam, sa->sa_len);
 388         }
 389         UNP_LINK_RUNLOCK();
 390         return (0);
 391 }
 392
 393 static int
 394 uipc_attach(struct socket *so, int proto, struct thread *td)
 395 {
 396         u_long sendspace, recvspace;
 397         struct unpcb *unp;
 398         int error;
 399
 400         KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
 401         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
 402                 switch (so->so_type) {
 403                 case SOCK_STREAM:
 404                         sendspace = unpst_sendspace;
 405                         recvspace = unpst_recvspace;
 406                         break;
 407
 408                 case SOCK_DGRAM:
 409                         sendspace = unpdg_sendspace;
 410                         recvspace = unpdg_recvspace;
 411                         break;
 412
 413                 case SOCK_SEQPACKET:
 414                         sendspace = unpsp_sendspace;
 415                         recvspace = unpsp_recvspace;
 416                         break;
 417
 418                 default:
 419                         panic("uipc_attach");
 420                 }
 421                 error = soreserve(so, sendspace, recvspace);
 422                 if (error)
 423                         return (error);
 424         }
 425         unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
 426         if (unp == NULL)
 427                 return (ENOBUFS);
 428         LIST_INIT(&unp->unp_refs);
 429         UNP_PCB_LOCK_INIT(unp);
 430         unp->unp_socket = so;
 431         so->so_pcb = unp;
 432         unp->unp_refcount = 1;
 433         if (so->so_head != NULL)
 434                 unp->unp_flags |= UNP_NASCENT;
 435
 436         UNP_LIST_LOCK();
 437         unp->unp_gencnt = ++unp_gencnt;
 438         unp_count++;
 439         switch (so->so_type) {
 440         case SOCK_STREAM:
 441                 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
 442                 break;
 443
 444         case SOCK_DGRAM:
 445                 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
 446                 break;
 447
 448         case SOCK_SEQPACKET:
 449                 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
 450                 break;
 451
 452         default:
 453                 panic("uipc_attach");
 454         }
 455         UNP_LIST_UNLOCK();
 456
 457         return (0);
 458 }
 459
 460 static int
 461 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
 462 {
 463         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
 464         struct vattr vattr;
 465         int error, namelen;
 466         struct nameidata nd;
 467         struct unpcb *unp;
 468         struct vnode *vp;
 469         struct mount *mp;
 470         cap_rights_t rights;
 471         char *buf;
 472
 473         if (nam->sa_family != AF_UNIX)
 474                 return (EAFNOSUPPORT);
 475
 476         unp = sotounpcb(so);
 477         KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
 478
 479         if (soun->sun_len > sizeof(struct sockaddr_un))
 480                 return (EINVAL);
 481         namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
 482         if (namelen <= 0)
 483                 return (EINVAL);
 484
 485         /*
 486          * We don't allow simultaneous bind() calls on a single UNIX domain
 487          * socket, so flag in-progress operations, and return an error if an
 488          * operation is already in progress.
 489          *
 490          * Historically, we have not allowed a socket to be rebound, so this
 491          * also returns an error.  Not allowing re-binding simplifies the
 492          * implementation and avoids a great many possible failure modes.
 493          */
 494         UNP_PCB_LOCK(unp);
 495         if (unp->unp_vnode != NULL) {
 496                 UNP_PCB_UNLOCK(unp);
 497                 return (EINVAL);
 498         }
 499         if (unp->unp_flags & UNP_BINDING) {
 500                 UNP_PCB_UNLOCK(unp);
 501                 return (EALREADY);
 502         }
 503         unp->unp_flags |= UNP_BINDING;
 504         UNP_PCB_UNLOCK(unp);
 505
 506         buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
 507         bcopy(soun->sun_path, buf, namelen);
 508         buf[namelen] = 0;
 509
 510 restart:
 511         NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE,
 512             UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td);
 513 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
 514         error = namei(&nd);
 515         if (error)
 516                 goto error;
 517         vp = nd.ni_vp;
 518         if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
 519                 NDFREE(&nd, NDF_ONLY_PNBUF);
 520                 if (nd.ni_dvp == vp)
 521                         vrele(nd.ni_dvp);
 522                 else
 523                         vput(nd.ni_dvp);
 524                 if (vp != NULL) {
 525                         vrele(vp);
 526                         error = EADDRINUSE;
 527                         goto error;
 528                 }
 529                 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
 530                 if (error)
 531                         goto error;
 532                 goto restart;
 533         }
 534         VATTR_NULL(&vattr);
 535         vattr.va_type = VSOCK;
 536         vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
 537 #ifdef MAC
 538         error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
 539             &vattr);
 540 #endif
 541         if (error == 0)
 542                 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
 543         NDFREE(&nd, NDF_ONLY_PNBUF);
 544         vput(nd.ni_dvp);
 545         if (error) {
 546                 vn_finished_write(mp);
 547                 goto error;
 548         }
 549         vp = nd.ni_vp;
 550         ASSERT_VOP_ELOCKED(vp, "uipc_bind");
 551         soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
 552
 553         UNP_LINK_WLOCK();
 554         UNP_PCB_LOCK(unp);
 555         VOP_UNP_BIND(vp, unp->unp_socket);
 556         unp->unp_vnode = vp;
 557         unp->unp_addr = soun;
 558         unp->unp_flags &= ~UNP_BINDING;
 559         UNP_PCB_UNLOCK(unp);
 560         UNP_LINK_WUNLOCK();
 561         VOP_UNLOCK(vp, 0);
 562         vn_finished_write(mp);
 563         free(buf, M_TEMP);
 564         return (0);
 565
 566 error:
 567         UNP_PCB_LOCK(unp);
 568         unp->unp_flags &= ~UNP_BINDING;
 569         UNP_PCB_UNLOCK(unp);
 570         free(buf, M_TEMP);
 571         return (error);
 572 }
 573
 574 static int
 575 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 576 {
 577
 578         return (uipc_bindat(AT_FDCWD, so, nam, td));
 579 }
 580
 581 static int
 582 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 583 {
 584         int error;
 585
 586         KASSERT(td == curthread, ("uipc_connect: td != curthread"));
 587         UNP_LINK_WLOCK();
 588         error = unp_connect(so, nam, td);
 589         UNP_LINK_WUNLOCK();
 590         return (error);
 591 }
 592
 593 static int
 594 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
 595     struct thread *td)
 596 {
 597         int error;
 598
 599         KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
 600         UNP_LINK_WLOCK();
 601         error = unp_connectat(fd, so, nam, td);
 602         UNP_LINK_WUNLOCK();
 603         return (error);
 604 }
 605
 606 static void
 607 uipc_close(struct socket *so)
 608 {
 609         struct unpcb *unp, *unp2;
 610
 611         unp = sotounpcb(so);
 612         KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
 613
 614         UNP_LINK_WLOCK();
 615         UNP_PCB_LOCK(unp);
 616         unp2 = unp->unp_conn;
 617         if (unp2 != NULL) {
 618                 UNP_PCB_LOCK(unp2);
 619                 unp_disconnect(unp, unp2);
 620                 UNP_PCB_UNLOCK(unp2);
 621         }
 622         UNP_PCB_UNLOCK(unp);
 623         UNP_LINK_WUNLOCK();
 624 }
 625
 626 static int
 627 uipc_connect2(struct socket *so1, struct socket *so2)
 628 {
 629         struct unpcb *unp, *unp2;
 630         int error;
 631
 632         UNP_LINK_WLOCK();
 633         unp = so1->so_pcb;
 634         KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
 635         UNP_PCB_LOCK(unp);
 636         unp2 = so2->so_pcb;
 637         KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
 638         UNP_PCB_LOCK(unp2);
 639         error = unp_connect2(so1, so2, PRU_CONNECT2);
 640         UNP_PCB_UNLOCK(unp2);
 641         UNP_PCB_UNLOCK(unp);
 642         UNP_LINK_WUNLOCK();
 643         return (error);
 644 }
 645
 646 static void
 647 uipc_detach(struct socket *so)
 648 {
 649         struct unpcb *unp, *unp2;
 650         struct sockaddr_un *saved_unp_addr;
 651         struct vnode *vp;
 652         int freeunp, local_unp_rights;
 653
 654         unp = sotounpcb(so);
 655         KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
 656
 657         vp = NULL;
 658         local_unp_rights = 0;
 659
 660         UNP_LIST_LOCK();
 661         LIST_REMOVE(unp, unp_link);
 662         unp->unp_gencnt = ++unp_gencnt;
 663         --unp_count;
 664         UNP_LIST_UNLOCK();
 665
 666         if ((unp->unp_flags & UNP_NASCENT) != 0) {
 667                 UNP_PCB_LOCK(unp);
 668                 goto teardown;
 669         }
 670         UNP_LINK_WLOCK();
 671         UNP_PCB_LOCK(unp);
 672
 673         /*
 674          * XXXRW: Should assert vp->v_socket == so.
 675          */
 676         if ((vp = unp->unp_vnode) != NULL) {
 677                 VOP_UNP_DETACH(vp);
 678                 unp->unp_vnode = NULL;
 679         }
 680         unp2 = unp->unp_conn;
 681         if (unp2 != NULL) {
 682                 UNP_PCB_LOCK(unp2);
 683                 unp_disconnect(unp, unp2);
 684                 UNP_PCB_UNLOCK(unp2);
 685         }
 686
 687         /*
 688          * We hold the linkage lock exclusively, so it's OK to acquire
 689          * multiple pcb locks at a time.
 690          */
 691         while (!LIST_EMPTY(&unp->unp_refs)) {
 692                 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
 693
 694                 UNP_PCB_LOCK(ref);
 695                 unp_drop(ref);
 696                 UNP_PCB_UNLOCK(ref);
 697         }
 698         local_unp_rights = unp_rights;
 699         UNP_LINK_WUNLOCK();
 700 teardown:
 701         unp->unp_socket->so_pcb = NULL;
 702         saved_unp_addr = unp->unp_addr;
 703         unp->unp_addr = NULL;
 704         unp->unp_refcount--;
 705         freeunp = (unp->unp_refcount == 0);
 706         if (saved_unp_addr != NULL)
 707                 free(saved_unp_addr, M_SONAME);
 708         if (freeunp) {
 709                 UNP_PCB_LOCK_DESTROY(unp);
 710                 uma_zfree(unp_zone, unp);
 711         } else
 712                 UNP_PCB_UNLOCK(unp);
 713         if (vp)
 714                 vrele(vp);
 715         if (local_unp_rights)
 716                 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
 717 }
 718
 719 static int
 720 uipc_disconnect(struct socket *so)
 721 {
 722         struct unpcb *unp, *unp2;
 723
 724         unp = sotounpcb(so);
 725         KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
 726
 727         UNP_LINK_WLOCK();
 728         UNP_PCB_LOCK(unp);
 729         unp2 = unp->unp_conn;
 730         if (unp2 != NULL) {
 731                 UNP_PCB_LOCK(unp2);
 732                 unp_disconnect(unp, unp2);
 733                 UNP_PCB_UNLOCK(unp2);
 734         }
 735         UNP_PCB_UNLOCK(unp);
 736         UNP_LINK_WUNLOCK();
 737         return (0);
 738 }
 739
 740 static int
 741 uipc_listen(struct socket *so, int backlog, struct thread *td)
 742 {
 743         struct unpcb *unp;
 744         int error;
 745
 746         unp = sotounpcb(so);
 747         KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
 748
 749         UNP_PCB_LOCK(unp);
 750         if (unp->unp_vnode == NULL) {
 751                 /* Already connected or not bound to an address. */
 752                 error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ;
 753                 UNP_PCB_UNLOCK(unp);
 754                 return (error);
 755         }
 756
 757         SOCK_LOCK(so);
 758         error = solisten_proto_check(so);
 759         if (error == 0) {
 760                 cru2x(td->td_ucred, &unp->unp_peercred);
 761                 unp->unp_flags |= UNP_HAVEPCCACHED;
 762                 solisten_proto(so, backlog);
 763         }
 764         SOCK_UNLOCK(so);
 765         UNP_PCB_UNLOCK(unp);
 766         return (error);
 767 }
 768
 769 static int
 770 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
 771 {
 772         struct unpcb *unp, *unp2;
 773         const struct sockaddr *sa;
 774
 775         unp = sotounpcb(so);
 776         KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
 777
 778         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 779         UNP_LINK_RLOCK();
 780         /*
 781          * XXX: It seems that this test always fails even when connection is
 782          * established.  So, this else clause is added as workaround to
 783          * return PF_LOCAL sockaddr.
 784          */
 785         unp2 = unp->unp_conn;
 786         if (unp2 != NULL) {
 787                 UNP_PCB_LOCK(unp2);
 788                 if (unp2->unp_addr != NULL)
 789                         sa = (struct sockaddr *) unp2->unp_addr;
 790                 else
 791                         sa = &sun_noname;
 792                 bcopy(sa, *nam, sa->sa_len);
 793                 UNP_PCB_UNLOCK(unp2);
 794         } else {
 795                 sa = &sun_noname;
 796                 bcopy(sa, *nam, sa->sa_len);
 797         }
 798         UNP_LINK_RUNLOCK();
 799         return (0);
 800 }
 801
 802 static int
 803 uipc_rcvd(struct socket *so, int flags)
 804 {
 805         struct unpcb *unp, *unp2;
 806         struct socket *so2;
 807         u_int mbcnt, sbcc;
 808
 809         unp = sotounpcb(so);
 810         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
 811         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
 812             ("%s: socktype %d", __func__, so->so_type));
 813
 814         /*
 815          * Adjust backpressure on sender and wakeup any waiting to write.
 816          *
 817          * The unp lock is acquired to maintain the validity of the unp_conn
 818          * pointer; no lock on unp2 is required as unp2->unp_socket will be
 819          * static as long as we don't permit unp2 to disconnect from unp,
 820          * which is prevented by the lock on unp.  We cache values from
 821          * so_rcv to avoid holding the so_rcv lock over the entire
 822          * transaction on the remote so_snd.
 823          */
 824         SOCKBUF_LOCK(&so->so_rcv);
 825         mbcnt = so->so_rcv.sb_mbcnt;
 826         sbcc = sbavail(&so->so_rcv);
 827         SOCKBUF_UNLOCK(&so->so_rcv);
 828         /*
 829          * There is a benign race condition at this point.  If we're planning to
 830          * clear SB_STOP, but uipc_send is called on the connected socket at
 831          * this instant, it might add data to the sockbuf and set SB_STOP.  Then
 832          * we would erroneously clear SB_STOP below, even though the sockbuf is
 833          * full.  The race is benign because the only ill effect is to allow the
 834          * sockbuf to exceed its size limit, and the size limits are not
 835          * strictly guaranteed anyway.
 836          */
 837         UNP_PCB_LOCK(unp);
 838         unp2 = unp->unp_conn;
 839         if (unp2 == NULL) {
 840                 UNP_PCB_UNLOCK(unp);
 841                 return (0);
 842         }
 843         so2 = unp2->unp_socket;
 844         SOCKBUF_LOCK(&so2->so_snd);
 845         if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
 846                 so2->so_snd.sb_flags &= ~SB_STOP;
 847         sowwakeup_locked(so2);
 848         UNP_PCB_UNLOCK(unp);
 849         return (0);
 850 }
 851
 852 static int
 853 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
 854     struct mbuf *control, struct thread *td)
 855 {
 856         struct unpcb *unp, *unp2;
 857         struct socket *so2;
 858         u_int mbcnt, sbcc;
 859         int error = 0;
 860
 861         unp = sotounpcb(so);
 862         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
 863         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
 864             so->so_type == SOCK_SEQPACKET,
 865             ("%s: socktype %d", __func__, so->so_type));
 866
 867         if (flags & PRUS_OOB) {
 868                 error = EOPNOTSUPP;
 869                 goto release;
 870         }
 871         if (control != NULL && (error = unp_internalize(&control, td)))
 872                 goto release;
 873         if ((nam != NULL) || (flags & PRUS_EOF))
 874                 UNP_LINK_WLOCK();
 875         else
 876                 UNP_LINK_RLOCK();
 877         switch (so->so_type) {
 878         case SOCK_DGRAM:
 879         {
 880                 const struct sockaddr *from;
 881
 882                 unp2 = unp->unp_conn;
 883                 if (nam != NULL) {
 884                         UNP_LINK_WLOCK_ASSERT();
 885                         if (unp2 != NULL) {
 886                                 error = EISCONN;
 887                                 break;
 888                         }
 889                         error = unp_connect(so, nam, td);
 890                         if (error)
 891                                 break;
 892                         unp2 = unp->unp_conn;
 893                 }
 894
 895                 /*
 896                  * Because connect() and send() are non-atomic in a sendto()
 897                  * with a target address, it's possible that the socket will
 898                  * have disconnected before the send() can run.  In that case
 899                  * return the slightly counter-intuitive but otherwise
 900                  * correct error that the socket is not connected.
 901                  */
 902                 if (unp2 == NULL) {
 903                         error = ENOTCONN;
 904                         break;
 905                 }
 906                 /* Lockless read. */
 907                 if (unp2->unp_flags & UNP_WANTCRED)
 908                         control = unp_addsockcred(td, control);
 909                 UNP_PCB_LOCK(unp);
 910                 if (unp->unp_addr != NULL)
 911                         from = (struct sockaddr *)unp->unp_addr;
 912                 else
 913                         from = &sun_noname;
 914                 so2 = unp2->unp_socket;
 915                 SOCKBUF_LOCK(&so2->so_rcv);
 916                 if (sbappendaddr_locked(&so2->so_rcv, from, m,
 917                     control)) {
 918                         sorwakeup_locked(so2);
 919                         m = NULL;
 920                         control = NULL;
 921                 } else {
 922                         SOCKBUF_UNLOCK(&so2->so_rcv);
 923                         error = ENOBUFS;
 924                 }
 925                 if (nam != NULL) {
 926                         UNP_LINK_WLOCK_ASSERT();
 927                         UNP_PCB_LOCK(unp2);
 928                         unp_disconnect(unp, unp2);
 929                         UNP_PCB_UNLOCK(unp2);
 930                 }
 931                 UNP_PCB_UNLOCK(unp);
 932                 break;
 933         }
 934
 935         case SOCK_SEQPACKET:
 936         case SOCK_STREAM:
 937                 if ((so->so_state & SS_ISCONNECTED) == 0) {
 938                         if (nam != NULL) {
 939                                 UNP_LINK_WLOCK_ASSERT();
 940                                 error = unp_connect(so, nam, td);
 941                                 if (error)
 942                                         break;  /* XXX */
 943                         } else {
 944                                 error = ENOTCONN;
 945                                 break;
 946                         }
 947                 }
 948
 949                 /* Lockless read. */
 950                 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
 951                         error = EPIPE;
 952                         break;
 953                 }
 954
 955                 /*
 956                  * Because connect() and send() are non-atomic in a sendto()
 957                  * with a target address, it's possible that the socket will
 958                  * have disconnected before the send() can run.  In that case
 959                  * return the slightly counter-intuitive but otherwise
 960                  * correct error that the socket is not connected.
 961                  *
 962                  * Locking here must be done carefully: the linkage lock
 963                  * prevents interconnections between unpcbs from changing, so
 964                  * we can traverse from unp to unp2 without acquiring unp's
 965                  * lock.  Socket buffer locks follow unpcb locks, so we can
 966                  * acquire both remote and lock socket buffer locks.
 967                  */
 968                 unp2 = unp->unp_conn;
 969                 if (unp2 == NULL) {
 970                         error = ENOTCONN;
 971                         break;
 972                 }
 973                 so2 = unp2->unp_socket;
 974                 UNP_PCB_LOCK(unp2);
 975                 SOCKBUF_LOCK(&so2->so_rcv);
 976                 if (unp2->unp_flags & UNP_WANTCRED) {
 977                         /*
 978                          * Credentials are passed only once on SOCK_STREAM
 979                          * and SOCK_SEQPACKET.
 980                          */
 981                         unp2->unp_flags &= ~UNP_WANTCRED;
 982                         control = unp_addsockcred(td, control);
 983                 }
 984
 985                 /*
 986                  * Send to paired receive port and wake up readers.  Don't
 987                  * check for space available in the receive buffer if we're
 988                  * attaching ancillary data; Unix domain sockets only check
 989                  * for space in the sending sockbuf, and that check is
 990                  * performed one level up the stack.  At that level we cannot
 991                  * precisely account for the amount of buffer space used
 992                  * (e.g., because control messages are not yet internalized).
 993                  */
 994                 switch (so->so_type) {
 995                 case SOCK_STREAM:
 996                         if (control != NULL) {
 997                                 sbappendcontrol_locked(&so2->so_rcv, m,
 998                                     control);
 999                                 control = NULL;
1000                         } else
1001                                 sbappend_locked(&so2->so_rcv, m, flags);
1002                         break;
1003
1004                 case SOCK_SEQPACKET: {
1005                         const struct sockaddr *from;
1006
1007                         from = &sun_noname;
1008                         if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1009                             from, m, control))
1010                                 control = NULL;
1011                         break;
1012                         }
1013                 }
1014
1015                 mbcnt = so2->so_rcv.sb_mbcnt;
1016                 sbcc = sbavail(&so2->so_rcv);
1017                 if (sbcc)
1018                         sorwakeup_locked(so2);
1019                 else
1020                         SOCKBUF_UNLOCK(&so2->so_rcv);
1021
1022                 /*
1023                  * The PCB lock on unp2 protects the SB_STOP flag.  Without it,
1024                  * it would be possible for uipc_rcvd to be called at this
1025                  * point, drain the receiving sockbuf, clear SB_STOP, and then
1026                  * we would set SB_STOP below.  That could lead to an empty
1027                  * sockbuf having SB_STOP set
1028                  */
1029                 SOCKBUF_LOCK(&so->so_snd);
1030                 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1031                         so->so_snd.sb_flags |= SB_STOP;
1032                 SOCKBUF_UNLOCK(&so->so_snd);
1033                 UNP_PCB_UNLOCK(unp2);
1034                 m = NULL;
1035                 break;
1036         }
1037
1038         /*
1039          * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1040          */
1041         if (flags & PRUS_EOF) {
1042                 UNP_PCB_LOCK(unp);
1043                 socantsendmore(so);
1044                 unp_shutdown(unp);
1045                 UNP_PCB_UNLOCK(unp);
1046         }
1047
1048         if ((nam != NULL) || (flags & PRUS_EOF))
1049                 UNP_LINK_WUNLOCK();
1050         else
1051                 UNP_LINK_RUNLOCK();
1052
1053         if (control != NULL && error != 0)
1054                 unp_dispose(control);
1055
1056 release:
1057         if (control != NULL)
1058                 m_freem(control);
1059         /*
1060          * In case of PRUS_NOTREADY, uipc_ready() is responsible
1061          * for freeing memory.
1062          */
1063         if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1064                 m_freem(m);
1065         return (error);
1066 }
1067
1068 static int
1069 uipc_ready(struct socket *so, struct mbuf *m, int count)
1070 {
1071         struct unpcb *unp, *unp2;
1072         struct socket *so2;
1073         int error;
1074
1075         unp = sotounpcb(so);
1076
1077         UNP_LINK_RLOCK();
1078         if ((unp2 = unp->unp_conn) == NULL) {
1079                 UNP_LINK_RUNLOCK();
1080                 for (int i = 0; i < count; i++)
1081                         m = m_free(m);
1082                 return (ECONNRESET);
1083         }
1084         UNP_PCB_LOCK(unp2);
1085         so2 = unp2->unp_socket;
1086
1087         SOCKBUF_LOCK(&so2->so_rcv);
1088         if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1089                 sorwakeup_locked(so2);
1090         else
1091                 SOCKBUF_UNLOCK(&so2->so_rcv);
1092
1093         UNP_PCB_UNLOCK(unp2);
1094         UNP_LINK_RUNLOCK();
1095
1096         return (error);
1097 }
1098
1099 static int
1100 uipc_sense(struct socket *so, struct stat *sb)
1101 {
1102         struct unpcb *unp;
1103
1104         unp = sotounpcb(so);
1105         KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1106
1107         sb->st_blksize = so->so_snd.sb_hiwat;
1108         UNP_PCB_LOCK(unp);
1109         sb->st_dev = NODEV;
1110         if (unp->unp_ino == 0)
1111                 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1112         sb->st_ino = unp->unp_ino;
1113         UNP_PCB_UNLOCK(unp);
1114         return (0);
1115 }
1116
1117 static int
1118 uipc_shutdown(struct socket *so)
1119 {
1120         struct unpcb *unp;
1121
1122         unp = sotounpcb(so);
1123         KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1124
1125         UNP_LINK_WLOCK();
1126         UNP_PCB_LOCK(unp);
1127         socantsendmore(so);
1128         unp_shutdown(unp);
1129         UNP_PCB_UNLOCK(unp);
1130         UNP_LINK_WUNLOCK();
1131         return (0);
1132 }
1133
1134 static int
1135 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1136 {
1137         struct unpcb *unp;
1138         const struct sockaddr *sa;
1139
1140         unp = sotounpcb(so);
1141         KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1142
1143         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1144         UNP_PCB_LOCK(unp);
1145         if (unp->unp_addr != NULL)
1146                 sa = (struct sockaddr *) unp->unp_addr;
1147         else
1148                 sa = &sun_noname;
1149         bcopy(sa, *nam, sa->sa_len);
1150         UNP_PCB_UNLOCK(unp);
1151         return (0);
1152 }
1153
1154 static struct pr_usrreqs uipc_usrreqs_dgram = {
1155         .pru_abort =            uipc_abort,
1156         .pru_accept =           uipc_accept,
1157         .pru_attach =           uipc_attach,
1158         .pru_bind =             uipc_bind,
1159         .pru_bindat =           uipc_bindat,
1160         .pru_connect =          uipc_connect,
1161         .pru_connectat =        uipc_connectat,
1162         .pru_connect2 =         uipc_connect2,
1163         .pru_detach =           uipc_detach,
1164         .pru_disconnect =       uipc_disconnect,
1165         .pru_listen =           uipc_listen,
1166         .pru_peeraddr =         uipc_peeraddr,
1167         .pru_rcvd =             uipc_rcvd,
1168         .pru_send =             uipc_send,
1169         .pru_sense =            uipc_sense,
1170         .pru_shutdown =         uipc_shutdown,
1171         .pru_sockaddr =         uipc_sockaddr,
1172         .pru_soreceive =        soreceive_dgram,
1173         .pru_close =            uipc_close,
1174 };
1175
1176 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1177         .pru_abort =            uipc_abort,
1178         .pru_accept =           uipc_accept,
1179         .pru_attach =           uipc_attach,
1180         .pru_bind =             uipc_bind,
1181         .pru_bindat =           uipc_bindat,
1182         .pru_connect =          uipc_connect,
1183         .pru_connectat =        uipc_connectat,
1184         .pru_connect2 =         uipc_connect2,
1185         .pru_detach =           uipc_detach,
1186         .pru_disconnect =       uipc_disconnect,
1187         .pru_listen =           uipc_listen,
1188         .pru_peeraddr =         uipc_peeraddr,
1189         .pru_rcvd =             uipc_rcvd,
1190         .pru_send =             uipc_send,
1191         .pru_sense =            uipc_sense,
1192         .pru_shutdown =         uipc_shutdown,
1193         .pru_sockaddr =         uipc_sockaddr,
1194         .pru_soreceive =        soreceive_generic,      /* XXX: or...? */
1195         .pru_close =            uipc_close,
1196 };
1197
1198 static struct pr_usrreqs uipc_usrreqs_stream = {
1199         .pru_abort =            uipc_abort,
1200         .pru_accept =           uipc_accept,
1201         .pru_attach =           uipc_attach,
1202         .pru_bind =             uipc_bind,
1203         .pru_bindat =           uipc_bindat,
1204         .pru_connect =          uipc_connect,
1205         .pru_connectat =        uipc_connectat,
1206         .pru_connect2 =         uipc_connect2,
1207         .pru_detach =           uipc_detach,
1208         .pru_disconnect =       uipc_disconnect,
1209         .pru_listen =           uipc_listen,
1210         .pru_peeraddr =         uipc_peeraddr,
1211         .pru_rcvd =             uipc_rcvd,
1212         .pru_send =             uipc_send,
1213         .pru_ready =            uipc_ready,
1214         .pru_sense =            uipc_sense,
1215         .pru_shutdown =         uipc_shutdown,
1216         .pru_sockaddr =         uipc_sockaddr,
1217         .pru_soreceive =        soreceive_generic,
1218         .pru_close =            uipc_close,
1219 };
1220
1221 static int
1222 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1223 {
1224         struct unpcb *unp;
1225         struct xucred xu;
1226         int error, optval;
1227
1228         if (sopt->sopt_level != 0)
1229                 return (EINVAL);
1230
1231         unp = sotounpcb(so);
1232         KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1233         error = 0;
1234         switch (sopt->sopt_dir) {
1235         case SOPT_GET:
1236                 switch (sopt->sopt_name) {
1237                 case LOCAL_PEERCRED:
1238                         UNP_PCB_LOCK(unp);
1239                         if (unp->unp_flags & UNP_HAVEPC)
1240                                 xu = unp->unp_peercred;
1241                         else {
1242                                 if (so->so_type == SOCK_STREAM)
1243                                         error = ENOTCONN;
1244                                 else
1245                                         error = EINVAL;
1246                         }
1247                         UNP_PCB_UNLOCK(unp);
1248                         if (error == 0)
1249                                 error = sooptcopyout(sopt, &xu, sizeof(xu));
1250                         break;
1251
1252                 case LOCAL_CREDS:
1253                         /* Unlocked read. */
1254                         optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1255                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1256                         break;
1257
1258                 case LOCAL_CONNWAIT:
1259                         /* Unlocked read. */
1260                         optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1261                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1262                         break;
1263
1264                 default:
1265                         error = EOPNOTSUPP;
1266                         break;
1267                 }
1268                 break;
1269
1270         case SOPT_SET:
1271                 switch (sopt->sopt_name) {
1272                 case LOCAL_CREDS:
1273                 case LOCAL_CONNWAIT:
1274                         error = sooptcopyin(sopt, &optval, sizeof(optval),
1275                                             sizeof(optval));
1276                         if (error)
1277                                 break;
1278
1279 #define OPTSET(bit) do {                                                \
1280         UNP_PCB_LOCK(unp);                                              \
1281         if (optval)                                                     \
1282                 unp->unp_flags |= bit;                                  \
1283         else                                                            \
1284                 unp->unp_flags &= ~bit;                                 \
1285         UNP_PCB_UNLOCK(unp);                                            \
1286 } while (0)
1287
1288                         switch (sopt->sopt_name) {
1289                         case LOCAL_CREDS:
1290                                 OPTSET(UNP_WANTCRED);
1291                                 break;
1292
1293                         case LOCAL_CONNWAIT:
1294                                 OPTSET(UNP_CONNWAIT);
1295                                 break;
1296
1297                         default:
1298                                 break;
1299                         }
1300                         break;
1301 #undef  OPTSET
1302                 default:
1303                         error = ENOPROTOOPT;
1304                         break;
1305                 }
1306                 break;
1307
1308         default:
1309                 error = EOPNOTSUPP;
1310                 break;
1311         }
1312         return (error);
1313 }
1314
1315 static int
1316 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1317 {
1318
1319         return (unp_connectat(AT_FDCWD, so, nam, td));
1320 }
1321
1322 static int
1323 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1324     struct thread *td)
1325 {
1326         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1327         struct vnode *vp;
1328         struct socket *so2, *so3;
1329         struct unpcb *unp, *unp2, *unp3;
1330         struct nameidata nd;
1331         char buf[SOCK_MAXADDRLEN];
1332         struct sockaddr *sa;
1333         cap_rights_t rights;
1334         int error, len;
1335
1336         if (nam->sa_family != AF_UNIX)
1337                 return (EAFNOSUPPORT);
1338
1339         UNP_LINK_WLOCK_ASSERT();
1340
1341         unp = sotounpcb(so);
1342         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1343
1344         if (nam->sa_len > sizeof(struct sockaddr_un))
1345                 return (EINVAL);
1346         len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1347         if (len <= 0)
1348                 return (EINVAL);
1349         bcopy(soun->sun_path, buf, len);
1350         buf[len] = 0;
1351
1352         UNP_PCB_LOCK(unp);
1353         if (unp->unp_flags & UNP_CONNECTING) {
1354                 UNP_PCB_UNLOCK(unp);
1355                 return (EALREADY);
1356         }
1357         UNP_LINK_WUNLOCK();
1358         unp->unp_flags |= UNP_CONNECTING;
1359         UNP_PCB_UNLOCK(unp);
1360
1361         sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1362         NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1363             UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1364         error = namei(&nd);
1365         if (error)
1366                 vp = NULL;
1367         else
1368                 vp = nd.ni_vp;
1369         ASSERT_VOP_LOCKED(vp, "unp_connect");
1370         NDFREE(&nd, NDF_ONLY_PNBUF);
1371         if (error)
1372                 goto bad;
1373
1374         if (vp->v_type != VSOCK) {
1375                 error = ENOTSOCK;
1376                 goto bad;
1377         }
1378 #ifdef MAC
1379         error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1380         if (error)
1381                 goto bad;
1382 #endif
1383         error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1384         if (error)
1385                 goto bad;
1386
1387         unp = sotounpcb(so);
1388         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1389
1390         /*
1391          * Lock linkage lock for two reasons: make sure v_socket is stable,
1392          * and to protect simultaneous locking of multiple pcbs.
1393          */
1394         UNP_LINK_WLOCK();
1395         VOP_UNP_CONNECT(vp, &so2);
1396         if (so2 == NULL) {
1397                 error = ECONNREFUSED;
1398                 goto bad2;
1399         }
1400         if (so->so_type != so2->so_type) {
1401                 error = EPROTOTYPE;
1402                 goto bad2;
1403         }
1404         if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1405                 if (so2->so_options & SO_ACCEPTCONN) {
1406                         CURVNET_SET(so2->so_vnet);
1407                         so3 = sonewconn(so2, 0);
1408                         CURVNET_RESTORE();
1409                 } else
1410                         so3 = NULL;
1411                 if (so3 == NULL) {
1412                         error = ECONNREFUSED;
1413                         goto bad2;
1414                 }
1415                 unp = sotounpcb(so);
1416                 unp2 = sotounpcb(so2);
1417                 unp3 = sotounpcb(so3);
1418                 UNP_PCB_LOCK(unp);
1419                 UNP_PCB_LOCK(unp2);
1420                 UNP_PCB_LOCK(unp3);
1421                 if (unp2->unp_addr != NULL) {
1422                         bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1423                         unp3->unp_addr = (struct sockaddr_un *) sa;
1424                         sa = NULL;
1425                 }
1426
1427                 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1428                     ("unp_connect: listener without cached peercred"));
1429                 unp_copy_peercred(td, unp3, unp, unp2);
1430
1431                 UNP_PCB_UNLOCK(unp3);
1432                 UNP_PCB_UNLOCK(unp2);
1433                 UNP_PCB_UNLOCK(unp);
1434 #ifdef MAC
1435                 mac_socketpeer_set_from_socket(so, so3);
1436                 mac_socketpeer_set_from_socket(so3, so);
1437 #endif
1438
1439                 so2 = so3;
1440         }
1441         unp = sotounpcb(so);
1442         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1443         unp2 = sotounpcb(so2);
1444         KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1445         UNP_PCB_LOCK(unp);
1446         UNP_PCB_LOCK(unp2);
1447         error = unp_connect2(so, so2, PRU_CONNECT);
1448         UNP_PCB_UNLOCK(unp2);
1449         UNP_PCB_UNLOCK(unp);
1450 bad2:
1451         UNP_LINK_WUNLOCK();
1452 bad:
1453         if (vp != NULL)
1454                 vput(vp);
1455         free(sa, M_SONAME);
1456         UNP_LINK_WLOCK();
1457         UNP_PCB_LOCK(unp);
1458         unp->unp_flags &= ~UNP_CONNECTING;
1459         UNP_PCB_UNLOCK(unp);
1460         return (error);
1461 }
1462
1463 /*
1464  * Set socket peer credentials at connection time.
1465  *
1466  * The client's PCB credentials are copied from its process structure.  The
1467  * server's PCB credentials are copied from the socket on which it called
1468  * listen(2).  uipc_listen cached that process's credentials at the time.
1469  */
1470 void
1471 unp_copy_peercred(struct thread *td, struct unpcb *client_unp,
1472     struct unpcb *server_unp, struct unpcb *listen_unp)
1473 {
1474         cru2x(td->td_ucred, &client_unp->unp_peercred);
1475         client_unp->unp_flags |= UNP_HAVEPC;
1476
1477         memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred,
1478             sizeof(server_unp->unp_peercred));
1479         server_unp->unp_flags |= UNP_HAVEPC;
1480         if (listen_unp->unp_flags & UNP_WANTCRED)
1481                 client_unp->unp_flags |= UNP_WANTCRED;
1482 }
1483
1484 static int
1485 unp_connect2(struct socket *so, struct socket *so2, int req)
1486 {
1487         struct unpcb *unp;
1488         struct unpcb *unp2;
1489
1490         unp = sotounpcb(so);
1491         KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1492         unp2 = sotounpcb(so2);
1493         KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1494
1495         UNP_LINK_WLOCK_ASSERT();
1496         UNP_PCB_LOCK_ASSERT(unp);
1497         UNP_PCB_LOCK_ASSERT(unp2);
1498
1499         if (so2->so_type != so->so_type)
1500                 return (EPROTOTYPE);
1501         unp2->unp_flags &= ~UNP_NASCENT;
1502         unp->unp_conn = unp2;
1503
1504         switch (so->so_type) {
1505         case SOCK_DGRAM:
1506                 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1507                 soisconnected(so);
1508                 break;
1509
1510         case SOCK_STREAM:
1511         case SOCK_SEQPACKET:
1512                 unp2->unp_conn = unp;
1513                 if (req == PRU_CONNECT &&
1514                     ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1515                         soisconnecting(so);
1516                 else
1517                         soisconnected(so);
1518                 soisconnected(so2);
1519                 break;
1520
1521         default:
1522                 panic("unp_connect2");
1523         }
1524         return (0);
1525 }
1526
1527 static void
1528 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1529 {
1530         struct socket *so;
1531
1532         KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1533
1534         UNP_LINK_WLOCK_ASSERT();
1535         UNP_PCB_LOCK_ASSERT(unp);
1536         UNP_PCB_LOCK_ASSERT(unp2);
1537
1538         unp->unp_conn = NULL;
1539         switch (unp->unp_socket->so_type) {
1540         case SOCK_DGRAM:
1541                 LIST_REMOVE(unp, unp_reflink);
1542                 so = unp->unp_socket;
1543                 SOCK_LOCK(so);
1544                 so->so_state &= ~SS_ISCONNECTED;
1545                 SOCK_UNLOCK(so);
1546                 break;
1547
1548         case SOCK_STREAM:
1549         case SOCK_SEQPACKET:
1550                 soisdisconnected(unp->unp_socket);
1551                 unp2->unp_conn = NULL;
1552                 soisdisconnected(unp2->unp_socket);
1553                 break;
1554         }
1555 }
1556
1557 /*
1558  * unp_pcblist() walks the global list of struct unpcb's to generate a
1559  * pointer list, bumping the refcount on each unpcb.  It then copies them out
1560  * sequentially, validating the generation number on each to see if it has
1561  * been detached.  All of this is necessary because copyout() may sleep on
1562  * disk I/O.
1563  */
1564 static int
1565 unp_pcblist(SYSCTL_HANDLER_ARGS)
1566 {
1567         int error, i, n;
1568         int freeunp;
1569         struct unpcb *unp, **unp_list;
1570         unp_gen_t gencnt;
1571         struct xunpgen *xug;
1572         struct unp_head *head;
1573         struct xunpcb *xu;
1574
1575         switch ((intptr_t)arg1) {
1576         case SOCK_STREAM:
1577                 head = &unp_shead;
1578                 break;
1579
1580         case SOCK_DGRAM:
1581                 head = &unp_dhead;
1582                 break;
1583
1584         case SOCK_SEQPACKET:
1585                 head = &unp_sphead;
1586                 break;
1587
1588         default:
1589                 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1590         }
1591
1592         /*
1593          * The process of preparing the PCB list is too time-consuming and
1594          * resource-intensive to repeat twice on every request.
1595          */
1596         if (req->oldptr == NULL) {
1597                 n = unp_count;
1598                 req->oldidx = 2 * (sizeof *xug)
1599                         + (n + n/8) * sizeof(struct xunpcb);
1600                 return (0);
1601         }
1602
1603         if (req->newptr != NULL)
1604                 return (EPERM);
1605
1606         /*
1607          * OK, now we're committed to doing something.
1608          */
1609         xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO);
1610         UNP_LIST_LOCK();
1611         gencnt = unp_gencnt;
1612         n = unp_count;
1613         UNP_LIST_UNLOCK();
1614
1615         xug->xug_len = sizeof *xug;
1616         xug->xug_count = n;
1617         xug->xug_gen = gencnt;
1618         xug->xug_sogen = so_gencnt;
1619         error = SYSCTL_OUT(req, xug, sizeof *xug);
1620         if (error) {
1621                 free(xug, M_TEMP);
1622                 return (error);
1623         }
1624
1625         unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1626
1627         UNP_LIST_LOCK();
1628         for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1629              unp = LIST_NEXT(unp, unp_link)) {
1630                 UNP_PCB_LOCK(unp);
1631                 if (unp->unp_gencnt <= gencnt) {
1632                         if (cr_cansee(req->td->td_ucred,
1633                             unp->unp_socket->so_cred)) {
1634                                 UNP_PCB_UNLOCK(unp);
1635                                 continue;
1636                         }
1637                         unp_list[i++] = unp;
1638                         unp->unp_refcount++;
1639                 }
1640                 UNP_PCB_UNLOCK(unp);
1641         }
1642         UNP_LIST_UNLOCK();
1643         n = i;                  /* In case we lost some during malloc. */
1644
1645         error = 0;
1646         xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1647         for (i = 0; i < n; i++) {
1648                 unp = unp_list[i];
1649                 UNP_PCB_LOCK(unp);
1650                 unp->unp_refcount--;
1651                 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1652                         xu->xu_len = sizeof *xu;
1653                         xu->xu_unpp = unp;
1654                         /*
1655                          * XXX - need more locking here to protect against
1656                          * connect/disconnect races for SMP.
1657                          */
1658                         if (unp->unp_addr != NULL)
1659                                 bcopy(unp->unp_addr, &xu->xu_addr,
1660                                       unp->unp_addr->sun_len);
1661                         if (unp->unp_conn != NULL &&
1662                             unp->unp_conn->unp_addr != NULL)
1663                                 bcopy(unp->unp_conn->unp_addr,
1664                                       &xu->xu_caddr,
1665                                       unp->unp_conn->unp_addr->sun_len);
1666                         bcopy(unp, &xu->xu_unp, sizeof *unp);
1667                         sotoxsocket(unp->unp_socket, &xu->xu_socket);
1668                         UNP_PCB_UNLOCK(unp);
1669                         error = SYSCTL_OUT(req, xu, sizeof *xu);
1670                 } else {
1671                         freeunp = (unp->unp_refcount == 0);
1672                         UNP_PCB_UNLOCK(unp);
1673                         if (freeunp) {
1674                                 UNP_PCB_LOCK_DESTROY(unp);
1675                                 uma_zfree(unp_zone, unp);
1676                         }
1677                 }
1678         }
1679         free(xu, M_TEMP);
1680         if (!error) {
1681                 /*
1682                  * Give the user an updated idea of our state.  If the
1683                  * generation differs from what we told her before, she knows
1684                  * that something happened while we were processing this
1685                  * request, and it might be necessary to retry.
1686                  */
1687                 xug->xug_gen = unp_gencnt;
1688                 xug->xug_sogen = so_gencnt;
1689                 xug->xug_count = unp_count;
1690                 error = SYSCTL_OUT(req, xug, sizeof *xug);
1691         }
1692         free(unp_list, M_TEMP);
1693         free(xug, M_TEMP);
1694         return (error);
1695 }
1696
1697 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1698     (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1699     "List of active local datagram sockets");
1700 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1701     (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1702     "List of active local stream sockets");
1703 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1704     CTLTYPE_OPAQUE | CTLFLAG_RD,
1705     (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1706     "List of active local seqpacket sockets");
1707
1708 static void
1709 unp_shutdown(struct unpcb *unp)
1710 {
1711         struct unpcb *unp2;
1712         struct socket *so;
1713
1714         UNP_LINK_WLOCK_ASSERT();
1715         UNP_PCB_LOCK_ASSERT(unp);
1716
1717         unp2 = unp->unp_conn;
1718         if ((unp->unp_socket->so_type == SOCK_STREAM ||
1719             (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1720                 so = unp2->unp_socket;
1721                 if (so != NULL)
1722                         socantrcvmore(so);
1723         }
1724 }
1725
1726 static void
1727 unp_drop(struct unpcb *unp)
1728 {
1729         struct socket *so = unp->unp_socket;
1730         struct unpcb *unp2;
1731
1732         UNP_LINK_WLOCK_ASSERT();
1733         UNP_PCB_LOCK_ASSERT(unp);
1734
1735         /*
1736          * Regardless of whether the socket's peer dropped the connection
1737          * with this socket by aborting or disconnecting, POSIX requires
1738          * that ECONNRESET is returned.
1739          */
1740         so->so_error = ECONNRESET;
1741         unp2 = unp->unp_conn;
1742         if (unp2 == NULL)
1743                 return;
1744         UNP_PCB_LOCK(unp2);
1745         unp_disconnect(unp, unp2);
1746         UNP_PCB_UNLOCK(unp2);
1747 }
1748
1749 static void
1750 unp_freerights(struct filedescent **fdep, int fdcount)
1751 {
1752         struct file *fp;
1753         int i;
1754
1755         KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1756
1757         for (i = 0; i < fdcount; i++) {
1758                 fp = fdep[i]->fde_file;
1759                 filecaps_free(&fdep[i]->fde_caps);
1760                 unp_discard(fp);
1761         }
1762         free(fdep[0], M_FILECAPS);
1763 }
1764
1765 static int
1766 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1767 {
1768         struct thread *td = curthread;          /* XXX */
1769         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1770         int i;
1771         int *fdp;
1772         struct filedesc *fdesc = td->td_proc->p_fd;
1773         struct filedescent **fdep;
1774         void *data;
1775         socklen_t clen = control->m_len, datalen;
1776         int error, newfds;
1777         u_int newlen;
1778
1779         UNP_LINK_UNLOCK_ASSERT();
1780
1781         error = 0;
1782         if (controlp != NULL) /* controlp == NULL => free control messages */
1783                 *controlp = NULL;
1784         while (cm != NULL) {
1785                 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1786                         error = EINVAL;
1787                         break;
1788                 }
1789                 data = CMSG_DATA(cm);
1790                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1791                 if (cm->cmsg_level == SOL_SOCKET
1792                     && cm->cmsg_type == SCM_RIGHTS) {
1793                         newfds = datalen / sizeof(*fdep);
1794                         if (newfds == 0)
1795                                 goto next;
1796                         fdep = data;
1797
1798                         /* If we're not outputting the descriptors free them. */
1799                         if (error || controlp == NULL) {
1800                                 unp_freerights(fdep, newfds);
1801                                 goto next;
1802                         }
1803                         FILEDESC_XLOCK(fdesc);
1804
1805                         /*
1806                          * Now change each pointer to an fd in the global
1807                          * table to an integer that is the index to the local
1808                          * fd table entry that we set up to point to the
1809                          * global one we are transferring.
1810                          */
1811                         newlen = newfds * sizeof(int);
1812                         *controlp = sbcreatecontrol(NULL, newlen,
1813                             SCM_RIGHTS, SOL_SOCKET);
1814                         if (*controlp == NULL) {
1815                                 FILEDESC_XUNLOCK(fdesc);
1816                                 error = E2BIG;
1817                                 unp_freerights(fdep, newfds);
1818                                 goto next;
1819                         }
1820
1821                         fdp = (int *)
1822                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1823                         if (fdallocn(td, 0, fdp, newfds) != 0) {
1824                                 FILEDESC_XUNLOCK(fdesc);
1825                                 error = EMSGSIZE;
1826                                 unp_freerights(fdep, newfds);
1827                                 m_freem(*controlp);
1828                                 *controlp = NULL;
1829                                 goto next;
1830                         }
1831                         for (i = 0; i < newfds; i++, fdp++) {
1832                                 _finstall(fdesc, fdep[i]->fde_file, *fdp,
1833                                     (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0,
1834                                     &fdep[i]->fde_caps);
1835                                 unp_externalize_fp(fdep[i]->fde_file);
1836                         }
1837
1838                         /*
1839                          * The new type indicates that the mbuf data refers to
1840                          * kernel resources that may need to be released before
1841                          * the mbuf is freed.
1842                          */
1843                         m_chtype(*controlp, MT_EXTCONTROL);
1844                         FILEDESC_XUNLOCK(fdesc);
1845                         free(fdep[0], M_FILECAPS);
1846                 } else {
1847                         /* We can just copy anything else across. */
1848                         if (error || controlp == NULL)
1849                                 goto next;
1850                         *controlp = sbcreatecontrol(NULL, datalen,
1851                             cm->cmsg_type, cm->cmsg_level);
1852                         if (*controlp == NULL) {
1853                                 error = ENOBUFS;
1854                                 goto next;
1855                         }
1856                         bcopy(data,
1857                             CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1858                             datalen);
1859                 }
1860                 controlp = &(*controlp)->m_next;
1861
1862 next:
1863                 if (CMSG_SPACE(datalen) < clen) {
1864                         clen -= CMSG_SPACE(datalen);
1865                         cm = (struct cmsghdr *)
1866                             ((caddr_t)cm + CMSG_SPACE(datalen));
1867                 } else {
1868                         clen = 0;
1869                         cm = NULL;
1870                 }
1871         }
1872
1873         m_freem(control);
1874         return (error);
1875 }
1876
1877 static void
1878 unp_zone_change(void *tag)
1879 {
1880
1881         uma_zone_set_max(unp_zone, maxsockets);
1882 }
1883
1884 static void
1885 unp_init(void)
1886 {
1887
1888 #ifdef VIMAGE
1889         if (!IS_DEFAULT_VNET(curvnet))
1890                 return;
1891 #endif
1892         unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1893             NULL, NULL, UMA_ALIGN_PTR, 0);
1894         if (unp_zone == NULL)
1895                 panic("unp_init");
1896         uma_zone_set_max(unp_zone, maxsockets);
1897         uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
1898         EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1899             NULL, EVENTHANDLER_PRI_ANY);
1900         LIST_INIT(&unp_dhead);
1901         LIST_INIT(&unp_shead);
1902         LIST_INIT(&unp_sphead);
1903         SLIST_INIT(&unp_defers);
1904         TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1905         TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1906         UNP_LINK_LOCK_INIT();
1907         UNP_LIST_LOCK_INIT();
1908         UNP_DEFERRED_LOCK_INIT();
1909 }
1910
1911 static void
1912 unp_internalize_cleanup_rights(struct mbuf *control)
1913 {
1914         struct cmsghdr *cp;
1915         struct mbuf *m;
1916         void *data;
1917         socklen_t datalen;
1918
1919         for (m = control; m != NULL; m = m->m_next) {
1920                 cp = mtod(m, struct cmsghdr *);
1921                 if (cp->cmsg_level != SOL_SOCKET ||
1922                     cp->cmsg_type != SCM_RIGHTS)
1923                         continue;
1924                 data = CMSG_DATA(cp);
1925                 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data;
1926                 unp_freerights(data, datalen / sizeof(struct filedesc *));
1927         }
1928 }
1929
1930 static int
1931 unp_internalize(struct mbuf **controlp, struct thread *td)
1932 {
1933         struct mbuf *control, **initial_controlp;
1934         struct proc *p;
1935         struct filedesc *fdesc;
1936         struct bintime *bt;
1937         struct cmsghdr *cm;
1938         struct cmsgcred *cmcred;
1939         struct filedescent *fde, **fdep, *fdev;
1940         struct file *fp;
1941         struct timeval *tv;
1942         struct timespec *ts;
1943         void *data;
1944         socklen_t clen, datalen;
1945         int i, error, *fdp, oldfds;
1946         u_int newlen;
1947
1948         UNP_LINK_UNLOCK_ASSERT();
1949
1950         p = td->td_proc;
1951         fdesc = p->p_fd;
1952         error = 0;
1953         control = *controlp;
1954         clen = control->m_len;
1955         *controlp = NULL;
1956         initial_controlp = controlp;
1957         for (cm = mtod(control, struct cmsghdr *); cm != NULL;) {
1958                 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1959                     || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
1960                         error = EINVAL;
1961                         goto out;
1962                 }
1963                 data = CMSG_DATA(cm);
1964                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1965
1966                 switch (cm->cmsg_type) {
1967                 /*
1968                  * Fill in credential information.
1969                  */
1970                 case SCM_CREDS:
1971                         *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1972                             SCM_CREDS, SOL_SOCKET);
1973                         if (*controlp == NULL) {
1974                                 error = ENOBUFS;
1975                                 goto out;
1976                         }
1977                         cmcred = (struct cmsgcred *)
1978                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1979                         cmcred->cmcred_pid = p->p_pid;
1980                         cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1981                         cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1982                         cmcred->cmcred_euid = td->td_ucred->cr_uid;
1983                         cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1984                             CMGROUP_MAX);
1985                         for (i = 0; i < cmcred->cmcred_ngroups; i++)
1986                                 cmcred->cmcred_groups[i] =
1987                                     td->td_ucred->cr_groups[i];
1988                         break;
1989
1990                 case SCM_RIGHTS:
1991                         oldfds = datalen / sizeof (int);
1992                         if (oldfds == 0)
1993                                 break;
1994                         /*
1995                          * Check that all the FDs passed in refer to legal
1996                          * files.  If not, reject the entire operation.
1997                          */
1998                         fdp = data;
1999                         FILEDESC_SLOCK(fdesc);
2000                         for (i = 0; i < oldfds; i++, fdp++) {
2001                                 fp = fget_locked(fdesc, *fdp);
2002                                 if (fp == NULL) {
2003                                         FILEDESC_SUNLOCK(fdesc);
2004                                         error = EBADF;
2005                                         goto out;
2006                                 }
2007                                 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
2008                                         FILEDESC_SUNLOCK(fdesc);
2009                                         error = EOPNOTSUPP;
2010                                         goto out;
2011                                 }
2012
2013                         }
2014
2015                         /*
2016                          * Now replace the integer FDs with pointers to the
2017                          * file structure and capability rights.
2018                          */
2019                         newlen = oldfds * sizeof(fdep[0]);
2020                         *controlp = sbcreatecontrol(NULL, newlen,
2021                             SCM_RIGHTS, SOL_SOCKET);
2022                         if (*controlp == NULL) {
2023                                 FILEDESC_SUNLOCK(fdesc);
2024                                 error = E2BIG;
2025                                 goto out;
2026                         }
2027                         fdp = data;
2028                         fdep = (struct filedescent **)
2029                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2030                         fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
2031                             M_WAITOK);
2032                         for (i = 0; i < oldfds; i++, fdev++, fdp++) {
2033                                 fde = &fdesc->fd_ofiles[*fdp];
2034                                 fdep[i] = fdev;
2035                                 fdep[i]->fde_file = fde->fde_file;
2036                                 filecaps_copy(&fde->fde_caps,
2037                                     &fdep[i]->fde_caps, true);
2038                                 unp_internalize_fp(fdep[i]->fde_file);
2039                         }
2040                         FILEDESC_SUNLOCK(fdesc);
2041                         break;
2042
2043                 case SCM_TIMESTAMP:
2044                         *controlp = sbcreatecontrol(NULL, sizeof(*tv),
2045                             SCM_TIMESTAMP, SOL_SOCKET);
2046                         if (*controlp == NULL) {
2047                                 error = ENOBUFS;
2048                                 goto out;
2049                         }
2050                         tv = (struct timeval *)
2051                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2052                         microtime(tv);
2053                         break;
2054
2055                 case SCM_BINTIME:
2056                         *controlp = sbcreatecontrol(NULL, sizeof(*bt),
2057                             SCM_BINTIME, SOL_SOCKET);
2058                         if (*controlp == NULL) {
2059                                 error = ENOBUFS;
2060                                 goto out;
2061                         }
2062                         bt = (struct bintime *)
2063                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2064                         bintime(bt);
2065                         break;
2066
2067                 case SCM_REALTIME:
2068                         *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2069                             SCM_REALTIME, SOL_SOCKET);
2070                         if (*controlp == NULL) {
2071                                 error = ENOBUFS;
2072                                 goto out;
2073                         }
2074                         ts = (struct timespec *)
2075                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2076                         nanotime(ts);
2077                         break;
2078
2079                 case SCM_MONOTONIC:
2080                         *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2081                             SCM_MONOTONIC, SOL_SOCKET);
2082                         if (*controlp == NULL) {
2083                                 error = ENOBUFS;
2084                                 goto out;
2085                         }
2086                         ts = (struct timespec *)
2087                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2088                         nanouptime(ts);
2089                         break;
2090
2091                 default:
2092                         error = EINVAL;
2093                         goto out;
2094                 }
2095
2096                 controlp = &(*controlp)->m_next;
2097                 if (CMSG_SPACE(datalen) < clen) {
2098                         clen -= CMSG_SPACE(datalen);
2099                         cm = (struct cmsghdr *)
2100                             ((caddr_t)cm + CMSG_SPACE(datalen));
2101                 } else {
2102                         clen = 0;
2103                         cm = NULL;
2104                 }
2105         }
2106
2107 out:
2108         if (error != 0 && initial_controlp != NULL)
2109                 unp_internalize_cleanup_rights(*initial_controlp);
2110         m_freem(control);
2111         return (error);
2112 }
2113
2114 static struct mbuf *
2115 unp_addsockcred(struct thread *td, struct mbuf *control)
2116 {
2117         struct mbuf *m, *n, *n_prev;
2118         struct sockcred *sc;
2119         const struct cmsghdr *cm;
2120         int ngroups;
2121         int i;
2122
2123         ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2124         m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2125         if (m == NULL)
2126                 return (control);
2127
2128         sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2129         sc->sc_uid = td->td_ucred->cr_ruid;
2130         sc->sc_euid = td->td_ucred->cr_uid;
2131         sc->sc_gid = td->td_ucred->cr_rgid;
2132         sc->sc_egid = td->td_ucred->cr_gid;
2133         sc->sc_ngroups = ngroups;
2134         for (i = 0; i < sc->sc_ngroups; i++)
2135                 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2136
2137         /*
2138          * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2139          * created SCM_CREDS control message (struct sockcred) has another
2140          * format.
2141          */
2142         if (control != NULL)
2143                 for (n = control, n_prev = NULL; n != NULL;) {
2144                         cm = mtod(n, struct cmsghdr *);
2145                         if (cm->cmsg_level == SOL_SOCKET &&
2146                             cm->cmsg_type == SCM_CREDS) {
2147                                 if (n_prev == NULL)
2148                                         control = n->m_next;
2149                                 else
2150                                         n_prev->m_next = n->m_next;
2151                                 n = m_free(n);
2152                         } else {
2153                                 n_prev = n;
2154                                 n = n->m_next;
2155                         }
2156                 }
2157
2158         /* Prepend it to the head. */
2159         m->m_next = control;
2160         return (m);
2161 }
2162
2163 static struct unpcb *
2164 fptounp(struct file *fp)
2165 {
2166         struct socket *so;
2167
2168         if (fp->f_type != DTYPE_SOCKET)
2169                 return (NULL);
2170         if ((so = fp->f_data) == NULL)
2171                 return (NULL);
2172         if (so->so_proto->pr_domain != &localdomain)
2173                 return (NULL);
2174         return sotounpcb(so);
2175 }
2176
2177 static void
2178 unp_discard(struct file *fp)
2179 {
2180         struct unp_defer *dr;
2181
2182         if (unp_externalize_fp(fp)) {
2183                 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2184                 dr->ud_fp = fp;
2185                 UNP_DEFERRED_LOCK();
2186                 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2187                 UNP_DEFERRED_UNLOCK();
2188                 atomic_add_int(&unp_defers_count, 1);
2189                 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2190         } else
2191                 (void) closef(fp, (struct thread *)NULL);
2192 }
2193
2194 static void
2195 unp_process_defers(void *arg __unused, int pending)
2196 {
2197         struct unp_defer *dr;
2198         SLIST_HEAD(, unp_defer) drl;
2199         int count;
2200
2201         SLIST_INIT(&drl);
2202         for (;;) {
2203                 UNP_DEFERRED_LOCK();
2204                 if (SLIST_FIRST(&unp_defers) == NULL) {
2205                         UNP_DEFERRED_UNLOCK();
2206                         break;
2207                 }
2208                 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2209                 UNP_DEFERRED_UNLOCK();
2210                 count = 0;
2211                 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2212                         SLIST_REMOVE_HEAD(&drl, ud_link);
2213                         closef(dr->ud_fp, NULL);
2214                         free(dr, M_TEMP);
2215                         count++;
2216                 }
2217                 atomic_add_int(&unp_defers_count, -count);
2218         }
2219 }
2220
2221 static void
2222 unp_internalize_fp(struct file *fp)
2223 {
2224         struct unpcb *unp;
2225
2226         UNP_LINK_WLOCK();
2227         if ((unp = fptounp(fp)) != NULL) {
2228                 unp->unp_file = fp;
2229                 unp->unp_msgcount++;
2230         }
2231         fhold(fp);
2232         unp_rights++;
2233         UNP_LINK_WUNLOCK();
2234 }
2235
2236 static int
2237 unp_externalize_fp(struct file *fp)
2238 {
2239         struct unpcb *unp;
2240         int ret;
2241
2242         UNP_LINK_WLOCK();
2243         if ((unp = fptounp(fp)) != NULL) {
2244                 unp->unp_msgcount--;
2245                 ret = 1;
2246         } else
2247                 ret = 0;
2248         unp_rights--;
2249         UNP_LINK_WUNLOCK();
2250         return (ret);
2251 }
2252
2253 /*
2254  * unp_defer indicates whether additional work has been defered for a future
2255  * pass through unp_gc().  It is thread local and does not require explicit
2256  * synchronization.
2257  */
2258 static int      unp_marked;
2259 static int      unp_unreachable;
2260
2261 static void
2262 unp_accessable(struct filedescent **fdep, int fdcount)
2263 {
2264         struct unpcb *unp;
2265         struct file *fp;
2266         int i;
2267
2268         for (i = 0; i < fdcount; i++) {
2269                 fp = fdep[i]->fde_file;
2270                 if ((unp = fptounp(fp)) == NULL)
2271                         continue;
2272                 if (unp->unp_gcflag & UNPGC_REF)
2273                         continue;
2274                 unp->unp_gcflag &= ~UNPGC_DEAD;
2275                 unp->unp_gcflag |= UNPGC_REF;
2276                 unp_marked++;
2277         }
2278 }
2279
2280 static void
2281 unp_gc_process(struct unpcb *unp)
2282 {
2283         struct socket *soa;
2284         struct socket *so;
2285         struct file *fp;
2286
2287         /* Already processed. */
2288         if (unp->unp_gcflag & UNPGC_SCANNED)
2289                 return;
2290         fp = unp->unp_file;
2291
2292         /*
2293          * Check for a socket potentially in a cycle.  It must be in a
2294          * queue as indicated by msgcount, and this must equal the file
2295          * reference count.  Note that when msgcount is 0 the file is NULL.
2296          */
2297         if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2298             unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2299                 unp->unp_gcflag |= UNPGC_DEAD;
2300                 unp_unreachable++;
2301                 return;
2302         }
2303
2304         /*
2305          * Mark all sockets we reference with RIGHTS.
2306          */
2307         so = unp->unp_socket;
2308         if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2309                 SOCKBUF_LOCK(&so->so_rcv);
2310                 unp_scan(so->so_rcv.sb_mb, unp_accessable);
2311                 SOCKBUF_UNLOCK(&so->so_rcv);
2312         }
2313
2314         /*
2315          * Mark all sockets in our accept queue.
2316          */
2317         ACCEPT_LOCK();
2318         TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2319                 if ((sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS) != 0)
2320                         continue;
2321                 SOCKBUF_LOCK(&soa->so_rcv);
2322                 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2323                 SOCKBUF_UNLOCK(&soa->so_rcv);
2324         }
2325         ACCEPT_UNLOCK();
2326         unp->unp_gcflag |= UNPGC_SCANNED;
2327 }
2328
2329 static int unp_recycled;
2330 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2331     "Number of unreachable sockets claimed by the garbage collector.");
2332
2333 static int unp_taskcount;
2334 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2335     "Number of times the garbage collector has run.");
2336
2337 static void
2338 unp_gc(__unused void *arg, int pending)
2339 {
2340         struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2341                                     NULL };
2342         struct unp_head **head;
2343         struct file *f, **unref;
2344         struct unpcb *unp;
2345         int i, total;
2346
2347         unp_taskcount++;
2348         UNP_LIST_LOCK();
2349         /*
2350          * First clear all gc flags from previous runs, apart from
2351          * UNPGC_IGNORE_RIGHTS.
2352          */
2353         for (head = heads; *head != NULL; head++)
2354                 LIST_FOREACH(unp, *head, unp_link)
2355                         unp->unp_gcflag =
2356                             (unp->unp_gcflag & UNPGC_IGNORE_RIGHTS);
2357
2358         /*
2359          * Scan marking all reachable sockets with UNPGC_REF.  Once a socket
2360          * is reachable all of the sockets it references are reachable.
2361          * Stop the scan once we do a complete loop without discovering
2362          * a new reachable socket.
2363          */
2364         do {
2365                 unp_unreachable = 0;
2366                 unp_marked = 0;
2367                 for (head = heads; *head != NULL; head++)
2368                         LIST_FOREACH(unp, *head, unp_link)
2369                                 unp_gc_process(unp);
2370         } while (unp_marked);
2371         UNP_LIST_UNLOCK();
2372         if (unp_unreachable == 0)
2373                 return;
2374
2375         /*
2376          * Allocate space for a local list of dead unpcbs.
2377          */
2378         unref = malloc(unp_unreachable * sizeof(struct file *),
2379             M_TEMP, M_WAITOK);
2380
2381         /*
2382          * Iterate looking for sockets which have been specifically marked
2383          * as as unreachable and store them locally.
2384          */
2385         UNP_LINK_RLOCK();
2386         UNP_LIST_LOCK();
2387         for (total = 0, head = heads; *head != NULL; head++)
2388                 LIST_FOREACH(unp, *head, unp_link)
2389                         if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2390                                 f = unp->unp_file;
2391                                 if (unp->unp_msgcount == 0 || f == NULL ||
2392                                     f->f_count != unp->unp_msgcount)
2393                                         continue;
2394                                 unref[total++] = f;
2395                                 fhold(f);
2396                                 KASSERT(total <= unp_unreachable,
2397                                     ("unp_gc: incorrect unreachable count."));
2398                         }
2399         UNP_LIST_UNLOCK();
2400         UNP_LINK_RUNLOCK();
2401
2402         /*
2403          * Now flush all sockets, free'ing rights.  This will free the
2404          * struct files associated with these sockets but leave each socket
2405          * with one remaining ref.
2406          */
2407         for (i = 0; i < total; i++) {
2408                 struct socket *so;
2409
2410                 so = unref[i]->f_data;
2411                 CURVNET_SET(so->so_vnet);
2412                 sorflush(so);
2413                 CURVNET_RESTORE();
2414         }
2415
2416         /*
2417          * And finally release the sockets so they can be reclaimed.
2418          */
2419         for (i = 0; i < total; i++)
2420                 fdrop(unref[i], NULL);
2421         unp_recycled += total;
2422         free(unref, M_TEMP);
2423 }
2424
2425 static void
2426 unp_dispose(struct mbuf *m)
2427 {
2428
2429         if (m)
2430                 unp_scan(m, unp_freerights);
2431 }
2432
2433 /*
2434  * Synchronize against unp_gc, which can trip over data as we are freeing it.
2435  */
2436 static void
2437 unp_dispose_so(struct socket *so)
2438 {
2439         struct unpcb *unp;
2440
2441         unp = sotounpcb(so);
2442         UNP_LIST_LOCK();
2443         unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2444         UNP_LIST_UNLOCK();
2445         unp_dispose(so->so_rcv.sb_mb);
2446 }
2447
2448 static void
2449 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2450 {
2451         struct mbuf *m;
2452         struct cmsghdr *cm;
2453         void *data;
2454         socklen_t clen, datalen;
2455
2456         while (m0 != NULL) {
2457                 for (m = m0; m; m = m->m_next) {
2458                         if (m->m_type != MT_CONTROL)
2459                                 continue;
2460
2461                         cm = mtod(m, struct cmsghdr *);
2462                         clen = m->m_len;
2463
2464                         while (cm != NULL) {
2465                                 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2466                                         break;
2467
2468                                 data = CMSG_DATA(cm);
2469                                 datalen = (caddr_t)cm + cm->cmsg_len
2470                                     - (caddr_t)data;
2471
2472                                 if (cm->cmsg_level == SOL_SOCKET &&
2473                                     cm->cmsg_type == SCM_RIGHTS) {
2474                                         (*op)(data, datalen /
2475                                             sizeof(struct filedescent *));
2476                                 }
2477
2478                                 if (CMSG_SPACE(datalen) < clen) {
2479                                         clen -= CMSG_SPACE(datalen);
2480                                         cm = (struct cmsghdr *)
2481                                             ((caddr_t)cm + CMSG_SPACE(datalen));
2482                                 } else {
2483                                         clen = 0;
2484                                         cm = NULL;
2485                                 }
2486                         }
2487                 }
2488                 m0 = m0->m_nextpkt;
2489         }
2490 }
2491
2492 /*
2493  * A helper function called by VFS before socket-type vnode reclamation.
2494  * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2495  * use count.
2496  */
2497 void
2498 vfs_unp_reclaim(struct vnode *vp)
2499 {
2500         struct socket *so;
2501         struct unpcb *unp;
2502         int active;
2503
2504         ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2505         KASSERT(vp->v_type == VSOCK,
2506             ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2507
2508         active = 0;
2509         UNP_LINK_WLOCK();
2510         VOP_UNP_CONNECT(vp, &so);
2511         if (so == NULL)
2512                 goto done;
2513         unp = sotounpcb(so);
2514         if (unp == NULL)
2515                 goto done;
2516         UNP_PCB_LOCK(unp);
2517         if (unp->unp_vnode == vp) {
2518                 VOP_UNP_DETACH(vp);
2519                 unp->unp_vnode = NULL;
2520                 active = 1;
2521         }
2522         UNP_PCB_UNLOCK(unp);
2523 done:
2524         UNP_LINK_WUNLOCK();
2525         if (active)
2526                 vunref(vp);
2527 }
2528
2529 #ifdef DDB
2530 static void
2531 db_print_indent(int indent)
2532 {
2533         int i;
2534
2535         for (i = 0; i < indent; i++)
2536                 db_printf(" ");
2537 }
2538
2539 static void
2540 db_print_unpflags(int unp_flags)
2541 {
2542         int comma;
2543
2544         comma = 0;
2545         if (unp_flags & UNP_HAVEPC) {
2546                 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2547                 comma = 1;
2548         }
2549         if (unp_flags & UNP_HAVEPCCACHED) {
2550                 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2551                 comma = 1;
2552         }
2553         if (unp_flags & UNP_WANTCRED) {
2554                 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2555                 comma = 1;
2556         }
2557         if (unp_flags & UNP_CONNWAIT) {
2558                 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2559                 comma = 1;
2560         }
2561         if (unp_flags & UNP_CONNECTING) {
2562                 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2563                 comma = 1;
2564         }
2565         if (unp_flags & UNP_BINDING) {
2566                 db_printf("%sUNP_BINDING", comma ? ", " : "");
2567                 comma = 1;
2568         }
2569 }
2570
2571 static void
2572 db_print_xucred(int indent, struct xucred *xu)
2573 {
2574         int comma, i;
2575
2576         db_print_indent(indent);
2577         db_printf("cr_version: %u   cr_uid: %u   cr_ngroups: %d\n",
2578             xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2579         db_print_indent(indent);
2580         db_printf("cr_groups: ");
2581         comma = 0;
2582         for (i = 0; i < xu->cr_ngroups; i++) {
2583                 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2584                 comma = 1;
2585         }
2586         db_printf("\n");
2587 }
2588
2589 static void
2590 db_print_unprefs(int indent, struct unp_head *uh)
2591 {
2592         struct unpcb *unp;
2593         int counter;
2594
2595         counter = 0;
2596         LIST_FOREACH(unp, uh, unp_reflink) {
2597                 if (counter % 4 == 0)
2598                         db_print_indent(indent);
2599                 db_printf("%p  ", unp);
2600                 if (counter % 4 == 3)
2601                         db_printf("\n");
2602                 counter++;
2603         }
2604         if (counter != 0 && counter % 4 != 0)
2605                 db_printf("\n");
2606 }
2607
2608 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2609 {
2610         struct unpcb *unp;
2611
2612         if (!have_addr) {
2613                 db_printf("usage: show unpcb <addr>\n");
2614                 return;
2615         }
2616         unp = (struct unpcb *)addr;
2617
2618         db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
2619             unp->unp_vnode);
2620
2621         db_printf("unp_ino: %ju   unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2622             unp->unp_conn);
2623
2624         db_printf("unp_refs:\n");
2625         db_print_unprefs(2, &unp->unp_refs);
2626
2627         /* XXXRW: Would be nice to print the full address, if any. */
2628         db_printf("unp_addr: %p\n", unp->unp_addr);
2629
2630         db_printf("unp_gencnt: %llu\n",
2631             (unsigned long long)unp->unp_gencnt);
2632
2633         db_printf("unp_flags: %x (", unp->unp_flags);
2634         db_print_unpflags(unp->unp_flags);
2635         db_printf(")\n");
2636
2637         db_printf("unp_peercred:\n");
2638         db_print_xucred(2, &unp->unp_peercred);
2639
2640         db_printf("unp_refcount: %u\n", unp->unp_refcount);
2641 }
2642 #endif