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) 2004-2008 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  *      SEQPACKET, 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 #include "opt_mac.h"
  63
  64 #include <sys/param.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/jail.h>
  72 #include <sys/kernel.h>
  73 #include <sys/lock.h>
  74 #include <sys/mbuf.h>
  75 #include <sys/mount.h>
  76 #include <sys/mutex.h>
  77 #include <sys/namei.h>
  78 #include <sys/proc.h>
  79 #include <sys/protosw.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 #ifdef DDB
  95 #include <ddb/ddb.h>
  96 #endif
  97
  98 #include <security/mac/mac_framework.h>
  99
 100 #include <vm/uma.h>
 101
 102 static uma_zone_t       unp_zone;
 103 static unp_gen_t        unp_gencnt;
 104 static u_int            unp_count;      /* Count of local sockets. */
 105 static ino_t            unp_ino;        /* Prototype for fake inode numbers. */
 106 static int              unp_rights;     /* File descriptors in flight. */
 107 static struct unp_head  unp_shead;      /* List of local stream sockets. */
 108 static struct unp_head  unp_dhead;      /* List of local datagram sockets. */
 109
 110 static const struct sockaddr    sun_noname = { sizeof(sun_noname), AF_LOCAL };
 111
 112 /*
 113  * Garbage collection of cyclic file descriptor/socket references occurs
 114  * asynchronously in a taskqueue context in order to avoid recursion and
 115  * reentrance in the UNIX domain socket, file descriptor, and socket layer
 116  * code.  See unp_gc() for a full description.
 117  */
 118 static struct task      unp_gc_task;
 119
 120 /*
 121  * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
 122  * stream sockets, although the total for sender and receiver is actually
 123  * only PIPSIZ.
 124  *
 125  * Datagram sockets really use the sendspace as the maximum datagram size,
 126  * and don't really want to reserve the sendspace.  Their recvspace should be
 127  * large enough for at least one max-size datagram plus address.
 128  */
 129 #ifndef PIPSIZ
 130 #define PIPSIZ  8192
 131 #endif
 132 static u_long   unpst_sendspace = PIPSIZ;
 133 static u_long   unpst_recvspace = PIPSIZ;
 134 static u_long   unpdg_sendspace = 2*1024;       /* really max datagram size */
 135 static u_long   unpdg_recvspace = 4*1024;
 136
 137 SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
 138 SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, "SOCK_STREAM");
 139 SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
 140
 141 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
 142            &unpst_sendspace, 0, "");
 143 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
 144            &unpst_recvspace, 0, "");
 145 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
 146            &unpdg_sendspace, 0, "");
 147 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
 148            &unpdg_recvspace, 0, "");
 149 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
 150
 151 /*-
 152  * Locking and synchronization:
 153  *
 154  * The global UNIX domain socket rwlock (unp_global_rwlock) protects all
 155  * global variables, including the linked lists tracking the set of allocated
 156  * UNIX domain sockets.  The global rwlock also serves to prevent deadlock
 157  * when more than one PCB lock is acquired at a time (i.e., during
 158  * connect()).  Finally, the global rwlock protects uncounted references from
 159  * vnodes to sockets bound to those vnodes: to safely dereference the
 160  * v_socket pointer, the global rwlock must be held while a full reference is
 161  * acquired.
 162  *
 163  * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
 164  * allocated in pru_attach() and freed in pru_detach().  The validity of that
 165  * pointer is an invariant, so no lock is required to dereference the so_pcb
 166  * pointer if a valid socket reference is held by the caller.  In practice,
 167  * this is always true during operations performed on a socket.  Each unpcb
 168  * has a back-pointer to its socket, unp_socket, which will be stable under
 169  * the same circumstances.
 170  *
 171  * This pointer may only be safely dereferenced as long as a valid reference
 172  * to the unpcb is held.  Typically, this reference will be from the socket,
 173  * or from another unpcb when the referring unpcb's lock is held (in order
 174  * that the reference not be invalidated during use).  For example, to follow
 175  * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
 176  * as unp_socket remains valid as long as the reference to unp_conn is valid.
 177  *
 178  * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx.  Individual
 179  * atomic reads without the lock may be performed "lockless", but more
 180  * complex reads and read-modify-writes require the mutex to be held.  No
 181  * lock order is defined between unpcb locks -- multiple unpcb locks may be
 182  * acquired at the same time only when holding the global UNIX domain socket
 183  * rwlock exclusively, which prevents deadlocks.
 184  *
 185  * Blocking with UNIX domain sockets is a tricky issue: unlike most network
 186  * protocols, bind() is a non-atomic operation, and connect() requires
 187  * potential sleeping in the protocol, due to potentially waiting on local or
 188  * distributed file systems.  We try to separate "lookup" operations, which
 189  * may sleep, and the IPC operations themselves, which typically can occur
 190  * with relative atomicity as locks can be held over the entire operation.
 191  *
 192  * Another tricky issue is simultaneous multi-threaded or multi-process
 193  * access to a single UNIX domain socket.  These are handled by the flags
 194  * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
 195  * binding, both of which involve dropping UNIX domain socket locks in order
 196  * to perform namei() and other file system operations.
 197  */
 198 static struct rwlock    unp_global_rwlock;
 199
 200 #define UNP_GLOBAL_LOCK_INIT()          rw_init(&unp_global_rwlock,     \
 201                                             "unp_global_rwlock")
 202
 203 #define UNP_GLOBAL_LOCK_ASSERT()        rw_assert(&unp_global_rwlock,   \
 204                                             RA_LOCKED)
 205 #define UNP_GLOBAL_UNLOCK_ASSERT()      rw_assert(&unp_global_rwlock,   \
 206                                             RA_UNLOCKED)
 207
 208 #define UNP_GLOBAL_WLOCK()              rw_wlock(&unp_global_rwlock)
 209 #define UNP_GLOBAL_WUNLOCK()            rw_wunlock(&unp_global_rwlock)
 210 #define UNP_GLOBAL_WLOCK_ASSERT()       rw_assert(&unp_global_rwlock,   \
 211                                             RA_WLOCKED)
 212 #define UNP_GLOBAL_WOWNED()             rw_wowned(&unp_global_rwlock)
 213
 214 #define UNP_GLOBAL_RLOCK()              rw_rlock(&unp_global_rwlock)
 215 #define UNP_GLOBAL_RUNLOCK()            rw_runlock(&unp_global_rwlock)
 216 #define UNP_GLOBAL_RLOCK_ASSERT()       rw_assert(&unp_global_rwlock,   \
 217                                             RA_RLOCKED)
 218
 219 #define UNP_PCB_LOCK_INIT(unp)          mtx_init(&(unp)->unp_mtx,       \
 220                                             "unp_mtx", "unp_mtx",       \
 221                                             MTX_DUPOK|MTX_DEF|MTX_RECURSE)
 222 #define UNP_PCB_LOCK_DESTROY(unp)       mtx_destroy(&(unp)->unp_mtx)
 223 #define UNP_PCB_LOCK(unp)               mtx_lock(&(unp)->unp_mtx)
 224 #define UNP_PCB_UNLOCK(unp)             mtx_unlock(&(unp)->unp_mtx)
 225 #define UNP_PCB_LOCK_ASSERT(unp)        mtx_assert(&(unp)->unp_mtx, MA_OWNED)
 226
 227 static int      uipc_connect2(struct socket *, struct socket *);
 228 static int      uipc_ctloutput(struct socket *, struct sockopt *);
 229 static int      unp_connect(struct socket *, struct sockaddr *,
 230                     struct thread *);
 231 static int      unp_connect2(struct socket *so, struct socket *so2, int);
 232 static void     unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
 233 static void     unp_dispose(struct mbuf *);
 234 static void     unp_shutdown(struct unpcb *);
 235 static void     unp_drop(struct unpcb *, int);
 236 static void     unp_gc(__unused void *, int);
 237 static void     unp_scan(struct mbuf *, void (*)(struct file *));
 238 static void     unp_mark(struct file *);
 239 static void     unp_discard(struct file *);
 240 static void     unp_freerights(struct file **, int);
 241 static void     unp_init(void);
 242 static int      unp_internalize(struct mbuf **, struct thread *);
 243 static int      unp_externalize(struct mbuf *, struct mbuf **);
 244 static struct mbuf      *unp_addsockcred(struct thread *, struct mbuf *);
 245
 246 /*
 247  * Definitions of protocols supported in the LOCAL domain.
 248  */
 249 static struct domain localdomain;
 250 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
 251 static struct protosw localsw[] = {
 252 {
 253         .pr_type =              SOCK_STREAM,
 254         .pr_domain =            &localdomain,
 255         .pr_flags =             PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
 256         .pr_ctloutput =         &uipc_ctloutput,
 257         .pr_usrreqs =           &uipc_usrreqs_stream
 258 },
 259 {
 260         .pr_type =              SOCK_DGRAM,
 261         .pr_domain =            &localdomain,
 262         .pr_flags =             PR_ATOMIC|PR_ADDR|PR_RIGHTS,
 263         .pr_usrreqs =           &uipc_usrreqs_dgram
 264 },
 265 };
 266
 267 static struct domain localdomain = {
 268         .dom_family =           AF_LOCAL,
 269         .dom_name =             "local",
 270         .dom_init =             unp_init,
 271         .dom_externalize =      unp_externalize,
 272         .dom_dispose =          unp_dispose,
 273         .dom_protosw =          localsw,
 274         .dom_protoswNPROTOSW =  &localsw[sizeof(localsw)/sizeof(localsw[0])]
 275 };
 276 DOMAIN_SET(local);
 277
 278 static void
 279 uipc_abort(struct socket *so)
 280 {
 281         struct unpcb *unp, *unp2;
 282
 283         unp = sotounpcb(so);
 284         KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
 285
 286         UNP_GLOBAL_WLOCK();
 287         UNP_PCB_LOCK(unp);
 288         unp2 = unp->unp_conn;
 289         if (unp2 != NULL) {
 290                 UNP_PCB_LOCK(unp2);
 291                 unp_drop(unp2, ECONNABORTED);
 292                 UNP_PCB_UNLOCK(unp2);
 293         }
 294         UNP_PCB_UNLOCK(unp);
 295         UNP_GLOBAL_WUNLOCK();
 296 }
 297
 298 static int
 299 uipc_accept(struct socket *so, struct sockaddr **nam)
 300 {
 301         struct unpcb *unp, *unp2;
 302         const struct sockaddr *sa;
 303
 304         /*
 305          * Pass back name of connected socket, if it was bound and we are
 306          * still connected (our peer may have closed already!).
 307          */
 308         unp = sotounpcb(so);
 309         KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
 310
 311         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 312         UNP_GLOBAL_RLOCK();
 313         unp2 = unp->unp_conn;
 314         if (unp2 != NULL && unp2->unp_addr != NULL) {
 315                 UNP_PCB_LOCK(unp2);
 316                 sa = (struct sockaddr *) unp2->unp_addr;
 317                 bcopy(sa, *nam, sa->sa_len);
 318                 UNP_PCB_UNLOCK(unp2);
 319         } else {
 320                 sa = &sun_noname;
 321                 bcopy(sa, *nam, sa->sa_len);
 322         }
 323         UNP_GLOBAL_RUNLOCK();
 324         return (0);
 325 }
 326
 327 static int
 328 uipc_attach(struct socket *so, int proto, struct thread *td)
 329 {
 330         u_long sendspace, recvspace;
 331         struct unpcb *unp;
 332         int error, locked;
 333
 334         KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
 335         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
 336                 switch (so->so_type) {
 337                 case SOCK_STREAM:
 338                         sendspace = unpst_sendspace;
 339                         recvspace = unpst_recvspace;
 340                         break;
 341
 342                 case SOCK_DGRAM:
 343                         sendspace = unpdg_sendspace;
 344                         recvspace = unpdg_recvspace;
 345                         break;
 346
 347                 default:
 348                         panic("uipc_attach");
 349                 }
 350                 error = soreserve(so, sendspace, recvspace);
 351                 if (error)
 352                         return (error);
 353         }
 354         unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
 355         if (unp == NULL)
 356                 return (ENOBUFS);
 357         LIST_INIT(&unp->unp_refs);
 358         UNP_PCB_LOCK_INIT(unp);
 359         unp->unp_socket = so;
 360         so->so_pcb = unp;
 361         unp->unp_refcount = 1;
 362
 363         /*
 364          * uipc_attach() may be called indirectly from within the UNIX domain
 365          * socket code via sonewconn() in unp_connect().  Since rwlocks can
 366          * not be recursed, we do the closest thing.
 367          */
 368         locked = 0;
 369         if (!UNP_GLOBAL_WOWNED()) {
 370                 UNP_GLOBAL_WLOCK();
 371                 locked = 1;
 372         }
 373         unp->unp_gencnt = ++unp_gencnt;
 374         unp_count++;
 375         LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead : &unp_shead,
 376             unp, unp_link);
 377         if (locked)
 378                 UNP_GLOBAL_WUNLOCK();
 379
 380         return (0);
 381 }
 382
 383 static int
 384 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 385 {
 386         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
 387         struct vattr vattr;
 388         int error, namelen, vfslocked;
 389         struct nameidata nd;
 390         struct unpcb *unp;
 391         struct vnode *vp;
 392         struct mount *mp;
 393         char *buf;
 394
 395         unp = sotounpcb(so);
 396         KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
 397
 398         namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
 399         if (namelen <= 0)
 400                 return (EINVAL);
 401
 402         /*
 403          * We don't allow simultaneous bind() calls on a single UNIX domain
 404          * socket, so flag in-progress operations, and return an error if an
 405          * operation is already in progress.
 406          *
 407          * Historically, we have not allowed a socket to be rebound, so this
 408          * also returns an error.  Not allowing re-binding simplifies the
 409          * implementation and avoids a great many possible failure modes.
 410          */
 411         UNP_PCB_LOCK(unp);
 412         if (unp->unp_vnode != NULL) {
 413                 UNP_PCB_UNLOCK(unp);
 414                 return (EINVAL);
 415         }
 416         if (unp->unp_flags & UNP_BINDING) {
 417                 UNP_PCB_UNLOCK(unp);
 418                 return (EALREADY);
 419         }
 420         unp->unp_flags |= UNP_BINDING;
 421         UNP_PCB_UNLOCK(unp);
 422
 423         buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
 424         strlcpy(buf, soun->sun_path, namelen + 1);
 425
 426 restart:
 427         vfslocked = 0;
 428         NDINIT(&nd, CREATE, MPSAFE | NOFOLLOW | LOCKPARENT | SAVENAME,
 429             UIO_SYSSPACE, buf, td);
 430 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
 431         error = namei(&nd);
 432         if (error)
 433                 goto error;
 434         vp = nd.ni_vp;
 435         vfslocked = NDHASGIANT(&nd);
 436         if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
 437                 NDFREE(&nd, NDF_ONLY_PNBUF);
 438                 if (nd.ni_dvp == vp)
 439                         vrele(nd.ni_dvp);
 440                 else
 441                         vput(nd.ni_dvp);
 442                 if (vp != NULL) {
 443                         vrele(vp);
 444                         error = EADDRINUSE;
 445                         goto error;
 446                 }
 447                 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
 448                 if (error)
 449                         goto error;
 450                 VFS_UNLOCK_GIANT(vfslocked);
 451                 goto restart;
 452         }
 453         VATTR_NULL(&vattr);
 454         vattr.va_type = VSOCK;
 455         vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
 456 #ifdef MAC
 457         error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
 458             &vattr);
 459 #endif
 460         if (error == 0) {
 461                 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE);
 462                 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
 463         }
 464         NDFREE(&nd, NDF_ONLY_PNBUF);
 465         vput(nd.ni_dvp);
 466         if (error) {
 467                 vn_finished_write(mp);
 468                 goto error;
 469         }
 470         vp = nd.ni_vp;
 471         ASSERT_VOP_ELOCKED(vp, "uipc_bind");
 472         soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
 473
 474         UNP_GLOBAL_WLOCK();
 475         UNP_PCB_LOCK(unp);
 476         vp->v_socket = unp->unp_socket;
 477         unp->unp_vnode = vp;
 478         unp->unp_addr = soun;
 479         unp->unp_flags &= ~UNP_BINDING;
 480         UNP_PCB_UNLOCK(unp);
 481         UNP_GLOBAL_WUNLOCK();
 482         VOP_UNLOCK(vp, 0, td);
 483         vn_finished_write(mp);
 484         VFS_UNLOCK_GIANT(vfslocked);
 485         free(buf, M_TEMP);
 486         return (0);
 487
 488 error:
 489         VFS_UNLOCK_GIANT(vfslocked);
 490         UNP_PCB_LOCK(unp);
 491         unp->unp_flags &= ~UNP_BINDING;
 492         UNP_PCB_UNLOCK(unp);
 493         free(buf, M_TEMP);
 494         return (error);
 495 }
 496
 497 static int
 498 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 499 {
 500         int error;
 501
 502         KASSERT(td == curthread, ("uipc_connect: td != curthread"));
 503         UNP_GLOBAL_WLOCK();
 504         error = unp_connect(so, nam, td);
 505         UNP_GLOBAL_WUNLOCK();
 506         return (error);
 507 }
 508
 509 static void
 510 uipc_close(struct socket *so)
 511 {
 512         struct unpcb *unp, *unp2;
 513
 514         unp = sotounpcb(so);
 515         KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
 516
 517         UNP_GLOBAL_WLOCK();
 518         UNP_PCB_LOCK(unp);
 519         unp2 = unp->unp_conn;
 520         if (unp2 != NULL) {
 521                 UNP_PCB_LOCK(unp2);
 522                 unp_disconnect(unp, unp2);
 523                 UNP_PCB_UNLOCK(unp2);
 524         }
 525         UNP_PCB_UNLOCK(unp);
 526         UNP_GLOBAL_WUNLOCK();
 527 }
 528
 529 static int
 530 uipc_connect2(struct socket *so1, struct socket *so2)
 531 {
 532         struct unpcb *unp, *unp2;
 533         int error;
 534
 535         UNP_GLOBAL_WLOCK();
 536         unp = so1->so_pcb;
 537         KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
 538         UNP_PCB_LOCK(unp);
 539         unp2 = so2->so_pcb;
 540         KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
 541         UNP_PCB_LOCK(unp2);
 542         error = unp_connect2(so1, so2, PRU_CONNECT2);
 543         UNP_PCB_UNLOCK(unp2);
 544         UNP_PCB_UNLOCK(unp);
 545         UNP_GLOBAL_WUNLOCK();
 546         return (error);
 547 }
 548
 549 static void
 550 uipc_detach(struct socket *so)
 551 {
 552         struct unpcb *unp, *unp2;
 553         struct sockaddr_un *saved_unp_addr;
 554         struct vnode *vp;
 555         int freeunp, local_unp_rights;
 556
 557         unp = sotounpcb(so);
 558         KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
 559
 560         UNP_GLOBAL_WLOCK();
 561         UNP_PCB_LOCK(unp);
 562
 563         LIST_REMOVE(unp, unp_link);
 564         unp->unp_gencnt = ++unp_gencnt;
 565         --unp_count;
 566
 567         /*
 568          * XXXRW: Should assert vp->v_socket == so.
 569          */
 570         if ((vp = unp->unp_vnode) != NULL) {
 571                 unp->unp_vnode->v_socket = NULL;
 572                 unp->unp_vnode = NULL;
 573         }
 574         unp2 = unp->unp_conn;
 575         if (unp2 != NULL) {
 576                 UNP_PCB_LOCK(unp2);
 577                 unp_disconnect(unp, unp2);
 578                 UNP_PCB_UNLOCK(unp2);
 579         }
 580
 581         /*
 582          * We hold the global lock exclusively, so it's OK to acquire
 583          * multiple pcb locks at a time.
 584          */
 585         while (!LIST_EMPTY(&unp->unp_refs)) {
 586                 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
 587
 588                 UNP_PCB_LOCK(ref);
 589                 unp_drop(ref, ECONNRESET);
 590                 UNP_PCB_UNLOCK(ref);
 591         }
 592         local_unp_rights = unp_rights;
 593         UNP_GLOBAL_WUNLOCK();
 594         unp->unp_socket->so_pcb = NULL;
 595         saved_unp_addr = unp->unp_addr;
 596         unp->unp_addr = NULL;
 597         unp->unp_refcount--;
 598         freeunp = (unp->unp_refcount == 0);
 599         if (saved_unp_addr != NULL)
 600                 FREE(saved_unp_addr, M_SONAME);
 601         if (freeunp) {
 602                 UNP_PCB_LOCK_DESTROY(unp);
 603                 uma_zfree(unp_zone, unp);
 604         } else
 605                 UNP_PCB_UNLOCK(unp);
 606         if (vp) {
 607                 int vfslocked;
 608
 609                 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 610                 vrele(vp);
 611                 VFS_UNLOCK_GIANT(vfslocked);
 612         }
 613         if (local_unp_rights)
 614                 taskqueue_enqueue(taskqueue_thread, &unp_gc_task);
 615 }
 616
 617 static int
 618 uipc_disconnect(struct socket *so)
 619 {
 620         struct unpcb *unp, *unp2;
 621
 622         unp = sotounpcb(so);
 623         KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
 624
 625         UNP_GLOBAL_WLOCK();
 626         UNP_PCB_LOCK(unp);
 627         unp2 = unp->unp_conn;
 628         if (unp2 != NULL) {
 629                 UNP_PCB_LOCK(unp2);
 630                 unp_disconnect(unp, unp2);
 631                 UNP_PCB_UNLOCK(unp2);
 632         }
 633         UNP_PCB_UNLOCK(unp);
 634         UNP_GLOBAL_WUNLOCK();
 635         return (0);
 636 }
 637
 638 static int
 639 uipc_listen(struct socket *so, int backlog, struct thread *td)
 640 {
 641         struct unpcb *unp;
 642         int error;
 643
 644         unp = sotounpcb(so);
 645         KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
 646
 647         UNP_PCB_LOCK(unp);
 648         if (unp->unp_vnode == NULL) {
 649                 UNP_PCB_UNLOCK(unp);
 650                 return (EINVAL);
 651         }
 652
 653         SOCK_LOCK(so);
 654         error = solisten_proto_check(so);
 655         if (error == 0) {
 656                 cru2x(td->td_ucred, &unp->unp_peercred);
 657                 unp->unp_flags |= UNP_HAVEPCCACHED;
 658                 solisten_proto(so, backlog);
 659         }
 660         SOCK_UNLOCK(so);
 661         UNP_PCB_UNLOCK(unp);
 662         return (error);
 663 }
 664
 665 static int
 666 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
 667 {
 668         struct unpcb *unp, *unp2;
 669         const struct sockaddr *sa;
 670
 671         unp = sotounpcb(so);
 672         KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
 673
 674         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 675         UNP_PCB_LOCK(unp);
 676         /*
 677          * XXX: It seems that this test always fails even when connection is
 678          * established.  So, this else clause is added as workaround to
 679          * return PF_LOCAL sockaddr.
 680          */
 681         unp2 = unp->unp_conn;
 682         if (unp2 != NULL) {
 683                 UNP_PCB_LOCK(unp2);
 684                 if (unp2->unp_addr != NULL)
 685                         sa = (struct sockaddr *) unp->unp_conn->unp_addr;
 686                 else
 687                         sa = &sun_noname;
 688                 bcopy(sa, *nam, sa->sa_len);
 689                 UNP_PCB_UNLOCK(unp2);
 690         } else {
 691                 sa = &sun_noname;
 692                 bcopy(sa, *nam, sa->sa_len);
 693         }
 694         UNP_PCB_UNLOCK(unp);
 695         return (0);
 696 }
 697
 698 static int
 699 uipc_rcvd(struct socket *so, int flags)
 700 {
 701         struct unpcb *unp, *unp2;
 702         struct socket *so2;
 703         u_int mbcnt, sbcc;
 704         u_long newhiwat;
 705
 706         unp = sotounpcb(so);
 707         KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
 708
 709         if (so->so_type == SOCK_DGRAM)
 710                 panic("uipc_rcvd DGRAM?");
 711
 712         if (so->so_type != SOCK_STREAM)
 713                 panic("uipc_rcvd unknown socktype");
 714
 715         /*
 716          * Adjust backpressure on sender and wakeup any waiting to write.
 717          *
 718          * The unp lock is acquired to maintain the validity of the unp_conn
 719          * pointer; no lock on unp2 is required as unp2->unp_socket will be
 720          * static as long as we don't permit unp2 to disconnect from unp,
 721          * which is prevented by the lock on unp.  We cache values from
 722          * so_rcv to avoid holding the so_rcv lock over the entire
 723          * transaction on the remote so_snd.
 724          */
 725         SOCKBUF_LOCK(&so->so_rcv);
 726         mbcnt = so->so_rcv.sb_mbcnt;
 727         sbcc = so->so_rcv.sb_cc;
 728         SOCKBUF_UNLOCK(&so->so_rcv);
 729         UNP_PCB_LOCK(unp);
 730         unp2 = unp->unp_conn;
 731         if (unp2 == NULL) {
 732                 UNP_PCB_UNLOCK(unp);
 733                 return (0);
 734         }
 735         so2 = unp2->unp_socket;
 736         SOCKBUF_LOCK(&so2->so_snd);
 737         so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt;
 738         newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc;
 739         (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
 740             newhiwat, RLIM_INFINITY);
 741         sowwakeup_locked(so2);
 742         unp->unp_mbcnt = mbcnt;
 743         unp->unp_cc = sbcc;
 744         UNP_PCB_UNLOCK(unp);
 745         return (0);
 746 }
 747
 748 static int
 749 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
 750     struct mbuf *control, struct thread *td)
 751 {
 752         struct unpcb *unp, *unp2;
 753         struct socket *so2;
 754         u_int mbcnt_delta, sbcc;
 755         u_long newhiwat;
 756         int error = 0;
 757
 758         unp = sotounpcb(so);
 759         KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
 760
 761         if (flags & PRUS_OOB) {
 762                 error = EOPNOTSUPP;
 763                 goto release;
 764         }
 765         if (control != NULL && (error = unp_internalize(&control, td)))
 766                 goto release;
 767         if ((nam != NULL) || (flags & PRUS_EOF))
 768                 UNP_GLOBAL_WLOCK();
 769         else
 770                 UNP_GLOBAL_RLOCK();
 771         switch (so->so_type) {
 772         case SOCK_DGRAM:
 773         {
 774                 const struct sockaddr *from;
 775
 776                 unp2 = unp->unp_conn;
 777                 if (nam != NULL) {
 778                         UNP_GLOBAL_WLOCK_ASSERT();
 779                         if (unp2 != NULL) {
 780                                 error = EISCONN;
 781                                 break;
 782                         }
 783                         error = unp_connect(so, nam, td);
 784                         if (error)
 785                                 break;
 786                         unp2 = unp->unp_conn;
 787                 }
 788
 789                 /*
 790                  * Because connect() and send() are non-atomic in a sendto()
 791                  * with a target address, it's possible that the socket will
 792                  * have disconnected before the send() can run.  In that case
 793                  * return the slightly counter-intuitive but otherwise
 794                  * correct error that the socket is not connected.
 795                  */
 796                 if (unp2 == NULL) {
 797                         error = ENOTCONN;
 798                         break;
 799                 }
 800                 /* Lockless read. */
 801                 if (unp2->unp_flags & UNP_WANTCRED)
 802                         control = unp_addsockcred(td, control);
 803                 UNP_PCB_LOCK(unp);
 804                 if (unp->unp_addr != NULL)
 805                         from = (struct sockaddr *)unp->unp_addr;
 806                 else
 807                         from = &sun_noname;
 808                 so2 = unp2->unp_socket;
 809                 SOCKBUF_LOCK(&so2->so_rcv);
 810                 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
 811                         sorwakeup_locked(so2);
 812                         m = NULL;
 813                         control = NULL;
 814                 } else {
 815                         SOCKBUF_UNLOCK(&so2->so_rcv);
 816                         error = ENOBUFS;
 817                 }
 818                 if (nam != NULL) {
 819                         UNP_GLOBAL_WLOCK_ASSERT();
 820                         UNP_PCB_LOCK(unp2);
 821                         unp_disconnect(unp, unp2);
 822                         UNP_PCB_UNLOCK(unp2);
 823                 }
 824                 UNP_PCB_UNLOCK(unp);
 825                 break;
 826         }
 827
 828         case SOCK_STREAM:
 829                 if ((so->so_state & SS_ISCONNECTED) == 0) {
 830                         if (nam != NULL) {
 831                                 UNP_GLOBAL_WLOCK_ASSERT();
 832                                 error = unp_connect(so, nam, td);
 833                                 if (error)
 834                                         break;  /* XXX */
 835                         } else {
 836                                 error = ENOTCONN;
 837                                 break;
 838                         }
 839                 }
 840
 841                 /* Lockless read. */
 842                 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
 843                         error = EPIPE;
 844                         break;
 845                 }
 846
 847                 /*
 848                  * Because connect() and send() are non-atomic in a sendto()
 849                  * with a target address, it's possible that the socket will
 850                  * have disconnected before the send() can run.  In that case
 851                  * return the slightly counter-intuitive but otherwise
 852                  * correct error that the socket is not connected.
 853                  *
 854                  * Locking here must be done carefully: the global lock
 855                  * prevents interconnections between unpcbs from changing, so
 856                  * we can traverse from unp to unp2 without acquiring unp's
 857                  * lock.  Socket buffer locks follow unpcb locks, so we can
 858                  * acquire both remote and lock socket buffer locks.
 859                  */
 860                 unp2 = unp->unp_conn;
 861                 if (unp2 == NULL) {
 862                         error = ENOTCONN;
 863                         break;
 864                 }
 865                 so2 = unp2->unp_socket;
 866                 UNP_PCB_LOCK(unp2);
 867                 SOCKBUF_LOCK(&so2->so_rcv);
 868                 if (unp2->unp_flags & UNP_WANTCRED) {
 869                         /*
 870                          * Credentials are passed only once on SOCK_STREAM.
 871                          */
 872                         unp2->unp_flags &= ~UNP_WANTCRED;
 873                         control = unp_addsockcred(td, control);
 874                 }
 875                 /*
 876                  * Send to paired receive port, and then reduce send buffer
 877                  * hiwater marks to maintain backpressure.  Wake up readers.
 878                  */
 879                 if (control != NULL) {
 880                         if (sbappendcontrol_locked(&so2->so_rcv, m, control))
 881                                 control = NULL;
 882                 } else
 883                         sbappend_locked(&so2->so_rcv, m);
 884                 mbcnt_delta = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt;
 885                 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt;
 886                 sbcc = so2->so_rcv.sb_cc;
 887                 sorwakeup_locked(so2);
 888
 889                 SOCKBUF_LOCK(&so->so_snd);
 890                 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc);
 891                 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
 892                     newhiwat, RLIM_INFINITY);
 893                 so->so_snd.sb_mbmax -= mbcnt_delta;
 894                 SOCKBUF_UNLOCK(&so->so_snd);
 895                 unp2->unp_cc = sbcc;
 896                 UNP_PCB_UNLOCK(unp2);
 897                 m = NULL;
 898                 break;
 899
 900         default:
 901                 panic("uipc_send unknown socktype");
 902         }
 903
 904         /*
 905          * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
 906          */
 907         if (flags & PRUS_EOF) {
 908                 UNP_PCB_LOCK(unp);
 909                 socantsendmore(so);
 910                 unp_shutdown(unp);
 911                 UNP_PCB_UNLOCK(unp);
 912         }
 913
 914         if ((nam != NULL) || (flags & PRUS_EOF))
 915                 UNP_GLOBAL_WUNLOCK();
 916         else
 917                 UNP_GLOBAL_RUNLOCK();
 918
 919         if (control != NULL && error != 0)
 920                 unp_dispose(control);
 921
 922 release:
 923         if (control != NULL)
 924                 m_freem(control);
 925         if (m != NULL)
 926                 m_freem(m);
 927         return (error);
 928 }
 929
 930 static int
 931 uipc_sense(struct socket *so, struct stat *sb)
 932 {
 933         struct unpcb *unp, *unp2;
 934         struct socket *so2;
 935
 936         unp = sotounpcb(so);
 937         KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
 938
 939         sb->st_blksize = so->so_snd.sb_hiwat;
 940         UNP_GLOBAL_RLOCK();
 941         UNP_PCB_LOCK(unp);
 942         unp2 = unp->unp_conn;
 943         if (so->so_type == SOCK_STREAM && unp2 != NULL) {
 944                 so2 = unp2->unp_socket;
 945                 sb->st_blksize += so2->so_rcv.sb_cc;
 946         }
 947         sb->st_dev = NODEV;
 948         if (unp->unp_ino == 0)
 949                 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
 950         sb->st_ino = unp->unp_ino;
 951         UNP_PCB_UNLOCK(unp);
 952         UNP_GLOBAL_RUNLOCK();
 953         return (0);
 954 }
 955
 956 static int
 957 uipc_shutdown(struct socket *so)
 958 {
 959         struct unpcb *unp;
 960
 961         unp = sotounpcb(so);
 962         KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
 963
 964         UNP_GLOBAL_WLOCK();
 965         UNP_PCB_LOCK(unp);
 966         socantsendmore(so);
 967         unp_shutdown(unp);
 968         UNP_PCB_UNLOCK(unp);
 969         UNP_GLOBAL_WUNLOCK();
 970         return (0);
 971 }
 972
 973 static int
 974 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
 975 {
 976         struct unpcb *unp;
 977         const struct sockaddr *sa;
 978
 979         unp = sotounpcb(so);
 980         KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
 981
 982         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 983         UNP_PCB_LOCK(unp);
 984         if (unp->unp_addr != NULL)
 985                 sa = (struct sockaddr *) unp->unp_addr;
 986         else
 987                 sa = &sun_noname;
 988         bcopy(sa, *nam, sa->sa_len);
 989         UNP_PCB_UNLOCK(unp);
 990         return (0);
 991 }
 992
 993 static struct pr_usrreqs uipc_usrreqs_dgram = {
 994         .pru_abort =            uipc_abort,
 995         .pru_accept =           uipc_accept,
 996         .pru_attach =           uipc_attach,
 997         .pru_bind =             uipc_bind,
 998         .pru_connect =          uipc_connect,
 999         .pru_connect2 =         uipc_connect2,
1000         .pru_detach =           uipc_detach,
1001         .pru_disconnect =       uipc_disconnect,
1002         .pru_listen =           uipc_listen,
1003         .pru_peeraddr =         uipc_peeraddr,
1004         .pru_rcvd =             uipc_rcvd,
1005         .pru_send =             uipc_send,
1006         .pru_sense =            uipc_sense,
1007         .pru_shutdown =         uipc_shutdown,
1008         .pru_sockaddr =         uipc_sockaddr,
1009         .pru_soreceive =        soreceive_dgram,
1010         .pru_close =            uipc_close,
1011 };
1012
1013 static struct pr_usrreqs uipc_usrreqs_stream = {
1014         .pru_abort =            uipc_abort,
1015         .pru_accept =           uipc_accept,
1016         .pru_attach =           uipc_attach,
1017         .pru_bind =             uipc_bind,
1018         .pru_connect =          uipc_connect,
1019         .pru_connect2 =         uipc_connect2,
1020         .pru_detach =           uipc_detach,
1021         .pru_disconnect =       uipc_disconnect,
1022         .pru_listen =           uipc_listen,
1023         .pru_peeraddr =         uipc_peeraddr,
1024         .pru_rcvd =             uipc_rcvd,
1025         .pru_send =             uipc_send,
1026         .pru_sense =            uipc_sense,
1027         .pru_shutdown =         uipc_shutdown,
1028         .pru_sockaddr =         uipc_sockaddr,
1029         .pru_soreceive =        soreceive_generic,
1030         .pru_close =            uipc_close,
1031 };
1032
1033 static int
1034 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1035 {
1036         struct unpcb *unp;
1037         struct xucred xu;
1038         int error, optval;
1039
1040         if (sopt->sopt_level != 0)
1041                 return (EINVAL);
1042
1043         unp = sotounpcb(so);
1044         KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1045         error = 0;
1046         switch (sopt->sopt_dir) {
1047         case SOPT_GET:
1048                 switch (sopt->sopt_name) {
1049                 case LOCAL_PEERCRED:
1050                         UNP_PCB_LOCK(unp);
1051                         if (unp->unp_flags & UNP_HAVEPC)
1052                                 xu = unp->unp_peercred;
1053                         else {
1054                                 if (so->so_type == SOCK_STREAM)
1055                                         error = ENOTCONN;
1056                                 else
1057                                         error = EINVAL;
1058                         }
1059                         UNP_PCB_UNLOCK(unp);
1060                         if (error == 0)
1061                                 error = sooptcopyout(sopt, &xu, sizeof(xu));
1062                         break;
1063
1064                 case LOCAL_CREDS:
1065                         /* Unlocked read. */
1066                         optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1067                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1068                         break;
1069
1070                 case LOCAL_CONNWAIT:
1071                         /* Unlocked read. */
1072                         optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1073                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1074                         break;
1075
1076                 default:
1077                         error = EOPNOTSUPP;
1078                         break;
1079                 }
1080                 break;
1081
1082         case SOPT_SET:
1083                 switch (sopt->sopt_name) {
1084                 case LOCAL_CREDS:
1085                 case LOCAL_CONNWAIT:
1086                         error = sooptcopyin(sopt, &optval, sizeof(optval),
1087                                             sizeof(optval));
1088                         if (error)
1089                                 break;
1090
1091 #define OPTSET(bit) do {                                                \
1092         UNP_PCB_LOCK(unp);                                              \
1093         if (optval)                                                     \
1094                 unp->unp_flags |= bit;                                  \
1095         else                                                            \
1096                 unp->unp_flags &= ~bit;                                 \
1097         UNP_PCB_UNLOCK(unp);                                            \
1098 } while (0)
1099
1100                         switch (sopt->sopt_name) {
1101                         case LOCAL_CREDS:
1102                                 OPTSET(UNP_WANTCRED);
1103                                 break;
1104
1105                         case LOCAL_CONNWAIT:
1106                                 OPTSET(UNP_CONNWAIT);
1107                                 break;
1108
1109                         default:
1110                                 break;
1111                         }
1112                         break;
1113 #undef  OPTSET
1114                 default:
1115                         error = ENOPROTOOPT;
1116                         break;
1117                 }
1118                 break;
1119
1120         default:
1121                 error = EOPNOTSUPP;
1122                 break;
1123         }
1124         return (error);
1125 }
1126
1127 static int
1128 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1129 {
1130         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1131         struct vnode *vp;
1132         struct socket *so2, *so3;
1133         struct unpcb *unp, *unp2, *unp3;
1134         int error, len, vfslocked;
1135         struct nameidata nd;
1136         char buf[SOCK_MAXADDRLEN];
1137         struct sockaddr *sa;
1138
1139         UNP_GLOBAL_WLOCK_ASSERT();
1140
1141         unp = sotounpcb(so);
1142         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1143
1144         len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1145         if (len <= 0)
1146                 return (EINVAL);
1147         strlcpy(buf, soun->sun_path, len + 1);
1148
1149         UNP_PCB_LOCK(unp);
1150         if (unp->unp_flags & UNP_CONNECTING) {
1151                 UNP_PCB_UNLOCK(unp);
1152                 return (EALREADY);
1153         }
1154         UNP_GLOBAL_WUNLOCK();
1155         unp->unp_flags |= UNP_CONNECTING;
1156         UNP_PCB_UNLOCK(unp);
1157
1158         sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1159         NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf,
1160             td);
1161         error = namei(&nd);
1162         if (error)
1163                 vp = NULL;
1164         else
1165                 vp = nd.ni_vp;
1166         ASSERT_VOP_LOCKED(vp, "unp_connect");
1167         vfslocked = NDHASGIANT(&nd);
1168         NDFREE(&nd, NDF_ONLY_PNBUF);
1169         if (error)
1170                 goto bad;
1171
1172         if (vp->v_type != VSOCK) {
1173                 error = ENOTSOCK;
1174                 goto bad;
1175         }
1176 #ifdef MAC
1177         error = mac_check_vnode_open(td->td_ucred, vp, VWRITE | VREAD);
1178         if (error)
1179                 goto bad;
1180 #endif
1181         error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1182         if (error)
1183                 goto bad;
1184         VFS_UNLOCK_GIANT(vfslocked);
1185
1186         unp = sotounpcb(so);
1187         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1188
1189         /*
1190          * Lock global lock for two reasons: make sure v_socket is stable,
1191          * and to protect simultaneous locking of multiple pcbs.
1192          */
1193         UNP_GLOBAL_WLOCK();
1194         so2 = vp->v_socket;
1195         if (so2 == NULL) {
1196                 error = ECONNREFUSED;
1197                 goto bad2;
1198         }
1199         if (so->so_type != so2->so_type) {
1200                 error = EPROTOTYPE;
1201                 goto bad2;
1202         }
1203         if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1204                 if (so2->so_options & SO_ACCEPTCONN) {
1205                         /*
1206                          * We can't drop the global lock here or 'so2' may
1207                          * become invalid.  As a result, we need to handle
1208                          * possibly lock recursion in uipc_attach.
1209                          */
1210                         so3 = sonewconn(so2, 0);
1211                 } else
1212                         so3 = NULL;
1213                 if (so3 == NULL) {
1214                         error = ECONNREFUSED;
1215                         goto bad2;
1216                 }
1217                 unp = sotounpcb(so);
1218                 unp2 = sotounpcb(so2);
1219                 unp3 = sotounpcb(so3);
1220                 UNP_PCB_LOCK(unp);
1221                 UNP_PCB_LOCK(unp2);
1222                 UNP_PCB_LOCK(unp3);
1223                 if (unp2->unp_addr != NULL) {
1224                         bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1225                         unp3->unp_addr = (struct sockaddr_un *) sa;
1226                         sa = NULL;
1227                 }
1228
1229                 /*
1230                  * The connecter's (client's) credentials are copied from its
1231                  * process structure at the time of connect() (which is now).
1232                  */
1233                 cru2x(td->td_ucred, &unp3->unp_peercred);
1234                 unp3->unp_flags |= UNP_HAVEPC;
1235
1236                 /*
1237                  * The receiver's (server's) credentials are copied from the
1238                  * unp_peercred member of socket on which the former called
1239                  * listen(); uipc_listen() cached that process's credentials
1240                  * at that time so we can use them now.
1241                  */
1242                 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1243                     ("unp_connect: listener without cached peercred"));
1244                 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1245                     sizeof(unp->unp_peercred));
1246                 unp->unp_flags |= UNP_HAVEPC;
1247                 if (unp2->unp_flags & UNP_WANTCRED)
1248                         unp3->unp_flags |= UNP_WANTCRED;
1249                 UNP_PCB_UNLOCK(unp3);
1250                 UNP_PCB_UNLOCK(unp2);
1251                 UNP_PCB_UNLOCK(unp);
1252 #ifdef MAC
1253                 SOCK_LOCK(so);
1254                 mac_set_socket_peer_from_socket(so, so3);
1255                 mac_set_socket_peer_from_socket(so3, so);
1256                 SOCK_UNLOCK(so);
1257 #endif
1258
1259                 so2 = so3;
1260         }
1261         unp = sotounpcb(so);
1262         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1263         unp2 = sotounpcb(so2);
1264         KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1265         UNP_PCB_LOCK(unp);
1266         UNP_PCB_LOCK(unp2);
1267         error = unp_connect2(so, so2, PRU_CONNECT);
1268         UNP_PCB_UNLOCK(unp2);
1269         UNP_PCB_UNLOCK(unp);
1270 bad2:
1271         UNP_GLOBAL_WUNLOCK();
1272         if (vfslocked)
1273                 /*
1274                  * Giant has been previously acquired. This means filesystem
1275                  * isn't MPSAFE.  Do it once again.
1276                  */
1277                 mtx_lock(&Giant);
1278 bad:
1279         if (vp != NULL)
1280                 vput(vp);
1281         VFS_UNLOCK_GIANT(vfslocked);
1282         free(sa, M_SONAME);
1283         UNP_GLOBAL_WLOCK();
1284         UNP_PCB_LOCK(unp);
1285         unp->unp_flags &= ~UNP_CONNECTING;
1286         UNP_PCB_UNLOCK(unp);
1287         return (error);
1288 }
1289
1290 static int
1291 unp_connect2(struct socket *so, struct socket *so2, int req)
1292 {
1293         struct unpcb *unp;
1294         struct unpcb *unp2;
1295
1296         unp = sotounpcb(so);
1297         KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1298         unp2 = sotounpcb(so2);
1299         KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1300
1301         UNP_GLOBAL_WLOCK_ASSERT();
1302         UNP_PCB_LOCK_ASSERT(unp);
1303         UNP_PCB_LOCK_ASSERT(unp2);
1304
1305         if (so2->so_type != so->so_type)
1306                 return (EPROTOTYPE);
1307         unp->unp_conn = unp2;
1308
1309         switch (so->so_type) {
1310         case SOCK_DGRAM:
1311                 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1312                 soisconnected(so);
1313                 break;
1314
1315         case SOCK_STREAM:
1316                 unp2->unp_conn = unp;
1317                 if (req == PRU_CONNECT &&
1318                     ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1319                         soisconnecting(so);
1320                 else
1321                         soisconnected(so);
1322                 soisconnected(so2);
1323                 break;
1324
1325         default:
1326                 panic("unp_connect2");
1327         }
1328         return (0);
1329 }
1330
1331 static void
1332 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1333 {
1334         struct socket *so;
1335
1336         KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1337
1338         UNP_GLOBAL_WLOCK_ASSERT();
1339         UNP_PCB_LOCK_ASSERT(unp);
1340         UNP_PCB_LOCK_ASSERT(unp2);
1341
1342         unp->unp_conn = NULL;
1343         switch (unp->unp_socket->so_type) {
1344         case SOCK_DGRAM:
1345                 LIST_REMOVE(unp, unp_reflink);
1346                 so = unp->unp_socket;
1347                 SOCK_LOCK(so);
1348                 so->so_state &= ~SS_ISCONNECTED;
1349                 SOCK_UNLOCK(so);
1350                 break;
1351
1352         case SOCK_STREAM:
1353                 soisdisconnected(unp->unp_socket);
1354                 unp2->unp_conn = NULL;
1355                 soisdisconnected(unp2->unp_socket);
1356                 break;
1357         }
1358 }
1359
1360 /*
1361  * unp_pcblist() walks the global list of struct unpcb's to generate a
1362  * pointer list, bumping the refcount on each unpcb.  It then copies them out
1363  * sequentially, validating the generation number on each to see if it has
1364  * been detached.  All of this is necessary because copyout() may sleep on
1365  * disk I/O.
1366  */
1367 static int
1368 unp_pcblist(SYSCTL_HANDLER_ARGS)
1369 {
1370         int error, i, n;
1371         int freeunp;
1372         struct unpcb *unp, **unp_list;
1373         unp_gen_t gencnt;
1374         struct xunpgen *xug;
1375         struct unp_head *head;
1376         struct xunpcb *xu;
1377
1378         head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
1379
1380         /*
1381          * The process of preparing the PCB list is too time-consuming and
1382          * resource-intensive to repeat twice on every request.
1383          */
1384         if (req->oldptr == NULL) {
1385                 n = unp_count;
1386                 req->oldidx = 2 * (sizeof *xug)
1387                         + (n + n/8) * sizeof(struct xunpcb);
1388                 return (0);
1389         }
1390
1391         if (req->newptr != NULL)
1392                 return (EPERM);
1393
1394         /*
1395          * OK, now we're committed to doing something.
1396          */
1397         xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1398         UNP_GLOBAL_RLOCK();
1399         gencnt = unp_gencnt;
1400         n = unp_count;
1401         UNP_GLOBAL_RUNLOCK();
1402
1403         xug->xug_len = sizeof *xug;
1404         xug->xug_count = n;
1405         xug->xug_gen = gencnt;
1406         xug->xug_sogen = so_gencnt;
1407         error = SYSCTL_OUT(req, xug, sizeof *xug);
1408         if (error) {
1409                 free(xug, M_TEMP);
1410                 return (error);
1411         }
1412
1413         unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1414
1415         UNP_GLOBAL_RLOCK();
1416         for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1417              unp = LIST_NEXT(unp, unp_link)) {
1418                 UNP_PCB_LOCK(unp);
1419                 if (unp->unp_gencnt <= gencnt) {
1420                         if (cr_cansee(req->td->td_ucred,
1421                             unp->unp_socket->so_cred)) {
1422                                 UNP_PCB_UNLOCK(unp);
1423                                 continue;
1424                         }
1425                         unp_list[i++] = unp;
1426                         unp->unp_refcount++;
1427                 }
1428                 UNP_PCB_UNLOCK(unp);
1429         }
1430         UNP_GLOBAL_RUNLOCK();
1431         n = i;                  /* In case we lost some during malloc. */
1432
1433         error = 0;
1434         xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1435         for (i = 0; i < n; i++) {
1436                 unp = unp_list[i];
1437                 UNP_PCB_LOCK(unp);
1438                 unp->unp_refcount--;
1439                 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1440                         xu->xu_len = sizeof *xu;
1441                         xu->xu_unpp = unp;
1442                         /*
1443                          * XXX - need more locking here to protect against
1444                          * connect/disconnect races for SMP.
1445                          */
1446                         if (unp->unp_addr != NULL)
1447                                 bcopy(unp->unp_addr, &xu->xu_addr,
1448                                       unp->unp_addr->sun_len);
1449                         if (unp->unp_conn != NULL &&
1450                             unp->unp_conn->unp_addr != NULL)
1451                                 bcopy(unp->unp_conn->unp_addr,
1452                                       &xu->xu_caddr,
1453                                       unp->unp_conn->unp_addr->sun_len);
1454                         bcopy(unp, &xu->xu_unp, sizeof *unp);
1455                         sotoxsocket(unp->unp_socket, &xu->xu_socket);
1456                         UNP_PCB_UNLOCK(unp);
1457                         error = SYSCTL_OUT(req, xu, sizeof *xu);
1458                 } else {
1459                         freeunp = (unp->unp_refcount == 0);
1460                         UNP_PCB_UNLOCK(unp);
1461                         if (freeunp) {
1462                                 UNP_PCB_LOCK_DESTROY(unp);
1463                                 uma_zfree(unp_zone, unp);
1464                         }
1465                 }
1466         }
1467         free(xu, M_TEMP);
1468         if (!error) {
1469                 /*
1470                  * Give the user an updated idea of our state.  If the
1471                  * generation differs from what we told her before, she knows
1472                  * that something happened while we were processing this
1473                  * request, and it might be necessary to retry.
1474                  */
1475                 xug->xug_gen = unp_gencnt;
1476                 xug->xug_sogen = so_gencnt;
1477                 xug->xug_count = unp_count;
1478                 error = SYSCTL_OUT(req, xug, sizeof *xug);
1479         }
1480         free(unp_list, M_TEMP);
1481         free(xug, M_TEMP);
1482         return (error);
1483 }
1484
1485 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1486             (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1487             "List of active local datagram sockets");
1488 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1489             (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1490             "List of active local stream sockets");
1491
1492 static void
1493 unp_shutdown(struct unpcb *unp)
1494 {
1495         struct unpcb *unp2;
1496         struct socket *so;
1497
1498         UNP_GLOBAL_WLOCK_ASSERT();
1499         UNP_PCB_LOCK_ASSERT(unp);
1500
1501         unp2 = unp->unp_conn;
1502         if (unp->unp_socket->so_type == SOCK_STREAM && unp2 != NULL) {
1503                 so = unp2->unp_socket;
1504                 if (so != NULL)
1505                         socantrcvmore(so);
1506         }
1507 }
1508
1509 static void
1510 unp_drop(struct unpcb *unp, int errno)
1511 {
1512         struct socket *so = unp->unp_socket;
1513         struct unpcb *unp2;
1514
1515         UNP_GLOBAL_WLOCK_ASSERT();
1516         UNP_PCB_LOCK_ASSERT(unp);
1517
1518         so->so_error = errno;
1519         unp2 = unp->unp_conn;
1520         if (unp2 == NULL)
1521                 return;
1522         UNP_PCB_LOCK(unp2);
1523         unp_disconnect(unp, unp2);
1524         UNP_PCB_UNLOCK(unp2);
1525 }
1526
1527 static void
1528 unp_freerights(struct file **rp, int fdcount)
1529 {
1530         int i;
1531         struct file *fp;
1532
1533         for (i = 0; i < fdcount; i++) {
1534                 fp = *rp;
1535                 *rp++ = NULL;
1536                 unp_discard(fp);
1537         }
1538 }
1539
1540 static int
1541 unp_externalize(struct mbuf *control, struct mbuf **controlp)
1542 {
1543         struct thread *td = curthread;          /* XXX */
1544         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1545         int i;
1546         int *fdp;
1547         struct file **rp;
1548         struct file *fp;
1549         void *data;
1550         socklen_t clen = control->m_len, datalen;
1551         int error, newfds;
1552         int f;
1553         u_int newlen;
1554
1555         UNP_GLOBAL_UNLOCK_ASSERT();
1556
1557         error = 0;
1558         if (controlp != NULL) /* controlp == NULL => free control messages */
1559                 *controlp = NULL;
1560         while (cm != NULL) {
1561                 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1562                         error = EINVAL;
1563                         break;
1564                 }
1565                 data = CMSG_DATA(cm);
1566                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1567                 if (cm->cmsg_level == SOL_SOCKET
1568                     && cm->cmsg_type == SCM_RIGHTS) {
1569                         newfds = datalen / sizeof(struct file *);
1570                         rp = data;
1571
1572                         /* If we're not outputting the descriptors free them. */
1573                         if (error || controlp == NULL) {
1574                                 unp_freerights(rp, newfds);
1575                                 goto next;
1576                         }
1577                         FILEDESC_XLOCK(td->td_proc->p_fd);
1578                         /* if the new FD's will not fit free them.  */
1579                         if (!fdavail(td, newfds)) {
1580                                 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1581                                 error = EMSGSIZE;
1582                                 unp_freerights(rp, newfds);
1583                                 goto next;
1584                         }
1585
1586                         /*
1587                          * Now change each pointer to an fd in the global
1588                          * table to an integer that is the index to the local
1589                          * fd table entry that we set up to point to the
1590                          * global one we are transferring.
1591                          */
1592                         newlen = newfds * sizeof(int);
1593                         *controlp = sbcreatecontrol(NULL, newlen,
1594                             SCM_RIGHTS, SOL_SOCKET);
1595                         if (*controlp == NULL) {
1596                                 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1597                                 error = E2BIG;
1598                                 unp_freerights(rp, newfds);
1599                                 goto next;
1600                         }
1601
1602                         fdp = (int *)
1603                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1604                         for (i = 0; i < newfds; i++) {
1605                                 if (fdalloc(td, 0, &f))
1606                                         panic("unp_externalize fdalloc failed");
1607                                 fp = *rp++;
1608                                 td->td_proc->p_fd->fd_ofiles[f] = fp;
1609                                 FILE_LOCK(fp);
1610                                 fp->f_msgcount--;
1611                                 FILE_UNLOCK(fp);
1612                                 UNP_GLOBAL_WLOCK();
1613                                 unp_rights--;
1614                                 UNP_GLOBAL_WUNLOCK();
1615                                 *fdp++ = f;
1616                         }
1617                         FILEDESC_XUNLOCK(td->td_proc->p_fd);
1618                 } else {
1619                         /* We can just copy anything else across. */
1620                         if (error || controlp == NULL)
1621                                 goto next;
1622                         *controlp = sbcreatecontrol(NULL, datalen,
1623                             cm->cmsg_type, cm->cmsg_level);
1624                         if (*controlp == NULL) {
1625                                 error = ENOBUFS;
1626                                 goto next;
1627                         }
1628                         bcopy(data,
1629                             CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1630                             datalen);
1631                 }
1632                 controlp = &(*controlp)->m_next;
1633
1634 next:
1635                 if (CMSG_SPACE(datalen) < clen) {
1636                         clen -= CMSG_SPACE(datalen);
1637                         cm = (struct cmsghdr *)
1638                             ((caddr_t)cm + CMSG_SPACE(datalen));
1639                 } else {
1640                         clen = 0;
1641                         cm = NULL;
1642                 }
1643         }
1644
1645         m_freem(control);
1646         return (error);
1647 }
1648
1649 static void
1650 unp_zone_change(void *tag)
1651 {
1652
1653         uma_zone_set_max(unp_zone, maxsockets);
1654 }
1655
1656 static void
1657 unp_init(void)
1658 {
1659
1660         unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1661             NULL, NULL, UMA_ALIGN_PTR, 0);
1662         if (unp_zone == NULL)
1663                 panic("unp_init");
1664         uma_zone_set_max(unp_zone, maxsockets);
1665         EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1666             NULL, EVENTHANDLER_PRI_ANY);
1667         LIST_INIT(&unp_dhead);
1668         LIST_INIT(&unp_shead);
1669         TASK_INIT(&unp_gc_task, 0, unp_gc, NULL);
1670         UNP_GLOBAL_LOCK_INIT();
1671 }
1672
1673 static int
1674 unp_internalize(struct mbuf **controlp, struct thread *td)
1675 {
1676         struct mbuf *control = *controlp;
1677         struct proc *p = td->td_proc;
1678         struct filedesc *fdescp = p->p_fd;
1679         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1680         struct cmsgcred *cmcred;
1681         struct file **rp;
1682         struct file *fp;
1683         struct timeval *tv;
1684         int i, fd, *fdp;
1685         void *data;
1686         socklen_t clen = control->m_len, datalen;
1687         int error, oldfds;
1688         u_int newlen;
1689
1690         UNP_GLOBAL_UNLOCK_ASSERT();
1691
1692         error = 0;
1693         *controlp = NULL;
1694         while (cm != NULL) {
1695                 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1696                     || cm->cmsg_len > clen) {
1697                         error = EINVAL;
1698                         goto out;
1699                 }
1700                 data = CMSG_DATA(cm);
1701                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1702
1703                 switch (cm->cmsg_type) {
1704                 /*
1705                  * Fill in credential information.
1706                  */
1707                 case SCM_CREDS:
1708                         *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1709                             SCM_CREDS, SOL_SOCKET);
1710                         if (*controlp == NULL) {
1711                                 error = ENOBUFS;
1712                                 goto out;
1713                         }
1714                         cmcred = (struct cmsgcred *)
1715                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1716                         cmcred->cmcred_pid = p->p_pid;
1717                         cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1718                         cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1719                         cmcred->cmcred_euid = td->td_ucred->cr_uid;
1720                         cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1721                             CMGROUP_MAX);
1722                         for (i = 0; i < cmcred->cmcred_ngroups; i++)
1723                                 cmcred->cmcred_groups[i] =
1724                                     td->td_ucred->cr_groups[i];
1725                         break;
1726
1727                 case SCM_RIGHTS:
1728                         oldfds = datalen / sizeof (int);
1729                         /*
1730                          * Check that all the FDs passed in refer to legal
1731                          * files.  If not, reject the entire operation.
1732                          */
1733                         fdp = data;
1734                         FILEDESC_SLOCK(fdescp);
1735                         for (i = 0; i < oldfds; i++) {
1736                                 fd = *fdp++;
1737                                 if ((unsigned)fd >= fdescp->fd_nfiles ||
1738                                     fdescp->fd_ofiles[fd] == NULL) {
1739                                         FILEDESC_SUNLOCK(fdescp);
1740                                         error = EBADF;
1741                                         goto out;
1742                                 }
1743                                 fp = fdescp->fd_ofiles[fd];
1744                                 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1745                                         FILEDESC_SUNLOCK(fdescp);
1746                                         error = EOPNOTSUPP;
1747                                         goto out;
1748                                 }
1749
1750                         }
1751
1752                         /*
1753                          * Now replace the integer FDs with pointers to the
1754                          * associated global file table entry..
1755                          */
1756                         newlen = oldfds * sizeof(struct file *);
1757                         *controlp = sbcreatecontrol(NULL, newlen,
1758                             SCM_RIGHTS, SOL_SOCKET);
1759                         if (*controlp == NULL) {
1760                                 FILEDESC_SUNLOCK(fdescp);
1761                                 error = E2BIG;
1762                                 goto out;
1763                         }
1764                         fdp = data;
1765                         rp = (struct file **)
1766                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1767                         for (i = 0; i < oldfds; i++) {
1768                                 fp = fdescp->fd_ofiles[*fdp++];
1769                                 *rp++ = fp;
1770                                 FILE_LOCK(fp);
1771                                 fp->f_count++;
1772                                 fp->f_msgcount++;
1773                                 FILE_UNLOCK(fp);
1774                                 UNP_GLOBAL_WLOCK();
1775                                 unp_rights++;
1776                                 UNP_GLOBAL_WUNLOCK();
1777                         }
1778                         FILEDESC_SUNLOCK(fdescp);
1779                         break;
1780
1781                 case SCM_TIMESTAMP:
1782                         *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1783                             SCM_TIMESTAMP, SOL_SOCKET);
1784                         if (*controlp == NULL) {
1785                                 error = ENOBUFS;
1786                                 goto out;
1787                         }
1788                         tv = (struct timeval *)
1789                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1790                         microtime(tv);
1791                         break;
1792
1793                 default:
1794                         error = EINVAL;
1795                         goto out;
1796                 }
1797
1798                 controlp = &(*controlp)->m_next;
1799                 if (CMSG_SPACE(datalen) < clen) {
1800                         clen -= CMSG_SPACE(datalen);
1801                         cm = (struct cmsghdr *)
1802                             ((caddr_t)cm + CMSG_SPACE(datalen));
1803                 } else {
1804                         clen = 0;
1805                         cm = NULL;
1806                 }
1807         }
1808
1809 out:
1810         m_freem(control);
1811         return (error);
1812 }
1813
1814 static struct mbuf *
1815 unp_addsockcred(struct thread *td, struct mbuf *control)
1816 {
1817         struct mbuf *m, *n, *n_prev;
1818         struct sockcred *sc;
1819         const struct cmsghdr *cm;
1820         int ngroups;
1821         int i;
1822
1823         ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
1824         m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
1825         if (m == NULL)
1826                 return (control);
1827
1828         sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
1829         sc->sc_uid = td->td_ucred->cr_ruid;
1830         sc->sc_euid = td->td_ucred->cr_uid;
1831         sc->sc_gid = td->td_ucred->cr_rgid;
1832         sc->sc_egid = td->td_ucred->cr_gid;
1833         sc->sc_ngroups = ngroups;
1834         for (i = 0; i < sc->sc_ngroups; i++)
1835                 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
1836
1837         /*
1838          * Unlink SCM_CREDS control messages (struct cmsgcred), since just
1839          * created SCM_CREDS control message (struct sockcred) has another
1840          * format.
1841          */
1842         if (control != NULL)
1843                 for (n = control, n_prev = NULL; n != NULL;) {
1844                         cm = mtod(n, struct cmsghdr *);
1845                         if (cm->cmsg_level == SOL_SOCKET &&
1846                             cm->cmsg_type == SCM_CREDS) {
1847                                 if (n_prev == NULL)
1848                                         control = n->m_next;
1849                                 else
1850                                         n_prev->m_next = n->m_next;
1851                                 n = m_free(n);
1852                         } else {
1853                                 n_prev = n;
1854                                 n = n->m_next;
1855                         }
1856                 }
1857
1858         /* Prepend it to the head. */
1859         m->m_next = control;
1860         return (m);
1861 }
1862
1863 /*
1864  * unp_defer indicates whether additional work has been defered for a future
1865  * pass through unp_gc().  It is thread local and does not require explicit
1866  * synchronization.
1867  */
1868 static int      unp_defer;
1869
1870 static int unp_taskcount;
1871 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, "");
1872
1873 static int unp_recycled;
1874 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, "");
1875
1876 static void
1877 unp_gc(__unused void *arg, int pending)
1878 {
1879         struct file *fp, *nextfp;
1880         struct socket *so;
1881         struct socket *soa;
1882         struct file **extra_ref, **fpp;
1883         int nunref, i;
1884         int nfiles_snap;
1885         int nfiles_slack = 20;
1886
1887         unp_taskcount++;
1888         unp_defer = 0;
1889
1890         /*
1891          * Before going through all this, set all FDs to be NOT deferred and
1892          * NOT externally accessible.
1893          */
1894         sx_slock(&filelist_lock);
1895         LIST_FOREACH(fp, &filehead, f_list)
1896                 fp->f_gcflag &= ~(FMARK|FDEFER);
1897         do {
1898                 KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer));
1899                 LIST_FOREACH(fp, &filehead, f_list) {
1900                         FILE_LOCK(fp);
1901                         /*
1902                          * If the file is not open, skip it -- could be a
1903                          * file in the process of being opened, or in the
1904                          * process of being closed.  If the file is
1905                          * "closing", it may have been marked for deferred
1906                          * consideration.  Clear the flag now if so.
1907                          */
1908                         if (fp->f_count == 0) {
1909                                 if (fp->f_gcflag & FDEFER)
1910                                         unp_defer--;
1911                                 fp->f_gcflag &= ~(FMARK|FDEFER);
1912                                 FILE_UNLOCK(fp);
1913                                 continue;
1914                         }
1915
1916                         /*
1917                          * If we already marked it as 'defer' in a
1918                          * previous pass, then try to process it this
1919                          * time and un-mark it.
1920                          */
1921                         if (fp->f_gcflag & FDEFER) {
1922                                 fp->f_gcflag &= ~FDEFER;
1923                                 unp_defer--;
1924                         } else {
1925                                 /*
1926                                  * If it's not deferred, then check if it's
1927                                  * already marked.. if so skip it
1928                                  */
1929                                 if (fp->f_gcflag & FMARK) {
1930                                         FILE_UNLOCK(fp);
1931                                         continue;
1932                                 }
1933
1934                                 /*
1935                                  * If all references are from messages in
1936                                  * transit, then skip it. it's not externally
1937                                  * accessible.
1938                                  */
1939                                 if (fp->f_count == fp->f_msgcount) {
1940                                         FILE_UNLOCK(fp);
1941                                         continue;
1942                                 }
1943
1944                                 /*
1945                                  * If it got this far then it must be
1946                                  * externally accessible.
1947                                  */
1948                                 fp->f_gcflag |= FMARK;
1949                         }
1950
1951                         /*
1952                          * Either it was deferred, or it is externally
1953                          * accessible and not already marked so.  Now check
1954                          * if it is possibly one of OUR sockets.
1955                          */
1956                         if (fp->f_type != DTYPE_SOCKET ||
1957                             (so = fp->f_data) == NULL) {
1958                                 FILE_UNLOCK(fp);
1959                                 continue;
1960                         }
1961
1962                         if (so->so_proto->pr_domain != &localdomain ||
1963                             (so->so_proto->pr_flags & PR_RIGHTS) == 0) {
1964                                 FILE_UNLOCK(fp);
1965                                 continue;
1966                         }
1967
1968                         /*
1969                          * Tell any other threads that do a subsequent
1970                          * fdrop() that we are scanning the message
1971                          * buffers.
1972                          */
1973                         fp->f_gcflag |= FWAIT;
1974                         FILE_UNLOCK(fp);
1975
1976                         /*
1977                          * So, Ok, it's one of our sockets and it IS
1978                          * externally accessible (or was deferred).  Now we
1979                          * look to see if we hold any file descriptors in its
1980                          * message buffers. Follow those links and mark them
1981                          * as accessible too.
1982                          */
1983                         SOCKBUF_LOCK(&so->so_rcv);
1984                         unp_scan(so->so_rcv.sb_mb, unp_mark);
1985                         SOCKBUF_UNLOCK(&so->so_rcv);
1986
1987                         /*
1988                          * If socket is in listening state, then sockets
1989                          * in its accept queue are accessible, and so
1990                          * are any descriptors in those sockets' receive
1991                          * queues.
1992                          */
1993                         ACCEPT_LOCK();
1994                         TAILQ_FOREACH(soa, &so->so_comp, so_list) {
1995                             SOCKBUF_LOCK(&soa->so_rcv);
1996                             unp_scan(soa->so_rcv.sb_mb, unp_mark);
1997                             SOCKBUF_UNLOCK(&soa->so_rcv);
1998                         }
1999                         ACCEPT_UNLOCK();
2000
2001                         /*
2002                          * Wake up any threads waiting in fdrop().
2003                          */
2004                         FILE_LOCK(fp);
2005                         fp->f_gcflag &= ~FWAIT;
2006                         wakeup(&fp->f_gcflag);
2007                         FILE_UNLOCK(fp);
2008                 }
2009         } while (unp_defer);
2010         sx_sunlock(&filelist_lock);
2011
2012         /*
2013          * XXXRW: The following comments need updating for a post-SMPng and
2014          * deferred unp_gc() world, but are still generally accurate.
2015          *
2016          * We grab an extra reference to each of the file table entries that
2017          * are not otherwise accessible and then free the rights that are
2018          * stored in messages on them.
2019          *
2020          * The bug in the orginal code is a little tricky, so I'll describe
2021          * what's wrong with it here.
2022          *
2023          * It is incorrect to simply unp_discard each entry for f_msgcount
2024          * times -- consider the case of sockets A and B that contain
2025          * references to each other.  On a last close of some other socket,
2026          * we trigger a gc since the number of outstanding rights (unp_rights)
2027          * is non-zero.  If during the sweep phase the gc code unp_discards,
2028          * we end up doing a (full) closef on the descriptor.  A closef on A
2029          * results in the following chain.  Closef calls soo_close, which
2030          * calls soclose.   Soclose calls first (through the switch
2031          * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
2032          * returns because the previous instance had set unp_gcing, and we
2033          * return all the way back to soclose, which marks the socket with
2034          * SS_NOFDREF, and then calls sofree.  Sofree calls sorflush to free
2035          * up the rights that are queued in messages on the socket A, i.e.,
2036          * the reference on B.  The sorflush calls via the dom_dispose switch
2037          * unp_dispose, which unp_scans with unp_discard.  This second
2038          * instance of unp_discard just calls closef on B.
2039          *
2040          * Well, a similar chain occurs on B, resulting in a sorflush on B,
2041          * which results in another closef on A.  Unfortunately, A is already
2042          * being closed, and the descriptor has already been marked with
2043          * SS_NOFDREF, and soclose panics at this point.
2044          *
2045          * Here, we first take an extra reference to each inaccessible
2046          * descriptor.  Then, we call sorflush ourself, since we know it is a
2047          * Unix domain socket anyhow.  After we destroy all the rights
2048          * carried in messages, we do a last closef to get rid of our extra
2049          * reference.  This is the last close, and the unp_detach etc will
2050          * shut down the socket.
2051          *
2052          * 91/09/19, bsy@cs.cmu.edu
2053          */
2054 again:
2055         nfiles_snap = openfiles + nfiles_slack; /* some slack */
2056         extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP,
2057             M_WAITOK);
2058         sx_slock(&filelist_lock);
2059         if (nfiles_snap < openfiles) {
2060                 sx_sunlock(&filelist_lock);
2061                 free(extra_ref, M_TEMP);
2062                 nfiles_slack += 20;
2063                 goto again;
2064         }
2065         for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref;
2066             fp != NULL; fp = nextfp) {
2067                 nextfp = LIST_NEXT(fp, f_list);
2068                 FILE_LOCK(fp);
2069
2070                 /*
2071                  * If it's not open, skip it
2072                  */
2073                 if (fp->f_count == 0) {
2074                         FILE_UNLOCK(fp);
2075                         continue;
2076                 }
2077
2078                 /*
2079                  * If all refs are from msgs, and it's not marked accessible
2080                  * then it must be referenced from some unreachable cycle of
2081                  * (shut-down) FDs, so include it in our list of FDs to
2082                  * remove.
2083                  */
2084                 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) {
2085                         *fpp++ = fp;
2086                         nunref++;
2087                         fp->f_count++;
2088                 }
2089                 FILE_UNLOCK(fp);
2090         }
2091         sx_sunlock(&filelist_lock);
2092
2093         /*
2094          * For each FD on our hit list, do the following two things:
2095          */
2096         for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
2097                 struct file *tfp = *fpp;
2098                 FILE_LOCK(tfp);
2099                 if (tfp->f_type == DTYPE_SOCKET &&
2100                     tfp->f_data != NULL) {
2101                         FILE_UNLOCK(tfp);
2102                         sorflush(tfp->f_data);
2103                 } else {
2104                         FILE_UNLOCK(tfp);
2105                 }
2106         }
2107         for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
2108                 closef(*fpp, (struct thread *) NULL);
2109                 unp_recycled++;
2110         }
2111         free(extra_ref, M_TEMP);
2112 }
2113
2114 static void
2115 unp_dispose(struct mbuf *m)
2116 {
2117
2118         if (m)
2119                 unp_scan(m, unp_discard);
2120 }
2121
2122 static void
2123 unp_scan(struct mbuf *m0, void (*op)(struct file *))
2124 {
2125         struct mbuf *m;
2126         struct file **rp;
2127         struct cmsghdr *cm;
2128         void *data;
2129         int i;
2130         socklen_t clen, datalen;
2131         int qfds;
2132
2133         while (m0 != NULL) {
2134                 for (m = m0; m; m = m->m_next) {
2135                         if (m->m_type != MT_CONTROL)
2136                                 continue;
2137
2138                         cm = mtod(m, struct cmsghdr *);
2139                         clen = m->m_len;
2140
2141                         while (cm != NULL) {
2142                                 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2143                                         break;
2144
2145                                 data = CMSG_DATA(cm);
2146                                 datalen = (caddr_t)cm + cm->cmsg_len
2147                                     - (caddr_t)data;
2148
2149                                 if (cm->cmsg_level == SOL_SOCKET &&
2150                                     cm->cmsg_type == SCM_RIGHTS) {
2151                                         qfds = datalen / sizeof (struct file *);
2152                                         rp = data;
2153                                         for (i = 0; i < qfds; i++)
2154                                                 (*op)(*rp++);
2155                                 }
2156
2157                                 if (CMSG_SPACE(datalen) < clen) {
2158                                         clen -= CMSG_SPACE(datalen);
2159                                         cm = (struct cmsghdr *)
2160                                             ((caddr_t)cm + CMSG_SPACE(datalen));
2161                                 } else {
2162                                         clen = 0;
2163                                         cm = NULL;
2164                                 }
2165                         }
2166                 }
2167                 m0 = m0->m_act;
2168         }
2169 }
2170
2171 static void
2172 unp_mark(struct file *fp)
2173 {
2174
2175         /* XXXRW: Should probably assert file list lock here. */
2176
2177         if (fp->f_gcflag & FMARK)
2178                 return;
2179         unp_defer++;
2180         fp->f_gcflag |= (FMARK|FDEFER);
2181 }
2182
2183 static void
2184 unp_discard(struct file *fp)
2185 {
2186
2187         UNP_GLOBAL_WLOCK();
2188         FILE_LOCK(fp);
2189         fp->f_msgcount--;
2190         unp_rights--;
2191         FILE_UNLOCK(fp);
2192         UNP_GLOBAL_WUNLOCK();
2193         (void) closef(fp, (struct thread *)NULL);
2194 }
2195
2196 #ifdef DDB
2197 static void
2198 db_print_indent(int indent)
2199 {
2200         int i;
2201
2202         for (i = 0; i < indent; i++)
2203                 db_printf(" ");
2204 }
2205
2206 static void
2207 db_print_unpflags(int unp_flags)
2208 {
2209         int comma;
2210
2211         comma = 0;
2212         if (unp_flags & UNP_HAVEPC) {
2213                 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2214                 comma = 1;
2215         }
2216         if (unp_flags & UNP_HAVEPCCACHED) {
2217                 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2218                 comma = 1;
2219         }
2220         if (unp_flags & UNP_WANTCRED) {
2221                 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2222                 comma = 1;
2223         }
2224         if (unp_flags & UNP_CONNWAIT) {
2225                 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2226                 comma = 1;
2227         }
2228         if (unp_flags & UNP_CONNECTING) {
2229                 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2230                 comma = 1;
2231         }
2232         if (unp_flags & UNP_BINDING) {
2233                 db_printf("%sUNP_BINDING", comma ? ", " : "");
2234                 comma = 1;
2235         }
2236 }
2237
2238 static void
2239 db_print_xucred(int indent, struct xucred *xu)
2240 {
2241         int comma, i;
2242
2243         db_print_indent(indent);
2244         db_printf("cr_version: %u   cr_uid: %u   cr_ngroups: %d\n",
2245             xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2246         db_print_indent(indent);
2247         db_printf("cr_groups: ");
2248         comma = 0;
2249         for (i = 0; i < xu->cr_ngroups; i++) {
2250                 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2251                 comma = 1;
2252         }
2253         db_printf("\n");
2254 }
2255
2256 static void
2257 db_print_unprefs(int indent, struct unp_head *uh)
2258 {
2259         struct unpcb *unp;
2260         int counter;
2261
2262         counter = 0;
2263         LIST_FOREACH(unp, uh, unp_reflink) {
2264                 if (counter % 4 == 0)
2265                         db_print_indent(indent);
2266                 db_printf("%p  ", unp);
2267                 if (counter % 4 == 3)
2268                         db_printf("\n");
2269                 counter++;
2270         }
2271         if (counter != 0 && counter % 4 != 0)
2272                 db_printf("\n");
2273 }
2274
2275 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2276 {
2277         struct unpcb *unp;
2278
2279         if (!have_addr) {
2280                 db_printf("usage: show unpcb <addr>\n");
2281                 return;
2282         }
2283         unp = (struct unpcb *)addr;
2284
2285         db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
2286             unp->unp_vnode);
2287
2288         db_printf("unp_ino: %d   unp_conn: %p\n", unp->unp_ino,
2289             unp->unp_conn);
2290
2291         db_printf("unp_refs:\n");
2292         db_print_unprefs(2, &unp->unp_refs);
2293
2294         /* XXXRW: Would be nice to print the full address, if any. */
2295         db_printf("unp_addr: %p\n", unp->unp_addr);
2296
2297         db_printf("unp_cc: %d   unp_mbcnt: %d   unp_gencnt: %llu\n",
2298             unp->unp_cc, unp->unp_mbcnt,
2299             (unsigned long long)unp->unp_gencnt);
2300
2301         db_printf("unp_flags: %x (", unp->unp_flags);
2302         db_print_unpflags(unp->unp_flags);
2303         db_printf(")\n");
2304
2305         db_printf("unp_peercred:\n");
2306         db_print_xucred(2, &unp->unp_peercred);
2307
2308         db_printf("unp_refcount: %u\n", unp->unp_refcount);
2309 }
2310 #endif