sys/kern/uipc_usrreq.c

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