sys/nfsserver/nfs_syscalls.c

   1 /*-
   2  * Copyright (c) 1989, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to Berkeley by
   6  * Rick Macklem at The University of Guelph.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 4. Neither the name of the University nor the names of its contributors
  17  *    may be used to endorse or promote products derived from this software
  18  *    without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  *
  32  *      @(#)nfs_syscalls.c      8.5 (Berkeley) 3/30/95
  33  */
  34
  35 #include <sys/cdefs.h>
  36 __FBSDID("$FreeBSD$");
  37
  38 #include "opt_inet6.h"
  39 #include "opt_mac.h"
  40
  41 #include <sys/param.h>
  42 #include <sys/systm.h>
  43 #include <sys/sysproto.h>
  44 #include <sys/kernel.h>
  45 #include <sys/sysctl.h>
  46 #include <sys/file.h>
  47 #include <sys/filedesc.h>
  48 #include <sys/vnode.h>
  49 #include <sys/mac.h>
  50 #include <sys/malloc.h>
  51 #include <sys/mount.h>
  52 #include <sys/proc.h>
  53 #include <sys/bio.h>
  54 #include <sys/buf.h>
  55 #include <sys/mbuf.h>
  56 #include <sys/socket.h>
  57 #include <sys/socketvar.h>
  58 #include <sys/domain.h>
  59 #include <sys/protosw.h>
  60 #include <sys/namei.h>
  61 #include <sys/fcntl.h>
  62 #include <sys/lockf.h>
  63
  64 #include <netinet/in.h>
  65 #include <netinet/tcp.h>
  66 #ifdef INET6
  67 #include <net/if.h>
  68 #include <netinet6/in6_var.h>
  69 #endif
  70 #include <nfs/xdr_subs.h>
  71 #include <nfs/rpcv2.h>
  72 #include <nfs/nfsproto.h>
  73 #include <nfsserver/nfs.h>
  74 #include <nfsserver/nfsm_subs.h>
  75 #include <nfsserver/nfsrvcache.h>
  76
  77 static MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure");
  78
  79 MALLOC_DEFINE(M_NFSRVDESC, "NFSV3 srvdesc", "NFS server socket descriptor");
  80 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
  81
  82
  83 #define TRUE    1
  84 #define FALSE   0
  85
  86 SYSCTL_DECL(_vfs_nfsrv);
  87
  88 int             nfsd_waiting = 0;
  89 int             nfsrv_numnfsd = 0;
  90 static int      notstarted = 1;
  91
  92 static int      nfs_privport = 0;
  93 SYSCTL_INT(_vfs_nfsrv, NFS_NFSPRIVPORT, nfs_privport, CTLFLAG_RW,
  94             &nfs_privport, 0, "");
  95 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, gatherdelay, CTLFLAG_RW,
  96             &nfsrvw_procrastinate, 0, "");
  97 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, gatherdelay_v3, CTLFLAG_RW,
  98             &nfsrvw_procrastinate_v3, 0, "");
  99
 100 static int      nfssvc_addsock(struct file *, struct sockaddr *,
 101                     struct thread *);
 102 static void     nfsrv_zapsock(struct nfssvc_sock *slp);
 103 static int      nfssvc_nfsd(struct thread *);
 104
 105 /*
 106  * NFS server system calls
 107  */
 108
 109 /*
 110  * Nfs server psuedo system call for the nfsd's
 111  * Based on the flag value it either:
 112  * - adds a socket to the selection list
 113  * - remains in the kernel as an nfsd
 114  * - remains in the kernel as an nfsiod
 115  * For INET6 we suppose that nfsd provides only IN6P_IPV6_V6ONLY sockets
 116  * and that mountd provides
 117  *  - sockaddr with no IPv4-mapped addresses
 118  *  - mask for both INET and INET6 families if there is IPv4-mapped overlap
 119  */
 120 #ifndef _SYS_SYSPROTO_H_
 121 struct nfssvc_args {
 122         int flag;
 123         caddr_t argp;
 124 };
 125 #endif
 126 /*
 127  * MPSAFE
 128  */
 129 int
 130 nfssvc(struct thread *td, struct nfssvc_args *uap)
 131 {
 132         struct file *fp;
 133         struct sockaddr *nam;
 134         struct nfsd_args nfsdarg;
 135         int error;
 136
 137         KASSERT(!mtx_owned(&Giant), ("nfssvc(): called with Giant"));
 138
 139 #ifdef MAC
 140         error = mac_check_system_nfsd(td->td_ucred);
 141         if (error)
 142                 return (error);
 143 #endif
 144         error = suser(td);
 145         if (error)
 146                 return (error);
 147         NET_LOCK_GIANT();
 148         NFSD_LOCK();
 149         while (nfssvc_sockhead_flag & SLP_INIT) {
 150                  nfssvc_sockhead_flag |= SLP_WANTINIT;
 151                 (void) msleep(&nfssvc_sockhead, &nfsd_mtx, PSOCK,
 152                     "nfsd init", 0);
 153         }
 154         NFSD_UNLOCK();
 155         if (uap->flag & NFSSVC_ADDSOCK) {
 156                 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg));
 157                 if (error)
 158                         goto done2;
 159                 if ((error = fget(td, nfsdarg.sock, &fp)) != 0)
 160                         goto done2;
 161                 if (fp->f_type != DTYPE_SOCKET) {
 162                         fdrop(fp, td);
 163                         goto done2;
 164                 }
 165                 /*
 166                  * Get the client address for connected sockets.
 167                  */
 168                 if (nfsdarg.name == NULL || nfsdarg.namelen == 0)
 169                         nam = NULL;
 170                 else {
 171                         error = getsockaddr(&nam, nfsdarg.name,
 172                                             nfsdarg.namelen);
 173                         if (error) {
 174                                 fdrop(fp, td);
 175                                 goto done2;
 176                         }
 177                 }
 178                 error = nfssvc_addsock(fp, nam, td);
 179                 fdrop(fp, td);
 180         } else if (uap->flag & NFSSVC_NFSD) {
 181                 error = nfssvc_nfsd(td);
 182         } else {
 183                 error = ENXIO;
 184         }
 185         if (error == EINTR || error == ERESTART)
 186                 error = 0;
 187 done2:
 188         NET_UNLOCK_GIANT();
 189         return (error);
 190 }
 191
 192 /*
 193  * Adds a socket to the list for servicing by nfsds.
 194  */
 195 static int
 196 nfssvc_addsock(struct file *fp, struct sockaddr *mynam, struct thread *td)
 197 {
 198         int siz;
 199         struct nfssvc_sock *slp;
 200         struct socket *so;
 201         int error, s;
 202
 203         NET_ASSERT_GIANT();
 204
 205         so = fp->f_data;
 206 #if 0
 207         /*
 208          * XXXRW: If this code is ever enabled, there's a race when running
 209          * MPSAFE.
 210          */
 211         tslp = NULL;
 212         /*
 213          * Add it to the list, as required.
 214          */
 215         if (so->so_proto->pr_protocol == IPPROTO_UDP) {
 216                 tslp = nfs_udpsock;
 217                 if (tslp->ns_flag & SLP_VALID) {
 218                         if (mynam != NULL)
 219                                 FREE(mynam, M_SONAME);
 220                         return (EPERM);
 221                 }
 222         }
 223 #endif
 224         if (so->so_type == SOCK_STREAM)
 225                 siz = NFS_MAXPACKET + sizeof (u_long);
 226         else
 227                 siz = NFS_MAXPACKET;
 228         error = soreserve(so, siz, siz);
 229         if (error) {
 230                 if (mynam != NULL)
 231                         FREE(mynam, M_SONAME);
 232                 return (error);
 233         }
 234
 235         /*
 236          * Set protocol specific options { for now TCP only } and
 237          * reserve some space. For datagram sockets, this can get called
 238          * repeatedly for the same socket, but that isn't harmful.
 239          */
 240         if (so->so_type == SOCK_STREAM) {
 241                 struct sockopt sopt;
 242                 int val;
 243
 244                 bzero(&sopt, sizeof sopt);
 245                 sopt.sopt_dir = SOPT_SET;
 246                 sopt.sopt_level = SOL_SOCKET;
 247                 sopt.sopt_name = SO_KEEPALIVE;
 248                 sopt.sopt_val = &val;
 249                 sopt.sopt_valsize = sizeof val;
 250                 val = 1;
 251                 sosetopt(so, &sopt);
 252         }
 253         if (so->so_proto->pr_protocol == IPPROTO_TCP) {
 254                 struct sockopt sopt;
 255                 int val;
 256
 257                 bzero(&sopt, sizeof sopt);
 258                 sopt.sopt_dir = SOPT_SET;
 259                 sopt.sopt_level = IPPROTO_TCP;
 260                 sopt.sopt_name = TCP_NODELAY;
 261                 sopt.sopt_val = &val;
 262                 sopt.sopt_valsize = sizeof val;
 263                 val = 1;
 264                 sosetopt(so, &sopt);
 265         }
 266         SOCKBUF_LOCK(&so->so_rcv);
 267         so->so_rcv.sb_flags &= ~SB_NOINTR;
 268         so->so_rcv.sb_timeo = 0;
 269         SOCKBUF_UNLOCK(&so->so_rcv);
 270         SOCKBUF_LOCK(&so->so_snd);
 271         so->so_snd.sb_flags &= ~SB_NOINTR;
 272         so->so_snd.sb_timeo = 0;
 273         SOCKBUF_UNLOCK(&so->so_snd);
 274
 275         slp = (struct nfssvc_sock *)
 276                 malloc(sizeof (struct nfssvc_sock), M_NFSSVC,
 277                 M_WAITOK | M_ZERO);
 278         STAILQ_INIT(&slp->ns_rec);
 279         NFSD_LOCK();
 280         TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain);
 281
 282         slp->ns_so = so;
 283         slp->ns_nam = mynam;
 284         fhold(fp);
 285         slp->ns_fp = fp;
 286         /*
 287          * XXXRW: Socket locking here?
 288          */
 289         s = splnet();
 290         so->so_upcallarg = (caddr_t)slp;
 291         so->so_upcall = nfsrv_rcv;
 292         SOCKBUF_LOCK(&so->so_rcv);
 293         so->so_rcv.sb_flags |= SB_UPCALL;
 294         SOCKBUF_UNLOCK(&so->so_rcv);
 295         slp->ns_flag = (SLP_VALID | SLP_NEEDQ);
 296         nfsrv_wakenfsd(slp);
 297         splx(s);
 298         NFSD_UNLOCK();
 299         return (0);
 300 }
 301
 302 /*
 303  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
 304  * until it is killed by a signal.
 305  */
 306 static int
 307 nfssvc_nfsd(struct thread *td)
 308 {
 309         int siz;
 310         struct nfssvc_sock *slp;
 311         struct nfsd *nfsd;
 312         struct nfsrv_descript *nd = NULL;
 313         struct mbuf *m, *mreq;
 314         int error = 0, cacherep, s, sotype, writes_todo;
 315         int procrastinate;
 316         u_quad_t cur_usec;
 317
 318         NET_ASSERT_GIANT();
 319
 320 #ifndef nolint
 321         cacherep = RC_DOIT;
 322         writes_todo = 0;
 323 #endif
 324         nfsd = (struct nfsd *)
 325                 malloc(sizeof (struct nfsd), M_NFSD, M_WAITOK | M_ZERO);
 326         s = splnet();
 327         NFSD_LOCK();
 328
 329         nfsd->nfsd_td = td;
 330         TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
 331         nfsrv_numnfsd++;
 332
 333         /*
 334          * Loop getting rpc requests until SIGKILL.
 335          */
 336         for (;;) {
 337                 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) {
 338                         while (nfsd->nfsd_slp == NULL &&
 339                             (nfsd_head_flag & NFSD_CHECKSLP) == 0) {
 340                                 nfsd->nfsd_flag |= NFSD_WAITING;
 341                                 nfsd_waiting++;
 342                                 error = msleep(nfsd, &nfsd_mtx,
 343                                     PSOCK | PCATCH, "-", 0);
 344                                 nfsd_waiting--;
 345                                 if (error)
 346                                         goto done;
 347                         }
 348                         if (nfsd->nfsd_slp == NULL &&
 349                             (nfsd_head_flag & NFSD_CHECKSLP) != 0) {
 350                                 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
 351                                     if ((slp->ns_flag & (SLP_VALID | SLP_DOREC))
 352                                         == (SLP_VALID | SLP_DOREC)) {
 353                                             slp->ns_flag &= ~SLP_DOREC;
 354                                             slp->ns_sref++;
 355                                             nfsd->nfsd_slp = slp;
 356                                             break;
 357                                     }
 358                                 }
 359                                 if (slp == NULL)
 360                                         nfsd_head_flag &= ~NFSD_CHECKSLP;
 361                         }
 362                         if ((slp = nfsd->nfsd_slp) == NULL)
 363                                 continue;
 364                         if (slp->ns_flag & SLP_VALID) {
 365                                 if (slp->ns_flag & SLP_DISCONN)
 366                                         nfsrv_zapsock(slp);
 367                                 else if (slp->ns_flag & SLP_NEEDQ) {
 368                                         slp->ns_flag &= ~SLP_NEEDQ;
 369                                         (void) nfs_slplock(slp, 1);
 370                                         NFSD_UNLOCK();
 371                                         nfsrv_rcv(slp->ns_so, (caddr_t)slp,
 372                                                 M_TRYWAIT);
 373                                         NFSD_LOCK();
 374                                         nfs_slpunlock(slp);
 375                                 }
 376                                 error = nfsrv_dorec(slp, nfsd, &nd);
 377                                 cur_usec = nfs_curusec();
 378                                 if (error && LIST_FIRST(&slp->ns_tq) &&
 379                                     LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) {
 380                                         error = 0;
 381                                         cacherep = RC_DOIT;
 382                                         writes_todo = 1;
 383                                 } else
 384                                         writes_todo = 0;
 385                                 nfsd->nfsd_flag |= NFSD_REQINPROG;
 386                         }
 387                 } else {
 388                         error = 0;
 389                         slp = nfsd->nfsd_slp;
 390                 }
 391                 if (error || (slp->ns_flag & SLP_VALID) == 0) {
 392                         if (nd) {
 393                                 free((caddr_t)nd, M_NFSRVDESC);
 394                                 nd = NULL;
 395                         }
 396                         nfsd->nfsd_slp = NULL;
 397                         nfsd->nfsd_flag &= ~NFSD_REQINPROG;
 398                         nfsrv_slpderef(slp);
 399                         continue;
 400                 }
 401                 splx(s);
 402                 sotype = slp->ns_so->so_type;
 403                 if (nd) {
 404                     getmicrotime(&nd->nd_starttime);
 405                     if (nd->nd_nam2)
 406                         nd->nd_nam = nd->nd_nam2;
 407                     else
 408                         nd->nd_nam = slp->ns_nam;
 409
 410                     /*
 411                      * Check to see if authorization is needed.
 412                      */
 413                     cacherep = nfsrv_getcache(nd, &mreq);
 414
 415                     if (nfs_privport) {
 416                         /* Check if source port is privileged */
 417                         u_short port;
 418                         struct sockaddr *nam = nd->nd_nam;
 419                         struct sockaddr_in *sin;
 420
 421                         sin = (struct sockaddr_in *)nam;
 422                         /*
 423                          * INET/INET6 - same code:
 424                          *    sin_port and sin6_port are at same offset
 425                          */
 426                         port = ntohs(sin->sin_port);
 427                         if (port >= IPPORT_RESERVED &&
 428                             nd->nd_procnum != NFSPROC_NULL) {
 429 #if defined(INET6) && defined(KLD_MODULE)
 430         /* do not use ip6_sprintf: the nfs module should work without INET6 */
 431         char b6[INET6_ADDRSTRLEN];
 432 #define ip6_sprintf(a) \
 433          (sprintf(b6, "%x:%x:%x:%x:%x:%x:%x:%x", \
 434                   (a)->s6_addr16[0], (a)->s6_addr16[1], \
 435                   (a)->s6_addr16[2], (a)->s6_addr16[3], \
 436                   (a)->s6_addr16[4], (a)->s6_addr16[5], \
 437                   (a)->s6_addr16[6], (a)->s6_addr16[7]), \
 438           b6)
 439 #endif
 440                             nd->nd_procnum = NFSPROC_NOOP;
 441                             nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
 442                             cacherep = RC_DOIT;
 443                             printf("NFS request from unprivileged port (%s:%d)\n",
 444 #ifdef INET6
 445                                    sin->sin_family == AF_INET6 ?
 446                                         ip6_sprintf(&satosin6(sin)->sin6_addr) :
 447 #undef ip6_sprintf
 448 #endif
 449                                    inet_ntoa(sin->sin_addr), port);
 450                         }
 451                     }
 452
 453                 }
 454
 455                 /*
 456                  * Loop to get all the write rpc relies that have been
 457                  * gathered together.
 458                  */
 459                 do {
 460                     switch (cacherep) {
 461                     case RC_DOIT:
 462                         if (nd && (nd->nd_flag & ND_NFSV3))
 463                             procrastinate = nfsrvw_procrastinate_v3;
 464                         else
 465                             procrastinate = nfsrvw_procrastinate;
 466                         if (writes_todo || (!(nd->nd_flag & ND_NFSV3) &&
 467                             nd->nd_procnum == NFSPROC_WRITE &&
 468                             procrastinate > 0 && !notstarted))
 469                             error = nfsrv_writegather(&nd, slp,
 470                                 nfsd->nfsd_td, &mreq);
 471                         else
 472                             error = (*(nfsrv3_procs[nd->nd_procnum]))(nd,
 473                                 slp, nfsd->nfsd_td, &mreq);
 474                         if (mreq == NULL)
 475                                 break;
 476                         if (error != 0 && error != NFSERR_RETVOID) {
 477                                 nfsrvstats.srv_errs++;
 478                                 nfsrv_updatecache(nd, FALSE, mreq);
 479                                 if (nd->nd_nam2)
 480                                         FREE(nd->nd_nam2, M_SONAME);
 481                                 break;
 482                         }
 483                         nfsrvstats.srvrpccnt[nd->nd_procnum]++;
 484                         nfsrv_updatecache(nd, TRUE, mreq);
 485                         nd->nd_mrep = NULL;
 486                         /* FALLTHROUGH */
 487                     case RC_REPLY:
 488                         NFSD_UNLOCK();
 489                         siz = m_length(mreq, NULL);
 490                         if (siz <= 0 || siz > NFS_MAXPACKET) {
 491                                 printf("mbuf siz=%d\n",siz);
 492                                 panic("Bad nfs svc reply");
 493                         }
 494                         m = mreq;
 495                         m->m_pkthdr.len = siz;
 496                         m->m_pkthdr.rcvif = NULL;
 497                         /*
 498                          * For stream protocols, prepend a Sun RPC
 499                          * Record Mark.
 500                          */
 501                         if (sotype == SOCK_STREAM) {
 502                                 M_PREPEND(m, NFSX_UNSIGNED, M_TRYWAIT);
 503                                 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz);
 504                         }
 505                         NFSD_LOCK();
 506                         if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
 507                                 (void) nfs_slplock(slp, 1);
 508                         if (slp->ns_flag & SLP_VALID) {
 509                             NFSD_UNLOCK();
 510                             error = nfsrv_send(slp->ns_so, nd->nd_nam2, m);
 511                             NFSD_LOCK();
 512                         } else {
 513                             error = EPIPE;
 514                             m_freem(m);
 515                         }
 516                         if (nd->nd_nam2)
 517                                 FREE(nd->nd_nam2, M_SONAME);
 518                         if (nd->nd_mrep)
 519                                 m_freem(nd->nd_mrep);
 520                         if (error == EPIPE)
 521                                 nfsrv_zapsock(slp);
 522                         if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
 523                                 nfs_slpunlock(slp);
 524                         if (error == EINTR || error == ERESTART) {
 525                                 free((caddr_t)nd, M_NFSRVDESC);
 526                                 nfsrv_slpderef(slp);
 527                                 s = splnet();
 528                                 goto done;
 529                         }
 530                         break;
 531                     case RC_DROPIT:
 532                         m_freem(nd->nd_mrep);
 533                         if (nd->nd_nam2)
 534                                 FREE(nd->nd_nam2, M_SONAME);
 535                         break;
 536                     };
 537                     if (nd) {
 538                         FREE((caddr_t)nd, M_NFSRVDESC);
 539                         nd = NULL;
 540                     }
 541
 542                     /*
 543                      * Check to see if there are outstanding writes that
 544                      * need to be serviced.
 545                      */
 546                     cur_usec = nfs_curusec();
 547                     s = splsoftclock();
 548                     if (LIST_FIRST(&slp->ns_tq) &&
 549                         LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) {
 550                         cacherep = RC_DOIT;
 551                         writes_todo = 1;
 552                     } else
 553                         writes_todo = 0;
 554                     splx(s);
 555                 } while (writes_todo);
 556                 s = splnet();
 557                 if (nfsrv_dorec(slp, nfsd, &nd)) {
 558                         nfsd->nfsd_flag &= ~NFSD_REQINPROG;
 559                         nfsd->nfsd_slp = NULL;
 560                         nfsrv_slpderef(slp);
 561                 }
 562                 KASSERT(!(debug_mpsafenet == 0 && !mtx_owned(&Giant)),
 563                     ("nfssvc_nfsd(): debug.mpsafenet=0 && !Giant"));
 564                 KASSERT(!(debug_mpsafenet == 1 && mtx_owned(&Giant)),
 565                     ("nfssvc_nfsd(): debug.mpsafenet=1 && Giant"));
 566         }
 567 done:
 568         KASSERT(!(debug_mpsafenet == 0 && !mtx_owned(&Giant)),
 569             ("nfssvc_nfsd(): debug.mpsafenet=0 && !Giant"));
 570         KASSERT(!(debug_mpsafenet == 1 && mtx_owned(&Giant)),
 571             ("nfssvc_nfsd(): debug.mpsafenet=1 && Giant"));
 572         TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
 573         splx(s);
 574         free((caddr_t)nfsd, M_NFSD);
 575         if (--nfsrv_numnfsd == 0)
 576                 nfsrv_init(TRUE);       /* Reinitialize everything */
 577         NFSD_UNLOCK();
 578         return (error);
 579 }
 580
 581 /*
 582  * Shut down a socket associated with an nfssvc_sock structure.
 583  * Should be called with the send lock set, if required.
 584  * The trick here is to increment the sref at the start, so that the nfsds
 585  * will stop using it and clear ns_flag at the end so that it will not be
 586  * reassigned during cleanup.
 587  */
 588 static void
 589 nfsrv_zapsock(struct nfssvc_sock *slp)
 590 {
 591         struct nfsrv_descript *nwp, *nnwp;
 592         struct socket *so;
 593         struct file *fp;
 594         struct nfsrv_rec *rec;
 595         int s;
 596
 597         NET_ASSERT_GIANT();
 598         NFSD_LOCK_ASSERT();
 599
 600         /*
 601          * XXXRW: By clearing all flags, other threads/etc should ignore
 602          * this slp and we can safely release nfsd_mtx so we can clean
 603          * up the slp safely.
 604          */
 605         slp->ns_flag &= ~SLP_ALLFLAGS;
 606         fp = slp->ns_fp;
 607         if (fp) {
 608                 NFSD_UNLOCK();
 609                 slp->ns_fp = NULL;
 610                 so = slp->ns_so;
 611                 SOCKBUF_LOCK(&so->so_rcv);
 612                 so->so_rcv.sb_flags &= ~SB_UPCALL;
 613                 SOCKBUF_UNLOCK(&so->so_rcv);
 614                 so->so_upcall = NULL;
 615                 so->so_upcallarg = NULL;
 616                 soshutdown(so, SHUT_RDWR);
 617                 closef(fp, NULL);
 618                 NFSD_LOCK();
 619                 if (slp->ns_nam)
 620                         FREE(slp->ns_nam, M_SONAME);
 621                 m_freem(slp->ns_raw);
 622                 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) {
 623                         STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link);
 624                         if (rec->nr_address)
 625                                 FREE(rec->nr_address, M_SONAME);
 626                         m_freem(rec->nr_packet);
 627                         free(rec, M_NFSRVDESC);
 628                 }
 629                 s = splsoftclock();
 630                 for (nwp = LIST_FIRST(&slp->ns_tq); nwp; nwp = nnwp) {
 631                         nnwp = LIST_NEXT(nwp, nd_tq);
 632                         LIST_REMOVE(nwp, nd_tq);
 633                         free((caddr_t)nwp, M_NFSRVDESC);
 634                 }
 635                 LIST_INIT(&slp->ns_tq);
 636                 splx(s);
 637         }
 638 }
 639
 640 /*
 641  * Derefence a server socket structure. If it has no more references and
 642  * is no longer valid, you can throw it away.
 643  */
 644 void
 645 nfsrv_slpderef(struct nfssvc_sock *slp)
 646 {
 647
 648         NFSD_LOCK_ASSERT();
 649
 650         if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) {
 651                 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
 652                 free((caddr_t)slp, M_NFSSVC);
 653         }
 654 }
 655
 656 /*
 657  * Lock a socket against others.
 658  *
 659  * XXXRW: Wait argument is always 1 in the caller.  Replace with a real
 660  * sleep lock?
 661  */
 662 int
 663 nfs_slplock(struct nfssvc_sock *slp, int wait)
 664 {
 665         int *statep = &slp->ns_solock;
 666
 667         NFSD_LOCK_ASSERT();
 668
 669         if (!wait && (*statep & NFSRV_SNDLOCK))
 670                 return(0);      /* already locked, fail */
 671         while (*statep & NFSRV_SNDLOCK) {
 672                 *statep |= NFSRV_WANTSND;
 673                 (void) msleep(statep, &nfsd_mtx, PZERO - 1, "nfsslplck", 0);
 674         }
 675         *statep |= NFSRV_SNDLOCK;
 676         return (1);
 677 }
 678
 679 /*
 680  * Unlock the stream socket for others.
 681  */
 682 void
 683 nfs_slpunlock(struct nfssvc_sock *slp)
 684 {
 685         int *statep = &slp->ns_solock;
 686
 687         NFSD_LOCK_ASSERT();
 688
 689         if ((*statep & NFSRV_SNDLOCK) == 0)
 690                 panic("nfs slpunlock");
 691         *statep &= ~NFSRV_SNDLOCK;
 692         if (*statep & NFSRV_WANTSND) {
 693                 *statep &= ~NFSRV_WANTSND;
 694                 wakeup(statep);
 695         }
 696 }
 697
 698 /*
 699  * Initialize the data structures for the server.
 700  * Handshake with any new nfsds starting up to avoid any chance of
 701  * corruption.
 702  */
 703 void
 704 nfsrv_init(int terminating)
 705 {
 706         struct nfssvc_sock *slp, *nslp;
 707
 708         NET_ASSERT_GIANT();
 709         NFSD_LOCK_ASSERT();
 710
 711         if (nfssvc_sockhead_flag & SLP_INIT)
 712                 panic("nfsd init");
 713         nfssvc_sockhead_flag |= SLP_INIT;
 714         if (terminating) {
 715                 for (slp = TAILQ_FIRST(&nfssvc_sockhead); slp != NULL;
 716                     slp = nslp) {
 717                         nslp = TAILQ_NEXT(slp, ns_chain);
 718                         if (slp->ns_flag & SLP_VALID)
 719                                 nfsrv_zapsock(slp);
 720                         TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
 721                         free((caddr_t)slp, M_NFSSVC);
 722                 }
 723                 nfsrv_cleancache();     /* And clear out server cache */
 724         } else
 725                 nfs_pub.np_valid = 0;
 726
 727         TAILQ_INIT(&nfssvc_sockhead);
 728         nfssvc_sockhead_flag &= ~SLP_INIT;
 729         if (nfssvc_sockhead_flag & SLP_WANTINIT) {
 730                 nfssvc_sockhead_flag &= ~SLP_WANTINIT;
 731                 wakeup(&nfssvc_sockhead);
 732         }
 733
 734         TAILQ_INIT(&nfsd_head);
 735         nfsd_head_flag &= ~NFSD_CHECKSLP;
 736
 737 #if 0
 738         nfs_udpsock = (struct nfssvc_sock *)
 739             malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
 740         STAILQ_INIT(&nfs_udpsock->ns_rec);
 741         TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain);
 742
 743         nfs_cltpsock = (struct nfssvc_sock *)
 744             malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
 745         STAILQ_INIT(&nfs_cltpsock->ns_rec);
 746         TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain);
 747 #endif
 748 }