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/malloc.h>
  50 #include <sys/mount.h>
  51 #include <sys/priv.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 #include <security/mac/mac_framework.h>
  78
  79 static MALLOC_DEFINE(M_NFSSVC, "nfss_srvsock", "Nfs server structure");
  80
  81 MALLOC_DEFINE(M_NFSRVDESC, "nfss_srvdesc", "NFS server socket descriptor");
  82 MALLOC_DEFINE(M_NFSD, "nfss_daemon", "Nfs server daemon structure");
  83
  84 #define TRUE    1
  85 #define FALSE   0
  86
  87 SYSCTL_DECL(_vfs_nfsrv);
  88
  89 int             nfsd_waiting = 0;
  90 int             nfsrv_numnfsd = 0;
  91 static int      notstarted = 1;
  92
  93 static int      nfs_privport = 0;
  94 SYSCTL_INT(_vfs_nfsrv, NFS_NFSPRIVPORT, nfs_privport, CTLFLAG_RW,
  95             &nfs_privport, 0, "");
  96 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, gatherdelay, CTLFLAG_RW,
  97             &nfsrvw_procrastinate, 0, "");
  98 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, gatherdelay_v3, CTLFLAG_RW,
  99             &nfsrvw_procrastinate_v3, 0, "");
 100
 101 static int      nfssvc_addsock(struct file *, struct sockaddr *,
 102                     struct thread *);
 103 static void     nfsrv_zapsock(struct nfssvc_sock *slp);
 104 static int      nfssvc_nfsd(struct thread *);
 105
 106 /*
 107  * NFS server system calls
 108  */
 109
 110 /*
 111  * Nfs server psuedo system call for the nfsd's
 112  * Based on the flag value it either:
 113  * - adds a socket to the selection list
 114  * - remains in the kernel as an nfsd
 115  * - remains in the kernel as an nfsiod
 116  * For INET6 we suppose that nfsd provides only IN6P_IPV6_V6ONLY sockets
 117  * and that mountd provides
 118  *  - sockaddr with no IPv4-mapped addresses
 119  *  - mask for both INET and INET6 families if there is IPv4-mapped overlap
 120  */
 121 #ifndef _SYS_SYSPROTO_H_
 122 struct nfssvc_args {
 123         int flag;
 124         caddr_t argp;
 125 };
 126 #endif
 127 int
 128 nfssvc(struct thread *td, struct nfssvc_args *uap)
 129 {
 130         struct file *fp;
 131         struct sockaddr *nam;
 132         struct nfsd_args nfsdarg;
 133         int error;
 134
 135         KASSERT(!mtx_owned(&Giant), ("nfssvc(): called with Giant"));
 136
 137 #ifdef MAC
 138         error = mac_check_system_nfsd(td->td_ucred);
 139         if (error)
 140                 return (error);
 141 #endif
 142         error = priv_check(td, PRIV_NFSD);
 143         if (error)
 144                 return (error);
 145         NET_LOCK_GIANT();
 146         NFSD_LOCK();
 147         while (nfssvc_sockhead_flag & SLP_INIT) {
 148                  nfssvc_sockhead_flag |= SLP_WANTINIT;
 149                 (void) msleep(&nfssvc_sockhead, &nfsd_mtx, PSOCK,
 150                     "nfsd init", 0);
 151         }
 152         NFSD_UNLOCK();
 153         if (uap->flag & NFSSVC_ADDSOCK) {
 154                 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg));
 155                 if (error)
 156                         goto done2;
 157                 if ((error = fget(td, nfsdarg.sock, &fp)) != 0)
 158                         goto done2;
 159                 if (fp->f_type != DTYPE_SOCKET) {
 160                         fdrop(fp, td);
 161                         goto done2;
 162                 }
 163                 /*
 164                  * Get the client address for connected sockets.
 165                  */
 166                 if (nfsdarg.name == NULL || nfsdarg.namelen == 0)
 167                         nam = NULL;
 168                 else {
 169                         error = getsockaddr(&nam, nfsdarg.name,
 170                                             nfsdarg.namelen);
 171                         if (error) {
 172                                 fdrop(fp, td);
 173                                 goto done2;
 174                         }
 175                 }
 176                 error = nfssvc_addsock(fp, nam, td);
 177                 fdrop(fp, td);
 178         } else if (uap->flag & NFSSVC_NFSD) {
 179                 error = nfssvc_nfsd(td);
 180         } else {
 181                 error = ENXIO;
 182         }
 183         if (error == EINTR || error == ERESTART)
 184                 error = 0;
 185 done2:
 186         NET_UNLOCK_GIANT();
 187         return (error);
 188 }
 189
 190 /*
 191  * Adds a socket to the list for servicing by nfsds.
 192  */
 193 static int
 194 nfssvc_addsock(struct file *fp, struct sockaddr *mynam, struct thread *td)
 195 {
 196         int siz;
 197         struct nfssvc_sock *slp;
 198         struct socket *so;
 199         int error, s;
 200
 201         NET_ASSERT_GIANT();
 202
 203         so = fp->f_data;
 204 #if 0
 205         /*
 206          * XXXRW: If this code is ever enabled, there's a race when running
 207          * MPSAFE.
 208          */
 209         tslp = NULL;
 210         /*
 211          * Add it to the list, as required.
 212          */
 213         if (so->so_proto->pr_protocol == IPPROTO_UDP) {
 214                 tslp = nfs_udpsock;
 215                 if (tslp->ns_flag & SLP_VALID) {
 216                         if (mynam != NULL)
 217                                 FREE(mynam, M_SONAME);
 218                         return (EPERM);
 219                 }
 220         }
 221 #endif
 222         if (so->so_type == SOCK_STREAM)
 223                 siz = NFS_MAXPACKET + sizeof (u_long);
 224         else
 225                 siz = NFS_MAXPACKET;
 226         error = soreserve(so, siz, siz);
 227         if (error) {
 228                 if (mynam != NULL)
 229                         FREE(mynam, M_SONAME);
 230                 return (error);
 231         }
 232
 233         /*
 234          * Set protocol specific options { for now TCP only } and
 235          * reserve some space. For datagram sockets, this can get called
 236          * repeatedly for the same socket, but that isn't harmful.
 237          */
 238         if (so->so_type == SOCK_STREAM) {
 239                 struct sockopt sopt;
 240                 int val;
 241
 242                 bzero(&sopt, sizeof sopt);
 243                 sopt.sopt_dir = SOPT_SET;
 244                 sopt.sopt_level = SOL_SOCKET;
 245                 sopt.sopt_name = SO_KEEPALIVE;
 246                 sopt.sopt_val = &val;
 247                 sopt.sopt_valsize = sizeof val;
 248                 val = 1;
 249                 sosetopt(so, &sopt);
 250         }
 251         if (so->so_proto->pr_protocol == IPPROTO_TCP) {
 252                 struct sockopt sopt;
 253                 int val;
 254
 255                 bzero(&sopt, sizeof sopt);
 256                 sopt.sopt_dir = SOPT_SET;
 257                 sopt.sopt_level = IPPROTO_TCP;
 258                 sopt.sopt_name = TCP_NODELAY;
 259                 sopt.sopt_val = &val;
 260                 sopt.sopt_valsize = sizeof val;
 261                 val = 1;
 262                 sosetopt(so, &sopt);
 263         }
 264         SOCKBUF_LOCK(&so->so_rcv);
 265         so->so_rcv.sb_flags &= ~SB_NOINTR;
 266         so->so_rcv.sb_timeo = 0;
 267         SOCKBUF_UNLOCK(&so->so_rcv);
 268         SOCKBUF_LOCK(&so->so_snd);
 269         so->so_snd.sb_flags &= ~SB_NOINTR;
 270         so->so_snd.sb_timeo = 0;
 271         SOCKBUF_UNLOCK(&so->so_snd);
 272
 273         slp = (struct nfssvc_sock *)
 274                 malloc(sizeof (struct nfssvc_sock), M_NFSSVC,
 275                 M_WAITOK | M_ZERO);
 276         STAILQ_INIT(&slp->ns_rec);
 277         NFSD_LOCK();
 278         TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain);
 279
 280         slp->ns_so = so;
 281         slp->ns_nam = mynam;
 282         fhold(fp);
 283         slp->ns_fp = fp;
 284         /*
 285          * XXXRW: Socket locking here?
 286          */
 287         s = splnet();
 288         so->so_upcallarg = (caddr_t)slp;
 289         so->so_upcall = nfsrv_rcv;
 290         SOCKBUF_LOCK(&so->so_rcv);
 291         so->so_rcv.sb_flags |= SB_UPCALL;
 292         SOCKBUF_UNLOCK(&so->so_rcv);
 293         slp->ns_flag = (SLP_VALID | SLP_NEEDQ);
 294         nfsrv_wakenfsd(slp);
 295         splx(s);
 296         NFSD_UNLOCK();
 297         return (0);
 298 }
 299
 300 /*
 301  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
 302  * until it is killed by a signal.
 303  */
 304 static int
 305 nfssvc_nfsd(struct thread *td)
 306 {
 307         int siz;
 308         struct nfssvc_sock *slp;
 309         struct nfsd *nfsd;
 310         struct nfsrv_descript *nd = NULL;
 311         struct mbuf *m, *mreq;
 312         int error = 0, cacherep, s, sotype, writes_todo;
 313         int procrastinate;
 314         u_quad_t cur_usec;
 315
 316         NET_ASSERT_GIANT();
 317
 318 #ifndef nolint
 319         cacherep = RC_DOIT;
 320         writes_todo = 0;
 321 #endif
 322         nfsd = (struct nfsd *)
 323                 malloc(sizeof (struct nfsd), M_NFSD, M_WAITOK | M_ZERO);
 324         s = splnet();
 325         NFSD_LOCK();
 326
 327         nfsd->nfsd_td = td;
 328         TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
 329         nfsrv_numnfsd++;
 330
 331         /*
 332          * Loop getting rpc requests until SIGKILL.
 333          */
 334         for (;;) {
 335                 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) {
 336                         while (nfsd->nfsd_slp == NULL &&
 337                             (nfsd_head_flag & NFSD_CHECKSLP) == 0) {
 338                                 nfsd->nfsd_flag |= NFSD_WAITING;
 339                                 nfsd_waiting++;
 340                                 error = msleep(nfsd, &nfsd_mtx,
 341                                     PSOCK | PCATCH, "-", 0);
 342                                 nfsd_waiting--;
 343                                 if (error)
 344                                         goto done;
 345                         }
 346                         if (nfsd->nfsd_slp == NULL &&
 347                             (nfsd_head_flag & NFSD_CHECKSLP) != 0) {
 348                                 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
 349                                     if ((slp->ns_flag & (SLP_VALID | SLP_DOREC))
 350                                         == (SLP_VALID | SLP_DOREC)) {
 351                                             slp->ns_flag &= ~SLP_DOREC;
 352                                             slp->ns_sref++;
 353                                             nfsd->nfsd_slp = slp;
 354                                             break;
 355                                     }
 356                                 }
 357                                 if (slp == NULL)
 358                                         nfsd_head_flag &= ~NFSD_CHECKSLP;
 359                         }
 360                         if ((slp = nfsd->nfsd_slp) == NULL)
 361                                 continue;
 362                         if (slp->ns_flag & SLP_VALID) {
 363                                 if (slp->ns_flag & SLP_DISCONN)
 364                                         nfsrv_zapsock(slp);
 365                                 else if (slp->ns_flag & SLP_NEEDQ) {
 366                                         slp->ns_flag &= ~SLP_NEEDQ;
 367                                         (void) nfs_slplock(slp, 1);
 368                                         NFSD_UNLOCK();
 369                                         nfsrv_rcv(slp->ns_so, (caddr_t)slp,
 370                                                 M_TRYWAIT);
 371                                         NFSD_LOCK();
 372                                         nfs_slpunlock(slp);
 373                                 }
 374                                 error = nfsrv_dorec(slp, nfsd, &nd);
 375                                 cur_usec = nfs_curusec();
 376                                 if (error && LIST_FIRST(&slp->ns_tq) &&
 377                                     LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) {
 378                                         error = 0;
 379                                         cacherep = RC_DOIT;
 380                                         writes_todo = 1;
 381                                 } else
 382                                         writes_todo = 0;
 383                                 nfsd->nfsd_flag |= NFSD_REQINPROG;
 384                         }
 385                 } else {
 386                         error = 0;
 387                         slp = nfsd->nfsd_slp;
 388                 }
 389                 if (error || (slp->ns_flag & SLP_VALID) == 0) {
 390                         if (nd) {
 391                                 if (nd->nd_cr != NULL)
 392                                         crfree(nd->nd_cr);
 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 #ifdef INET6
 430                             char b6[INET6_ADDRSTRLEN];
 431 #if defined(KLD_MODULE)
 432         /* Do not use ip6_sprintf: the nfs module should work without INET6. */
 433 #define ip6_sprintf(buf, a) \
 434          (sprintf((buf), "%x:%x:%x:%x:%x:%x:%x:%x", \
 435                   (a)->s6_addr16[0], (a)->s6_addr16[1], \
 436                   (a)->s6_addr16[2], (a)->s6_addr16[3], \
 437                   (a)->s6_addr16[4], (a)->s6_addr16[5], \
 438                   (a)->s6_addr16[6], (a)->s6_addr16[7]), \
 439          (buf))
 440 #endif
 441 #endif
 442                             nd->nd_procnum = NFSPROC_NOOP;
 443                             nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
 444                             cacherep = RC_DOIT;
 445                             printf("NFS request from unprivileged port (%s:%d)\n",
 446 #ifdef INET6
 447                                 sin->sin_family == AF_INET6 ?
 448                                     ip6_sprintf(b6, &satosin6(sin)->sin6_addr) :
 449 #if defined(KLD_MODULE)
 450 #undef ip6_sprintf
 451 #endif
 452 #endif
 453                                     inet_ntoa(sin->sin_addr), port);
 454                         }
 455                     }
 456
 457                 }
 458
 459                 /*
 460                  * Loop to get all the write rpc relies that have been
 461                  * gathered together.
 462                  */
 463                 do {
 464                     switch (cacherep) {
 465                     case RC_DOIT:
 466                         if (nd && (nd->nd_flag & ND_NFSV3))
 467                             procrastinate = nfsrvw_procrastinate_v3;
 468                         else
 469                             procrastinate = nfsrvw_procrastinate;
 470                         NFSD_UNLOCK();
 471                         if (writes_todo || (!(nd->nd_flag & ND_NFSV3) &&
 472                             nd->nd_procnum == NFSPROC_WRITE &&
 473                             procrastinate > 0 && !notstarted))
 474                             error = nfsrv_writegather(&nd, slp,
 475                                 nfsd->nfsd_td, &mreq);
 476                         else
 477                             error = (*(nfsrv3_procs[nd->nd_procnum]))(nd,
 478                                 slp, nfsd->nfsd_td, &mreq);
 479                         NFSD_LOCK();
 480                         if (mreq == NULL)
 481                                 break;
 482                         if (error != 0 && error != NFSERR_RETVOID) {
 483                                 nfsrvstats.srv_errs++;
 484                                 nfsrv_updatecache(nd, FALSE, mreq);
 485                                 if (nd->nd_nam2)
 486                                         FREE(nd->nd_nam2, M_SONAME);
 487                                 break;
 488                         }
 489                         nfsrvstats.srvrpccnt[nd->nd_procnum]++;
 490                         nfsrv_updatecache(nd, TRUE, mreq);
 491                         nd->nd_mrep = NULL;
 492                         /* FALLTHROUGH */
 493                     case RC_REPLY:
 494                         NFSD_UNLOCK();
 495                         siz = m_length(mreq, NULL);
 496                         if (siz <= 0 || siz > NFS_MAXPACKET) {
 497                                 printf("mbuf siz=%d\n",siz);
 498                                 panic("Bad nfs svc reply");
 499                         }
 500                         m = mreq;
 501                         m->m_pkthdr.len = siz;
 502                         m->m_pkthdr.rcvif = NULL;
 503                         /*
 504                          * For stream protocols, prepend a Sun RPC
 505                          * Record Mark.
 506                          */
 507                         if (sotype == SOCK_STREAM) {
 508                                 M_PREPEND(m, NFSX_UNSIGNED, M_TRYWAIT);
 509                                 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz);
 510                         }
 511                         NFSD_LOCK();
 512                         if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
 513                                 (void) nfs_slplock(slp, 1);
 514                         if (slp->ns_flag & SLP_VALID) {
 515                             NFSD_UNLOCK();
 516                             error = nfsrv_send(slp->ns_so, nd->nd_nam2, m);
 517                             NFSD_LOCK();
 518                         } else {
 519                             error = EPIPE;
 520                             m_freem(m);
 521                         }
 522                         if (nd->nd_nam2)
 523                                 FREE(nd->nd_nam2, M_SONAME);
 524                         if (nd->nd_mrep)
 525                                 m_freem(nd->nd_mrep);
 526                         if (error == EPIPE)
 527                                 nfsrv_zapsock(slp);
 528                         if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
 529                                 nfs_slpunlock(slp);
 530                         if (error == EINTR || error == ERESTART) {
 531                                 if (nd->nd_cr != NULL)
 532                                         crfree(nd->nd_cr);
 533                                 free((caddr_t)nd, M_NFSRVDESC);
 534                                 nfsrv_slpderef(slp);
 535                                 s = splnet();
 536                                 goto done;
 537                         }
 538                         break;
 539                     case RC_DROPIT:
 540                         m_freem(nd->nd_mrep);
 541                         if (nd->nd_nam2)
 542                                 FREE(nd->nd_nam2, M_SONAME);
 543                         break;
 544                     };
 545                     if (nd) {
 546                         if (nd->nd_cr != NULL)
 547                                 crfree(nd->nd_cr);
 548                         FREE((caddr_t)nd, M_NFSRVDESC);
 549                         nd = NULL;
 550                     }
 551
 552                     /*
 553                      * Check to see if there are outstanding writes that
 554                      * need to be serviced.
 555                      */
 556                     cur_usec = nfs_curusec();
 557                     s = splsoftclock();
 558                     if (LIST_FIRST(&slp->ns_tq) &&
 559                         LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) {
 560                         cacherep = RC_DOIT;
 561                         writes_todo = 1;
 562                     } else
 563                         writes_todo = 0;
 564                     splx(s);
 565                 } while (writes_todo);
 566                 s = splnet();
 567                 if (nfsrv_dorec(slp, nfsd, &nd)) {
 568                         nfsd->nfsd_flag &= ~NFSD_REQINPROG;
 569                         nfsd->nfsd_slp = NULL;
 570                         nfsrv_slpderef(slp);
 571                 }
 572                 KASSERT(!(debug_mpsafenet == 0 && !mtx_owned(&Giant)),
 573                     ("nfssvc_nfsd(): debug.mpsafenet=0 && !Giant"));
 574                 KASSERT(!(debug_mpsafenet == 1 && mtx_owned(&Giant)),
 575                     ("nfssvc_nfsd(): debug.mpsafenet=1 && Giant"));
 576         }
 577 done:
 578         KASSERT(!(debug_mpsafenet == 0 && !mtx_owned(&Giant)),
 579             ("nfssvc_nfsd(): debug.mpsafenet=0 && !Giant"));
 580         KASSERT(!(debug_mpsafenet == 1 && mtx_owned(&Giant)),
 581             ("nfssvc_nfsd(): debug.mpsafenet=1 && Giant"));
 582         TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
 583         splx(s);
 584         free((caddr_t)nfsd, M_NFSD);
 585         if (--nfsrv_numnfsd == 0)
 586                 nfsrv_init(TRUE);       /* Reinitialize everything */
 587         NFSD_UNLOCK();
 588         return (error);
 589 }
 590
 591 /*
 592  * Shut down a socket associated with an nfssvc_sock structure.
 593  * Should be called with the send lock set, if required.
 594  * The trick here is to increment the sref at the start, so that the nfsds
 595  * will stop using it and clear ns_flag at the end so that it will not be
 596  * reassigned during cleanup.
 597  */
 598 static void
 599 nfsrv_zapsock(struct nfssvc_sock *slp)
 600 {
 601         struct nfsrv_descript *nwp, *nnwp;
 602         struct socket *so;
 603         struct file *fp;
 604         struct nfsrv_rec *rec;
 605         int s;
 606
 607         NET_ASSERT_GIANT();
 608         NFSD_LOCK_ASSERT();
 609
 610         /*
 611          * XXXRW: By clearing all flags, other threads/etc should ignore
 612          * this slp and we can safely release nfsd_mtx so we can clean
 613          * up the slp safely.
 614          */
 615         slp->ns_flag &= ~SLP_ALLFLAGS;
 616         fp = slp->ns_fp;
 617         if (fp) {
 618                 NFSD_UNLOCK();
 619                 slp->ns_fp = NULL;
 620                 so = slp->ns_so;
 621                 SOCKBUF_LOCK(&so->so_rcv);
 622                 so->so_rcv.sb_flags &= ~SB_UPCALL;
 623                 SOCKBUF_UNLOCK(&so->so_rcv);
 624                 so->so_upcall = NULL;
 625                 so->so_upcallarg = NULL;
 626                 soshutdown(so, SHUT_RDWR);
 627                 closef(fp, NULL);
 628                 NFSD_LOCK();
 629                 if (slp->ns_nam)
 630                         FREE(slp->ns_nam, M_SONAME);
 631                 m_freem(slp->ns_raw);
 632                 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) {
 633                         STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link);
 634                         if (rec->nr_address)
 635                                 FREE(rec->nr_address, M_SONAME);
 636                         m_freem(rec->nr_packet);
 637                         free(rec, M_NFSRVDESC);
 638                 }
 639                 s = splsoftclock();
 640                 for (nwp = LIST_FIRST(&slp->ns_tq); nwp; nwp = nnwp) {
 641                         nnwp = LIST_NEXT(nwp, nd_tq);
 642                         LIST_REMOVE(nwp, nd_tq);
 643                         if (nwp->nd_cr != NULL)
 644                                 crfree(nwp->nd_cr);
 645                         free((caddr_t)nwp, M_NFSRVDESC);
 646                 }
 647                 LIST_INIT(&slp->ns_tq);
 648                 splx(s);
 649         }
 650 }
 651
 652 /*
 653  * Derefence a server socket structure. If it has no more references and
 654  * is no longer valid, you can throw it away.
 655  */
 656 void
 657 nfsrv_slpderef(struct nfssvc_sock *slp)
 658 {
 659
 660         NFSD_LOCK_ASSERT();
 661
 662         if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) {
 663                 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
 664                 free((caddr_t)slp, M_NFSSVC);
 665         }
 666 }
 667
 668 /*
 669  * Lock a socket against others.
 670  *
 671  * XXXRW: Wait argument is always 1 in the caller.  Replace with a real
 672  * sleep lock?
 673  */
 674 int
 675 nfs_slplock(struct nfssvc_sock *slp, int wait)
 676 {
 677         int *statep = &slp->ns_solock;
 678
 679         NFSD_LOCK_ASSERT();
 680
 681         if (!wait && (*statep & NFSRV_SNDLOCK))
 682                 return(0);      /* already locked, fail */
 683         while (*statep & NFSRV_SNDLOCK) {
 684                 *statep |= NFSRV_WANTSND;
 685                 (void) msleep(statep, &nfsd_mtx, PZERO - 1, "nfsslplck", 0);
 686         }
 687         *statep |= NFSRV_SNDLOCK;
 688         return (1);
 689 }
 690
 691 /*
 692  * Unlock the stream socket for others.
 693  */
 694 void
 695 nfs_slpunlock(struct nfssvc_sock *slp)
 696 {
 697         int *statep = &slp->ns_solock;
 698
 699         NFSD_LOCK_ASSERT();
 700
 701         if ((*statep & NFSRV_SNDLOCK) == 0)
 702                 panic("nfs slpunlock");
 703         *statep &= ~NFSRV_SNDLOCK;
 704         if (*statep & NFSRV_WANTSND) {
 705                 *statep &= ~NFSRV_WANTSND;
 706                 wakeup(statep);
 707         }
 708 }
 709
 710 /*
 711  * Initialize the data structures for the server.
 712  * Handshake with any new nfsds starting up to avoid any chance of
 713  * corruption.
 714  */
 715 void
 716 nfsrv_init(int terminating)
 717 {
 718         struct nfssvc_sock *slp, *nslp;
 719
 720         NET_ASSERT_GIANT();
 721         NFSD_LOCK_ASSERT();
 722
 723         if (nfssvc_sockhead_flag & SLP_INIT)
 724                 panic("nfsd init");
 725         nfssvc_sockhead_flag |= SLP_INIT;
 726         if (terminating) {
 727                 TAILQ_FOREACH_SAFE(slp, &nfssvc_sockhead, ns_chain, nslp) {
 728                         if (slp->ns_flag & SLP_VALID)
 729                                 nfsrv_zapsock(slp);
 730                         TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
 731                         free((caddr_t)slp, M_NFSSVC);
 732                 }
 733                 nfsrv_cleancache();     /* And clear out server cache */
 734         } else
 735                 nfs_pub.np_valid = 0;
 736
 737         TAILQ_INIT(&nfssvc_sockhead);
 738         nfssvc_sockhead_flag &= ~SLP_INIT;
 739         if (nfssvc_sockhead_flag & SLP_WANTINIT) {
 740                 nfssvc_sockhead_flag &= ~SLP_WANTINIT;
 741                 wakeup(&nfssvc_sockhead);
 742         }
 743
 744         TAILQ_INIT(&nfsd_head);
 745         nfsd_head_flag &= ~NFSD_CHECKSLP;
 746
 747 #if 0
 748         nfs_udpsock = (struct nfssvc_sock *)
 749             malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
 750         STAILQ_INIT(&nfs_udpsock->ns_rec);
 751         TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain);
 752
 753         nfs_cltpsock = (struct nfssvc_sock *)
 754             malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
 755         STAILQ_INIT(&nfs_cltpsock->ns_rec);
 756         TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain);
 757 #endif
 758 }