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, "nfsserver_srvsock", "Nfs server structure");
  80
  81 MALLOC_DEFINE(M_NFSRVDESC, "nfsserver_srvdesc", "NFS server socket descriptor");
  82 MALLOC_DEFINE(M_NFSD, "nfsserver_daemon", "Nfs server daemon structure");
  83
  84
  85 #define TRUE    1
  86 #define FALSE   0
  87
  88 SYSCTL_DECL(_vfs_nfsrv);
  89
  90 int             nfsd_waiting = 0;
  91 int             nfsrv_numnfsd = 0;
  92 static int      notstarted = 1;
  93
  94 static int      nfs_privport = 0;
  95 SYSCTL_INT(_vfs_nfsrv, NFS_NFSPRIVPORT, nfs_privport, CTLFLAG_RW,
  96             &nfs_privport, 0, "");
  97 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, gatherdelay, CTLFLAG_RW,
  98             &nfsrvw_procrastinate, 0, "");
  99 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, gatherdelay_v3, CTLFLAG_RW,
 100             &nfsrvw_procrastinate_v3, 0, "");
 101
 102 static int      nfssvc_addsock(struct file *, struct sockaddr *,
 103                     struct thread *);
 104 static void     nfsrv_zapsock(struct nfssvc_sock *slp);
 105 static int      nfssvc_nfsd(struct thread *);
 106
 107 /*
 108  * NFS server system calls
 109  */
 110
 111 /*
 112  * Nfs server psuedo system call for the nfsd's
 113  * Based on the flag value it either:
 114  * - adds a socket to the selection list
 115  * - remains in the kernel as an nfsd
 116  * - remains in the kernel as an nfsiod
 117  * For INET6 we suppose that nfsd provides only IN6P_IPV6_V6ONLY sockets
 118  * and that mountd provides
 119  *  - sockaddr with no IPv4-mapped addresses
 120  *  - mask for both INET and INET6 families if there is IPv4-mapped overlap
 121  */
 122 #ifndef _SYS_SYSPROTO_H_
 123 struct nfssvc_args {
 124         int flag;
 125         caddr_t argp;
 126 };
 127 #endif
 128 /*
 129  * MPSAFE
 130  */
 131 int
 132 nfssvc(struct thread *td, struct nfssvc_args *uap)
 133 {
 134         struct file *fp;
 135         struct sockaddr *nam;
 136         struct nfsd_args nfsdarg;
 137         int error;
 138
 139         KASSERT(!mtx_owned(&Giant), ("nfssvc(): called with Giant"));
 140
 141 #ifdef MAC
 142         error = mac_check_system_nfsd(td->td_ucred);
 143         if (error)
 144                 return (error);
 145 #endif
 146         error = priv_check(td, PRIV_NFSD);
 147         if (error)
 148                 return (error);
 149         NET_LOCK_GIANT();
 150         NFSD_LOCK();
 151         while (nfssvc_sockhead_flag & SLP_INIT) {
 152                  nfssvc_sockhead_flag |= SLP_WANTINIT;
 153                 (void) msleep(&nfssvc_sockhead, &nfsd_mtx, PSOCK,
 154                     "nfsd init", 0);
 155         }
 156         NFSD_UNLOCK();
 157         if (uap->flag & NFSSVC_ADDSOCK) {
 158                 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg));
 159                 if (error)
 160                         goto done2;
 161                 if ((error = fget(td, nfsdarg.sock, &fp)) != 0)
 162                         goto done2;
 163                 if (fp->f_type != DTYPE_SOCKET) {
 164                         fdrop(fp, td);
 165                         goto done2;
 166                 }
 167                 /*
 168                  * Get the client address for connected sockets.
 169                  */
 170                 if (nfsdarg.name == NULL || nfsdarg.namelen == 0)
 171                         nam = NULL;
 172                 else {
 173                         error = getsockaddr(&nam, nfsdarg.name,
 174                                             nfsdarg.namelen);
 175                         if (error) {
 176                                 fdrop(fp, td);
 177                                 goto done2;
 178                         }
 179                 }
 180                 error = nfssvc_addsock(fp, nam, td);
 181                 fdrop(fp, td);
 182         } else if (uap->flag & NFSSVC_NFSD) {
 183                 error = nfssvc_nfsd(td);
 184         } else {
 185                 error = ENXIO;
 186         }
 187         if (error == EINTR || error == ERESTART)
 188                 error = 0;
 189 done2:
 190         NET_UNLOCK_GIANT();
 191         return (error);
 192 }
 193
 194 /*
 195  * Adds a socket to the list for servicing by nfsds.
 196  */
 197 static int
 198 nfssvc_addsock(struct file *fp, struct sockaddr *mynam, struct thread *td)
 199 {
 200         int siz;
 201         struct nfssvc_sock *slp;
 202         struct socket *so;
 203         int error, s;
 204
 205         NET_ASSERT_GIANT();
 206
 207         so = fp->f_data;
 208 #if 0
 209         /*
 210          * XXXRW: If this code is ever enabled, there's a race when running
 211          * MPSAFE.
 212          */
 213         tslp = NULL;
 214         /*
 215          * Add it to the list, as required.
 216          */
 217         if (so->so_proto->pr_protocol == IPPROTO_UDP) {
 218                 tslp = nfs_udpsock;
 219                 if (tslp->ns_flag & SLP_VALID) {
 220                         if (mynam != NULL)
 221                                 FREE(mynam, M_SONAME);
 222                         return (EPERM);
 223                 }
 224         }
 225 #endif
 226         if (so->so_type == SOCK_STREAM)
 227                 siz = NFS_MAXPACKET + sizeof (u_long);
 228         else
 229                 siz = NFS_MAXPACKET;
 230         error = soreserve(so, siz, siz);
 231         if (error) {
 232                 if (mynam != NULL)
 233                         FREE(mynam, M_SONAME);
 234                 return (error);
 235         }
 236
 237         /*
 238          * Set protocol specific options { for now TCP only } and
 239          * reserve some space. For datagram sockets, this can get called
 240          * repeatedly for the same socket, but that isn't harmful.
 241          */
 242         if (so->so_type == SOCK_STREAM) {
 243                 struct sockopt sopt;
 244                 int val;
 245
 246                 bzero(&sopt, sizeof sopt);
 247                 sopt.sopt_dir = SOPT_SET;
 248                 sopt.sopt_level = SOL_SOCKET;
 249                 sopt.sopt_name = SO_KEEPALIVE;
 250                 sopt.sopt_val = &val;
 251                 sopt.sopt_valsize = sizeof val;
 252                 val = 1;
 253                 sosetopt(so, &sopt);
 254         }
 255         if (so->so_proto->pr_protocol == IPPROTO_TCP) {
 256                 struct sockopt sopt;
 257                 int val;
 258
 259                 bzero(&sopt, sizeof sopt);
 260                 sopt.sopt_dir = SOPT_SET;
 261                 sopt.sopt_level = IPPROTO_TCP;
 262                 sopt.sopt_name = TCP_NODELAY;
 263                 sopt.sopt_val = &val;
 264                 sopt.sopt_valsize = sizeof val;
 265                 val = 1;
 266                 sosetopt(so, &sopt);
 267         }
 268         SOCKBUF_LOCK(&so->so_rcv);
 269         so->so_rcv.sb_flags &= ~SB_NOINTR;
 270         so->so_rcv.sb_timeo = 0;
 271         SOCKBUF_UNLOCK(&so->so_rcv);
 272         SOCKBUF_LOCK(&so->so_snd);
 273         so->so_snd.sb_flags &= ~SB_NOINTR;
 274         so->so_snd.sb_timeo = 0;
 275         SOCKBUF_UNLOCK(&so->so_snd);
 276
 277         slp = (struct nfssvc_sock *)
 278                 malloc(sizeof (struct nfssvc_sock), M_NFSSVC,
 279                 M_WAITOK | M_ZERO);
 280         STAILQ_INIT(&slp->ns_rec);
 281         NFSD_LOCK();
 282         TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain);
 283
 284         slp->ns_so = so;
 285         slp->ns_nam = mynam;
 286         fhold(fp);
 287         slp->ns_fp = fp;
 288         /*
 289          * XXXRW: Socket locking here?
 290          */
 291         s = splnet();
 292         so->so_upcallarg = (caddr_t)slp;
 293         so->so_upcall = nfsrv_rcv;
 294         SOCKBUF_LOCK(&so->so_rcv);
 295         so->so_rcv.sb_flags |= SB_UPCALL;
 296         SOCKBUF_UNLOCK(&so->so_rcv);
 297         slp->ns_flag = (SLP_VALID | SLP_NEEDQ);
 298         nfsrv_wakenfsd(slp);
 299         splx(s);
 300         NFSD_UNLOCK();
 301         return (0);
 302 }
 303
 304 /*
 305  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
 306  * until it is killed by a signal.
 307  */
 308 static int
 309 nfssvc_nfsd(struct thread *td)
 310 {
 311         int siz;
 312         struct nfssvc_sock *slp;
 313         struct nfsd *nfsd;
 314         struct nfsrv_descript *nd = NULL;
 315         struct mbuf *m, *mreq;
 316         int error = 0, cacherep, s, sotype, writes_todo;
 317         int procrastinate;
 318         u_quad_t cur_usec;
 319
 320         NET_ASSERT_GIANT();
 321
 322 #ifndef nolint
 323         cacherep = RC_DOIT;
 324         writes_todo = 0;
 325 #endif
 326         nfsd = (struct nfsd *)
 327                 malloc(sizeof (struct nfsd), M_NFSD, M_WAITOK | M_ZERO);
 328         s = splnet();
 329         NFSD_LOCK();
 330
 331         nfsd->nfsd_td = td;
 332         TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
 333         nfsrv_numnfsd++;
 334
 335         /*
 336          * Loop getting rpc requests until SIGKILL.
 337          */
 338         for (;;) {
 339                 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) {
 340                         while (nfsd->nfsd_slp == NULL &&
 341                             (nfsd_head_flag & NFSD_CHECKSLP) == 0) {
 342                                 nfsd->nfsd_flag |= NFSD_WAITING;
 343                                 nfsd_waiting++;
 344                                 error = msleep(nfsd, &nfsd_mtx,
 345                                     PSOCK | PCATCH, "-", 0);
 346                                 nfsd_waiting--;
 347                                 if (error)
 348                                         goto done;
 349                         }
 350                         if (nfsd->nfsd_slp == NULL &&
 351                             (nfsd_head_flag & NFSD_CHECKSLP) != 0) {
 352                                 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
 353                                     if ((slp->ns_flag & (SLP_VALID | SLP_DOREC))
 354                                         == (SLP_VALID | SLP_DOREC)) {
 355                                             slp->ns_flag &= ~SLP_DOREC;
 356                                             slp->ns_sref++;
 357                                             nfsd->nfsd_slp = slp;
 358                                             break;
 359                                     }
 360                                 }
 361                                 if (slp == NULL)
 362                                         nfsd_head_flag &= ~NFSD_CHECKSLP;
 363                         }
 364                         if ((slp = nfsd->nfsd_slp) == NULL)
 365                                 continue;
 366                         if (slp->ns_flag & SLP_VALID) {
 367                                 if (slp->ns_flag & SLP_DISCONN)
 368                                         nfsrv_zapsock(slp);
 369                                 else if (slp->ns_flag & SLP_NEEDQ) {
 370                                         slp->ns_flag &= ~SLP_NEEDQ;
 371                                         (void) nfs_slplock(slp, 1);
 372                                         NFSD_UNLOCK();
 373                                         nfsrv_rcv(slp->ns_so, (caddr_t)slp,
 374                                                 M_TRYWAIT);
 375                                         NFSD_LOCK();
 376                                         nfs_slpunlock(slp);
 377                                 }
 378                                 error = nfsrv_dorec(slp, nfsd, &nd);
 379                                 cur_usec = nfs_curusec();
 380                                 if (error && LIST_FIRST(&slp->ns_tq) &&
 381                                     LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) {
 382                                         error = 0;
 383                                         cacherep = RC_DOIT;
 384                                         writes_todo = 1;
 385                                 } else
 386                                         writes_todo = 0;
 387                                 nfsd->nfsd_flag |= NFSD_REQINPROG;
 388                         }
 389                 } else {
 390                         error = 0;
 391                         slp = nfsd->nfsd_slp;
 392                 }
 393                 if (error || (slp->ns_flag & SLP_VALID) == 0) {
 394                         if (nd) {
 395                                 if (nd->nd_cr != NULL)
 396                                         crfree(nd->nd_cr);
 397                                 free((caddr_t)nd, M_NFSRVDESC);
 398                                 nd = NULL;
 399                         }
 400                         nfsd->nfsd_slp = NULL;
 401                         nfsd->nfsd_flag &= ~NFSD_REQINPROG;
 402                         nfsrv_slpderef(slp);
 403                         continue;
 404                 }
 405                 splx(s);
 406                 sotype = slp->ns_so->so_type;
 407                 if (nd) {
 408                     getmicrotime(&nd->nd_starttime);
 409                     if (nd->nd_nam2)
 410                         nd->nd_nam = nd->nd_nam2;
 411                     else
 412                         nd->nd_nam = slp->ns_nam;
 413
 414                     /*
 415                      * Check to see if authorization is needed.
 416                      */
 417                     cacherep = nfsrv_getcache(nd, &mreq);
 418
 419                     if (nfs_privport) {
 420                         /* Check if source port is privileged */
 421                         u_short port;
 422                         struct sockaddr *nam = nd->nd_nam;
 423                         struct sockaddr_in *sin;
 424
 425                         sin = (struct sockaddr_in *)nam;
 426                         /*
 427                          * INET/INET6 - same code:
 428                          *    sin_port and sin6_port are at same offset
 429                          */
 430                         port = ntohs(sin->sin_port);
 431                         if (port >= IPPORT_RESERVED &&
 432                             nd->nd_procnum != NFSPROC_NULL) {
 433 #ifdef INET6
 434                             char b6[INET6_ADDRSTRLEN];
 435 #if defined(KLD_MODULE)
 436         /* Do not use ip6_sprintf: the nfs module should work without INET6. */
 437 #define ip6_sprintf(buf, a) \
 438          (sprintf((buf), "%x:%x:%x:%x:%x:%x:%x:%x", \
 439                   (a)->s6_addr16[0], (a)->s6_addr16[1], \
 440                   (a)->s6_addr16[2], (a)->s6_addr16[3], \
 441                   (a)->s6_addr16[4], (a)->s6_addr16[5], \
 442                   (a)->s6_addr16[6], (a)->s6_addr16[7]), \
 443          (buf))
 444 #endif
 445 #endif
 446                             nd->nd_procnum = NFSPROC_NOOP;
 447                             nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
 448                             cacherep = RC_DOIT;
 449                             printf("NFS request from unprivileged port (%s:%d)\n",
 450 #ifdef INET6
 451                                 sin->sin_family == AF_INET6 ?
 452                                     ip6_sprintf(b6, &satosin6(sin)->sin6_addr) :
 453 #if defined(KLD_MODULE)
 454 #undef ip6_sprintf
 455 #endif
 456 #endif
 457                                     inet_ntoa(sin->sin_addr), port);
 458                         }
 459                     }
 460
 461                 }
 462
 463                 /*
 464                  * Loop to get all the write rpc relies that have been
 465                  * gathered together.
 466                  */
 467                 do {
 468                     switch (cacherep) {
 469                     case RC_DOIT:
 470                         if (nd && (nd->nd_flag & ND_NFSV3))
 471                             procrastinate = nfsrvw_procrastinate_v3;
 472                         else
 473                             procrastinate = nfsrvw_procrastinate;
 474                         if (writes_todo || (!(nd->nd_flag & ND_NFSV3) &&
 475                             nd->nd_procnum == NFSPROC_WRITE &&
 476                             procrastinate > 0 && !notstarted))
 477                             error = nfsrv_writegather(&nd, slp,
 478                                 nfsd->nfsd_td, &mreq);
 479                         else
 480                             error = (*(nfsrv3_procs[nd->nd_procnum]))(nd,
 481                                 slp, nfsd->nfsd_td, &mreq);
 482                         if (mreq == NULL)
 483                                 break;
 484                         if (error != 0 && error != NFSERR_RETVOID) {
 485                                 nfsrvstats.srv_errs++;
 486                                 nfsrv_updatecache(nd, FALSE, mreq);
 487                                 if (nd->nd_nam2)
 488                                         FREE(nd->nd_nam2, M_SONAME);
 489                                 break;
 490                         }
 491                         nfsrvstats.srvrpccnt[nd->nd_procnum]++;
 492                         nfsrv_updatecache(nd, TRUE, mreq);
 493                         nd->nd_mrep = NULL;
 494                         /* FALLTHROUGH */
 495                     case RC_REPLY:
 496                         NFSD_UNLOCK();
 497                         siz = m_length(mreq, NULL);
 498                         if (siz <= 0 || siz > NFS_MAXPACKET) {
 499                                 printf("mbuf siz=%d\n",siz);
 500                                 panic("Bad nfs svc reply");
 501                         }
 502                         m = mreq;
 503                         m->m_pkthdr.len = siz;
 504                         m->m_pkthdr.rcvif = NULL;
 505                         /*
 506                          * For stream protocols, prepend a Sun RPC
 507                          * Record Mark.
 508                          */
 509                         if (sotype == SOCK_STREAM) {
 510                                 M_PREPEND(m, NFSX_UNSIGNED, M_TRYWAIT);
 511                                 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz);
 512                         }
 513                         NFSD_LOCK();
 514                         if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
 515                                 (void) nfs_slplock(slp, 1);
 516                         if (slp->ns_flag & SLP_VALID) {
 517                             NFSD_UNLOCK();
 518                             error = nfsrv_send(slp->ns_so, nd->nd_nam2, m);
 519                             NFSD_LOCK();
 520                         } else {
 521                             error = EPIPE;
 522                             m_freem(m);
 523                         }
 524                         if (nd->nd_nam2)
 525                                 FREE(nd->nd_nam2, M_SONAME);
 526                         if (nd->nd_mrep)
 527                                 m_freem(nd->nd_mrep);
 528                         if (error == EPIPE)
 529                                 nfsrv_zapsock(slp);
 530                         if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
 531                                 nfs_slpunlock(slp);
 532                         if (error == EINTR || error == ERESTART) {
 533                                 if (nd->nd_cr != NULL)
 534                                         crfree(nd->nd_cr);
 535                                 free((caddr_t)nd, M_NFSRVDESC);
 536                                 nfsrv_slpderef(slp);
 537                                 s = splnet();
 538                                 goto done;
 539                         }
 540                         break;
 541                     case RC_DROPIT:
 542                         m_freem(nd->nd_mrep);
 543                         if (nd->nd_nam2)
 544                                 FREE(nd->nd_nam2, M_SONAME);
 545                         break;
 546                     };
 547                     if (nd) {
 548                         if (nd->nd_cr != NULL)
 549                                 crfree(nd->nd_cr);
 550                         FREE((caddr_t)nd, M_NFSRVDESC);
 551                         nd = NULL;
 552                     }
 553
 554                     /*
 555                      * Check to see if there are outstanding writes that
 556                      * need to be serviced.
 557                      */
 558                     cur_usec = nfs_curusec();
 559                     s = splsoftclock();
 560                     if (LIST_FIRST(&slp->ns_tq) &&
 561                         LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) {
 562                         cacherep = RC_DOIT;
 563                         writes_todo = 1;
 564                     } else
 565                         writes_todo = 0;
 566                     splx(s);
 567                 } while (writes_todo);
 568                 s = splnet();
 569                 if (nfsrv_dorec(slp, nfsd, &nd)) {
 570                         nfsd->nfsd_flag &= ~NFSD_REQINPROG;
 571                         nfsd->nfsd_slp = NULL;
 572                         nfsrv_slpderef(slp);
 573                 }
 574                 KASSERT(!(debug_mpsafenet == 0 && !mtx_owned(&Giant)),
 575                     ("nfssvc_nfsd(): debug.mpsafenet=0 && !Giant"));
 576                 KASSERT(!(debug_mpsafenet == 1 && mtx_owned(&Giant)),
 577                     ("nfssvc_nfsd(): debug.mpsafenet=1 && Giant"));
 578         }
 579 done:
 580         KASSERT(!(debug_mpsafenet == 0 && !mtx_owned(&Giant)),
 581             ("nfssvc_nfsd(): debug.mpsafenet=0 && !Giant"));
 582         KASSERT(!(debug_mpsafenet == 1 && mtx_owned(&Giant)),
 583             ("nfssvc_nfsd(): debug.mpsafenet=1 && Giant"));
 584         TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
 585         splx(s);
 586         free((caddr_t)nfsd, M_NFSD);
 587         if (--nfsrv_numnfsd == 0)
 588                 nfsrv_init(TRUE);       /* Reinitialize everything */
 589         NFSD_UNLOCK();
 590         return (error);
 591 }
 592
 593 /*
 594  * Shut down a socket associated with an nfssvc_sock structure.
 595  * Should be called with the send lock set, if required.
 596  * The trick here is to increment the sref at the start, so that the nfsds
 597  * will stop using it and clear ns_flag at the end so that it will not be
 598  * reassigned during cleanup.
 599  */
 600 static void
 601 nfsrv_zapsock(struct nfssvc_sock *slp)
 602 {
 603         struct nfsrv_descript *nwp, *nnwp;
 604         struct socket *so;
 605         struct file *fp;
 606         struct nfsrv_rec *rec;
 607         int s;
 608
 609         NET_ASSERT_GIANT();
 610         NFSD_LOCK_ASSERT();
 611
 612         /*
 613          * XXXRW: By clearing all flags, other threads/etc should ignore
 614          * this slp and we can safely release nfsd_mtx so we can clean
 615          * up the slp safely.
 616          */
 617         slp->ns_flag &= ~SLP_ALLFLAGS;
 618         fp = slp->ns_fp;
 619         if (fp) {
 620                 NFSD_UNLOCK();
 621                 slp->ns_fp = NULL;
 622                 so = slp->ns_so;
 623                 SOCKBUF_LOCK(&so->so_rcv);
 624                 so->so_rcv.sb_flags &= ~SB_UPCALL;
 625                 SOCKBUF_UNLOCK(&so->so_rcv);
 626                 so->so_upcall = NULL;
 627                 so->so_upcallarg = NULL;
 628                 soshutdown(so, SHUT_RDWR);
 629                 closef(fp, NULL);
 630                 NFSD_LOCK();
 631                 if (slp->ns_nam)
 632                         FREE(slp->ns_nam, M_SONAME);
 633                 m_freem(slp->ns_raw);
 634                 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) {
 635                         STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link);
 636                         if (rec->nr_address)
 637                                 FREE(rec->nr_address, M_SONAME);
 638                         m_freem(rec->nr_packet);
 639                         free(rec, M_NFSRVDESC);
 640                 }
 641                 s = splsoftclock();
 642                 for (nwp = LIST_FIRST(&slp->ns_tq); nwp; nwp = nnwp) {
 643                         nnwp = LIST_NEXT(nwp, nd_tq);
 644                         LIST_REMOVE(nwp, nd_tq);
 645                         if (nwp->nd_cr != NULL)
 646                                 crfree(nwp->nd_cr);
 647                         free((caddr_t)nwp, M_NFSRVDESC);
 648                 }
 649                 LIST_INIT(&slp->ns_tq);
 650                 splx(s);
 651         }
 652 }
 653
 654 /*
 655  * Derefence a server socket structure. If it has no more references and
 656  * is no longer valid, you can throw it away.
 657  */
 658 void
 659 nfsrv_slpderef(struct nfssvc_sock *slp)
 660 {
 661
 662         NFSD_LOCK_ASSERT();
 663
 664         if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) {
 665                 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
 666                 free((caddr_t)slp, M_NFSSVC);
 667         }
 668 }
 669
 670 /*
 671  * Lock a socket against others.
 672  *
 673  * XXXRW: Wait argument is always 1 in the caller.  Replace with a real
 674  * sleep lock?
 675  */
 676 int
 677 nfs_slplock(struct nfssvc_sock *slp, int wait)
 678 {
 679         int *statep = &slp->ns_solock;
 680
 681         NFSD_LOCK_ASSERT();
 682
 683         if (!wait && (*statep & NFSRV_SNDLOCK))
 684                 return(0);      /* already locked, fail */
 685         while (*statep & NFSRV_SNDLOCK) {
 686                 *statep |= NFSRV_WANTSND;
 687                 (void) msleep(statep, &nfsd_mtx, PZERO - 1, "nfsslplck", 0);
 688         }
 689         *statep |= NFSRV_SNDLOCK;
 690         return (1);
 691 }
 692
 693 /*
 694  * Unlock the stream socket for others.
 695  */
 696 void
 697 nfs_slpunlock(struct nfssvc_sock *slp)
 698 {
 699         int *statep = &slp->ns_solock;
 700
 701         NFSD_LOCK_ASSERT();
 702
 703         if ((*statep & NFSRV_SNDLOCK) == 0)
 704                 panic("nfs slpunlock");
 705         *statep &= ~NFSRV_SNDLOCK;
 706         if (*statep & NFSRV_WANTSND) {
 707                 *statep &= ~NFSRV_WANTSND;
 708                 wakeup(statep);
 709         }
 710 }
 711
 712 /*
 713  * Initialize the data structures for the server.
 714  * Handshake with any new nfsds starting up to avoid any chance of
 715  * corruption.
 716  */
 717 void
 718 nfsrv_init(int terminating)
 719 {
 720         struct nfssvc_sock *slp, *nslp;
 721
 722         NET_ASSERT_GIANT();
 723         NFSD_LOCK_ASSERT();
 724
 725         if (nfssvc_sockhead_flag & SLP_INIT)
 726                 panic("nfsd init");
 727         nfssvc_sockhead_flag |= SLP_INIT;
 728         if (terminating) {
 729                 TAILQ_FOREACH_SAFE(slp, &nfssvc_sockhead, ns_chain, nslp) {
 730                         if (slp->ns_flag & SLP_VALID)
 731                                 nfsrv_zapsock(slp);
 732                         TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
 733                         free((caddr_t)slp, M_NFSSVC);
 734                 }
 735                 nfsrv_cleancache();     /* And clear out server cache */
 736         } else
 737                 nfs_pub.np_valid = 0;
 738
 739         TAILQ_INIT(&nfssvc_sockhead);
 740         nfssvc_sockhead_flag &= ~SLP_INIT;
 741         if (nfssvc_sockhead_flag & SLP_WANTINIT) {
 742                 nfssvc_sockhead_flag &= ~SLP_WANTINIT;
 743                 wakeup(&nfssvc_sockhead);
 744         }
 745
 746         TAILQ_INIT(&nfsd_head);
 747         nfsd_head_flag &= ~NFSD_CHECKSLP;
 748
 749 #if 0
 750         nfs_udpsock = (struct nfssvc_sock *)
 751             malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
 752         STAILQ_INIT(&nfs_udpsock->ns_rec);
 753         TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain);
 754
 755         nfs_cltpsock = (struct nfssvc_sock *)
 756             malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
 757         STAILQ_INIT(&nfs_cltpsock->ns_rec);
 758         TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain);
 759 #endif
 760 }