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