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