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, s;
 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         /*
 271          * XXXRW: Socket locking here?
 272          */
 273         s = splnet();
 274         so->so_upcallarg = (caddr_t)slp;
 275         so->so_upcall = nfsrv_rcv;
 276         SOCKBUF_LOCK(&so->so_rcv);
 277         so->so_rcv.sb_flags |= SB_UPCALL;
 278         SOCKBUF_UNLOCK(&so->so_rcv);
 279         slp->ns_flag = (SLP_VALID | SLP_NEEDQ);
 280         nfsrv_wakenfsd(slp);
 281         splx(s);
 282         NFSD_UNLOCK();
 283         return (0);
 284 }
 285
 286 /*
 287  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
 288  * until it is killed by a signal.
 289  */
 290 static int
 291 nfssvc_nfsd(struct thread *td)
 292 {
 293         int siz;
 294         struct nfssvc_sock *slp;
 295         struct nfsd *nfsd;
 296         struct nfsrv_descript *nd = NULL;
 297         struct mbuf *m, *mreq;
 298         int error = 0, cacherep, s, sotype, writes_todo;
 299         int procrastinate;
 300         u_quad_t cur_usec;
 301
 302 #ifndef nolint
 303         cacherep = RC_DOIT;
 304         writes_todo = 0;
 305 #endif
 306         nfsd = (struct nfsd *)
 307                 malloc(sizeof (struct nfsd), M_NFSD, M_WAITOK | M_ZERO);
 308         s = splnet();
 309         NFSD_LOCK();
 310
 311         nfsd->nfsd_td = td;
 312         TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
 313         nfsrv_numnfsd++;
 314
 315         /*
 316          * Loop getting rpc requests until SIGKILL.
 317          */
 318         for (;;) {
 319                 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) {
 320                         while (nfsd->nfsd_slp == NULL &&
 321                             (nfsd_head_flag & NFSD_CHECKSLP) == 0) {
 322                                 nfsd->nfsd_flag |= NFSD_WAITING;
 323                                 nfsd_waiting++;
 324                                 error = msleep(nfsd, &nfsd_mtx,
 325                                     PSOCK | PCATCH, "-", 0);
 326                                 nfsd_waiting--;
 327                                 if (error)
 328                                         goto done;
 329                         }
 330                         if (nfsd->nfsd_slp == NULL &&
 331                             (nfsd_head_flag & NFSD_CHECKSLP) != 0) {
 332                                 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
 333                                     if ((slp->ns_flag & (SLP_VALID | SLP_DOREC))
 334                                         == (SLP_VALID | SLP_DOREC)) {
 335                                             slp->ns_flag &= ~SLP_DOREC;
 336                                             slp->ns_sref++;
 337                                             nfsd->nfsd_slp = slp;
 338                                             break;
 339                                     }
 340                                 }
 341                                 if (slp == NULL)
 342                                         nfsd_head_flag &= ~NFSD_CHECKSLP;
 343                         }
 344                         if ((slp = nfsd->nfsd_slp) == NULL)
 345                                 continue;
 346                         if (slp->ns_flag & SLP_VALID) {
 347                                 if (slp->ns_flag & SLP_DISCONN)
 348                                         nfsrv_zapsock(slp);
 349                                 else if (slp->ns_flag & SLP_NEEDQ) {
 350                                         slp->ns_flag &= ~SLP_NEEDQ;
 351                                         (void) nfs_slplock(slp, 1);
 352                                         NFSD_UNLOCK();
 353                                         nfsrv_rcv(slp->ns_so, (caddr_t)slp,
 354                                                 M_TRYWAIT);
 355                                         NFSD_LOCK();
 356                                         nfs_slpunlock(slp);
 357                                 }
 358                                 error = nfsrv_dorec(slp, nfsd, &nd);
 359                                 cur_usec = nfs_curusec();
 360                                 if (error && LIST_FIRST(&slp->ns_tq) &&
 361                                     LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) {
 362                                         error = 0;
 363                                         cacherep = RC_DOIT;
 364                                         writes_todo = 1;
 365                                 } else
 366                                         writes_todo = 0;
 367                                 nfsd->nfsd_flag |= NFSD_REQINPROG;
 368                         }
 369                 } else {
 370                         error = 0;
 371                         slp = nfsd->nfsd_slp;
 372                 }
 373                 if (error || (slp->ns_flag & SLP_VALID) == 0) {
 374                         if (nd) {
 375                                 if (nd->nd_cr != NULL)
 376                                         crfree(nd->nd_cr);
 377                                 free((caddr_t)nd, M_NFSRVDESC);
 378                                 nd = NULL;
 379                         }
 380                         nfsd->nfsd_slp = NULL;
 381                         nfsd->nfsd_flag &= ~NFSD_REQINPROG;
 382                         nfsrv_slpderef(slp);
 383                         continue;
 384                 }
 385                 splx(s);
 386                 sotype = slp->ns_so->so_type;
 387                 if (nd) {
 388                     getmicrotime(&nd->nd_starttime);
 389                     if (nd->nd_nam2)
 390                         nd->nd_nam = nd->nd_nam2;
 391                     else
 392                         nd->nd_nam = slp->ns_nam;
 393
 394                     /*
 395                      * Check to see if authorization is needed.
 396                      */
 397                     cacherep = nfsrv_getcache(nd, &mreq);
 398
 399                     if (nfs_privport) {
 400                         /* Check if source port is privileged */
 401                         u_short port;
 402                         struct sockaddr *nam = nd->nd_nam;
 403                         struct sockaddr_in *sin;
 404
 405                         sin = (struct sockaddr_in *)nam;
 406                         /*
 407                          * INET/INET6 - same code:
 408                          *    sin_port and sin6_port are at same offset
 409                          */
 410                         port = ntohs(sin->sin_port);
 411                         if (port >= IPPORT_RESERVED &&
 412                             nd->nd_procnum != NFSPROC_NULL) {
 413 #ifdef INET6
 414                             char b6[INET6_ADDRSTRLEN];
 415 #if defined(KLD_MODULE)
 416         /* Do not use ip6_sprintf: the nfs module should work without INET6. */
 417 #define ip6_sprintf(buf, a) \
 418          (sprintf((buf), "%x:%x:%x:%x:%x:%x:%x:%x", \
 419                   (a)->s6_addr16[0], (a)->s6_addr16[1], \
 420                   (a)->s6_addr16[2], (a)->s6_addr16[3], \
 421                   (a)->s6_addr16[4], (a)->s6_addr16[5], \
 422                   (a)->s6_addr16[6], (a)->s6_addr16[7]), \
 423          (buf))
 424 #endif
 425 #endif
 426                             nd->nd_procnum = NFSPROC_NOOP;
 427                             nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
 428                             cacherep = RC_DOIT;
 429                             printf("NFS request from unprivileged port (%s:%d)\n",
 430 #ifdef INET6
 431                                 sin->sin_family == AF_INET6 ?
 432                                     ip6_sprintf(b6, &satosin6(sin)->sin6_addr) :
 433 #if defined(KLD_MODULE)
 434 #undef ip6_sprintf
 435 #endif
 436 #endif
 437                                     inet_ntoa(sin->sin_addr), port);
 438                         }
 439                     }
 440
 441                 }
 442
 443                 /*
 444                  * Loop to get all the write rpc relies that have been
 445                  * gathered together.
 446                  */
 447                 do {
 448                     switch (cacherep) {
 449                     case RC_DOIT:
 450                         if (nd && (nd->nd_flag & ND_NFSV3))
 451                             procrastinate = nfsrvw_procrastinate_v3;
 452                         else
 453                             procrastinate = nfsrvw_procrastinate;
 454                         NFSD_UNLOCK();
 455                         if (writes_todo || (!(nd->nd_flag & ND_NFSV3) &&
 456                             nd->nd_procnum == NFSPROC_WRITE &&
 457                             procrastinate > 0 && !notstarted))
 458                             error = nfsrv_writegather(&nd, slp,
 459                                 nfsd->nfsd_td, &mreq);
 460                         else
 461                             error = (*(nfsrv3_procs[nd->nd_procnum]))(nd,
 462                                 slp, nfsd->nfsd_td, &mreq);
 463                         NFSD_LOCK();
 464                         if (mreq == NULL)
 465                                 break;
 466                         if (error != 0 && error != NFSERR_RETVOID) {
 467                                 nfsrvstats.srv_errs++;
 468                                 nfsrv_updatecache(nd, FALSE, mreq);
 469                                 if (nd->nd_nam2)
 470                                         FREE(nd->nd_nam2, M_SONAME);
 471                                 break;
 472                         }
 473                         nfsrvstats.srvrpccnt[nd->nd_procnum]++;
 474                         nfsrv_updatecache(nd, TRUE, mreq);
 475                         nd->nd_mrep = NULL;
 476                         /* FALLTHROUGH */
 477                     case RC_REPLY:
 478                         NFSD_UNLOCK();
 479                         siz = m_length(mreq, NULL);
 480                         if (siz <= 0 || siz > NFS_MAXPACKET) {
 481                                 printf("mbuf siz=%d\n",siz);
 482                                 panic("Bad nfs svc reply");
 483                         }
 484                         m = mreq;
 485                         m->m_pkthdr.len = siz;
 486                         m->m_pkthdr.rcvif = NULL;
 487                         /*
 488                          * For stream protocols, prepend a Sun RPC
 489                          * Record Mark.
 490                          */
 491                         if (sotype == SOCK_STREAM) {
 492                                 M_PREPEND(m, NFSX_UNSIGNED, M_TRYWAIT);
 493                                 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz);
 494                         }
 495                         NFSD_LOCK();
 496                         if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
 497                                 (void) nfs_slplock(slp, 1);
 498                         if (slp->ns_flag & SLP_VALID) {
 499                             NFSD_UNLOCK();
 500                             error = nfsrv_send(slp->ns_so, nd->nd_nam2, m);
 501                             NFSD_LOCK();
 502                         } else {
 503                             error = EPIPE;
 504                             m_freem(m);
 505                         }
 506                         if (nd->nd_nam2)
 507                                 FREE(nd->nd_nam2, M_SONAME);
 508                         if (nd->nd_mrep)
 509                                 m_freem(nd->nd_mrep);
 510                         if (error == EPIPE)
 511                                 nfsrv_zapsock(slp);
 512                         if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
 513                                 nfs_slpunlock(slp);
 514                         if (error == EINTR || error == ERESTART) {
 515                                 if (nd->nd_cr != NULL)
 516                                         crfree(nd->nd_cr);
 517                                 free((caddr_t)nd, M_NFSRVDESC);
 518                                 nfsrv_slpderef(slp);
 519                                 s = splnet();
 520                                 goto done;
 521                         }
 522                         break;
 523                     case RC_DROPIT:
 524                         m_freem(nd->nd_mrep);
 525                         if (nd->nd_nam2)
 526                                 FREE(nd->nd_nam2, M_SONAME);
 527                         break;
 528                     };
 529                     if (nd) {
 530                         if (nd->nd_cr != NULL)
 531                                 crfree(nd->nd_cr);
 532                         FREE((caddr_t)nd, M_NFSRVDESC);
 533                         nd = NULL;
 534                     }
 535
 536                     /*
 537                      * Check to see if there are outstanding writes that
 538                      * need to be serviced.
 539                      */
 540                     cur_usec = nfs_curusec();
 541                     s = splsoftclock();
 542                     if (LIST_FIRST(&slp->ns_tq) &&
 543                         LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) {
 544                         cacherep = RC_DOIT;
 545                         writes_todo = 1;
 546                     } else
 547                         writes_todo = 0;
 548                     splx(s);
 549                 } while (writes_todo);
 550                 s = splnet();
 551                 if (nfsrv_dorec(slp, nfsd, &nd)) {
 552                         nfsd->nfsd_flag &= ~NFSD_REQINPROG;
 553                         nfsd->nfsd_slp = NULL;
 554                         nfsrv_slpderef(slp);
 555                 }
 556                 mtx_assert(&Giant, MA_NOTOWNED);
 557         }
 558 done:
 559         mtx_assert(&Giant, MA_NOTOWNED);
 560         TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
 561         splx(s);
 562         free((caddr_t)nfsd, M_NFSD);
 563         if (--nfsrv_numnfsd == 0)
 564                 nfsrv_init(TRUE);       /* Reinitialize everything */
 565         NFSD_UNLOCK();
 566         return (error);
 567 }
 568
 569 /*
 570  * Shut down a socket associated with an nfssvc_sock structure.
 571  * Should be called with the send lock set, if required.
 572  * The trick here is to increment the sref at the start, so that the nfsds
 573  * will stop using it and clear ns_flag at the end so that it will not be
 574  * reassigned during cleanup.
 575  */
 576 static void
 577 nfsrv_zapsock(struct nfssvc_sock *slp)
 578 {
 579         struct nfsrv_descript *nwp, *nnwp;
 580         struct socket *so;
 581         struct file *fp;
 582         struct nfsrv_rec *rec;
 583         int s;
 584
 585         NFSD_LOCK_ASSERT();
 586
 587         /*
 588          * XXXRW: By clearing all flags, other threads/etc should ignore
 589          * this slp and we can safely release nfsd_mtx so we can clean
 590          * up the slp safely.
 591          */
 592         slp->ns_flag &= ~SLP_ALLFLAGS;
 593         fp = slp->ns_fp;
 594         if (fp) {
 595                 NFSD_UNLOCK();
 596                 slp->ns_fp = NULL;
 597                 so = slp->ns_so;
 598                 SOCKBUF_LOCK(&so->so_rcv);
 599                 so->so_rcv.sb_flags &= ~SB_UPCALL;
 600                 SOCKBUF_UNLOCK(&so->so_rcv);
 601                 so->so_upcall = NULL;
 602                 so->so_upcallarg = NULL;
 603                 soshutdown(so, SHUT_RDWR);
 604                 closef(fp, NULL);
 605                 NFSD_LOCK();
 606                 if (slp->ns_nam)
 607                         FREE(slp->ns_nam, M_SONAME);
 608                 m_freem(slp->ns_raw);
 609                 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) {
 610                         STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link);
 611                         if (rec->nr_address)
 612                                 FREE(rec->nr_address, M_SONAME);
 613                         m_freem(rec->nr_packet);
 614                         free(rec, M_NFSRVDESC);
 615                 }
 616                 s = splsoftclock();
 617                 for (nwp = LIST_FIRST(&slp->ns_tq); nwp; nwp = nnwp) {
 618                         nnwp = LIST_NEXT(nwp, nd_tq);
 619                         LIST_REMOVE(nwp, nd_tq);
 620                         if (nwp->nd_cr != NULL)
 621                                 crfree(nwp->nd_cr);
 622                         free((caddr_t)nwp, M_NFSRVDESC);
 623                 }
 624                 LIST_INIT(&slp->ns_tq);
 625                 splx(s);
 626         }
 627 }
 628
 629 /*
 630  * Derefence a server socket structure. If it has no more references and
 631  * is no longer valid, you can throw it away.
 632  */
 633 void
 634 nfsrv_slpderef(struct nfssvc_sock *slp)
 635 {
 636
 637         NFSD_LOCK_ASSERT();
 638
 639         if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) {
 640                 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
 641                 free((caddr_t)slp, M_NFSSVC);
 642         }
 643 }
 644
 645 /*
 646  * Lock a socket against others.
 647  *
 648  * XXXRW: Wait argument is always 1 in the caller.  Replace with a real
 649  * sleep lock?
 650  */
 651 int
 652 nfs_slplock(struct nfssvc_sock *slp, int wait)
 653 {
 654         int *statep = &slp->ns_solock;
 655
 656         NFSD_LOCK_ASSERT();
 657
 658         if (!wait && (*statep & NFSRV_SNDLOCK))
 659                 return(0);      /* already locked, fail */
 660         while (*statep & NFSRV_SNDLOCK) {
 661                 *statep |= NFSRV_WANTSND;
 662                 (void) msleep(statep, &nfsd_mtx, PZERO - 1, "nfsslplck", 0);
 663         }
 664         *statep |= NFSRV_SNDLOCK;
 665         return (1);
 666 }
 667
 668 /*
 669  * Unlock the stream socket for others.
 670  */
 671 void
 672 nfs_slpunlock(struct nfssvc_sock *slp)
 673 {
 674         int *statep = &slp->ns_solock;
 675
 676         NFSD_LOCK_ASSERT();
 677
 678         if ((*statep & NFSRV_SNDLOCK) == 0)
 679                 panic("nfs slpunlock");
 680         *statep &= ~NFSRV_SNDLOCK;
 681         if (*statep & NFSRV_WANTSND) {
 682                 *statep &= ~NFSRV_WANTSND;
 683                 wakeup(statep);
 684         }
 685 }
 686
 687 /*
 688  * Initialize the data structures for the server.
 689  * Handshake with any new nfsds starting up to avoid any chance of
 690  * corruption.
 691  */
 692 void
 693 nfsrv_init(int terminating)
 694 {
 695         struct nfssvc_sock *slp, *nslp;
 696
 697         NFSD_LOCK_ASSERT();
 698
 699         if (nfssvc_sockhead_flag & SLP_INIT)
 700                 panic("nfsd init");
 701         nfssvc_sockhead_flag |= SLP_INIT;
 702         if (terminating) {
 703                 TAILQ_FOREACH_SAFE(slp, &nfssvc_sockhead, ns_chain, nslp) {
 704                         if (slp->ns_flag & SLP_VALID)
 705                                 nfsrv_zapsock(slp);
 706                         TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
 707                         free((caddr_t)slp, M_NFSSVC);
 708                 }
 709                 nfsrv_cleancache();     /* And clear out server cache */
 710         } else
 711                 nfs_pub.np_valid = 0;
 712
 713         TAILQ_INIT(&nfssvc_sockhead);
 714         nfssvc_sockhead_flag &= ~SLP_INIT;
 715         if (nfssvc_sockhead_flag & SLP_WANTINIT) {
 716                 nfssvc_sockhead_flag &= ~SLP_WANTINIT;
 717                 wakeup(&nfssvc_sockhead);
 718         }
 719
 720         TAILQ_INIT(&nfsd_head);
 721         nfsd_head_flag &= ~NFSD_CHECKSLP;
 722
 723 #if 0
 724         nfs_udpsock = (struct nfssvc_sock *)
 725             malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
 726         STAILQ_INIT(&nfs_udpsock->ns_rec);
 727         TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain);
 728
 729         nfs_cltpsock = (struct nfssvc_sock *)
 730             malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
 731         STAILQ_INIT(&nfs_cltpsock->ns_rec);
 732         TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain);
 733 #endif
 734 }