sys/vfs/nfs/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  * 3. All advertising materials mentioning features or use of this software
  17  *    must display the following acknowledgement:
  18  *      This product includes software developed by the University of
  19  *      California, Berkeley and its contributors.
  20  * 4. Neither the name of the University nor the names of its contributors
  21  *    may be used to endorse or promote products derived from this software
  22  *    without specific prior written permission.
  23  *
  24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  34  * SUCH DAMAGE.
  35  *
  36  *      @(#)nfs_syscalls.c      8.5 (Berkeley) 3/30/95
  37  * $FreeBSD: src/sys/nfs/nfs_syscalls.c,v 1.58.2.1 2000/11/26 02:30:06 dillon Exp $
  38  * $DragonFly: src/sys/vfs/nfs/nfs_syscalls.c,v 1.31 2008/01/05 14:02:41 swildner Exp $
  39  */
  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/proc.h>
  52 #include <sys/priv.h>
  53 #include <sys/buf.h>
  54 #include <sys/mbuf.h>
  55 #include <sys/resourcevar.h>
  56 #include <sys/socket.h>
  57 #include <sys/socketvar.h>
  58 #include <sys/domain.h>
  59 #include <sys/protosw.h>
  60 #include <sys/nlookup.h>
  61 #include <sys/mutex.h>
  62 #include <vm/vm_zone.h>
  63
  64 #include <sys/mutex2.h>
  65 #include <sys/mplock2.h>
  66
  67 #include <netinet/in.h>
  68 #include <netinet/tcp.h>
  69 #include "xdr_subs.h"
  70 #include "rpcv2.h"
  71 #include "nfsproto.h"
  72 #include "nfs.h"
  73 #include "nfsm_subs.h"
  74 #include "nfsrvcache.h"
  75 #include "nfsmount.h"
  76 #include "nfsnode.h"
  77 #include "nfsrtt.h"
  78
  79 #include <sys/thread2.h>
  80
  81 static MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure");
  82
  83 static int nuidhash_max = NFS_MAXUIDHASH;
  84
  85 #ifndef NFS_NOSERVER
  86 static void     nfsrv_zapsock (struct nfssvc_sock *slp);
  87 #endif
  88
  89 #define TRUE    1
  90 #define FALSE   0
  91
  92 SYSCTL_DECL(_vfs_nfs);
  93
  94 #ifndef NFS_NOSERVER
  95 int nfsd_waiting = 0;
  96 static struct nfsdrt nfsdrt;
  97 static int nfs_numnfsd = 0;
  98 static void     nfsd_rt (int sotype, struct nfsrv_descript *nd,
  99                              int cacherep);
 100 static int      nfssvc_addsock (struct file *, struct sockaddr *,
 101                                     struct thread *);
 102 static int      nfssvc_nfsd (struct nfsd_srvargs *,caddr_t,struct thread *);
 103
 104 static int nfs_privport = 0;
 105 SYSCTL_INT(_vfs_nfs, NFS_NFSPRIVPORT, nfs_privport, CTLFLAG_RW, &nfs_privport, 0, "");
 106 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay, CTLFLAG_RW, &nfsrvw_procrastinate, 0, "");
 107 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay_v3, CTLFLAG_RW, &nfsrvw_procrastinate_v3, 0, "");
 108 int     nfs_soreserve = NFS_MAXPACKET * NFS_MAXASYNCBIO;
 109 SYSCTL_INT(_vfs_nfs, OID_AUTO, soreserve, CTLFLAG_RW, &nfs_soreserve, 0, "");
 110
 111 /*
 112  * NFS server system calls
 113  */
 114
 115 #endif /* NFS_NOSERVER */
 116 /*
 117  * nfssvc_args(int flag, caddr_t argp)
 118  *
 119  * Nfs server psuedo system call for the nfsd's
 120  * Based on the flag value it either:
 121  * - adds a socket to the selection list
 122  * - remains in the kernel as an nfsd
 123  * - remains in the kernel as an nfsiod
 124  *
 125  * MPALMOSTSAFE
 126  */
 127 int
 128 sys_nfssvc(struct nfssvc_args *uap)
 129 {
 130 #ifndef NFS_NOSERVER
 131         struct nlookupdata nd;
 132         struct file *fp;
 133         struct sockaddr *nam;
 134         struct nfsd_args nfsdarg;
 135         struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs;
 136         struct nfsd_cargs ncd;
 137         struct nfsd *nfsd;
 138         struct nfssvc_sock *slp;
 139         struct nfsuid *nuidp;
 140         struct nfsmount *nmp;
 141         struct vnode *vp;
 142 #endif /* NFS_NOSERVER */
 143         int error;
 144         struct thread *td = curthread;
 145
 146         /*
 147          * Must be super user
 148          */
 149         error = priv_check(td, PRIV_ROOT);
 150         if (error)
 151                 return (error);
 152
 153         get_mplock();
 154         while (nfssvc_sockhead_flag & SLP_INIT) {
 155                 nfssvc_sockhead_flag |= SLP_WANTINIT;
 156                 tsleep((caddr_t)&nfssvc_sockhead, 0, "nfsd init", 0);
 157         }
 158         if (uap->flag & NFSSVC_BIOD)
 159                 error = ENXIO;          /* no longer need nfsiod's */
 160 #ifdef NFS_NOSERVER
 161         else
 162                 error = ENXIO;
 163 #else /* !NFS_NOSERVER */
 164         else if (uap->flag & NFSSVC_MNTD) {
 165                 error = copyin(uap->argp, (caddr_t)&ncd, sizeof (ncd));
 166                 if (error)
 167                         goto done;
 168                 vp = NULL;
 169                 error = nlookup_init(&nd, ncd.ncd_dirp, UIO_USERSPACE,
 170                                         NLC_FOLLOW);
 171                 if (error == 0)
 172                         error = nlookup(&nd);
 173                 if (error == 0)
 174                         error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
 175                 nlookup_done(&nd);
 176                 if (error)
 177                         goto done;
 178
 179                 if ((vp->v_flag & VROOT) == 0)
 180                         error = EINVAL;
 181                 nmp = VFSTONFS(vp->v_mount);
 182                 vput(vp);
 183                 if (error)
 184                         goto done;
 185                 if ((nmp->nm_state & NFSSTA_MNTD) &&
 186                         (uap->flag & NFSSVC_GOTAUTH) == 0) {
 187                         error = 0;
 188                         goto done;
 189                 }
 190                 nmp->nm_state |= NFSSTA_MNTD;
 191                 error = nfs_clientd(nmp, td->td_ucred, &ncd, uap->flag,
 192                                     uap->argp, td);
 193         } else if (uap->flag & NFSSVC_ADDSOCK) {
 194                 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg));
 195                 if (error)
 196                         goto done;
 197                 error = holdsock(td->td_proc->p_fd, nfsdarg.sock, &fp);
 198                 if (error)
 199                         goto done;
 200                 /*
 201                  * Get the client address for connected sockets.
 202                  */
 203                 if (nfsdarg.name == NULL || nfsdarg.namelen == 0)
 204                         nam = NULL;
 205                 else {
 206                         error = getsockaddr(&nam, nfsdarg.name,
 207                                             nfsdarg.namelen);
 208                         if (error) {
 209                                 fdrop(fp);
 210                                 goto done;
 211                         }
 212                 }
 213                 error = nfssvc_addsock(fp, nam, td);
 214                 fdrop(fp);
 215         } else {
 216                 error = copyin(uap->argp, (caddr_t)nsd, sizeof (*nsd));
 217                 if (error)
 218                         goto done;
 219                 if ((uap->flag & NFSSVC_AUTHIN) &&
 220                     ((nfsd = nsd->nsd_nfsd)) != NULL &&
 221                     (nfsd->nfsd_slp->ns_flag & SLP_VALID)) {
 222                         slp = nfsd->nfsd_slp;
 223
 224                         /*
 225                          * First check to see if another nfsd has already
 226                          * added this credential.
 227                          */
 228                         for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first;
 229                             nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
 230                                 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid &&
 231                                     (!nfsd->nfsd_nd->nd_nam2 ||
 232                                      netaddr_match(NU_NETFAM(nuidp),
 233                                      &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2)))
 234                                         break;
 235                         }
 236                         if (nuidp) {
 237                             nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr);
 238                             nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
 239                         } else {
 240                             /*
 241                              * Nope, so we will.
 242                              */
 243                             if (slp->ns_numuids < nuidhash_max) {
 244                                 slp->ns_numuids++;
 245                                 nuidp = (struct nfsuid *)
 246                                    kmalloc(sizeof (struct nfsuid), M_NFSUID,
 247                                         M_WAITOK);
 248                             } else
 249                                 nuidp = NULL;
 250                             if ((slp->ns_flag & SLP_VALID) == 0) {
 251                                 if (nuidp)
 252                                     kfree((caddr_t)nuidp, M_NFSUID);
 253                             } else {
 254                                 if (nuidp == NULL) {
 255                                     nuidp = TAILQ_FIRST(&slp->ns_uidlruhead);
 256                                     LIST_REMOVE(nuidp, nu_hash);
 257                                     TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp,
 258                                         nu_lru);
 259                                     if (nuidp->nu_flag & NU_NAM)
 260                                         FREE(nuidp->nu_nam, M_SONAME);
 261                                 }
 262                                 nuidp->nu_flag = 0;
 263                                 nuidp->nu_cr = nsd->nsd_cr;
 264                                 if (nuidp->nu_cr.cr_ngroups > NGROUPS)
 265                                     nuidp->nu_cr.cr_ngroups = NGROUPS;
 266                                 nuidp->nu_cr.cr_ref = 1;
 267                                 nuidp->nu_timestamp = nsd->nsd_timestamp;
 268                                 nuidp->nu_expire = time_second + nsd->nsd_ttl;
 269                                 /*
 270                                  * and save the session key in nu_key.
 271                                  */
 272                                 bcopy(nsd->nsd_key, nuidp->nu_key,
 273                                     sizeof (nsd->nsd_key));
 274                                 if (nfsd->nfsd_nd->nd_nam2) {
 275                                     struct sockaddr_in *saddr;
 276
 277                                     saddr = (struct sockaddr_in *)
 278                                             nfsd->nfsd_nd->nd_nam2;
 279                                     switch (saddr->sin_family) {
 280                                     case AF_INET:
 281                                         nuidp->nu_flag |= NU_INETADDR;
 282                                         nuidp->nu_inetaddr =
 283                                              saddr->sin_addr.s_addr;
 284                                         break;
 285                                     case AF_ISO:
 286                                     default:
 287                                         nuidp->nu_flag |= NU_NAM;
 288                                         nuidp->nu_nam =
 289                                           dup_sockaddr(nfsd->nfsd_nd->nd_nam2);
 290                                         break;
 291                                     };
 292                                 }
 293                                 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp,
 294                                         nu_lru);
 295                                 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid),
 296                                         nuidp, nu_hash);
 297                                 nfsrv_setcred(&nuidp->nu_cr,
 298                                     &nfsd->nfsd_nd->nd_cr);
 299                                 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
 300                             }
 301                         }
 302                 }
 303                 if ((uap->flag & NFSSVC_AUTHINFAIL) && (nfsd = nsd->nsd_nfsd))
 304                         nfsd->nfsd_flag |= NFSD_AUTHFAIL;
 305                 error = nfssvc_nfsd(nsd, uap->argp, td);
 306         }
 307 #endif /* NFS_NOSERVER */
 308         if (error == EINTR || error == ERESTART)
 309                 error = 0;
 310 done:
 311         rel_mplock();
 312         return (error);
 313 }
 314
 315 #ifndef NFS_NOSERVER
 316 /*
 317  * Adds a socket to the list for servicing by nfsds.
 318  */
 319 static int
 320 nfssvc_addsock(struct file *fp, struct sockaddr *mynam, struct thread *td)
 321 {
 322         int siz;
 323         struct nfssvc_sock *slp;
 324         struct socket *so;
 325         int error;
 326
 327         so = (struct socket *)fp->f_data;
 328 #if 0
 329         tslp = NULL;
 330         /*
 331          * Add it to the list, as required.
 332          */
 333         if (so->so_proto->pr_protocol == IPPROTO_UDP) {
 334                 tslp = nfs_udpsock;
 335                 if (tslp->ns_flag & SLP_VALID) {
 336                         if (mynam != NULL)
 337                                 FREE(mynam, M_SONAME);
 338                         return (EPERM);
 339                 }
 340         }
 341 #endif
 342         /*
 343          * Reserve buffer space in the socket.  Note that due to bugs in
 344          * Linux's delayed-ack code, serious performance degredation may
 345          * occur with linux hosts if the minimum is used.
 346          *
 347          * NFS sockets are not limited to the standard sb_max or by
 348          * resource limits.
 349          */
 350         if (so->so_type == SOCK_STREAM)
 351                 siz = NFS_MAXPACKET + sizeof (u_long);
 352         else
 353                 siz = NFS_MAXPACKET;
 354         if (siz < nfs_soreserve)
 355             siz = nfs_soreserve;
 356
 357         error = soreserve(so, siz, siz, NULL);
 358         if (error) {
 359                 if (mynam != NULL)
 360                         FREE(mynam, M_SONAME);
 361                 return (error);
 362         }
 363
 364         /*
 365          * Set protocol specific options { for now TCP only } and
 366          * reserve some space. For datagram sockets, this can get called
 367          * repeatedly for the same socket, but that isn't harmful.
 368          */
 369         if (so->so_type == SOCK_STREAM) {
 370                 struct sockopt sopt;
 371                 int val;
 372
 373                 bzero(&sopt, sizeof sopt);
 374                 sopt.sopt_level = SOL_SOCKET;
 375                 sopt.sopt_name = SO_KEEPALIVE;
 376                 sopt.sopt_val = &val;
 377                 sopt.sopt_valsize = sizeof val;
 378                 val = 1;
 379                 sosetopt(so, &sopt);
 380         }
 381         if (so->so_proto->pr_domain->dom_family == AF_INET &&
 382             so->so_proto->pr_protocol == IPPROTO_TCP) {
 383                 struct sockopt sopt;
 384                 int val;
 385
 386                 bzero(&sopt, sizeof sopt);
 387                 sopt.sopt_level = IPPROTO_TCP;
 388                 sopt.sopt_name = TCP_NODELAY;
 389                 sopt.sopt_val = &val;
 390                 sopt.sopt_valsize = sizeof val;
 391                 val = 1;
 392                 sosetopt(so, &sopt);
 393         }
 394         so->so_rcv.ssb_flags &= ~SSB_NOINTR;
 395         so->so_rcv.ssb_timeo = 0;
 396         so->so_snd.ssb_flags &= ~SSB_NOINTR;
 397         so->so_snd.ssb_timeo = 0;
 398
 399         slp = (struct nfssvc_sock *)kmalloc(sizeof (struct nfssvc_sock),
 400             M_NFSSVC, M_WAITOK | M_ZERO);
 401         mtx_init(&slp->ns_solock);
 402         STAILQ_INIT(&slp->ns_rec);
 403         TAILQ_INIT(&slp->ns_uidlruhead);
 404         TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain);
 405
 406         slp->ns_so = so;
 407         slp->ns_nam = mynam;
 408         fp->f_count++;
 409         slp->ns_fp = fp;
 410         crit_enter();
 411         so->so_upcallarg = (caddr_t)slp;
 412         so->so_upcall = nfsrv_rcv;
 413         so->so_rcv.ssb_flags |= SSB_UPCALL;
 414         slp->ns_flag = (SLP_VALID | SLP_NEEDQ);
 415         nfsrv_wakenfsd(slp, 1);
 416         crit_exit();
 417         return (0);
 418 }
 419
 420 /*
 421  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
 422  * until it is killed by a signal.
 423  */
 424 static int
 425 nfssvc_nfsd(struct nfsd_srvargs *nsd, caddr_t argp, struct thread *td)
 426 {
 427         int siz;
 428         struct nfssvc_sock *slp;
 429         struct nfsd *nfsd = nsd->nsd_nfsd;
 430         struct nfsrv_descript *nd = NULL;
 431         struct mbuf *m, *mreq;
 432         int error = 0, cacherep, sotype, writes_todo;
 433         int procrastinate;
 434         u_quad_t cur_usec;
 435
 436 #ifndef nolint
 437         cacherep = RC_DOIT;
 438         writes_todo = 0;
 439 #endif
 440         if (nfsd == NULL) {
 441                 nsd->nsd_nfsd = nfsd = (struct nfsd *)
 442                         kmalloc(sizeof (struct nfsd), M_NFSD, M_WAITOK|M_ZERO);
 443                 crit_enter();
 444                 nfsd->nfsd_td = td;
 445                 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
 446                 nfs_numnfsd++;
 447         } else
 448                 crit_enter();
 449
 450         /*
 451          * Loop getting rpc requests until SIGKILL.
 452          */
 453         for (;;) {
 454                 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) {
 455                         while (nfsd->nfsd_slp == NULL &&
 456                             (nfsd_head_flag & NFSD_CHECKSLP) == 0) {
 457                                 nfsd->nfsd_flag |= NFSD_WAITING;
 458                                 nfsd_waiting++;
 459                                 error = tsleep((caddr_t)nfsd, PCATCH, "nfsd", 0);
 460                                 nfsd_waiting--;
 461                                 if (error)
 462                                         goto done;
 463                         }
 464                         if (nfsd->nfsd_slp == NULL &&
 465                             (nfsd_head_flag & NFSD_CHECKSLP) != 0) {
 466                                 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
 467                                     if ((slp->ns_flag & (SLP_VALID | SLP_DOREC))
 468                                         == (SLP_VALID | SLP_DOREC)) {
 469                                             slp->ns_flag &= ~SLP_DOREC;
 470                                             slp->ns_sref++;
 471                                             nfsd->nfsd_slp = slp;
 472                                             break;
 473                                     }
 474                                 }
 475                                 if (slp == 0)
 476                                         nfsd_head_flag &= ~NFSD_CHECKSLP;
 477                         }
 478                         if ((slp = nfsd->nfsd_slp) == NULL)
 479                                 continue;
 480                         if (slp->ns_flag & SLP_VALID) {
 481                                 if (slp->ns_flag & SLP_DISCONN)
 482                                         nfsrv_zapsock(slp);
 483                                 else if (slp->ns_flag & SLP_NEEDQ) {
 484                                         slp->ns_flag &= ~SLP_NEEDQ;
 485                                         (void) nfs_slplock(slp, 1);
 486                                         nfsrv_rcv(slp->ns_so, (caddr_t)slp,
 487                                                 MB_WAIT);
 488                                         nfs_slpunlock(slp);
 489                                 }
 490                                 error = nfsrv_dorec(slp, nfsd, &nd);
 491                                 cur_usec = nfs_curusec();
 492                                 if (error && slp->ns_tq.lh_first &&
 493                                     slp->ns_tq.lh_first->nd_time <= cur_usec) {
 494                                         error = 0;
 495                                         cacherep = RC_DOIT;
 496                                         writes_todo = 1;
 497                                 } else
 498                                         writes_todo = 0;
 499                                 nfsd->nfsd_flag |= NFSD_REQINPROG;
 500                         }
 501                 } else {
 502                         error = 0;
 503                         slp = nfsd->nfsd_slp;
 504                 }
 505                 if (error || (slp->ns_flag & SLP_VALID) == 0) {
 506                         if (nd) {
 507                                 kfree((caddr_t)nd, M_NFSRVDESC);
 508                                 nd = NULL;
 509                         }
 510                         nfsd->nfsd_slp = NULL;
 511                         nfsd->nfsd_flag &= ~NFSD_REQINPROG;
 512                         nfsrv_slpderef(slp);
 513                         continue;
 514                 }
 515                 crit_exit();
 516                 sotype = slp->ns_so->so_type;
 517                 if (nd) {
 518                     getmicrotime(&nd->nd_starttime);
 519                     if (nd->nd_nam2)
 520                         nd->nd_nam = nd->nd_nam2;
 521                     else
 522                         nd->nd_nam = slp->ns_nam;
 523
 524                     /*
 525                      * Check to see if authorization is needed.
 526                      */
 527                     if (nfsd->nfsd_flag & NFSD_NEEDAUTH) {
 528                         nfsd->nfsd_flag &= ~NFSD_NEEDAUTH;
 529                         nsd->nsd_haddr =
 530                                 ((struct sockaddr_in *)
 531                                  nd->nd_nam)->sin_addr.s_addr;
 532                         nsd->nsd_authlen = nfsd->nfsd_authlen;
 533                         nsd->nsd_verflen = nfsd->nfsd_verflen;
 534                         if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr,
 535                                 nfsd->nfsd_authlen) &&
 536                             !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr,
 537                                 nfsd->nfsd_verflen) &&
 538                             !copyout((caddr_t)nsd, argp, sizeof (*nsd)))
 539                             return (ENEEDAUTH);
 540                         cacherep = RC_DROPIT;
 541                     } else {
 542                         cacherep = nfsrv_getcache(nd, slp, &mreq);
 543                     }
 544
 545                     if (nfsd->nfsd_flag & NFSD_AUTHFAIL) {
 546                         nfsd->nfsd_flag &= ~NFSD_AUTHFAIL;
 547                         nd->nd_procnum = NFSPROC_NOOP;
 548                         nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
 549                         cacherep = RC_DOIT;
 550                     } else if (nfs_privport) {
 551                         /* Check if source port is privileged */
 552                         u_short port;
 553                         struct sockaddr *nam = nd->nd_nam;
 554                         struct sockaddr_in *sin;
 555
 556                         sin = (struct sockaddr_in *)nam;
 557                         port = ntohs(sin->sin_port);
 558                         if (port >= IPPORT_RESERVED &&
 559                             nd->nd_procnum != NFSPROC_NULL) {
 560                             nd->nd_procnum = NFSPROC_NOOP;
 561                             nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
 562                             cacherep = RC_DOIT;
 563                             kprintf("NFS request from unprivileged port (%s:%d)\n",
 564                                    inet_ntoa(sin->sin_addr), port);
 565                         }
 566                     }
 567
 568                 }
 569
 570                 /*
 571                  * Loop to get all the write rpc relies that have been
 572                  * gathered together.
 573                  */
 574                 do {
 575                     switch (cacherep) {
 576                     case RC_DOIT:
 577                         if (nd && (nd->nd_flag & ND_NFSV3))
 578                             procrastinate = nfsrvw_procrastinate_v3;
 579                         else
 580                             procrastinate = nfsrvw_procrastinate;
 581                         if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE &&
 582                             procrastinate > 0)
 583                         ) {
 584                             error = nfsrv_writegather(&nd, slp,
 585                                 nfsd->nfsd_td, &mreq);
 586                         } else {
 587                             error = (*(nfsrv3_procs[nd->nd_procnum]))(nd,
 588                                 slp, nfsd->nfsd_td, &mreq);
 589                         }
 590                         if (mreq == NULL)
 591                                 break;
 592                         if (error != 0 && error != NFSERR_RETVOID) {
 593                                 if (nd->nd_procnum != NQNFSPROC_VACATED)
 594                                         nfsstats.srv_errs++;
 595                                 nfsrv_updatecache(nd, FALSE, mreq);
 596                                 if (nd->nd_nam2)
 597                                         FREE(nd->nd_nam2, M_SONAME);
 598                                 break;
 599                         }
 600                         nfsstats.srvrpccnt[nd->nd_procnum]++;
 601                         nfsrv_updatecache(nd, TRUE, mreq);
 602                         nd->nd_mrep = NULL;
 603                     case RC_REPLY:
 604                         m = mreq;
 605                         siz = 0;
 606                         while (m) {
 607                                 siz += m->m_len;
 608                                 m = m->m_next;
 609                         }
 610                         if (siz <= 0 || siz > NFS_MAXPACKET) {
 611                                 kprintf("mbuf siz=%d\n",siz);
 612                                 panic("Bad nfs svc reply");
 613                         }
 614                         m = mreq;
 615                         m->m_pkthdr.len = siz;
 616                         m->m_pkthdr.rcvif = NULL;
 617                         /*
 618                          * For stream protocols, prepend a Sun RPC
 619                          * Record Mark.
 620                          */
 621                         if (sotype == SOCK_STREAM) {
 622                                 M_PREPEND(m, NFSX_UNSIGNED, MB_WAIT);
 623                                 if (m == NULL)
 624                                         return (ENOBUFS);
 625                                 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz);
 626                         }
 627                         if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
 628                                 (void) nfs_slplock(slp, 1);
 629                         if (slp->ns_flag & SLP_VALID)
 630                             error = nfs_send(slp->ns_so, nd->nd_nam2, m, NULL);
 631                         else {
 632                             error = EPIPE;
 633                             m_freem(m);
 634                         }
 635                         if (nfsrtton)
 636                                 nfsd_rt(sotype, nd, cacherep);
 637                         if (nd->nd_nam2)
 638                                 FREE(nd->nd_nam2, M_SONAME);
 639                         if (nd->nd_mrep)
 640                                 m_freem(nd->nd_mrep);
 641                         if (error == EPIPE)
 642                                 nfsrv_zapsock(slp);
 643                         if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
 644                                 nfs_slpunlock(slp);
 645                         if (error == EINTR || error == ERESTART) {
 646                                 kfree((caddr_t)nd, M_NFSRVDESC);
 647                                 nfsrv_slpderef(slp);
 648                                 crit_enter();
 649                                 goto done;
 650                         }
 651                         break;
 652                     case RC_DROPIT:
 653                         if (nfsrtton)
 654                                 nfsd_rt(sotype, nd, cacherep);
 655                         m_freem(nd->nd_mrep);
 656                         if (nd->nd_nam2)
 657                                 FREE(nd->nd_nam2, M_SONAME);
 658                         break;
 659                     };
 660                     if (nd) {
 661                         FREE((caddr_t)nd, M_NFSRVDESC);
 662                         nd = NULL;
 663                     }
 664
 665                     /*
 666                      * Check to see if there are outstanding writes that
 667                      * need to be serviced.
 668                      */
 669                     cur_usec = nfs_curusec();
 670                     crit_enter();
 671                     if (slp->ns_tq.lh_first &&
 672                         slp->ns_tq.lh_first->nd_time <= cur_usec) {
 673                         cacherep = RC_DOIT;
 674                         writes_todo = 1;
 675                     } else
 676                         writes_todo = 0;
 677                     crit_exit();
 678                 } while (writes_todo);
 679                 crit_enter();
 680                 if (nfsrv_dorec(slp, nfsd, &nd)) {
 681                         nfsd->nfsd_flag &= ~NFSD_REQINPROG;
 682                         nfsd->nfsd_slp = NULL;
 683                         nfsrv_slpderef(slp);
 684                 }
 685         }
 686 done:
 687         TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
 688         crit_exit();
 689         kfree((caddr_t)nfsd, M_NFSD);
 690         nsd->nsd_nfsd = NULL;
 691         if (--nfs_numnfsd == 0)
 692                 nfsrv_init(TRUE);       /* Reinitialize everything */
 693         return (error);
 694 }
 695
 696 /*
 697  * Shut down a socket associated with an nfssvc_sock structure.
 698  * Should be called with the send lock set, if required.
 699  * The trick here is to increment the sref at the start, so that the nfsds
 700  * will stop using it and clear ns_flag at the end so that it will not be
 701  * reassigned during cleanup.
 702  */
 703 static void
 704 nfsrv_zapsock(struct nfssvc_sock *slp)
 705 {
 706         struct nfsuid *nuidp, *nnuidp;
 707         struct nfsrv_descript *nwp, *nnwp;
 708         struct socket *so;
 709         struct file *fp;
 710         struct nfsrv_rec *rec;
 711
 712         slp->ns_flag &= ~SLP_ALLFLAGS;
 713         fp = slp->ns_fp;
 714         if (fp) {
 715                 slp->ns_fp = NULL;
 716                 so = slp->ns_so;
 717                 so->so_rcv.ssb_flags &= ~SSB_UPCALL;
 718                 so->so_upcall = NULL;
 719                 so->so_upcallarg = NULL;
 720                 soshutdown(so, SHUT_RDWR);
 721                 closef(fp, NULL);
 722                 if (slp->ns_nam)
 723                         FREE(slp->ns_nam, M_SONAME);
 724                 m_freem(slp->ns_raw);
 725                 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) {
 726                         --slp->ns_numrec;
 727                         STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link);
 728                         if (rec->nr_address)
 729                                 FREE(rec->nr_address, M_SONAME);
 730                         m_freem(rec->nr_packet);
 731                         kfree(rec, M_NFSRVDESC);
 732                 }
 733                 TAILQ_FOREACH_MUTABLE(nuidp, &slp->ns_uidlruhead, nu_lru,
 734                                       nnuidp) {
 735                         LIST_REMOVE(nuidp, nu_hash);
 736                         TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru);
 737                         if (nuidp->nu_flag & NU_NAM)
 738                                 FREE(nuidp->nu_nam, M_SONAME);
 739                         kfree((caddr_t)nuidp, M_NFSUID);
 740                 }
 741                 crit_enter();
 742                 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) {
 743                         nnwp = nwp->nd_tq.le_next;
 744                         LIST_REMOVE(nwp, nd_tq);
 745                         kfree((caddr_t)nwp, M_NFSRVDESC);
 746                 }
 747                 LIST_INIT(&slp->ns_tq);
 748                 crit_exit();
 749         }
 750 }
 751
 752 /*
 753  * Derefence a server socket structure. If it has no more references and
 754  * is no longer valid, you can throw it away.
 755  */
 756 void
 757 nfsrv_slpderef(struct nfssvc_sock *slp)
 758 {
 759         if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) {
 760                 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
 761                 kfree((caddr_t)slp, M_NFSSVC);
 762         }
 763 }
 764
 765 /*
 766  * Lock a socket against others.
 767  *
 768  * Returns 0 on failure, 1 on success.
 769  */
 770 int
 771 nfs_slplock(struct nfssvc_sock *slp, int wait)
 772 {
 773         mtx_t mtx = &slp->ns_solock;
 774
 775         if (wait) {
 776                 mtx_lock_ex(mtx, "nfsslplck", 0, 0);
 777                 return(1);
 778         } else if (mtx_lock_ex_try(mtx) == 0) {
 779                 return(1);
 780         } else {
 781                 return(0);
 782         }
 783 }
 784
 785 /*
 786  * Unlock the stream socket for others.
 787  */
 788 void
 789 nfs_slpunlock(struct nfssvc_sock *slp)
 790 {
 791         mtx_t mtx = &slp->ns_solock;
 792
 793         mtx_unlock(mtx);
 794 }
 795
 796 /*
 797  * Initialize the data structures for the server.
 798  * Handshake with any new nfsds starting up to avoid any chance of
 799  * corruption.
 800  */
 801 void
 802 nfsrv_init(int terminating)
 803 {
 804         struct nfssvc_sock *slp, *nslp;
 805
 806         if (nfssvc_sockhead_flag & SLP_INIT)
 807                 panic("nfsd init");
 808         nfssvc_sockhead_flag |= SLP_INIT;
 809         if (terminating) {
 810                 TAILQ_FOREACH_MUTABLE(slp, &nfssvc_sockhead, ns_chain, nslp) {
 811                         if (slp->ns_flag & SLP_VALID)
 812                                 nfsrv_zapsock(slp);
 813                         TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
 814                         kfree((caddr_t)slp, M_NFSSVC);
 815                 }
 816                 nfsrv_cleancache();     /* And clear out server cache */
 817         } else
 818                 nfs_pub.np_valid = 0;
 819
 820         TAILQ_INIT(&nfssvc_sockhead);
 821         nfssvc_sockhead_flag &= ~SLP_INIT;
 822         if (nfssvc_sockhead_flag & SLP_WANTINIT) {
 823                 nfssvc_sockhead_flag &= ~SLP_WANTINIT;
 824                 wakeup((caddr_t)&nfssvc_sockhead);
 825         }
 826
 827         TAILQ_INIT(&nfsd_head);
 828         nfsd_head_flag &= ~NFSD_CHECKSLP;
 829
 830 #if 0
 831         nfs_udpsock = (struct nfssvc_sock *)
 832             kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
 833         mtx_init(&nfs_udpsock->ns_solock);
 834         STAILQ_INIT(&nfs_udpsock->ns_rec);
 835         TAILQ_INIT(&nfs_udpsock->ns_uidlruhead);
 836         TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain);
 837
 838         nfs_cltpsock = (struct nfssvc_sock *)
 839             kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
 840         mtx_init(&nfs_cltpsock->ns_solock);
 841         STAILQ_INIT(&nfs_cltpsock->ns_rec);
 842         TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead);
 843         TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain);
 844 #endif
 845 }
 846
 847 /*
 848  * Add entries to the server monitor log.
 849  */
 850 static void
 851 nfsd_rt(int sotype, struct nfsrv_descript *nd, int cacherep)
 852 {
 853         struct drt *rt;
 854
 855         rt = &nfsdrt.drt[nfsdrt.pos];
 856         if (cacherep == RC_DOIT)
 857                 rt->flag = 0;
 858         else if (cacherep == RC_REPLY)
 859                 rt->flag = DRT_CACHEREPLY;
 860         else
 861                 rt->flag = DRT_CACHEDROP;
 862         if (sotype == SOCK_STREAM)
 863                 rt->flag |= DRT_TCP;
 864         if (nd->nd_flag & ND_NFSV3)
 865                 rt->flag |= DRT_NFSV3;
 866         rt->proc = nd->nd_procnum;
 867         if (nd->nd_nam->sa_family == AF_INET)
 868             rt->ipadr = ((struct sockaddr_in *)nd->nd_nam)->sin_addr.s_addr;
 869         else
 870             rt->ipadr = INADDR_ANY;
 871         rt->resptime = nfs_curusec() - (nd->nd_starttime.tv_sec * 1000000 + nd->nd_starttime.tv_usec);
 872         getmicrotime(&rt->tstamp);
 873         nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ;
 874 }
 875 #endif /* NFS_NOSERVER */
 876
 877 static int nfs_defect = 0;
 878 SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, "");
 879
 880 /*
 881  * Get an authorization string for the uid by having the mount_nfs sitting
 882  * on this mount point porpous out of the kernel and do it.
 883  */
 884 int
 885 nfs_getauth(struct nfsmount *nmp, struct nfsreq *rep,
 886             struct ucred *cred, char **auth_str, int *auth_len, char *verf_str,
 887             int *verf_len, NFSKERBKEY_T key /* return session key */)
 888 {
 889         int error = 0;
 890
 891         while ((nmp->nm_state & NFSSTA_WAITAUTH) == 0) {
 892                 nmp->nm_state |= NFSSTA_WANTAUTH;
 893                 (void) tsleep((caddr_t)&nmp->nm_authtype, 0,
 894                         "nfsauth1", 2 * hz);
 895                 error = nfs_sigintr(nmp, rep, rep->r_td);
 896                 if (error) {
 897                         nmp->nm_state &= ~NFSSTA_WANTAUTH;
 898                         return (error);
 899                 }
 900         }
 901         nmp->nm_state &= ~(NFSSTA_WAITAUTH | NFSSTA_WANTAUTH);
 902         nmp->nm_authstr = *auth_str = (char *)kmalloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK);
 903         nmp->nm_authlen = RPCAUTH_MAXSIZ;
 904         nmp->nm_verfstr = verf_str;
 905         nmp->nm_verflen = *verf_len;
 906         nmp->nm_authuid = cred->cr_uid;
 907         wakeup((caddr_t)&nmp->nm_authstr);
 908
 909         /*
 910          * And wait for mount_nfs to do its stuff.
 911          */
 912         while ((nmp->nm_state & NFSSTA_HASAUTH) == 0 && error == 0) {
 913                 (void) tsleep((caddr_t)&nmp->nm_authlen, 0,
 914                         "nfsauth2", 2 * hz);
 915                 error = nfs_sigintr(nmp, rep, rep->r_td);
 916         }
 917         if (nmp->nm_state & NFSSTA_AUTHERR) {
 918                 nmp->nm_state &= ~NFSSTA_AUTHERR;
 919                 error = EAUTH;
 920         }
 921         if (error)
 922                 kfree((caddr_t)*auth_str, M_TEMP);
 923         else {
 924                 *auth_len = nmp->nm_authlen;
 925                 *verf_len = nmp->nm_verflen;
 926                 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (key));
 927         }
 928         nmp->nm_state &= ~NFSSTA_HASAUTH;
 929         nmp->nm_state |= NFSSTA_WAITAUTH;
 930         if (nmp->nm_state & NFSSTA_WANTAUTH) {
 931                 nmp->nm_state &= ~NFSSTA_WANTAUTH;
 932                 wakeup((caddr_t)&nmp->nm_authtype);
 933         }
 934         return (error);
 935 }
 936
 937 /*
 938  * Get a nickname authenticator and verifier.
 939  */
 940 int
 941 nfs_getnickauth(struct nfsmount *nmp, struct ucred *cred, char **auth_str,
 942                 int *auth_len, char *verf_str, int verf_len)
 943 {
 944         struct nfsuid *nuidp;
 945         u_int32_t *nickp, *verfp;
 946         struct timeval ktvin, ktvout;
 947
 948 #ifdef DIAGNOSTIC
 949         if (verf_len < (4 * NFSX_UNSIGNED))
 950                 panic("nfs_getnickauth verf too small");
 951 #endif
 952         for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first;
 953             nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
 954                 if (nuidp->nu_cr.cr_uid == cred->cr_uid)
 955                         break;
 956         }
 957         if (!nuidp || nuidp->nu_expire < time_second)
 958                 return (EACCES);
 959
 960         /*
 961          * Move to the end of the lru list (end of lru == most recently used).
 962          */
 963         TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru);
 964         TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru);
 965
 966         nickp = (u_int32_t *)kmalloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK);
 967         *nickp++ = txdr_unsigned(RPCAKN_NICKNAME);
 968         *nickp = txdr_unsigned(nuidp->nu_nickname);
 969         *auth_str = (char *)nickp;
 970         *auth_len = 2 * NFSX_UNSIGNED;
 971
 972         /*
 973          * Now we must encrypt the verifier and package it up.
 974          */
 975         verfp = (u_int32_t *)verf_str;
 976         *verfp++ = txdr_unsigned(RPCAKN_NICKNAME);
 977         if (time_second > nuidp->nu_timestamp.tv_sec ||
 978             (time_second == nuidp->nu_timestamp.tv_sec &&
 979              time_second > nuidp->nu_timestamp.tv_usec))
 980                 getmicrotime(&nuidp->nu_timestamp);
 981         else
 982                 nuidp->nu_timestamp.tv_usec++;
 983         ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec);
 984         ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec);
 985
 986         /*
 987          * Now encrypt the timestamp verifier in ecb mode using the session
 988          * key.
 989          */
 990 #ifdef NFSKERB
 991         XXX
 992 #else
 993         ktvout.tv_sec = 0;
 994         ktvout.tv_usec = 0;
 995 #endif
 996
 997         *verfp++ = ktvout.tv_sec;
 998         *verfp++ = ktvout.tv_usec;
 999         *verfp = 0;
1000         return (0);
1001 }
1002
1003 /*
1004  * Save the current nickname in a hash list entry on the mount point.
1005  */
1006 int
1007 nfs_savenickauth(struct nfsmount *nmp, struct ucred *cred, int len,
1008                  NFSKERBKEY_T key, struct mbuf **mdp, char **dposp,
1009                  struct mbuf *mrep)
1010 {
1011         struct nfsuid *nuidp;
1012         u_int32_t *tl;
1013         struct timeval ktvin, ktvout;
1014         u_int32_t nick;
1015         int deltasec, error = 0;
1016         struct nfsm_info info;
1017
1018         info.md = *mdp;
1019         info.dpos = *dposp;
1020         info.mrep = mrep;
1021
1022         if (len == (3 * NFSX_UNSIGNED)) {
1023                 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
1024                 ktvin.tv_sec = *tl++;
1025                 ktvin.tv_usec = *tl++;
1026                 nick = fxdr_unsigned(u_int32_t, *tl);
1027
1028                 /*
1029                  * Decrypt the timestamp in ecb mode.
1030                  */
1031 #ifdef NFSKERB
1032                 XXX
1033 #else
1034                 ktvout.tv_sec = 0;
1035                 ktvout.tv_usec = 0;
1036 #endif
1037                 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec);
1038                 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec);
1039                 deltasec = time_second - ktvout.tv_sec;
1040                 if (deltasec < 0)
1041                         deltasec = -deltasec;
1042                 /*
1043                  * If ok, add it to the hash list for the mount point.
1044                  */
1045                 if (deltasec <= NFS_KERBCLOCKSKEW) {
1046                         if (nmp->nm_numuids < nuidhash_max) {
1047                                 nmp->nm_numuids++;
1048                                 nuidp = (struct nfsuid *)
1049                                    kmalloc(sizeof (struct nfsuid), M_NFSUID,
1050                                         M_WAITOK);
1051                         } else {
1052                                 nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead);
1053                                 LIST_REMOVE(nuidp, nu_hash);
1054                                 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp,
1055                                         nu_lru);
1056                         }
1057                         nuidp->nu_flag = 0;
1058                         nuidp->nu_cr.cr_uid = cred->cr_uid;
1059                         nuidp->nu_expire = time_second + NFS_KERBTTL;
1060                         nuidp->nu_timestamp = ktvout;
1061                         nuidp->nu_nickname = nick;
1062                         bcopy(key, nuidp->nu_key, sizeof (key));
1063                         TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp,
1064                                 nu_lru);
1065                         LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid),
1066                                 nuidp, nu_hash);
1067                 }
1068         } else {
1069                 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
1070         }
1071 nfsmout:
1072         *mdp = info.md;
1073         *dposp = info.dpos;
1074         return (error);
1075 }