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
  66 #include <netinet/in.h>
  67 #include <netinet/tcp.h>
  68 #include "xdr_subs.h"
  69 #include "rpcv2.h"
  70 #include "nfsproto.h"
  71 #include "nfs.h"
  72 #include "nfsm_subs.h"
  73 #include "nfsrvcache.h"
  74 #include "nfsmount.h"
  75 #include "nfsnode.h"
  76 #include "nfsrtt.h"
  77
  78 #include <sys/thread2.h>
  79
  80 static MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure");
  81
  82 static int nuidhash_max = NFS_MAXUIDHASH;
  83
  84 #ifndef NFS_NOSERVER
  85 static void     nfsrv_zapsock (struct nfssvc_sock *slp);
  86 #endif
  87
  88 #define TRUE    1
  89 #define FALSE   0
  90
  91 SYSCTL_DECL(_vfs_nfs);
  92
  93 #ifndef NFS_NOSERVER
  94 int nfsd_waiting = 0;
  95 static struct nfsdrt nfsdrt;
  96 static int nfs_numnfsd = 0;
  97 static void     nfsd_rt (int sotype, struct nfsrv_descript *nd,
  98                              int cacherep);
  99 static int      nfssvc_addsock (struct file *, struct sockaddr *,
 100                                     struct thread *);
 101 static int      nfssvc_nfsd (struct nfsd_srvargs *,caddr_t,struct thread *);
 102
 103 static int nfs_privport = 0;
 104 SYSCTL_INT(_vfs_nfs, NFS_NFSPRIVPORT, nfs_privport, CTLFLAG_RW, &nfs_privport, 0, "");
 105 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay, CTLFLAG_RW, &nfsrvw_procrastinate, 0, "");
 106 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay_v3, CTLFLAG_RW, &nfsrvw_procrastinate_v3, 0, "");
 107 static int      nfs_soreserve = 65535;
 108 SYSCTL_INT(_vfs_nfs, OID_AUTO, soreserve, CTLFLAG_RW, &nfs_soreserve, 0, "");
 109
 110 /*
 111  * NFS server system calls
 112  */
 113
 114 #endif /* NFS_NOSERVER */
 115 /*
 116  * nfssvc_args(int flag, caddr_t argp)
 117  *
 118  * Nfs server psuedo system call for the nfsd's
 119  * Based on the flag value it either:
 120  * - adds a socket to the selection list
 121  * - remains in the kernel as an nfsd
 122  * - remains in the kernel as an nfsiod
 123  */
 124 int
 125 sys_nfssvc(struct nfssvc_args *uap)
 126 {
 127 #ifndef NFS_NOSERVER
 128         struct nlookupdata nd;
 129         struct file *fp;
 130         struct sockaddr *nam;
 131         struct nfsd_args nfsdarg;
 132         struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs;
 133         struct nfsd_cargs ncd;
 134         struct nfsd *nfsd;
 135         struct nfssvc_sock *slp;
 136         struct nfsuid *nuidp;
 137         struct nfsmount *nmp;
 138         struct vnode *vp;
 139 #endif /* NFS_NOSERVER */
 140         int error;
 141         struct thread *td = curthread;
 142
 143         /*
 144          * Must be super user
 145          */
 146         error = priv_check(td, PRIV_ROOT);
 147         if(error)
 148                 return (error);
 149         KKASSERT(td->td_proc);  /* for ucred and p_fd */
 150         while (nfssvc_sockhead_flag & SLP_INIT) {
 151                  nfssvc_sockhead_flag |= SLP_WANTINIT;
 152                 (void) tsleep((caddr_t)&nfssvc_sockhead, 0, "nfsd init", 0);
 153         }
 154         if (uap->flag & NFSSVC_BIOD)
 155                 error = ENXIO;          /* no longer need nfsiod's */
 156 #ifdef NFS_NOSERVER
 157         else
 158                 error = ENXIO;
 159 #else /* !NFS_NOSERVER */
 160         else if (uap->flag & NFSSVC_MNTD) {
 161                 error = copyin(uap->argp, (caddr_t)&ncd, sizeof (ncd));
 162                 if (error)
 163                         return (error);
 164                 vp = NULL;
 165                 error = nlookup_init(&nd, ncd.ncd_dirp, UIO_USERSPACE,
 166                                         NLC_FOLLOW);
 167                 if (error == 0)
 168                         error = nlookup(&nd);
 169                 if (error == 0)
 170                         error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
 171                 nlookup_done(&nd);
 172                 if (error)
 173                         return (error);
 174
 175                 if ((vp->v_flag & VROOT) == 0)
 176                         error = EINVAL;
 177                 nmp = VFSTONFS(vp->v_mount);
 178                 vput(vp);
 179                 if (error)
 180                         return (error);
 181                 if ((nmp->nm_state & NFSSTA_MNTD) &&
 182                         (uap->flag & NFSSVC_GOTAUTH) == 0)
 183                         return (0);
 184                 nmp->nm_state |= NFSSTA_MNTD;
 185                 error = nfs_clientd(nmp, td->td_proc->p_ucred, &ncd, uap->flag,
 186                                     uap->argp, td);
 187         } else if (uap->flag & NFSSVC_ADDSOCK) {
 188                 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg));
 189                 if (error)
 190                         return (error);
 191                 error = holdsock(td->td_proc->p_fd, nfsdarg.sock, &fp);
 192                 if (error)
 193                         return (error);
 194                 /*
 195                  * Get the client address for connected sockets.
 196                  */
 197                 if (nfsdarg.name == NULL || nfsdarg.namelen == 0)
 198                         nam = NULL;
 199                 else {
 200                         error = getsockaddr(&nam, nfsdarg.name,
 201                                             nfsdarg.namelen);
 202                         if (error) {
 203                                 fdrop(fp);
 204                                 return (error);
 205                         }
 206                 }
 207                 error = nfssvc_addsock(fp, nam, td);
 208                 fdrop(fp);
 209         } else {
 210                 error = copyin(uap->argp, (caddr_t)nsd, sizeof (*nsd));
 211                 if (error)
 212                         return (error);
 213                 if ((uap->flag & NFSSVC_AUTHIN) &&
 214                     ((nfsd = nsd->nsd_nfsd)) != NULL &&
 215                     (nfsd->nfsd_slp->ns_flag & SLP_VALID)) {
 216                         slp = nfsd->nfsd_slp;
 217
 218                         /*
 219                          * First check to see if another nfsd has already
 220                          * added this credential.
 221                          */
 222                         for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first;
 223                             nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
 224                                 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid &&
 225                                     (!nfsd->nfsd_nd->nd_nam2 ||
 226                                      netaddr_match(NU_NETFAM(nuidp),
 227                                      &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2)))
 228                                         break;
 229                         }
 230                         if (nuidp) {
 231                             nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr);
 232                             nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
 233                         } else {
 234                             /*
 235                              * Nope, so we will.
 236                              */
 237                             if (slp->ns_numuids < nuidhash_max) {
 238                                 slp->ns_numuids++;
 239                                 nuidp = (struct nfsuid *)
 240                                    kmalloc(sizeof (struct nfsuid), M_NFSUID,
 241                                         M_WAITOK);
 242                             } else
 243                                 nuidp = NULL;
 244                             if ((slp->ns_flag & SLP_VALID) == 0) {
 245                                 if (nuidp)
 246                                     kfree((caddr_t)nuidp, M_NFSUID);
 247                             } else {
 248                                 if (nuidp == NULL) {
 249                                     nuidp = TAILQ_FIRST(&slp->ns_uidlruhead);
 250                                     LIST_REMOVE(nuidp, nu_hash);
 251                                     TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp,
 252                                         nu_lru);
 253                                     if (nuidp->nu_flag & NU_NAM)
 254                                         FREE(nuidp->nu_nam, M_SONAME);
 255                                 }
 256                                 nuidp->nu_flag = 0;
 257                                 nuidp->nu_cr = nsd->nsd_cr;
 258                                 if (nuidp->nu_cr.cr_ngroups > NGROUPS)
 259                                     nuidp->nu_cr.cr_ngroups = NGROUPS;
 260                                 nuidp->nu_cr.cr_ref = 1;
 261                                 nuidp->nu_timestamp = nsd->nsd_timestamp;
 262                                 nuidp->nu_expire = time_second + nsd->nsd_ttl;
 263                                 /*
 264                                  * and save the session key in nu_key.
 265                                  */
 266                                 bcopy(nsd->nsd_key, nuidp->nu_key,
 267                                     sizeof (nsd->nsd_key));
 268                                 if (nfsd->nfsd_nd->nd_nam2) {
 269                                     struct sockaddr_in *saddr;
 270
 271                                     saddr = (struct sockaddr_in *)
 272                                             nfsd->nfsd_nd->nd_nam2;
 273                                     switch (saddr->sin_family) {
 274                                     case AF_INET:
 275                                         nuidp->nu_flag |= NU_INETADDR;
 276                                         nuidp->nu_inetaddr =
 277                                              saddr->sin_addr.s_addr;
 278                                         break;
 279                                     case AF_ISO:
 280                                     default:
 281                                         nuidp->nu_flag |= NU_NAM;
 282                                         nuidp->nu_nam =
 283                                           dup_sockaddr(nfsd->nfsd_nd->nd_nam2);
 284                                         break;
 285                                     };
 286                                 }
 287                                 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp,
 288                                         nu_lru);
 289                                 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid),
 290                                         nuidp, nu_hash);
 291                                 nfsrv_setcred(&nuidp->nu_cr,
 292                                     &nfsd->nfsd_nd->nd_cr);
 293                                 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
 294                             }
 295                         }
 296                 }
 297                 if ((uap->flag & NFSSVC_AUTHINFAIL) && (nfsd = nsd->nsd_nfsd))
 298                         nfsd->nfsd_flag |= NFSD_AUTHFAIL;
 299                 error = nfssvc_nfsd(nsd, uap->argp, td);
 300         }
 301 #endif /* NFS_NOSERVER */
 302         if (error == EINTR || error == ERESTART)
 303                 error = 0;
 304         return (error);
 305 }
 306
 307 #ifndef NFS_NOSERVER
 308 /*
 309  * Adds a socket to the list for servicing by nfsds.
 310  */
 311 static int
 312 nfssvc_addsock(struct file *fp, struct sockaddr *mynam, struct thread *td)
 313 {
 314         int siz;
 315         struct nfssvc_sock *slp;
 316         struct socket *so;
 317         int error;
 318
 319         so = (struct socket *)fp->f_data;
 320 #if 0
 321         tslp = NULL;
 322         /*
 323          * Add it to the list, as required.
 324          */
 325         if (so->so_proto->pr_protocol == IPPROTO_UDP) {
 326                 tslp = nfs_udpsock;
 327                 if (tslp->ns_flag & SLP_VALID) {
 328                         if (mynam != NULL)
 329                                 FREE(mynam, M_SONAME);
 330                         return (EPERM);
 331                 }
 332         }
 333 #endif
 334         /*
 335          * Reserve buffer space in the socket.  Note that due to bugs in
 336          * Linux's delayed-ack code, serious performance degredation may
 337          * occur with linux hosts if the minimum is used.
 338          */
 339         if (so->so_type == SOCK_STREAM)
 340                 siz = NFS_MAXPACKET + sizeof (u_long);
 341         else
 342                 siz = NFS_MAXPACKET;
 343         if (siz < nfs_soreserve)
 344             siz = nfs_soreserve;
 345         if (siz > sb_max_adj) {
 346             kprintf("Warning: vfs.nfs.soreserve (%d) "
 347                 "limited to adjusted sb_max (%ld)\n",
 348                 nfs_soreserve, sb_max_adj);
 349             siz = sb_max_adj;
 350         }
 351
 352         error = soreserve(so, siz, siz, &td->td_proc->p_rlimit[RLIMIT_SBSIZE]);
 353         if (error) {
 354                 if (mynam != NULL)
 355                         FREE(mynam, M_SONAME);
 356                 return (error);
 357         }
 358
 359         /*
 360          * Set protocol specific options { for now TCP only } and
 361          * reserve some space. For datagram sockets, this can get called
 362          * repeatedly for the same socket, but that isn't harmful.
 363          */
 364         if (so->so_type == SOCK_STREAM) {
 365                 struct sockopt sopt;
 366                 int val;
 367
 368                 bzero(&sopt, sizeof sopt);
 369                 sopt.sopt_level = SOL_SOCKET;
 370                 sopt.sopt_name = SO_KEEPALIVE;
 371                 sopt.sopt_val = &val;
 372                 sopt.sopt_valsize = sizeof val;
 373                 val = 1;
 374                 sosetopt(so, &sopt);
 375         }
 376         if (so->so_proto->pr_domain->dom_family == AF_INET &&
 377             so->so_proto->pr_protocol == IPPROTO_TCP) {
 378                 struct sockopt sopt;
 379                 int val;
 380
 381                 bzero(&sopt, sizeof sopt);
 382                 sopt.sopt_level = IPPROTO_TCP;
 383                 sopt.sopt_name = TCP_NODELAY;
 384                 sopt.sopt_val = &val;
 385                 sopt.sopt_valsize = sizeof val;
 386                 val = 1;
 387                 sosetopt(so, &sopt);
 388         }
 389         so->so_rcv.ssb_flags &= ~SSB_NOINTR;
 390         so->so_rcv.ssb_timeo = 0;
 391         so->so_snd.ssb_flags &= ~SSB_NOINTR;
 392         so->so_snd.ssb_timeo = 0;
 393
 394         slp = (struct nfssvc_sock *)kmalloc(sizeof (struct nfssvc_sock),
 395             M_NFSSVC, M_WAITOK | M_ZERO);
 396         mtx_init(&slp->ns_solock);
 397         STAILQ_INIT(&slp->ns_rec);
 398         TAILQ_INIT(&slp->ns_uidlruhead);
 399         TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain);
 400
 401         slp->ns_so = so;
 402         slp->ns_nam = mynam;
 403         fp->f_count++;
 404         slp->ns_fp = fp;
 405         crit_enter();
 406         so->so_upcallarg = (caddr_t)slp;
 407         so->so_upcall = nfsrv_rcv;
 408         so->so_rcv.ssb_flags |= SSB_UPCALL;
 409         slp->ns_flag = (SLP_VALID | SLP_NEEDQ);
 410         nfsrv_wakenfsd(slp, 1);
 411         crit_exit();
 412         return (0);
 413 }
 414
 415 /*
 416  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
 417  * until it is killed by a signal.
 418  */
 419 static int
 420 nfssvc_nfsd(struct nfsd_srvargs *nsd, caddr_t argp, struct thread *td)
 421 {
 422         int siz;
 423         struct nfssvc_sock *slp;
 424         struct nfsd *nfsd = nsd->nsd_nfsd;
 425         struct nfsrv_descript *nd = NULL;
 426         struct mbuf *m, *mreq;
 427         int error = 0, cacherep, sotype, writes_todo;
 428         int procrastinate;
 429         u_quad_t cur_usec;
 430
 431 #ifndef nolint
 432         cacherep = RC_DOIT;
 433         writes_todo = 0;
 434 #endif
 435         if (nfsd == NULL) {
 436                 nsd->nsd_nfsd = nfsd = (struct nfsd *)
 437                         kmalloc(sizeof (struct nfsd), M_NFSD, M_WAITOK|M_ZERO);
 438                 crit_enter();
 439                 nfsd->nfsd_td = td;
 440                 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
 441                 nfs_numnfsd++;
 442         } else
 443                 crit_enter();
 444
 445         /*
 446          * Loop getting rpc requests until SIGKILL.
 447          */
 448         for (;;) {
 449                 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) {
 450                         while (nfsd->nfsd_slp == NULL &&
 451                             (nfsd_head_flag & NFSD_CHECKSLP) == 0) {
 452                                 nfsd->nfsd_flag |= NFSD_WAITING;
 453                                 nfsd_waiting++;
 454                                 error = tsleep((caddr_t)nfsd, PCATCH, "nfsd", 0);
 455                                 nfsd_waiting--;
 456                                 if (error)
 457                                         goto done;
 458                         }
 459                         if (nfsd->nfsd_slp == NULL &&
 460                             (nfsd_head_flag & NFSD_CHECKSLP) != 0) {
 461                                 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
 462                                     if ((slp->ns_flag & (SLP_VALID | SLP_DOREC))
 463                                         == (SLP_VALID | SLP_DOREC)) {
 464                                             slp->ns_flag &= ~SLP_DOREC;
 465                                             slp->ns_sref++;
 466                                             nfsd->nfsd_slp = slp;
 467                                             break;
 468                                     }
 469                                 }
 470                                 if (slp == 0)
 471                                         nfsd_head_flag &= ~NFSD_CHECKSLP;
 472                         }
 473                         if ((slp = nfsd->nfsd_slp) == NULL)
 474                                 continue;
 475                         if (slp->ns_flag & SLP_VALID) {
 476                                 if (slp->ns_flag & SLP_DISCONN)
 477                                         nfsrv_zapsock(slp);
 478                                 else if (slp->ns_flag & SLP_NEEDQ) {
 479                                         slp->ns_flag &= ~SLP_NEEDQ;
 480                                         (void) nfs_slplock(slp, 1);
 481                                         nfsrv_rcv(slp->ns_so, (caddr_t)slp,
 482                                                 MB_WAIT);
 483                                         nfs_slpunlock(slp);
 484                                 }
 485                                 error = nfsrv_dorec(slp, nfsd, &nd);
 486                                 cur_usec = nfs_curusec();
 487                                 if (error && slp->ns_tq.lh_first &&
 488                                     slp->ns_tq.lh_first->nd_time <= cur_usec) {
 489                                         error = 0;
 490                                         cacherep = RC_DOIT;
 491                                         writes_todo = 1;
 492                                 } else
 493                                         writes_todo = 0;
 494                                 nfsd->nfsd_flag |= NFSD_REQINPROG;
 495                         }
 496                 } else {
 497                         error = 0;
 498                         slp = nfsd->nfsd_slp;
 499                 }
 500                 if (error || (slp->ns_flag & SLP_VALID) == 0) {
 501                         if (nd) {
 502                                 kfree((caddr_t)nd, M_NFSRVDESC);
 503                                 nd = NULL;
 504                         }
 505                         nfsd->nfsd_slp = NULL;
 506                         nfsd->nfsd_flag &= ~NFSD_REQINPROG;
 507                         nfsrv_slpderef(slp);
 508                         continue;
 509                 }
 510                 crit_exit();
 511                 sotype = slp->ns_so->so_type;
 512                 if (nd) {
 513                     getmicrotime(&nd->nd_starttime);
 514                     if (nd->nd_nam2)
 515                         nd->nd_nam = nd->nd_nam2;
 516                     else
 517                         nd->nd_nam = slp->ns_nam;
 518
 519                     /*
 520                      * Check to see if authorization is needed.
 521                      */
 522                     if (nfsd->nfsd_flag & NFSD_NEEDAUTH) {
 523                         nfsd->nfsd_flag &= ~NFSD_NEEDAUTH;
 524                         nsd->nsd_haddr =
 525                                 ((struct sockaddr_in *)
 526                                  nd->nd_nam)->sin_addr.s_addr;
 527                         nsd->nsd_authlen = nfsd->nfsd_authlen;
 528                         nsd->nsd_verflen = nfsd->nfsd_verflen;
 529                         if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr,
 530                                 nfsd->nfsd_authlen) &&
 531                             !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr,
 532                                 nfsd->nfsd_verflen) &&
 533                             !copyout((caddr_t)nsd, argp, sizeof (*nsd)))
 534                             return (ENEEDAUTH);
 535                         cacherep = RC_DROPIT;
 536                     } else {
 537                         cacherep = nfsrv_getcache(nd, slp, &mreq);
 538                     }
 539
 540                     if (nfsd->nfsd_flag & NFSD_AUTHFAIL) {
 541                         nfsd->nfsd_flag &= ~NFSD_AUTHFAIL;
 542                         nd->nd_procnum = NFSPROC_NOOP;
 543                         nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
 544                         cacherep = RC_DOIT;
 545                     } else if (nfs_privport) {
 546                         /* Check if source port is privileged */
 547                         u_short port;
 548                         struct sockaddr *nam = nd->nd_nam;
 549                         struct sockaddr_in *sin;
 550
 551                         sin = (struct sockaddr_in *)nam;
 552                         port = ntohs(sin->sin_port);
 553                         if (port >= IPPORT_RESERVED &&
 554                             nd->nd_procnum != NFSPROC_NULL) {
 555                             nd->nd_procnum = NFSPROC_NOOP;
 556                             nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
 557                             cacherep = RC_DOIT;
 558                             kprintf("NFS request from unprivileged port (%s:%d)\n",
 559                                    inet_ntoa(sin->sin_addr), port);
 560                         }
 561                     }
 562
 563                 }
 564
 565                 /*
 566                  * Loop to get all the write rpc relies that have been
 567                  * gathered together.
 568                  */
 569                 do {
 570                     switch (cacherep) {
 571                     case RC_DOIT:
 572                         if (nd && (nd->nd_flag & ND_NFSV3))
 573                             procrastinate = nfsrvw_procrastinate_v3;
 574                         else
 575                             procrastinate = nfsrvw_procrastinate;
 576                         if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE &&
 577                             procrastinate > 0)
 578                         ) {
 579                             error = nfsrv_writegather(&nd, slp,
 580                                 nfsd->nfsd_td, &mreq);
 581                         } else {
 582                             error = (*(nfsrv3_procs[nd->nd_procnum]))(nd,
 583                                 slp, nfsd->nfsd_td, &mreq);
 584                         }
 585                         if (mreq == NULL)
 586                                 break;
 587                         if (error != 0 && error != NFSERR_RETVOID) {
 588                                 if (nd->nd_procnum != NQNFSPROC_VACATED)
 589                                         nfsstats.srv_errs++;
 590                                 nfsrv_updatecache(nd, FALSE, mreq);
 591                                 if (nd->nd_nam2)
 592                                         FREE(nd->nd_nam2, M_SONAME);
 593                                 break;
 594                         }
 595                         nfsstats.srvrpccnt[nd->nd_procnum]++;
 596                         nfsrv_updatecache(nd, TRUE, mreq);
 597                         nd->nd_mrep = NULL;
 598                     case RC_REPLY:
 599                         m = mreq;
 600                         siz = 0;
 601                         while (m) {
 602                                 siz += m->m_len;
 603                                 m = m->m_next;
 604                         }
 605                         if (siz <= 0 || siz > NFS_MAXPACKET) {
 606                                 kprintf("mbuf siz=%d\n",siz);
 607                                 panic("Bad nfs svc reply");
 608                         }
 609                         m = mreq;
 610                         m->m_pkthdr.len = siz;
 611                         m->m_pkthdr.rcvif = NULL;
 612                         /*
 613                          * For stream protocols, prepend a Sun RPC
 614                          * Record Mark.
 615                          */
 616                         if (sotype == SOCK_STREAM) {
 617                                 M_PREPEND(m, NFSX_UNSIGNED, MB_WAIT);
 618                                 if (m == NULL)
 619                                         return (ENOBUFS);
 620                                 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz);
 621                         }
 622                         if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
 623                                 (void) nfs_slplock(slp, 1);
 624                         if (slp->ns_flag & SLP_VALID)
 625                             error = nfs_send(slp->ns_so, nd->nd_nam2, m, NULL);
 626                         else {
 627                             error = EPIPE;
 628                             m_freem(m);
 629                         }
 630                         if (nfsrtton)
 631                                 nfsd_rt(sotype, nd, cacherep);
 632                         if (nd->nd_nam2)
 633                                 FREE(nd->nd_nam2, M_SONAME);
 634                         if (nd->nd_mrep)
 635                                 m_freem(nd->nd_mrep);
 636                         if (error == EPIPE)
 637                                 nfsrv_zapsock(slp);
 638                         if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
 639                                 nfs_slpunlock(slp);
 640                         if (error == EINTR || error == ERESTART) {
 641                                 kfree((caddr_t)nd, M_NFSRVDESC);
 642                                 nfsrv_slpderef(slp);
 643                                 crit_enter();
 644                                 goto done;
 645                         }
 646                         break;
 647                     case RC_DROPIT:
 648                         if (nfsrtton)
 649                                 nfsd_rt(sotype, nd, cacherep);
 650                         m_freem(nd->nd_mrep);
 651                         if (nd->nd_nam2)
 652                                 FREE(nd->nd_nam2, M_SONAME);
 653                         break;
 654                     };
 655                     if (nd) {
 656                         FREE((caddr_t)nd, M_NFSRVDESC);
 657                         nd = NULL;
 658                     }
 659
 660                     /*
 661                      * Check to see if there are outstanding writes that
 662                      * need to be serviced.
 663                      */
 664                     cur_usec = nfs_curusec();
 665                     crit_enter();
 666                     if (slp->ns_tq.lh_first &&
 667                         slp->ns_tq.lh_first->nd_time <= cur_usec) {
 668                         cacherep = RC_DOIT;
 669                         writes_todo = 1;
 670                     } else
 671                         writes_todo = 0;
 672                     crit_exit();
 673                 } while (writes_todo);
 674                 crit_enter();
 675                 if (nfsrv_dorec(slp, nfsd, &nd)) {
 676                         nfsd->nfsd_flag &= ~NFSD_REQINPROG;
 677                         nfsd->nfsd_slp = NULL;
 678                         nfsrv_slpderef(slp);
 679                 }
 680         }
 681 done:
 682         TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
 683         crit_exit();
 684         kfree((caddr_t)nfsd, M_NFSD);
 685         nsd->nsd_nfsd = NULL;
 686         if (--nfs_numnfsd == 0)
 687                 nfsrv_init(TRUE);       /* Reinitialize everything */
 688         return (error);
 689 }
 690
 691 /*
 692  * Shut down a socket associated with an nfssvc_sock structure.
 693  * Should be called with the send lock set, if required.
 694  * The trick here is to increment the sref at the start, so that the nfsds
 695  * will stop using it and clear ns_flag at the end so that it will not be
 696  * reassigned during cleanup.
 697  */
 698 static void
 699 nfsrv_zapsock(struct nfssvc_sock *slp)
 700 {
 701         struct nfsuid *nuidp, *nnuidp;
 702         struct nfsrv_descript *nwp, *nnwp;
 703         struct socket *so;
 704         struct file *fp;
 705         struct nfsrv_rec *rec;
 706
 707         slp->ns_flag &= ~SLP_ALLFLAGS;
 708         fp = slp->ns_fp;
 709         if (fp) {
 710                 slp->ns_fp = NULL;
 711                 so = slp->ns_so;
 712                 so->so_rcv.ssb_flags &= ~SSB_UPCALL;
 713                 so->so_upcall = NULL;
 714                 so->so_upcallarg = NULL;
 715                 soshutdown(so, SHUT_RDWR);
 716                 closef(fp, NULL);
 717                 if (slp->ns_nam)
 718                         FREE(slp->ns_nam, M_SONAME);
 719                 m_freem(slp->ns_raw);
 720                 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) {
 721                         --slp->ns_numrec;
 722                         STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link);
 723                         if (rec->nr_address)
 724                                 FREE(rec->nr_address, M_SONAME);
 725                         m_freem(rec->nr_packet);
 726                         kfree(rec, M_NFSRVDESC);
 727                 }
 728                 TAILQ_FOREACH_MUTABLE(nuidp, &slp->ns_uidlruhead, nu_lru,
 729                                       nnuidp) {
 730                         LIST_REMOVE(nuidp, nu_hash);
 731                         TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru);
 732                         if (nuidp->nu_flag & NU_NAM)
 733                                 FREE(nuidp->nu_nam, M_SONAME);
 734                         kfree((caddr_t)nuidp, M_NFSUID);
 735                 }
 736                 crit_enter();
 737                 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) {
 738                         nnwp = nwp->nd_tq.le_next;
 739                         LIST_REMOVE(nwp, nd_tq);
 740                         kfree((caddr_t)nwp, M_NFSRVDESC);
 741                 }
 742                 LIST_INIT(&slp->ns_tq);
 743                 crit_exit();
 744         }
 745 }
 746
 747 /*
 748  * Derefence a server socket structure. If it has no more references and
 749  * is no longer valid, you can throw it away.
 750  */
 751 void
 752 nfsrv_slpderef(struct nfssvc_sock *slp)
 753 {
 754         if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) {
 755                 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
 756                 kfree((caddr_t)slp, M_NFSSVC);
 757         }
 758 }
 759
 760 /*
 761  * Lock a socket against others.
 762  *
 763  * Returns 0 on failure, 1 on success.
 764  */
 765 int
 766 nfs_slplock(struct nfssvc_sock *slp, int wait)
 767 {
 768         mtx_t mtx = &slp->ns_solock;
 769
 770         if (wait) {
 771                 mtx_lock_ex(mtx, "nfsslplck", 0, 0);
 772                 return(1);
 773         } else if (mtx_lock_ex_try(mtx) == 0) {
 774                 return(1);
 775         } else {
 776                 return(0);
 777         }
 778 }
 779
 780 /*
 781  * Unlock the stream socket for others.
 782  */
 783 void
 784 nfs_slpunlock(struct nfssvc_sock *slp)
 785 {
 786         mtx_t mtx = &slp->ns_solock;
 787
 788         mtx_unlock(mtx);
 789 }
 790
 791 /*
 792  * Initialize the data structures for the server.
 793  * Handshake with any new nfsds starting up to avoid any chance of
 794  * corruption.
 795  */
 796 void
 797 nfsrv_init(int terminating)
 798 {
 799         struct nfssvc_sock *slp, *nslp;
 800
 801         if (nfssvc_sockhead_flag & SLP_INIT)
 802                 panic("nfsd init");
 803         nfssvc_sockhead_flag |= SLP_INIT;
 804         if (terminating) {
 805                 TAILQ_FOREACH_MUTABLE(slp, &nfssvc_sockhead, ns_chain, nslp) {
 806                         if (slp->ns_flag & SLP_VALID)
 807                                 nfsrv_zapsock(slp);
 808                         TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
 809                         kfree((caddr_t)slp, M_NFSSVC);
 810                 }
 811                 nfsrv_cleancache();     /* And clear out server cache */
 812         } else
 813                 nfs_pub.np_valid = 0;
 814
 815         TAILQ_INIT(&nfssvc_sockhead);
 816         nfssvc_sockhead_flag &= ~SLP_INIT;
 817         if (nfssvc_sockhead_flag & SLP_WANTINIT) {
 818                 nfssvc_sockhead_flag &= ~SLP_WANTINIT;
 819                 wakeup((caddr_t)&nfssvc_sockhead);
 820         }
 821
 822         TAILQ_INIT(&nfsd_head);
 823         nfsd_head_flag &= ~NFSD_CHECKSLP;
 824
 825 #if 0
 826         nfs_udpsock = (struct nfssvc_sock *)
 827             kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
 828         mtx_init(&nfs_udpsock->ns_solock);
 829         STAILQ_INIT(&nfs_udpsock->ns_rec);
 830         TAILQ_INIT(&nfs_udpsock->ns_uidlruhead);
 831         TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain);
 832
 833         nfs_cltpsock = (struct nfssvc_sock *)
 834             kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
 835         mtx_init(&nfs_cltpsock->ns_solock);
 836         STAILQ_INIT(&nfs_cltpsock->ns_rec);
 837         TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead);
 838         TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain);
 839 #endif
 840 }
 841
 842 /*
 843  * Add entries to the server monitor log.
 844  */
 845 static void
 846 nfsd_rt(int sotype, struct nfsrv_descript *nd, int cacherep)
 847 {
 848         struct drt *rt;
 849
 850         rt = &nfsdrt.drt[nfsdrt.pos];
 851         if (cacherep == RC_DOIT)
 852                 rt->flag = 0;
 853         else if (cacherep == RC_REPLY)
 854                 rt->flag = DRT_CACHEREPLY;
 855         else
 856                 rt->flag = DRT_CACHEDROP;
 857         if (sotype == SOCK_STREAM)
 858                 rt->flag |= DRT_TCP;
 859         if (nd->nd_flag & ND_NFSV3)
 860                 rt->flag |= DRT_NFSV3;
 861         rt->proc = nd->nd_procnum;
 862         if (nd->nd_nam->sa_family == AF_INET)
 863             rt->ipadr = ((struct sockaddr_in *)nd->nd_nam)->sin_addr.s_addr;
 864         else
 865             rt->ipadr = INADDR_ANY;
 866         rt->resptime = nfs_curusec() - (nd->nd_starttime.tv_sec * 1000000 + nd->nd_starttime.tv_usec);
 867         getmicrotime(&rt->tstamp);
 868         nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ;
 869 }
 870 #endif /* NFS_NOSERVER */
 871
 872 static int nfs_defect = 0;
 873 SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, "");
 874
 875 /*
 876  * Get an authorization string for the uid by having the mount_nfs sitting
 877  * on this mount point porpous out of the kernel and do it.
 878  */
 879 int
 880 nfs_getauth(struct nfsmount *nmp, struct nfsreq *rep,
 881             struct ucred *cred, char **auth_str, int *auth_len, char *verf_str,
 882             int *verf_len, NFSKERBKEY_T key /* return session key */)
 883 {
 884         int error = 0;
 885
 886         while ((nmp->nm_state & NFSSTA_WAITAUTH) == 0) {
 887                 nmp->nm_state |= NFSSTA_WANTAUTH;
 888                 (void) tsleep((caddr_t)&nmp->nm_authtype, 0,
 889                         "nfsauth1", 2 * hz);
 890                 error = nfs_sigintr(nmp, rep, rep->r_td);
 891                 if (error) {
 892                         nmp->nm_state &= ~NFSSTA_WANTAUTH;
 893                         return (error);
 894                 }
 895         }
 896         nmp->nm_state &= ~(NFSSTA_WAITAUTH | NFSSTA_WANTAUTH);
 897         nmp->nm_authstr = *auth_str = (char *)kmalloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK);
 898         nmp->nm_authlen = RPCAUTH_MAXSIZ;
 899         nmp->nm_verfstr = verf_str;
 900         nmp->nm_verflen = *verf_len;
 901         nmp->nm_authuid = cred->cr_uid;
 902         wakeup((caddr_t)&nmp->nm_authstr);
 903
 904         /*
 905          * And wait for mount_nfs to do its stuff.
 906          */
 907         while ((nmp->nm_state & NFSSTA_HASAUTH) == 0 && error == 0) {
 908                 (void) tsleep((caddr_t)&nmp->nm_authlen, 0,
 909                         "nfsauth2", 2 * hz);
 910                 error = nfs_sigintr(nmp, rep, rep->r_td);
 911         }
 912         if (nmp->nm_state & NFSSTA_AUTHERR) {
 913                 nmp->nm_state &= ~NFSSTA_AUTHERR;
 914                 error = EAUTH;
 915         }
 916         if (error)
 917                 kfree((caddr_t)*auth_str, M_TEMP);
 918         else {
 919                 *auth_len = nmp->nm_authlen;
 920                 *verf_len = nmp->nm_verflen;
 921                 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (key));
 922         }
 923         nmp->nm_state &= ~NFSSTA_HASAUTH;
 924         nmp->nm_state |= NFSSTA_WAITAUTH;
 925         if (nmp->nm_state & NFSSTA_WANTAUTH) {
 926                 nmp->nm_state &= ~NFSSTA_WANTAUTH;
 927                 wakeup((caddr_t)&nmp->nm_authtype);
 928         }
 929         return (error);
 930 }
 931
 932 /*
 933  * Get a nickname authenticator and verifier.
 934  */
 935 int
 936 nfs_getnickauth(struct nfsmount *nmp, struct ucred *cred, char **auth_str,
 937                 int *auth_len, char *verf_str, int verf_len)
 938 {
 939         struct nfsuid *nuidp;
 940         u_int32_t *nickp, *verfp;
 941         struct timeval ktvin, ktvout;
 942
 943 #ifdef DIAGNOSTIC
 944         if (verf_len < (4 * NFSX_UNSIGNED))
 945                 panic("nfs_getnickauth verf too small");
 946 #endif
 947         for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first;
 948             nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
 949                 if (nuidp->nu_cr.cr_uid == cred->cr_uid)
 950                         break;
 951         }
 952         if (!nuidp || nuidp->nu_expire < time_second)
 953                 return (EACCES);
 954
 955         /*
 956          * Move to the end of the lru list (end of lru == most recently used).
 957          */
 958         TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru);
 959         TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru);
 960
 961         nickp = (u_int32_t *)kmalloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK);
 962         *nickp++ = txdr_unsigned(RPCAKN_NICKNAME);
 963         *nickp = txdr_unsigned(nuidp->nu_nickname);
 964         *auth_str = (char *)nickp;
 965         *auth_len = 2 * NFSX_UNSIGNED;
 966
 967         /*
 968          * Now we must encrypt the verifier and package it up.
 969          */
 970         verfp = (u_int32_t *)verf_str;
 971         *verfp++ = txdr_unsigned(RPCAKN_NICKNAME);
 972         if (time_second > nuidp->nu_timestamp.tv_sec ||
 973             (time_second == nuidp->nu_timestamp.tv_sec &&
 974              time_second > nuidp->nu_timestamp.tv_usec))
 975                 getmicrotime(&nuidp->nu_timestamp);
 976         else
 977                 nuidp->nu_timestamp.tv_usec++;
 978         ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec);
 979         ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec);
 980
 981         /*
 982          * Now encrypt the timestamp verifier in ecb mode using the session
 983          * key.
 984          */
 985 #ifdef NFSKERB
 986         XXX
 987 #endif
 988
 989         *verfp++ = ktvout.tv_sec;
 990         *verfp++ = ktvout.tv_usec;
 991         *verfp = 0;
 992         return (0);
 993 }
 994
 995 /*
 996  * Save the current nickname in a hash list entry on the mount point.
 997  */
 998 int
 999 nfs_savenickauth(struct nfsmount *nmp, struct ucred *cred, int len,
1000                  NFSKERBKEY_T key, struct mbuf **mdp, char **dposp,
1001                  struct mbuf *mrep)
1002 {
1003         struct nfsuid *nuidp;
1004         u_int32_t *tl;
1005         struct timeval ktvin, ktvout;
1006         u_int32_t nick;
1007         int deltasec, error = 0;
1008         struct nfsm_info info;
1009
1010         info.md = *mdp;
1011         info.dpos = *dposp;
1012         info.mrep = mrep;
1013
1014         if (len == (3 * NFSX_UNSIGNED)) {
1015                 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
1016                 ktvin.tv_sec = *tl++;
1017                 ktvin.tv_usec = *tl++;
1018                 nick = fxdr_unsigned(u_int32_t, *tl);
1019
1020                 /*
1021                  * Decrypt the timestamp in ecb mode.
1022                  */
1023 #ifdef NFSKERB
1024                 XXX
1025 #endif
1026                 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec);
1027                 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec);
1028                 deltasec = time_second - ktvout.tv_sec;
1029                 if (deltasec < 0)
1030                         deltasec = -deltasec;
1031                 /*
1032                  * If ok, add it to the hash list for the mount point.
1033                  */
1034                 if (deltasec <= NFS_KERBCLOCKSKEW) {
1035                         if (nmp->nm_numuids < nuidhash_max) {
1036                                 nmp->nm_numuids++;
1037                                 nuidp = (struct nfsuid *)
1038                                    kmalloc(sizeof (struct nfsuid), M_NFSUID,
1039                                         M_WAITOK);
1040                         } else {
1041                                 nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead);
1042                                 LIST_REMOVE(nuidp, nu_hash);
1043                                 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp,
1044                                         nu_lru);
1045                         }
1046                         nuidp->nu_flag = 0;
1047                         nuidp->nu_cr.cr_uid = cred->cr_uid;
1048                         nuidp->nu_expire = time_second + NFS_KERBTTL;
1049                         nuidp->nu_timestamp = ktvout;
1050                         nuidp->nu_nickname = nick;
1051                         bcopy(key, nuidp->nu_key, sizeof (key));
1052                         TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp,
1053                                 nu_lru);
1054                         LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid),
1055                                 nuidp, nu_hash);
1056                 }
1057         } else {
1058                 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
1059         }
1060 nfsmout:
1061         *mdp = info.md;
1062         *dposp = info.dpos;
1063         return (error);
1064 }