sys/fs/nfsserver/nfs_nfsdport.c

   1 /*-
   2  * Copyright (c) 1989, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to Berkeley by
   6  * Rick Macklem at The University of Guelph.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 4. Neither the name of the University nor the names of its contributors
  17  *    may be used to endorse or promote products derived from this software
  18  *    without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  *
  32  */
  33
  34 #include <sys/cdefs.h>
  35 __FBSDID("$FreeBSD$");
  36
  37 /*
  38  * Functions that perform the vfs operations required by the routines in
  39  * nfsd_serv.c. It is hoped that this change will make the server more
  40  * portable.
  41  */
  42
  43 #include <fs/nfs/nfsport.h>
  44 #include <sys/sysctl.h>
  45 #include <nlm/nlm_prot.h>
  46 #include <nlm/nlm.h>
  47
  48 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
  49 extern int nfsv4root_set;
  50 extern int nfsrv_useacl;
  51 extern int newnfs_numnfsd;
  52 extern struct mount nfsv4root_mnt;
  53 extern struct nfsrv_stablefirst nfsrv_stablefirst;
  54 extern void (*nfsd_call_servertimer)(void);
  55 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
  56 NFSDLOCKMUTEX;
  57 struct mtx nfs_cache_mutex;
  58 struct mtx nfs_v4root_mutex;
  59 struct nfsrvfh nfs_rootfh, nfs_pubfh;
  60 int nfs_pubfhset = 0, nfs_rootfhset = 0;
  61 static uint32_t nfsv4_sysid = 0;
  62
  63 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
  64     struct ucred *);
  65
  66 static int enable_crossmntpt = 1;
  67 static int nfs_commit_blks;
  68 static int nfs_commit_miss;
  69 extern int nfsrv_issuedelegs;
  70 extern int nfsrv_dolocallocks;
  71
  72 SYSCTL_DECL(_vfs_newnfs);
  73 SYSCTL_INT(_vfs_newnfs, OID_AUTO, mirrormnt, CTLFLAG_RW, &enable_crossmntpt,
  74     0, "Enable nfsd to cross mount points");
  75 SYSCTL_INT(_vfs_newnfs, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
  76     0, "");
  77 SYSCTL_INT(_vfs_newnfs, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
  78     0, "");
  79 SYSCTL_INT(_vfs_newnfs, OID_AUTO, issue_delegations, CTLFLAG_RW,
  80     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
  81 SYSCTL_INT(_vfs_newnfs, OID_AUTO, enable_locallocks, CTLFLAG_RW,
  82     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
  83
  84 #define NUM_HEURISTIC           1017
  85 #define NHUSE_INIT              64
  86 #define NHUSE_INC               16
  87 #define NHUSE_MAX               2048
  88
  89 static struct nfsheur {
  90         struct vnode *nh_vp;    /* vp to match (unreferenced pointer) */
  91         off_t nh_nextr;         /* next offset for sequential detection */
  92         int nh_use;             /* use count for selection */
  93         int nh_seqcount;        /* heuristic */
  94 } nfsheur[NUM_HEURISTIC];
  95
  96
  97 /*
  98  * Get attributes into nfsvattr structure.
  99  */
 100 int
 101 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
 102     struct thread *p)
 103 {
 104         int error, lockedit = 0;
 105
 106         /* Since FreeBSD insists the vnode be locked... */
 107         if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
 108                 lockedit = 1;
 109                 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
 110         }
 111         error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
 112         if (lockedit)
 113                 NFSVOPUNLOCK(vp, 0, p);
 114         return (error);
 115 }
 116
 117 /*
 118  * Get a file handle for a vnode.
 119  */
 120 int
 121 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
 122 {
 123         int error;
 124
 125         NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
 126         fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
 127         error = VOP_VPTOFH(vp, &fhp->fh_fid);
 128         return (error);
 129 }
 130
 131 /*
 132  * Perform access checking for vnodes obtained from file handles that would
 133  * refer to files already opened by a Unix client. You cannot just use
 134  * vn_writechk() and VOP_ACCESSX() for two reasons.
 135  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
 136  *     case.
 137  * 2 - The owner is to be given access irrespective of mode bits for some
 138  *     operations, so that processes that chmod after opening a file don't
 139  *     break.
 140  */
 141 int
 142 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
 143     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
 144     u_int32_t *supportedtypep)
 145 {
 146         struct vattr vattr;
 147         int error = 0, getret = 0;
 148
 149         if (accmode & VWRITE) {
 150                 /* Just vn_writechk() changed to check rdonly */
 151                 /*
 152                  * Disallow write attempts on read-only file systems;
 153                  * unless the file is a socket or a block or character
 154                  * device resident on the file system.
 155                  */
 156                 if (NFSVNO_EXRDONLY(exp) ||
 157                     (vp->v_mount->mnt_flag & MNT_RDONLY)) {
 158                         switch (vp->v_type) {
 159                         case VREG:
 160                         case VDIR:
 161                         case VLNK:
 162                                 return (EROFS);
 163                         default:
 164                                 break;
 165                         }
 166                 }
 167                 /*
 168                  * If there's shared text associated with
 169                  * the inode, try to free it up once.  If
 170                  * we fail, we can't allow writing.
 171                  */
 172                 if (vp->v_vflag & VV_TEXT)
 173                         return (ETXTBSY);
 174         }
 175         if (vpislocked == 0)
 176                 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
 177
 178         /*
 179          * Should the override still be applied when ACLs are enabled?
 180          */
 181         error = VOP_ACCESSX(vp, accmode, cred, p);
 182         if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
 183                 /*
 184                  * Try again with VEXPLICIT_DENY, to see if the test for
 185                  * deletion is supported.
 186                  */
 187                 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
 188                 if (error == 0) {
 189                         if (vp->v_type == VDIR) {
 190                                 accmode &= ~(VDELETE | VDELETE_CHILD);
 191                                 accmode |= VWRITE;
 192                                 error = VOP_ACCESSX(vp, accmode, cred, p);
 193                         } else if (supportedtypep != NULL) {
 194                                 *supportedtypep &= ~NFSACCESS_DELETE;
 195                         }
 196                 }
 197         }
 198
 199         /*
 200          * Allow certain operations for the owner (reads and writes
 201          * on files that are already open).
 202          */
 203         if (override != NFSACCCHK_NOOVERRIDE &&
 204             (error == EPERM || error == EACCES)) {
 205                 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
 206                         error = 0;
 207                 else if (override & NFSACCCHK_ALLOWOWNER) {
 208                         getret = VOP_GETATTR(vp, &vattr, cred);
 209                         if (getret == 0 && cred->cr_uid == vattr.va_uid)
 210                                 error = 0;
 211                 }
 212         }
 213         if (vpislocked == 0)
 214                 NFSVOPUNLOCK(vp, 0, p);
 215         return (error);
 216 }
 217
 218 /*
 219  * Set attribute(s) vnop.
 220  */
 221 int
 222 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
 223     struct thread *p, struct nfsexstuff *exp)
 224 {
 225         int error;
 226
 227         error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
 228         return (error);
 229 }
 230
 231 /*
 232  * Set up nameidata for a lookup() call and do it
 233  * For the cases where we are crossing mount points
 234  * (looking up the public fh path or the v4 root path when
 235  *  not using a pseudo-root fs), set/release the Giant lock,
 236  * as required.
 237  */
 238 int
 239 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
 240     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
 241     struct vnode **retdirp)
 242 {
 243         struct componentname *cnp = &ndp->ni_cnd;
 244         int i;
 245         struct iovec aiov;
 246         struct uio auio;
 247         int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
 248         int error = 0, crossmnt;
 249         char *cp;
 250
 251         *retdirp = NULL;
 252         cnp->cn_nameptr = cnp->cn_pnbuf;
 253         /*
 254          * Extract and set starting directory.
 255          */
 256         if (dp->v_type != VDIR) {
 257                 if (islocked)
 258                         vput(dp);
 259                 else
 260                         vrele(dp);
 261                 nfsvno_relpathbuf(ndp);
 262                 return (ENOTDIR);
 263         }
 264         if (islocked)
 265                 NFSVOPUNLOCK(dp, 0, p);
 266         VREF(dp);
 267         *retdirp = dp;
 268         if (NFSVNO_EXRDONLY(exp))
 269                 cnp->cn_flags |= RDONLY;
 270         ndp->ni_segflg = UIO_SYSSPACE;
 271         crossmnt = 1;
 272
 273         if (nd->nd_flag & ND_PUBLOOKUP) {
 274                 ndp->ni_loopcnt = 0;
 275                 if (cnp->cn_pnbuf[0] == '/') {
 276                         vrele(dp);
 277                         /*
 278                          * Check for degenerate pathnames here, since lookup()
 279                          * panics on them.
 280                          */
 281                         for (i = 1; i < ndp->ni_pathlen; i++)
 282                                 if (cnp->cn_pnbuf[i] != '/')
 283                                         break;
 284                         if (i == ndp->ni_pathlen) {
 285                                 error = NFSERR_ACCES;
 286                                 goto out;
 287                         }
 288                         dp = rootvnode;
 289                         VREF(dp);
 290                 }
 291         } else if ((enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
 292             (nd->nd_flag & ND_NFSV4) == 0) {
 293                 /*
 294                  * Only cross mount points for NFSv4 when doing a
 295                  * mount while traversing the file system above
 296                  * the mount point, unless enable_crossmntpt is set.
 297                  */
 298                 cnp->cn_flags |= NOCROSSMOUNT;
 299                 crossmnt = 0;
 300         }
 301
 302         /*
 303          * Initialize for scan, set ni_startdir and bump ref on dp again
 304          * becuase lookup() will dereference ni_startdir.
 305          */
 306
 307         cnp->cn_thread = p;
 308         ndp->ni_startdir = dp;
 309         ndp->ni_rootdir = rootvnode;
 310
 311         if (!lockleaf)
 312                 cnp->cn_flags |= LOCKLEAF;
 313         for (;;) {
 314                 cnp->cn_nameptr = cnp->cn_pnbuf;
 315                 /*
 316                  * Call lookup() to do the real work.  If an error occurs,
 317                  * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
 318                  * we do not have to dereference anything before returning.
 319                  * In either case ni_startdir will be dereferenced and NULLed
 320                  * out.
 321                  */
 322                 if (exp->nes_vfslocked)
 323                         ndp->ni_cnd.cn_flags |= GIANTHELD;
 324                 error = lookup(ndp);
 325                 /*
 326                  * The Giant lock should only change when
 327                  * crossing mount points.
 328                  */
 329                 if (crossmnt) {
 330                         exp->nes_vfslocked =
 331                             (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
 332                         ndp->ni_cnd.cn_flags &= ~GIANTHELD;
 333                 }
 334                 if (error)
 335                         break;
 336
 337                 /*
 338                  * Check for encountering a symbolic link.  Trivial
 339                  * termination occurs if no symlink encountered.
 340                  */
 341                 if ((cnp->cn_flags & ISSYMLINK) == 0) {
 342                         if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
 343                                 nfsvno_relpathbuf(ndp);
 344                         if (ndp->ni_vp && !lockleaf)
 345                                 NFSVOPUNLOCK(ndp->ni_vp, 0, p);
 346                         break;
 347                 }
 348
 349                 /*
 350                  * Validate symlink
 351                  */
 352                 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
 353                         NFSVOPUNLOCK(ndp->ni_dvp, 0, p);
 354                 if (!(nd->nd_flag & ND_PUBLOOKUP)) {
 355                         error = EINVAL;
 356                         goto badlink2;
 357                 }
 358
 359                 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
 360                         error = ELOOP;
 361                         goto badlink2;
 362                 }
 363                 if (ndp->ni_pathlen > 1)
 364                         cp = uma_zalloc(namei_zone, M_WAITOK);
 365                 else
 366                         cp = cnp->cn_pnbuf;
 367                 aiov.iov_base = cp;
 368                 aiov.iov_len = MAXPATHLEN;
 369                 auio.uio_iov = &aiov;
 370                 auio.uio_iovcnt = 1;
 371                 auio.uio_offset = 0;
 372                 auio.uio_rw = UIO_READ;
 373                 auio.uio_segflg = UIO_SYSSPACE;
 374                 auio.uio_td = NULL;
 375                 auio.uio_resid = MAXPATHLEN;
 376                 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
 377                 if (error) {
 378                 badlink1:
 379                         if (ndp->ni_pathlen > 1)
 380                                 uma_zfree(namei_zone, cp);
 381                 badlink2:
 382                         vrele(ndp->ni_dvp);
 383                         vput(ndp->ni_vp);
 384                         break;
 385                 }
 386                 linklen = MAXPATHLEN - auio.uio_resid;
 387                 if (linklen == 0) {
 388                         error = ENOENT;
 389                         goto badlink1;
 390                 }
 391                 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
 392                         error = ENAMETOOLONG;
 393                         goto badlink1;
 394                 }
 395
 396                 /*
 397                  * Adjust or replace path
 398                  */
 399                 if (ndp->ni_pathlen > 1) {
 400                         NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
 401                         uma_zfree(namei_zone, cnp->cn_pnbuf);
 402                         cnp->cn_pnbuf = cp;
 403                 } else
 404                         cnp->cn_pnbuf[linklen] = '\0';
 405                 ndp->ni_pathlen += linklen;
 406
 407                 /*
 408                  * Cleanup refs for next loop and check if root directory
 409                  * should replace current directory.  Normally ni_dvp
 410                  * becomes the new base directory and is cleaned up when
 411                  * we loop.  Explicitly null pointers after invalidation
 412                  * to clarify operation.
 413                  */
 414                 vput(ndp->ni_vp);
 415                 ndp->ni_vp = NULL;
 416
 417                 if (cnp->cn_pnbuf[0] == '/') {
 418                         vrele(ndp->ni_dvp);
 419                         ndp->ni_dvp = ndp->ni_rootdir;
 420                         VREF(ndp->ni_dvp);
 421                 }
 422                 ndp->ni_startdir = ndp->ni_dvp;
 423                 ndp->ni_dvp = NULL;
 424         }
 425         if (!lockleaf)
 426                 cnp->cn_flags &= ~LOCKLEAF;
 427
 428 out:
 429         if (error) {
 430                 uma_zfree(namei_zone, cnp->cn_pnbuf);
 431                 ndp->ni_vp = NULL;
 432                 ndp->ni_dvp = NULL;
 433                 ndp->ni_startdir = NULL;
 434                 cnp->cn_flags &= ~HASBUF;
 435         } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
 436                 ndp->ni_dvp = NULL;
 437         }
 438         return (error);
 439 }
 440
 441 /*
 442  * Set up a pathname buffer and return a pointer to it and, optionally
 443  * set a hash pointer.
 444  */
 445 void
 446 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
 447 {
 448         struct componentname *cnp = &ndp->ni_cnd;
 449
 450         cnp->cn_flags |= (NOMACCHECK | HASBUF);
 451         cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
 452         if (hashpp != NULL)
 453                 *hashpp = NULL;
 454         *bufpp = cnp->cn_pnbuf;
 455 }
 456
 457 /*
 458  * Release the above path buffer, if not released by nfsvno_namei().
 459  */
 460 void
 461 nfsvno_relpathbuf(struct nameidata *ndp)
 462 {
 463
 464         if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
 465                 panic("nfsrelpath");
 466         uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
 467         ndp->ni_cnd.cn_flags &= ~HASBUF;
 468 }
 469
 470 /*
 471  * Readlink vnode op into an mbuf list.
 472  */
 473 int
 474 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
 475     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
 476 {
 477         struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
 478         struct iovec *ivp = iv;
 479         struct uio io, *uiop = &io;
 480         struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
 481         int i, len, tlen, error;
 482
 483         len = 0;
 484         i = 0;
 485         while (len < NFS_MAXPATHLEN) {
 486                 NFSMGET(mp);
 487                 MCLGET(mp, M_WAIT);
 488                 mp->m_len = NFSMSIZ(mp);
 489                 if (len == 0) {
 490                         mp3 = mp2 = mp;
 491                 } else {
 492                         mp2->m_next = mp;
 493                         mp2 = mp;
 494                 }
 495                 if ((len + mp->m_len) > NFS_MAXPATHLEN) {
 496                         mp->m_len = NFS_MAXPATHLEN - len;
 497                         len = NFS_MAXPATHLEN;
 498                 } else {
 499                         len += mp->m_len;
 500                 }
 501                 ivp->iov_base = mtod(mp, caddr_t);
 502                 ivp->iov_len = mp->m_len;
 503                 i++;
 504                 ivp++;
 505         }
 506         uiop->uio_iov = iv;
 507         uiop->uio_iovcnt = i;
 508         uiop->uio_offset = 0;
 509         uiop->uio_resid = len;
 510         uiop->uio_rw = UIO_READ;
 511         uiop->uio_segflg = UIO_SYSSPACE;
 512         uiop->uio_td = NULL;
 513         error = VOP_READLINK(vp, uiop, cred);
 514         if (error) {
 515                 m_freem(mp3);
 516                 *lenp = 0;
 517                 return (error);
 518         }
 519         if (uiop->uio_resid > 0) {
 520                 len -= uiop->uio_resid;
 521                 tlen = NFSM_RNDUP(len);
 522                 nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
 523         }
 524         *lenp = len;
 525         *mpp = mp3;
 526         *mpendp = mp;
 527         return (0);
 528 }
 529
 530 /*
 531  * Read vnode op call into mbuf list.
 532  */
 533 int
 534 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
 535     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
 536 {
 537         struct mbuf *m;
 538         int i;
 539         struct iovec *iv;
 540         struct iovec *iv2;
 541         int error = 0, len, left, siz, tlen, ioflag = 0, hi, try = 32;
 542         struct mbuf *m2 = NULL, *m3;
 543         struct uio io, *uiop = &io;
 544         struct nfsheur *nh;
 545
 546         /*
 547          * Calculate seqcount for heuristic
 548          */
 549         /*
 550          * Locate best candidate
 551          */
 552
 553         hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
 554         nh = &nfsheur[hi];
 555
 556         while (try--) {
 557                 if (nfsheur[hi].nh_vp == vp) {
 558                         nh = &nfsheur[hi];
 559                         break;
 560                 }
 561                 if (nfsheur[hi].nh_use > 0)
 562                         --nfsheur[hi].nh_use;
 563                 hi = (hi + 1) % NUM_HEURISTIC;
 564                 if (nfsheur[hi].nh_use < nh->nh_use)
 565                         nh = &nfsheur[hi];
 566         }
 567
 568         if (nh->nh_vp != vp) {
 569                 nh->nh_vp = vp;
 570                 nh->nh_nextr = off;
 571                 nh->nh_use = NHUSE_INIT;
 572                 if (off == 0)
 573                         nh->nh_seqcount = 4;
 574                 else
 575                         nh->nh_seqcount = 1;
 576         }
 577
 578         /*
 579          * Calculate heuristic
 580          */
 581
 582         if ((off == 0 && nh->nh_seqcount > 0) || off == nh->nh_nextr) {
 583                 if (++nh->nh_seqcount > IO_SEQMAX)
 584                         nh->nh_seqcount = IO_SEQMAX;
 585         } else if (nh->nh_seqcount > 1) {
 586                 nh->nh_seqcount = 1;
 587         } else {
 588                 nh->nh_seqcount = 0;
 589         }
 590         nh->nh_use += NHUSE_INC;
 591         if (nh->nh_use > NHUSE_MAX)
 592                 nh->nh_use = NHUSE_MAX;
 593         ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
 594
 595         len = left = NFSM_RNDUP(cnt);
 596         m3 = NULL;
 597         /*
 598          * Generate the mbuf list with the uio_iov ref. to it.
 599          */
 600         i = 0;
 601         while (left > 0) {
 602                 NFSMGET(m);
 603                 MCLGET(m, M_WAIT);
 604                 m->m_len = 0;
 605                 siz = min(M_TRAILINGSPACE(m), left);
 606                 left -= siz;
 607                 i++;
 608                 if (m3)
 609                         m2->m_next = m;
 610                 else
 611                         m3 = m;
 612                 m2 = m;
 613         }
 614         MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
 615             M_TEMP, M_WAITOK);
 616         uiop->uio_iov = iv2 = iv;
 617         m = m3;
 618         left = len;
 619         i = 0;
 620         while (left > 0) {
 621                 if (m == NULL)
 622                         panic("nfsvno_read iov");
 623                 siz = min(M_TRAILINGSPACE(m), left);
 624                 if (siz > 0) {
 625                         iv->iov_base = mtod(m, caddr_t) + m->m_len;
 626                         iv->iov_len = siz;
 627                         m->m_len += siz;
 628                         left -= siz;
 629                         iv++;
 630                         i++;
 631                 }
 632                 m = m->m_next;
 633         }
 634         uiop->uio_iovcnt = i;
 635         uiop->uio_offset = off;
 636         uiop->uio_resid = len;
 637         uiop->uio_rw = UIO_READ;
 638         uiop->uio_segflg = UIO_SYSSPACE;
 639         error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
 640         FREE((caddr_t)iv2, M_TEMP);
 641         if (error) {
 642                 m_freem(m3);
 643                 *mpp = NULL;
 644                 return (error);
 645         }
 646         tlen = len - uiop->uio_resid;
 647         cnt = cnt < tlen ? cnt : tlen;
 648         tlen = NFSM_RNDUP(cnt);
 649         if (tlen == 0) {
 650                 m_freem(m3);
 651                 m3 = NULL;
 652         } else if (len != tlen || tlen != cnt)
 653                 nfsrv_adj(m3, len - tlen, tlen - cnt);
 654         *mpp = m3;
 655         *mpendp = m2;
 656         return (0);
 657 }
 658
 659 /*
 660  * Write vnode op from an mbuf list.
 661  */
 662 int
 663 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
 664     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
 665 {
 666         struct iovec *ivp;
 667         int i, len;
 668         struct iovec *iv;
 669         int ioflags, error;
 670         struct uio io, *uiop = &io;
 671
 672         MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
 673             M_WAITOK);
 674         uiop->uio_iov = iv = ivp;
 675         uiop->uio_iovcnt = cnt;
 676         i = mtod(mp, caddr_t) + mp->m_len - cp;
 677         len = retlen;
 678         while (len > 0) {
 679                 if (mp == NULL)
 680                         panic("nfsvno_write");
 681                 if (i > 0) {
 682                         i = min(i, len);
 683                         ivp->iov_base = cp;
 684                         ivp->iov_len = i;
 685                         ivp++;
 686                         len -= i;
 687                 }
 688                 mp = mp->m_next;
 689                 if (mp) {
 690                         i = mp->m_len;
 691                         cp = mtod(mp, caddr_t);
 692                 }
 693         }
 694
 695         if (stable == NFSWRITE_UNSTABLE)
 696                 ioflags = IO_NODELOCKED;
 697         else
 698                 ioflags = (IO_SYNC | IO_NODELOCKED);
 699         uiop->uio_resid = retlen;
 700         uiop->uio_rw = UIO_WRITE;
 701         uiop->uio_segflg = UIO_SYSSPACE;
 702         NFSUIOPROC(uiop, p);
 703         uiop->uio_offset = off;
 704         error = VOP_WRITE(vp, uiop, ioflags, cred);
 705         FREE((caddr_t)iv, M_TEMP);
 706         return (error);
 707 }
 708
 709 /*
 710  * Common code for creating a regular file (plus special files for V2).
 711  */
 712 int
 713 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
 714     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
 715     int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
 716 {
 717         u_quad_t tempsize;
 718         int error;
 719
 720         error = nd->nd_repstat;
 721         if (!error && ndp->ni_vp == NULL) {
 722                 if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
 723                         vrele(ndp->ni_startdir);
 724                         error = VOP_CREATE(ndp->ni_dvp,
 725                             &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
 726                         vput(ndp->ni_dvp);
 727                         nfsvno_relpathbuf(ndp);
 728                         if (!error) {
 729                                 if (*exclusive_flagp) {
 730                                         *exclusive_flagp = 0;
 731                                         NFSVNO_ATTRINIT(nvap);
 732                                         nvap->na_atime.tv_sec = cverf[0];
 733                                         nvap->na_atime.tv_nsec = cverf[1];
 734                                         error = VOP_SETATTR(ndp->ni_vp,
 735                                             &nvap->na_vattr, nd->nd_cred);
 736                                 }
 737                         }
 738                 /*
 739                  * NFS V2 Only. nfsrvd_mknod() does this for V3.
 740                  * (This implies, just get out on an error.)
 741                  */
 742                 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
 743                         nvap->na_type == VFIFO) {
 744                         if (nvap->na_type == VCHR && rdev == 0xffffffff)
 745                                 nvap->na_type = VFIFO;
 746                         if (nvap->na_type != VFIFO &&
 747                             (error = priv_check_cred(nd->nd_cred,
 748                              PRIV_VFS_MKNOD_DEV, 0))) {
 749                                 vrele(ndp->ni_startdir);
 750                                 nfsvno_relpathbuf(ndp);
 751                                 vput(ndp->ni_dvp);
 752                                 return (error);
 753                         }
 754                         nvap->na_rdev = rdev;
 755                         error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
 756                             &ndp->ni_cnd, &nvap->na_vattr);
 757                         vput(ndp->ni_dvp);
 758                         nfsvno_relpathbuf(ndp);
 759                         if (error) {
 760                                 vrele(ndp->ni_startdir);
 761                                 return (error);
 762                         }
 763                 } else {
 764                         vrele(ndp->ni_startdir);
 765                         nfsvno_relpathbuf(ndp);
 766                         vput(ndp->ni_dvp);
 767                         return (ENXIO);
 768                 }
 769                 *vpp = ndp->ni_vp;
 770         } else {
 771                 /*
 772                  * Handle cases where error is already set and/or
 773                  * the file exists.
 774                  * 1 - clean up the lookup
 775                  * 2 - iff !error and na_size set, truncate it
 776                  */
 777                 vrele(ndp->ni_startdir);
 778                 nfsvno_relpathbuf(ndp);
 779                 *vpp = ndp->ni_vp;
 780                 if (ndp->ni_dvp == *vpp)
 781                         vrele(ndp->ni_dvp);
 782                 else
 783                         vput(ndp->ni_dvp);
 784                 if (!error && nvap->na_size != VNOVAL) {
 785                         error = nfsvno_accchk(*vpp, VWRITE,
 786                             nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
 787                             NFSACCCHK_VPISLOCKED, NULL);
 788                         if (!error) {
 789                                 tempsize = nvap->na_size;
 790                                 NFSVNO_ATTRINIT(nvap);
 791                                 nvap->na_size = tempsize;
 792                                 error = VOP_SETATTR(*vpp,
 793                                     &nvap->na_vattr, nd->nd_cred);
 794                         }
 795                 }
 796                 if (error)
 797                         vput(*vpp);
 798         }
 799         return (error);
 800 }
 801
 802 /*
 803  * Do a mknod vnode op.
 804  */
 805 int
 806 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
 807     struct thread *p)
 808 {
 809         int error = 0;
 810         enum vtype vtyp;
 811
 812         vtyp = nvap->na_type;
 813         /*
 814          * Iff doesn't exist, create it.
 815          */
 816         if (ndp->ni_vp) {
 817                 vrele(ndp->ni_startdir);
 818                 nfsvno_relpathbuf(ndp);
 819                 vput(ndp->ni_dvp);
 820                 vrele(ndp->ni_vp);
 821                 return (EEXIST);
 822         }
 823         if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
 824                 vrele(ndp->ni_startdir);
 825                 nfsvno_relpathbuf(ndp);
 826                 vput(ndp->ni_dvp);
 827                 return (NFSERR_BADTYPE);
 828         }
 829         if (vtyp == VSOCK) {
 830                 vrele(ndp->ni_startdir);
 831                 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
 832                     &ndp->ni_cnd, &nvap->na_vattr);
 833                 vput(ndp->ni_dvp);
 834                 nfsvno_relpathbuf(ndp);
 835         } else {
 836                 if (nvap->na_type != VFIFO &&
 837                     (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
 838                         vrele(ndp->ni_startdir);
 839                         nfsvno_relpathbuf(ndp);
 840                         vput(ndp->ni_dvp);
 841                         return (error);
 842                 }
 843                 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
 844                     &ndp->ni_cnd, &nvap->na_vattr);
 845                 vput(ndp->ni_dvp);
 846                 nfsvno_relpathbuf(ndp);
 847                 if (error)
 848                         vrele(ndp->ni_startdir);
 849                 /*
 850                  * Since VOP_MKNOD returns the ni_vp, I can't
 851                  * see any reason to do the lookup.
 852                  */
 853         }
 854         return (error);
 855 }
 856
 857 /*
 858  * Mkdir vnode op.
 859  */
 860 int
 861 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
 862     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
 863 {
 864         int error = 0;
 865
 866         if (ndp->ni_vp != NULL) {
 867                 if (ndp->ni_dvp == ndp->ni_vp)
 868                         vrele(ndp->ni_dvp);
 869                 else
 870                         vput(ndp->ni_dvp);
 871                 vrele(ndp->ni_vp);
 872                 nfsvno_relpathbuf(ndp);
 873                 return (EEXIST);
 874         }
 875         error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
 876             &nvap->na_vattr);
 877         vput(ndp->ni_dvp);
 878         nfsvno_relpathbuf(ndp);
 879         return (error);
 880 }
 881
 882 /*
 883  * symlink vnode op.
 884  */
 885 int
 886 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
 887     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
 888     struct nfsexstuff *exp)
 889 {
 890         int error = 0;
 891
 892         if (ndp->ni_vp) {
 893                 vrele(ndp->ni_startdir);
 894                 nfsvno_relpathbuf(ndp);
 895                 if (ndp->ni_dvp == ndp->ni_vp)
 896                         vrele(ndp->ni_dvp);
 897                 else
 898                         vput(ndp->ni_dvp);
 899                 vrele(ndp->ni_vp);
 900                 return (EEXIST);
 901         }
 902
 903         error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
 904             &nvap->na_vattr, pathcp);
 905         vput(ndp->ni_dvp);
 906         vrele(ndp->ni_startdir);
 907         nfsvno_relpathbuf(ndp);
 908         /*
 909          * Although FreeBSD still had the lookup code in
 910          * it for 7/current, there doesn't seem to be any
 911          * point, since VOP_SYMLINK() returns the ni_vp.
 912          * Just vput it for v2.
 913          */
 914         if (!not_v2 && !error)
 915                 vput(ndp->ni_vp);
 916         return (error);
 917 }
 918
 919 /*
 920  * Parse symbolic link arguments.
 921  * This function has an ugly side effect. It will MALLOC() an area for
 922  * the symlink and set iov_base to point to it, only if it succeeds.
 923  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
 924  * be FREE'd later.
 925  */
 926 int
 927 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
 928     struct thread *p, char **pathcpp, int *lenp)
 929 {
 930         u_int32_t *tl;
 931         char *pathcp = NULL;
 932         int error = 0, len;
 933         struct nfsv2_sattr *sp;
 934
 935         *pathcpp = NULL;
 936         *lenp = 0;
 937         if ((nd->nd_flag & ND_NFSV3) &&
 938             (error = nfsrv_sattr(nd, nvap, NULL, NULL, p)))
 939                 goto nfsmout;
 940         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 941         len = fxdr_unsigned(int, *tl);
 942         if (len > NFS_MAXPATHLEN || len <= 0) {
 943                 error = EBADRPC;
 944                 goto nfsmout;
 945         }
 946         MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK);
 947         error = nfsrv_mtostr(nd, pathcp, len);
 948         if (error)
 949                 goto nfsmout;
 950         if (nd->nd_flag & ND_NFSV2) {
 951                 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
 952                 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
 953         }
 954         *pathcpp = pathcp;
 955         *lenp = len;
 956         return (0);
 957 nfsmout:
 958         if (pathcp)
 959                 free(pathcp, M_TEMP);
 960         return (error);
 961 }
 962
 963 /*
 964  * Remove a non-directory object.
 965  */
 966 int
 967 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
 968     struct thread *p, struct nfsexstuff *exp)
 969 {
 970         struct vnode *vp;
 971         int error = 0;
 972
 973         vp = ndp->ni_vp;
 974         if (vp->v_type == VDIR)
 975                 error = NFSERR_ISDIR;
 976         else if (is_v4)
 977                 error = nfsrv_checkremove(vp, 1, p);
 978         if (!error)
 979                 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
 980         if (ndp->ni_dvp == vp)
 981                 vrele(ndp->ni_dvp);
 982         else
 983                 vput(ndp->ni_dvp);
 984         vput(vp);
 985         return (error);
 986 }
 987
 988 /*
 989  * Remove a directory.
 990  */
 991 int
 992 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
 993     struct thread *p, struct nfsexstuff *exp)
 994 {
 995         struct vnode *vp;
 996         int error = 0;
 997
 998         vp = ndp->ni_vp;
 999         if (vp->v_type != VDIR) {
1000                 error = ENOTDIR;
1001                 goto out;
1002         }
1003         /*
1004          * No rmdir "." please.
1005          */
1006         if (ndp->ni_dvp == vp) {
1007                 error = EINVAL;
1008                 goto out;
1009         }
1010         /*
1011          * The root of a mounted filesystem cannot be deleted.
1012          */
1013         if (vp->v_vflag & VV_ROOT)
1014                 error = EBUSY;
1015 out:
1016         if (!error)
1017                 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1018         if (ndp->ni_dvp == vp)
1019                 vrele(ndp->ni_dvp);
1020         else
1021                 vput(ndp->ni_dvp);
1022         vput(vp);
1023         return (error);
1024 }
1025
1026 /*
1027  * Rename vnode op.
1028  */
1029 int
1030 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1031     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1032 {
1033         struct vnode *fvp, *tvp, *tdvp;
1034         int error = 0;
1035
1036         fvp = fromndp->ni_vp;
1037         if (ndstat) {
1038                 vrele(fromndp->ni_dvp);
1039                 vrele(fvp);
1040                 error = ndstat;
1041                 goto out1;
1042         }
1043         tdvp = tondp->ni_dvp;
1044         tvp = tondp->ni_vp;
1045         if (tvp != NULL) {
1046                 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1047                         error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1048                         goto out;
1049                 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1050                         error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1051                         goto out;
1052                 }
1053                 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1054                         error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1055                         goto out;
1056                 }
1057
1058                 /*
1059                  * A rename to '.' or '..' results in a prematurely
1060                  * unlocked vnode on FreeBSD5, so I'm just going to fail that
1061                  * here.
1062                  */
1063                 if ((tondp->ni_cnd.cn_namelen == 1 &&
1064                      tondp->ni_cnd.cn_nameptr[0] == '.') ||
1065                     (tondp->ni_cnd.cn_namelen == 2 &&
1066                      tondp->ni_cnd.cn_nameptr[0] == '.' &&
1067                      tondp->ni_cnd.cn_nameptr[1] == '.')) {
1068                         error = EINVAL;
1069                         goto out;
1070                 }
1071         }
1072         if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1073                 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1074                 goto out;
1075         }
1076         if (fvp->v_mount != tdvp->v_mount) {
1077                 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1078                 goto out;
1079         }
1080         if (fvp == tdvp) {
1081                 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1082                 goto out;
1083         }
1084         if (fvp == tvp) {
1085                 /*
1086                  * If source and destination are the same, there is nothing to
1087                  * do. Set error to -1 to indicate this.
1088                  */
1089                 error = -1;
1090                 goto out;
1091         }
1092         if (ndflag & ND_NFSV4) {
1093                 NFSVOPLOCK(fvp, LK_EXCLUSIVE | LK_RETRY, p);
1094                 error = nfsrv_checkremove(fvp, 0, p);
1095                 NFSVOPUNLOCK(fvp, 0, p);
1096                 if (tvp && !error)
1097                         error = nfsrv_checkremove(tvp, 1, p);
1098         } else {
1099                 /*
1100                  * For NFSv2 and NFSv3, try to get rid of the delegation, so
1101                  * that the NFSv4 client won't be confused by the rename.
1102                  * Since nfsd_recalldelegation() can only be called on an
1103                  * unlocked vnode at this point and fvp is the file that will
1104                  * still exist after the rename, just do fvp.
1105                  */
1106                 nfsd_recalldelegation(fvp, p);
1107         }
1108 out:
1109         if (!error) {
1110                 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1111                     &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1112                     &tondp->ni_cnd);
1113         } else {
1114                 if (tdvp == tvp)
1115                         vrele(tdvp);
1116                 else
1117                         vput(tdvp);
1118                 if (tvp)
1119                         vput(tvp);
1120                 vrele(fromndp->ni_dvp);
1121                 vrele(fvp);
1122                 if (error == -1)
1123                         error = 0;
1124         }
1125         vrele(tondp->ni_startdir);
1126         nfsvno_relpathbuf(tondp);
1127 out1:
1128         vrele(fromndp->ni_startdir);
1129         nfsvno_relpathbuf(fromndp);
1130         return (error);
1131 }
1132
1133 /*
1134  * Link vnode op.
1135  */
1136 int
1137 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1138     struct thread *p, struct nfsexstuff *exp)
1139 {
1140         struct vnode *xp;
1141         int error = 0;
1142
1143         xp = ndp->ni_vp;
1144         if (xp != NULL) {
1145                 error = EEXIST;
1146         } else {
1147                 xp = ndp->ni_dvp;
1148                 if (vp->v_mount != xp->v_mount)
1149                         error = EXDEV;
1150         }
1151         if (!error) {
1152                 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
1153                 error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1154                 if (ndp->ni_dvp == vp)
1155                         vrele(ndp->ni_dvp);
1156                 else
1157                         vput(ndp->ni_dvp);
1158                 NFSVOPUNLOCK(vp, 0, p);
1159         } else {
1160                 if (ndp->ni_dvp == ndp->ni_vp)
1161                         vrele(ndp->ni_dvp);
1162                 else
1163                         vput(ndp->ni_dvp);
1164                 if (ndp->ni_vp)
1165                         vrele(ndp->ni_vp);
1166         }
1167         nfsvno_relpathbuf(ndp);
1168         return (error);
1169 }
1170
1171 /*
1172  * Do the fsync() appropriate for the commit.
1173  */
1174 int
1175 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1176     struct thread *td)
1177 {
1178         int error = 0;
1179
1180         if (cnt > MAX_COMMIT_COUNT) {
1181                 /*
1182                  * Give up and do the whole thing
1183                  */
1184                 if (vp->v_object &&
1185                    (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1186                         VM_OBJECT_LOCK(vp->v_object);
1187                         vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1188                         VM_OBJECT_UNLOCK(vp->v_object);
1189                 }
1190                 error = VOP_FSYNC(vp, MNT_WAIT, td);
1191         } else {
1192                 /*
1193                  * Locate and synchronously write any buffers that fall
1194                  * into the requested range.  Note:  we are assuming that
1195                  * f_iosize is a power of 2.
1196                  */
1197                 int iosize = vp->v_mount->mnt_stat.f_iosize;
1198                 int iomask = iosize - 1;
1199                 struct bufobj *bo;
1200                 daddr_t lblkno;
1201
1202                 /*
1203                  * Align to iosize boundry, super-align to page boundry.
1204                  */
1205                 if (off & iomask) {
1206                         cnt += off & iomask;
1207                         off &= ~(u_quad_t)iomask;
1208                 }
1209                 if (off & PAGE_MASK) {
1210                         cnt += off & PAGE_MASK;
1211                         off &= ~(u_quad_t)PAGE_MASK;
1212                 }
1213                 lblkno = off / iosize;
1214
1215                 if (vp->v_object &&
1216                    (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1217                         VM_OBJECT_LOCK(vp->v_object);
1218                         vm_object_page_clean(vp->v_object, off / PAGE_SIZE, (cnt + PAGE_MASK) / PAGE_SIZE, OBJPC_SYNC);
1219                         VM_OBJECT_UNLOCK(vp->v_object);
1220                 }
1221
1222                 bo = &vp->v_bufobj;
1223                 BO_LOCK(bo);
1224                 while (cnt > 0) {
1225                         struct buf *bp;
1226
1227                         /*
1228                          * If we have a buffer and it is marked B_DELWRI we
1229                          * have to lock and write it.  Otherwise the prior
1230                          * write is assumed to have already been committed.
1231                          *
1232                          * gbincore() can return invalid buffers now so we
1233                          * have to check that bit as well (though B_DELWRI
1234                          * should not be set if B_INVAL is set there could be
1235                          * a race here since we haven't locked the buffer).
1236                          */
1237                         if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1238                                 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1239                                     LK_INTERLOCK, BO_MTX(bo)) == ENOLCK) {
1240                                         BO_LOCK(bo);
1241                                         continue; /* retry */
1242                                 }
1243                                 if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1244                                     B_DELWRI) {
1245                                         bremfree(bp);
1246                                         bp->b_flags &= ~B_ASYNC;
1247                                         bwrite(bp);
1248                                         ++nfs_commit_miss;
1249                                 } else
1250                                         BUF_UNLOCK(bp);
1251                                 BO_LOCK(bo);
1252                         }
1253                         ++nfs_commit_blks;
1254                         if (cnt < iosize)
1255                                 break;
1256                         cnt -= iosize;
1257                         ++lblkno;
1258                 }
1259                 BO_UNLOCK(bo);
1260         }
1261         return (error);
1262 }
1263
1264 /*
1265  * Statfs vnode op.
1266  */
1267 int
1268 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1269 {
1270
1271         return (VFS_STATFS(vp->v_mount, sf));
1272 }
1273
1274 /*
1275  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1276  * must handle nfsrv_opencheck() calls after any other access checks.
1277  */
1278 void
1279 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1280     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1281     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1282     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1283     struct nfsexstuff *exp, struct vnode **vpp)
1284 {
1285         struct vnode *vp = NULL;
1286         u_quad_t tempsize;
1287         struct nfsexstuff nes;
1288
1289         if (ndp->ni_vp == NULL)
1290                 nd->nd_repstat = nfsrv_opencheck(clientid,
1291                     stateidp, stp, NULL, nd, p, nd->nd_repstat);
1292         if (!nd->nd_repstat) {
1293                 if (ndp->ni_vp == NULL) {
1294                         vrele(ndp->ni_startdir);
1295                         nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1296                             &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1297                         vput(ndp->ni_dvp);
1298                         nfsvno_relpathbuf(ndp);
1299                         if (!nd->nd_repstat) {
1300                                 if (*exclusive_flagp) {
1301                                         *exclusive_flagp = 0;
1302                                         NFSVNO_ATTRINIT(nvap);
1303                                         nvap->na_atime.tv_sec = cverf[0];
1304                                         nvap->na_atime.tv_nsec = cverf[1];
1305                                         nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1306                                             &nvap->na_vattr, cred);
1307                                 } else {
1308                                         nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1309                                             aclp, p, attrbitp, exp);
1310                                 }
1311                         }
1312                         vp = ndp->ni_vp;
1313                 } else {
1314                         if (ndp->ni_startdir)
1315                                 vrele(ndp->ni_startdir);
1316                         nfsvno_relpathbuf(ndp);
1317                         vp = ndp->ni_vp;
1318                         if (create == NFSV4OPEN_CREATE) {
1319                                 if (ndp->ni_dvp == vp)
1320                                         vrele(ndp->ni_dvp);
1321                                 else
1322                                         vput(ndp->ni_dvp);
1323                         }
1324                         if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1325                                 if (ndp->ni_cnd.cn_flags & RDONLY)
1326                                         NFSVNO_SETEXRDONLY(&nes);
1327                                 else
1328                                         NFSVNO_EXINIT(&nes);
1329                                 nd->nd_repstat = nfsvno_accchk(vp,
1330                                     VWRITE, cred, &nes, p,
1331                                     NFSACCCHK_NOOVERRIDE,
1332                                     NFSACCCHK_VPISLOCKED, NULL);
1333                                 nd->nd_repstat = nfsrv_opencheck(clientid,
1334                                     stateidp, stp, vp, nd, p, nd->nd_repstat);
1335                                 if (!nd->nd_repstat) {
1336                                         tempsize = nvap->na_size;
1337                                         NFSVNO_ATTRINIT(nvap);
1338                                         nvap->na_size = tempsize;
1339                                         nd->nd_repstat = VOP_SETATTR(vp,
1340                                             &nvap->na_vattr, cred);
1341                                 }
1342                         } else if (vp->v_type == VREG) {
1343                                 nd->nd_repstat = nfsrv_opencheck(clientid,
1344                                     stateidp, stp, vp, nd, p, nd->nd_repstat);
1345                         }
1346                 }
1347         } else {
1348                 if (ndp->ni_cnd.cn_flags & HASBUF)
1349                         nfsvno_relpathbuf(ndp);
1350                 if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1351                         vrele(ndp->ni_startdir);
1352                         if (ndp->ni_dvp == ndp->ni_vp)
1353                                 vrele(ndp->ni_dvp);
1354                         else
1355                                 vput(ndp->ni_dvp);
1356                         if (ndp->ni_vp)
1357                                 vput(ndp->ni_vp);
1358                 }
1359         }
1360         *vpp = vp;
1361 }
1362
1363 /*
1364  * Updates the file rev and sets the mtime and ctime
1365  * to the current clock time, returning the va_filerev and va_Xtime
1366  * values.
1367  */
1368 void
1369 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1370     struct ucred *cred, struct thread *p)
1371 {
1372         struct vattr va;
1373
1374         VATTR_NULL(&va);
1375         getnanotime(&va.va_mtime);
1376         (void) VOP_SETATTR(vp, &va, cred);
1377         (void) nfsvno_getattr(vp, nvap, cred, p);
1378 }
1379
1380 /*
1381  * Glue routine to nfsv4_fillattr().
1382  */
1383 int
1384 nfsvno_fillattr(struct nfsrv_descript *nd, struct vnode *vp,
1385     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1386     struct ucred *cred, struct thread *p, int isdgram, int reterr)
1387 {
1388         int error;
1389
1390         error = nfsv4_fillattr(nd, vp, NULL, &nvap->na_vattr, fhp, rderror,
1391             attrbitp, cred, p, isdgram, reterr);
1392         return (error);
1393 }
1394
1395 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1396 /*
1397  * nfs readdir service
1398  * - mallocs what it thinks is enough to read
1399  *      count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1400  * - calls VOP_READDIR()
1401  * - loops around building the reply
1402  *      if the output generated exceeds count break out of loop
1403  *      The NFSM_CLGET macro is used here so that the reply will be packed
1404  *      tightly in mbuf clusters.
1405  * - it trims out records with d_fileno == 0
1406  *      this doesn't matter for Unix clients, but they might confuse clients
1407  *      for other os'.
1408  * - it trims out records with d_type == DT_WHT
1409  *      these cannot be seen through NFS (unless we extend the protocol)
1410  *     The alternate call nfsrvd_readdirplus() does lookups as well.
1411  * PS: The NFS protocol spec. does not clarify what the "count" byte
1412  *      argument is a count of.. just name strings and file id's or the
1413  *      entire reply rpc or ...
1414  *      I tried just file name and id sizes and it confused the Sun client,
1415  *      so I am using the full rpc size now. The "paranoia.." comment refers
1416  *      to including the status longwords that are not a part of the dir.
1417  *      "entry" structures, but are in the rpc.
1418  */
1419 int
1420 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1421     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1422 {
1423         struct dirent *dp;
1424         u_int32_t *tl;
1425         int dirlen;
1426         char *cpos, *cend, *rbuf;
1427         struct nfsvattr at;
1428         int nlen, error = 0, getret = 1;
1429         int siz, cnt, fullsiz, eofflag, ncookies;
1430         u_int64_t off, toff, verf;
1431         u_long *cookies = NULL, *cookiep;
1432         struct uio io;
1433         struct iovec iv;
1434
1435         if (nd->nd_repstat) {
1436                 nfsrv_postopattr(nd, getret, &at);
1437                 return (0);
1438         }
1439         if (nd->nd_flag & ND_NFSV2) {
1440                 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1441                 off = fxdr_unsigned(u_quad_t, *tl++);
1442         } else {
1443                 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1444                 off = fxdr_hyper(tl);
1445                 tl += 2;
1446                 verf = fxdr_hyper(tl);
1447                 tl += 2;
1448         }
1449         toff = off;
1450         cnt = fxdr_unsigned(int, *tl);
1451         if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1452                 cnt = NFS_SRVMAXDATA(nd);
1453         siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1454         fullsiz = siz;
1455         if (nd->nd_flag & ND_NFSV3) {
1456                 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1457                     p);
1458 #if 0
1459                 /*
1460                  * va_filerev is not sufficient as a cookie verifier,
1461                  * since it is not supposed to change when entries are
1462                  * removed/added unless that offset cookies returned to
1463                  * the client are no longer valid.
1464                  */
1465                 if (!nd->nd_repstat && toff && verf != at.na_filerev)
1466                         nd->nd_repstat = NFSERR_BAD_COOKIE;
1467 #endif
1468         }
1469         if (nd->nd_repstat == 0 && cnt == 0) {
1470                 if (nd->nd_flag & ND_NFSV2)
1471                         /* NFSv2 does not have NFSERR_TOOSMALL */
1472                         nd->nd_repstat = EPERM;
1473                 else
1474                         nd->nd_repstat = NFSERR_TOOSMALL;
1475         }
1476         if (!nd->nd_repstat)
1477                 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1478                     nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1479                     NFSACCCHK_VPISLOCKED, NULL);
1480         if (nd->nd_repstat) {
1481                 vput(vp);
1482                 if (nd->nd_flag & ND_NFSV3)
1483                         nfsrv_postopattr(nd, getret, &at);
1484                 return (0);
1485         }
1486         NFSVOPUNLOCK(vp, 0, p);
1487         MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1488 again:
1489         eofflag = 0;
1490         if (cookies) {
1491                 free((caddr_t)cookies, M_TEMP);
1492                 cookies = NULL;
1493         }
1494
1495         iv.iov_base = rbuf;
1496         iv.iov_len = siz;
1497         io.uio_iov = &iv;
1498         io.uio_iovcnt = 1;
1499         io.uio_offset = (off_t)off;
1500         io.uio_resid = siz;
1501         io.uio_segflg = UIO_SYSSPACE;
1502         io.uio_rw = UIO_READ;
1503         io.uio_td = NULL;
1504         NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
1505         nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1506             &cookies);
1507         NFSVOPUNLOCK(vp, 0, p);
1508         off = (u_int64_t)io.uio_offset;
1509         if (io.uio_resid)
1510                 siz -= io.uio_resid;
1511
1512         if (!cookies && !nd->nd_repstat)
1513                 nd->nd_repstat = NFSERR_PERM;
1514         if (nd->nd_flag & ND_NFSV3) {
1515                 getret = nfsvno_getattr(vp, &at, nd->nd_cred, p);
1516                 if (!nd->nd_repstat)
1517                         nd->nd_repstat = getret;
1518         }
1519
1520         /*
1521          * Handles the failed cases. nd->nd_repstat == 0 past here.
1522          */
1523         if (nd->nd_repstat) {
1524                 vrele(vp);
1525                 free((caddr_t)rbuf, M_TEMP);
1526                 if (cookies)
1527                         free((caddr_t)cookies, M_TEMP);
1528                 if (nd->nd_flag & ND_NFSV3)
1529                         nfsrv_postopattr(nd, getret, &at);
1530                 return (0);
1531         }
1532         /*
1533          * If nothing read, return eof
1534          * rpc reply
1535          */
1536         if (siz == 0) {
1537                 vrele(vp);
1538                 if (nd->nd_flag & ND_NFSV2) {
1539                         NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1540                 } else {
1541                         nfsrv_postopattr(nd, getret, &at);
1542                         NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1543                         txdr_hyper(at.na_filerev, tl);
1544                         tl += 2;
1545                 }
1546                 *tl++ = newnfs_false;
1547                 *tl = newnfs_true;
1548                 FREE((caddr_t)rbuf, M_TEMP);
1549                 FREE((caddr_t)cookies, M_TEMP);
1550                 return (0);
1551         }
1552
1553         /*
1554          * Check for degenerate cases of nothing useful read.
1555          * If so go try again
1556          */
1557         cpos = rbuf;
1558         cend = rbuf + siz;
1559         dp = (struct dirent *)cpos;
1560         cookiep = cookies;
1561
1562         /*
1563          * For some reason FreeBSD's ufs_readdir() chooses to back the
1564          * directory offset up to a block boundary, so it is necessary to
1565          * skip over the records that precede the requested offset. This
1566          * requires the assumption that file offset cookies monotonically
1567          * increase.
1568          */
1569         while (cpos < cend && ncookies > 0 &&
1570             (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1571              ((u_quad_t)(*cookiep)) <= toff)) {
1572                 cpos += dp->d_reclen;
1573                 dp = (struct dirent *)cpos;
1574                 cookiep++;
1575                 ncookies--;
1576         }
1577         if (cpos >= cend || ncookies == 0) {
1578                 siz = fullsiz;
1579                 toff = off;
1580                 goto again;
1581         }
1582
1583         /*
1584          * dirlen is the size of the reply, including all XDR and must
1585          * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1586          * if the XDR should be included in "count", but to be safe, we do.
1587          * (Include the two booleans at the end of the reply in dirlen now.)
1588          */
1589         if (nd->nd_flag & ND_NFSV3) {
1590                 nfsrv_postopattr(nd, getret, &at);
1591                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1592                 txdr_hyper(at.na_filerev, tl);
1593                 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1594         } else {
1595                 dirlen = 2 * NFSX_UNSIGNED;
1596         }
1597
1598         /* Loop through the records and build reply */
1599         while (cpos < cend && ncookies > 0) {
1600                 nlen = dp->d_namlen;
1601                 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1602                         nlen <= NFS_MAXNAMLEN) {
1603                         if (nd->nd_flag & ND_NFSV3)
1604                                 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1605                         else
1606                                 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1607                         if (dirlen > cnt) {
1608                                 eofflag = 0;
1609                                 break;
1610                         }
1611
1612                         /*
1613                          * Build the directory record xdr from
1614                          * the dirent entry.
1615                          */
1616                         if (nd->nd_flag & ND_NFSV3) {
1617                                 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1618                                 *tl++ = newnfs_true;
1619                                 *tl++ = 0;
1620                         } else {
1621                                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1622                                 *tl++ = newnfs_true;
1623                         }
1624                         *tl = txdr_unsigned(dp->d_fileno);
1625                         (void) nfsm_strtom(nd, dp->d_name, nlen);
1626                         if (nd->nd_flag & ND_NFSV3) {
1627                                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1628                                 *tl++ = 0;
1629                         } else
1630                                 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1631                         *tl = txdr_unsigned(*cookiep);
1632                 }
1633                 cpos += dp->d_reclen;
1634                 dp = (struct dirent *)cpos;
1635                 cookiep++;
1636                 ncookies--;
1637         }
1638         if (cpos < cend)
1639                 eofflag = 0;
1640         vrele(vp);
1641         NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1642         *tl++ = newnfs_false;
1643         if (eofflag)
1644                 *tl = newnfs_true;
1645         else
1646                 *tl = newnfs_false;
1647         FREE((caddr_t)rbuf, M_TEMP);
1648         FREE((caddr_t)cookies, M_TEMP);
1649         return (0);
1650 nfsmout:
1651         vput(vp);
1652         return (error);
1653 }
1654
1655 /*
1656  * Readdirplus for V3 and Readdir for V4.
1657  */
1658 int
1659 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1660     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1661 {
1662         struct dirent *dp;
1663         u_int32_t *tl;
1664         int dirlen;
1665         char *cpos, *cend, *rbuf;
1666         struct vnode *nvp;
1667         fhandle_t nfh;
1668         struct nfsvattr nva, at, *nvap = &nva;
1669         struct mbuf *mb0, *mb1;
1670         struct nfsreferral *refp;
1671         int nlen, r, error = 0, getret = 1, usevget = 1;
1672         int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1673         caddr_t bpos0, bpos1;
1674         u_int64_t off, toff, verf;
1675         u_long *cookies = NULL, *cookiep;
1676         nfsattrbit_t attrbits, rderrbits, savbits;
1677         struct uio io;
1678         struct iovec iv;
1679         struct componentname cn;
1680
1681         if (nd->nd_repstat) {
1682                 nfsrv_postopattr(nd, getret, &at);
1683                 return (0);
1684         }
1685         NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1686         off = fxdr_hyper(tl);
1687         toff = off;
1688         tl += 2;
1689         verf = fxdr_hyper(tl);
1690         tl += 2;
1691         siz = fxdr_unsigned(int, *tl++);
1692         cnt = fxdr_unsigned(int, *tl);
1693
1694         /*
1695          * Use the server's maximum data transfer size as the upper bound
1696          * on reply datalen.
1697          */
1698         if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1699                 cnt = NFS_SRVMAXDATA(nd);
1700
1701         /*
1702          * siz is a "hint" of how much directory information (name, fileid,
1703          * cookie) should be in the reply. At least one client "hints" 0,
1704          * so I set it to cnt for that case. I also round it up to the
1705          * next multiple of DIRBLKSIZ.
1706          */
1707         if (siz <= 0)
1708                 siz = cnt;
1709         siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1710
1711         if (nd->nd_flag & ND_NFSV4) {
1712                 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1713                 if (error)
1714                         goto nfsmout;
1715                 NFSSET_ATTRBIT(&savbits, &attrbits);
1716                 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1717                 NFSZERO_ATTRBIT(&rderrbits);
1718                 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1719         } else {
1720                 NFSZERO_ATTRBIT(&attrbits);
1721         }
1722         fullsiz = siz;
1723         nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p);
1724         if (!nd->nd_repstat) {
1725             if (off && verf != at.na_filerev) {
1726                 /*
1727                  * va_filerev is not sufficient as a cookie verifier,
1728                  * since it is not supposed to change when entries are
1729                  * removed/added unless that offset cookies returned to
1730                  * the client are no longer valid.
1731                  */
1732 #if 0
1733                 if (nd->nd_flag & ND_NFSV4) {
1734                         nd->nd_repstat = NFSERR_NOTSAME;
1735                 } else {
1736                         nd->nd_repstat = NFSERR_BAD_COOKIE;
1737                 }
1738 #endif
1739             } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1740                 nd->nd_repstat = NFSERR_BAD_COOKIE;
1741             }
1742         }
1743         if (!nd->nd_repstat && vp->v_type != VDIR)
1744                 nd->nd_repstat = NFSERR_NOTDIR;
1745         if (!nd->nd_repstat && cnt == 0)
1746                 nd->nd_repstat = NFSERR_TOOSMALL;
1747         if (!nd->nd_repstat)
1748                 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1749                     nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1750                     NFSACCCHK_VPISLOCKED, NULL);
1751         if (nd->nd_repstat) {
1752                 vput(vp);
1753                 if (nd->nd_flag & ND_NFSV3)
1754                         nfsrv_postopattr(nd, getret, &at);
1755                 return (0);
1756         }
1757
1758         MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1759 again:
1760         eofflag = 0;
1761         if (cookies) {
1762                 free((caddr_t)cookies, M_TEMP);
1763                 cookies = NULL;
1764         }
1765
1766         iv.iov_base = rbuf;
1767         iv.iov_len = siz;
1768         io.uio_iov = &iv;
1769         io.uio_iovcnt = 1;
1770         io.uio_offset = (off_t)off;
1771         io.uio_resid = siz;
1772         io.uio_segflg = UIO_SYSSPACE;
1773         io.uio_rw = UIO_READ;
1774         io.uio_td = NULL;
1775         nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1776             &cookies);
1777         off = (u_int64_t)io.uio_offset;
1778         if (io.uio_resid)
1779                 siz -= io.uio_resid;
1780
1781         getret = nfsvno_getattr(vp, &at, nd->nd_cred, p);
1782
1783         if (!cookies && !nd->nd_repstat)
1784                 nd->nd_repstat = NFSERR_PERM;
1785         if (!nd->nd_repstat)
1786                 nd->nd_repstat = getret;
1787         if (nd->nd_repstat) {
1788                 vput(vp);
1789                 if (cookies)
1790                         free((caddr_t)cookies, M_TEMP);
1791                 free((caddr_t)rbuf, M_TEMP);
1792                 if (nd->nd_flag & ND_NFSV3)
1793                         nfsrv_postopattr(nd, getret, &at);
1794                 return (0);
1795         }
1796         /*
1797          * If nothing read, return eof
1798          * rpc reply
1799          */
1800         if (siz == 0) {
1801                 vput(vp);
1802                 if (nd->nd_flag & ND_NFSV3)
1803                         nfsrv_postopattr(nd, getret, &at);
1804                 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1805                 txdr_hyper(at.na_filerev, tl);
1806                 tl += 2;
1807                 *tl++ = newnfs_false;
1808                 *tl = newnfs_true;
1809                 free((caddr_t)cookies, M_TEMP);
1810                 free((caddr_t)rbuf, M_TEMP);
1811                 return (0);
1812         }
1813
1814         /*
1815          * Check for degenerate cases of nothing useful read.
1816          * If so go try again
1817          */
1818         cpos = rbuf;
1819         cend = rbuf + siz;
1820         dp = (struct dirent *)cpos;
1821         cookiep = cookies;
1822
1823         /*
1824          * For some reason FreeBSD's ufs_readdir() chooses to back the
1825          * directory offset up to a block boundary, so it is necessary to
1826          * skip over the records that precede the requested offset. This
1827          * requires the assumption that file offset cookies monotonically
1828          * increase.
1829          */
1830         while (cpos < cend && ncookies > 0 &&
1831           (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1832            ((u_quad_t)(*cookiep)) <= toff ||
1833            ((nd->nd_flag & ND_NFSV4) &&
1834             ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1835              (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1836                 cpos += dp->d_reclen;
1837                 dp = (struct dirent *)cpos;
1838                 cookiep++;
1839                 ncookies--;
1840         }
1841         if (cpos >= cend || ncookies == 0) {
1842                 siz = fullsiz;
1843                 toff = off;
1844                 goto again;
1845         }
1846         NFSVOPUNLOCK(vp, 0, p);
1847
1848         /*
1849          * Save this position, in case there is an error before one entry
1850          * is created.
1851          */
1852         mb0 = nd->nd_mb;
1853         bpos0 = nd->nd_bpos;
1854
1855         /*
1856          * Fill in the first part of the reply.
1857          * dirlen is the reply length in bytes and cannot exceed cnt.
1858          * (Include the two booleans at the end of the reply in dirlen now,
1859          *  so we recognize when we have exceeded cnt.)
1860          */
1861         if (nd->nd_flag & ND_NFSV3) {
1862                 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1863                 nfsrv_postopattr(nd, getret, &at);
1864         } else {
1865                 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
1866         }
1867         NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
1868         txdr_hyper(at.na_filerev, tl);
1869
1870         /*
1871          * Save this position, in case there is an empty reply needed.
1872          */
1873         mb1 = nd->nd_mb;
1874         bpos1 = nd->nd_bpos;
1875
1876         /* Loop through the records and build reply */
1877         entrycnt = 0;
1878         while (cpos < cend && ncookies > 0 && dirlen < cnt) {
1879                 nlen = dp->d_namlen;
1880                 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1881                     nlen <= NFS_MAXNAMLEN &&
1882                     ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
1883                      (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
1884                       || (nlen == 1 && dp->d_name[0] != '.'))) {
1885                         /*
1886                          * Save the current position in the reply, in case
1887                          * this entry exceeds cnt.
1888                          */
1889                         mb1 = nd->nd_mb;
1890                         bpos1 = nd->nd_bpos;
1891
1892                         /*
1893                          * For readdir_and_lookup get the vnode using
1894                          * the file number.
1895                          */
1896                         nvp = NULL;
1897                         refp = NULL;
1898                         r = 0;
1899                         if ((nd->nd_flag & ND_NFSV3) ||
1900                             NFSNONZERO_ATTRBIT(&savbits)) {
1901                                 if (nd->nd_flag & ND_NFSV4)
1902                                         refp = nfsv4root_getreferral(NULL,
1903                                             vp, dp->d_fileno);
1904                                 if (refp == NULL) {
1905                                         if (usevget)
1906                                                 r = VFS_VGET(vp->v_mount,
1907                                                     dp->d_fileno, LK_EXCLUSIVE,
1908                                                     &nvp);
1909                                         else
1910                                                 r = EOPNOTSUPP;
1911                                         if (r == EOPNOTSUPP) {
1912                                                 if (usevget) {
1913                                                         usevget = 0;
1914                                                         cn.cn_nameiop = LOOKUP;
1915                                                         cn.cn_lkflags =
1916                                                             LK_EXCLUSIVE |
1917                                                             LK_RETRY;
1918                                                         cn.cn_cred =
1919                                                             nd->nd_cred;
1920                                                         cn.cn_thread = p;
1921                                                 }
1922                                                 cn.cn_nameptr = dp->d_name;
1923                                                 cn.cn_namelen = nlen;
1924                                                 cn.cn_flags = ISLASTCN |
1925                                                     NOFOLLOW | LOCKLEAF |
1926                                                     MPSAFE;
1927                                                 if (nlen == 2 &&
1928                                                     dp->d_name[0] == '.' &&
1929                                                     dp->d_name[1] == '.')
1930                                                         cn.cn_flags |=
1931                                                             ISDOTDOT;
1932                                                 if (!VOP_ISLOCKED(vp))
1933                                                         vn_lock(vp,
1934                                                             LK_EXCLUSIVE |
1935                                                             LK_RETRY);
1936                                                 r = VOP_LOOKUP(vp, &nvp, &cn);
1937                                         }
1938                                 }
1939                                 if (!r) {
1940                                     if (refp == NULL &&
1941                                         ((nd->nd_flag & ND_NFSV3) ||
1942                                          NFSNONZERO_ATTRBIT(&attrbits))) {
1943                                         r = nfsvno_getfh(nvp, &nfh, p);
1944                                         if (!r)
1945                                             r = nfsvno_getattr(nvp, nvap,
1946                                                 nd->nd_cred, p);
1947                                     }
1948                                 } else {
1949                                     nvp = NULL;
1950                                 }
1951                                 if (r) {
1952                                         if (!NFSISSET_ATTRBIT(&attrbits,
1953                                             NFSATTRBIT_RDATTRERROR)) {
1954                                                 if (nvp != NULL)
1955                                                         vput(nvp);
1956                                                 nd->nd_repstat = r;
1957                                                 break;
1958                                         }
1959                                 }
1960                         }
1961
1962                         /*
1963                          * Build the directory record xdr
1964                          */
1965                         if (nd->nd_flag & ND_NFSV3) {
1966                                 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1967                                 *tl++ = newnfs_true;
1968                                 *tl++ = 0;
1969                                 *tl = txdr_unsigned(dp->d_fileno);
1970                                 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
1971                                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1972                                 *tl++ = 0;
1973                                 *tl = txdr_unsigned(*cookiep);
1974                                 nfsrv_postopattr(nd, 0, nvap);
1975                                 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
1976                                 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
1977                                 if (nvp != NULL)
1978                                         vput(nvp);
1979                         } else {
1980                                 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1981                                 *tl++ = newnfs_true;
1982                                 *tl++ = 0;
1983                                 *tl = txdr_unsigned(*cookiep);
1984                                 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
1985                                 if (nvp != NULL)
1986                                         NFSVOPUNLOCK(nvp, 0, p);
1987                                 if (refp != NULL) {
1988                                         dirlen += nfsrv_putreferralattr(nd,
1989                                             &savbits, refp, 0,
1990                                             &nd->nd_repstat);
1991                                         if (nd->nd_repstat) {
1992                                                 if (nvp != NULL)
1993                                                         vrele(nvp);
1994                                                 break;
1995                                         }
1996                                 } else if (r) {
1997                                         dirlen += nfsvno_fillattr(nd, nvp, nvap,
1998                                             &nfh, r, &rderrbits, nd->nd_cred,
1999                                             p, isdgram, 0);
2000                                 } else {
2001                                         dirlen += nfsvno_fillattr(nd, nvp, nvap,
2002                                             &nfh, r, &attrbits, nd->nd_cred,
2003                                             p, isdgram, 0);
2004                                 }
2005                                 if (nvp != NULL)
2006                                         vrele(nvp);
2007                                 dirlen += (3 * NFSX_UNSIGNED);
2008                         }
2009                         if (dirlen <= cnt)
2010                                 entrycnt++;
2011                 }
2012                 cpos += dp->d_reclen;
2013                 dp = (struct dirent *)cpos;
2014                 cookiep++;
2015                 ncookies--;
2016         }
2017         if (!usevget && VOP_ISLOCKED(vp))
2018                 vput(vp);
2019         else
2020                 vrele(vp);
2021
2022         /*
2023          * If dirlen > cnt, we must strip off the last entry. If that
2024          * results in an empty reply, report NFSERR_TOOSMALL.
2025          */
2026         if (dirlen > cnt || nd->nd_repstat) {
2027                 if (!nd->nd_repstat && entrycnt == 0)
2028                         nd->nd_repstat = NFSERR_TOOSMALL;
2029                 if (nd->nd_repstat)
2030                         newnfs_trimtrailing(nd, mb0, bpos0);
2031                 else
2032                         newnfs_trimtrailing(nd, mb1, bpos1);
2033                 eofflag = 0;
2034         } else if (cpos < cend)
2035                 eofflag = 0;
2036         if (!nd->nd_repstat) {
2037                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2038                 *tl++ = newnfs_false;
2039                 if (eofflag)
2040                         *tl = newnfs_true;
2041                 else
2042                         *tl = newnfs_false;
2043         }
2044         FREE((caddr_t)cookies, M_TEMP);
2045         FREE((caddr_t)rbuf, M_TEMP);
2046         return (0);
2047 nfsmout:
2048         vput(vp);
2049         return (error);
2050 }
2051
2052 /*
2053  * Get the settable attributes out of the mbuf list.
2054  * (Return 0 or EBADRPC)
2055  */
2056 int
2057 nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2058     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2059 {
2060         u_int32_t *tl;
2061         struct nfsv2_sattr *sp;
2062         struct timeval curtime;
2063         int error = 0, toclient = 0;
2064
2065         switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2066         case ND_NFSV2:
2067                 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2068                 /*
2069                  * Some old clients didn't fill in the high order 16bits.
2070                  * --> check the low order 2 bytes for 0xffff
2071                  */
2072                 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2073                         nvap->na_mode = nfstov_mode(sp->sa_mode);
2074                 if (sp->sa_uid != newnfs_xdrneg1)
2075                         nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2076                 if (sp->sa_gid != newnfs_xdrneg1)
2077                         nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2078                 if (sp->sa_size != newnfs_xdrneg1)
2079                         nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2080                 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2081 #ifdef notyet
2082                         fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2083 #else
2084                         nvap->na_atime.tv_sec =
2085                                 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2086                         nvap->na_atime.tv_nsec = 0;
2087 #endif
2088                 }
2089                 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2090                         fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2091                 break;
2092         case ND_NFSV3:
2093                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2094                 if (*tl == newnfs_true) {
2095                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2096                         nvap->na_mode = nfstov_mode(*tl);
2097                 }
2098                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2099                 if (*tl == newnfs_true) {
2100                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2101                         nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2102                 }
2103                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2104                 if (*tl == newnfs_true) {
2105                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2106                         nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2107                 }
2108                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2109                 if (*tl == newnfs_true) {
2110                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2111                         nvap->na_size = fxdr_hyper(tl);
2112                 }
2113                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2114                 switch (fxdr_unsigned(int, *tl)) {
2115                 case NFSV3SATTRTIME_TOCLIENT:
2116                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2117                         fxdr_nfsv3time(tl, &nvap->na_atime);
2118                         toclient = 1;
2119                         break;
2120                 case NFSV3SATTRTIME_TOSERVER:
2121                         NFSGETTIME(&curtime);
2122                         nvap->na_atime.tv_sec = curtime.tv_sec;
2123                         nvap->na_atime.tv_nsec = curtime.tv_usec * 1000;
2124                         nvap->na_vaflags |= VA_UTIMES_NULL;
2125                         break;
2126                 };
2127                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2128                 switch (fxdr_unsigned(int, *tl)) {
2129                 case NFSV3SATTRTIME_TOCLIENT:
2130                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2131                         fxdr_nfsv3time(tl, &nvap->na_mtime);
2132                         nvap->na_vaflags &= ~VA_UTIMES_NULL;
2133                         break;
2134                 case NFSV3SATTRTIME_TOSERVER:
2135                         NFSGETTIME(&curtime);
2136                         nvap->na_mtime.tv_sec = curtime.tv_sec;
2137                         nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000;
2138                         if (!toclient)
2139                                 nvap->na_vaflags |= VA_UTIMES_NULL;
2140                         break;
2141                 };
2142                 break;
2143         case ND_NFSV4:
2144                 error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p);
2145         };
2146 nfsmout:
2147         return (error);
2148 }
2149
2150 /*
2151  * Handle the setable attributes for V4.
2152  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2153  */
2154 int
2155 nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2156     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2157 {
2158         u_int32_t *tl;
2159         int attrsum = 0;
2160         int i, j;
2161         int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2162         int toclient = 0;
2163         u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2164         uid_t uid;
2165         gid_t gid;
2166         struct timeval curtime;
2167
2168         error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2169         if (error)
2170                 return (error);
2171         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2172         attrsize = fxdr_unsigned(int, *tl);
2173
2174         /*
2175          * Loop around getting the setable attributes. If an unsupported
2176          * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2177          */
2178         if (retnotsup) {
2179                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2180                 bitpos = NFSATTRBIT_MAX;
2181         } else {
2182                 bitpos = 0;
2183         }
2184         for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2185             if (attrsum > attrsize) {
2186                 error = NFSERR_BADXDR;
2187                 goto nfsmout;
2188             }
2189             if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2190                 switch (bitpos) {
2191                 case NFSATTRBIT_SIZE:
2192                         NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2193                         nvap->na_size = fxdr_hyper(tl);
2194                         attrsum += NFSX_HYPER;
2195                         break;
2196                 case NFSATTRBIT_ACL:
2197                         error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2198                             p);
2199                         if (error)
2200                                 goto nfsmout;
2201                         if (aceerr && !nd->nd_repstat)
2202                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2203                         attrsum += aclsize;
2204                         break;
2205                 case NFSATTRBIT_ARCHIVE:
2206                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2207                         if (!nd->nd_repstat)
2208                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2209                         attrsum += NFSX_UNSIGNED;
2210                         break;
2211                 case NFSATTRBIT_HIDDEN:
2212                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2213                         if (!nd->nd_repstat)
2214                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2215                         attrsum += NFSX_UNSIGNED;
2216                         break;
2217                 case NFSATTRBIT_MIMETYPE:
2218                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2219                         i = fxdr_unsigned(int, *tl);
2220                         error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2221                         if (error)
2222                                 goto nfsmout;
2223                         if (!nd->nd_repstat)
2224                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2225                         attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2226                         break;
2227                 case NFSATTRBIT_MODE:
2228                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2229                         nvap->na_mode = nfstov_mode(*tl);
2230                         attrsum += NFSX_UNSIGNED;
2231                         break;
2232                 case NFSATTRBIT_OWNER:
2233                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2234                         j = fxdr_unsigned(int, *tl);
2235                         if (j < 0)
2236                                 return (NFSERR_BADXDR);
2237                         if (j > NFSV4_SMALLSTR)
2238                                 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2239                         else
2240                                 cp = namestr;
2241                         error = nfsrv_mtostr(nd, cp, j);
2242                         if (error) {
2243                                 if (j > NFSV4_SMALLSTR)
2244                                         free(cp, M_NFSSTRING);
2245                                 return (error);
2246                         }
2247                         if (!nd->nd_repstat) {
2248                                 nd->nd_repstat = nfsv4_strtouid(cp,j,&uid,p);
2249                                 if (!nd->nd_repstat)
2250                                         nvap->na_uid = uid;
2251                         }
2252                         if (j > NFSV4_SMALLSTR)
2253                                 free(cp, M_NFSSTRING);
2254                         attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2255                         break;
2256                 case NFSATTRBIT_OWNERGROUP:
2257                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2258                         j = fxdr_unsigned(int, *tl);
2259                         if (j < 0)
2260                                 return (NFSERR_BADXDR);
2261                         if (j > NFSV4_SMALLSTR)
2262                                 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2263                         else
2264                                 cp = namestr;
2265                         error = nfsrv_mtostr(nd, cp, j);
2266                         if (error) {
2267                                 if (j > NFSV4_SMALLSTR)
2268                                         free(cp, M_NFSSTRING);
2269                                 return (error);
2270                         }
2271                         if (!nd->nd_repstat) {
2272                                 nd->nd_repstat = nfsv4_strtogid(cp,j,&gid,p);
2273                                 if (!nd->nd_repstat)
2274                                         nvap->na_gid = gid;
2275                         }
2276                         if (j > NFSV4_SMALLSTR)
2277                                 free(cp, M_NFSSTRING);
2278                         attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2279                         break;
2280                 case NFSATTRBIT_SYSTEM:
2281                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2282                         if (!nd->nd_repstat)
2283                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2284                         attrsum += NFSX_UNSIGNED;
2285                         break;
2286                 case NFSATTRBIT_TIMEACCESSSET:
2287                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2288                         attrsum += NFSX_UNSIGNED;
2289                         if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2290                             NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2291                             fxdr_nfsv4time(tl, &nvap->na_atime);
2292                             toclient = 1;
2293                             attrsum += NFSX_V4TIME;
2294                         } else {
2295                             NFSGETTIME(&curtime);
2296                             nvap->na_atime.tv_sec = curtime.tv_sec;
2297                             nvap->na_atime.tv_nsec = curtime.tv_usec * 1000;
2298                             nvap->na_vaflags |= VA_UTIMES_NULL;
2299                         }
2300                         break;
2301                 case NFSATTRBIT_TIMEBACKUP:
2302                         NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2303                         if (!nd->nd_repstat)
2304                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2305                         attrsum += NFSX_V4TIME;
2306                         break;
2307                 case NFSATTRBIT_TIMECREATE:
2308                         NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2309                         if (!nd->nd_repstat)
2310                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2311                         attrsum += NFSX_V4TIME;
2312                         break;
2313                 case NFSATTRBIT_TIMEMODIFYSET:
2314                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2315                         attrsum += NFSX_UNSIGNED;
2316                         if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2317                             NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2318                             fxdr_nfsv4time(tl, &nvap->na_mtime);
2319                             nvap->na_vaflags &= ~VA_UTIMES_NULL;
2320                             attrsum += NFSX_V4TIME;
2321                         } else {
2322                             NFSGETTIME(&curtime);
2323                             nvap->na_mtime.tv_sec = curtime.tv_sec;
2324                             nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000;
2325                             if (!toclient)
2326                                 nvap->na_vaflags |= VA_UTIMES_NULL;
2327                         }
2328                         break;
2329                 default:
2330                         nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2331                         /*
2332                          * set bitpos so we drop out of the loop.
2333                          */
2334                         bitpos = NFSATTRBIT_MAX;
2335                         break;
2336                 };
2337         }
2338
2339         /*
2340          * some clients pad the attrlist, so we need to skip over the
2341          * padding.
2342          */
2343         if (attrsum > attrsize) {
2344                 error = NFSERR_BADXDR;
2345         } else {
2346                 attrsize = NFSM_RNDUP(attrsize);
2347                 if (attrsum < attrsize)
2348                         error = nfsm_advance(nd, attrsize - attrsum, -1);
2349         }
2350 nfsmout:
2351         return (error);
2352 }
2353
2354 /*
2355  * Check/setup export credentials.
2356  */
2357 int
2358 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2359     struct ucred *credanon)
2360 {
2361         int error = 0;
2362
2363         /*
2364          * Check/setup credentials.
2365          */
2366         if (nd->nd_flag & ND_GSS)
2367                 exp->nes_exflag &= ~MNT_EXPORTANON;
2368
2369         /*
2370          * Check to see if the operation is allowed for this security flavor.
2371          * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2372          * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2373          * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2374          */
2375         if (nfsvno_testexp(nd, exp) &&
2376             nd->nd_procnum != NFSV4OP_SECINFO &&
2377             nd->nd_procnum != NFSPROC_FSINFO) {
2378                 if (nd->nd_flag & ND_NFSV4)
2379                         error = NFSERR_WRONGSEC;
2380                 else
2381                         error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2382                 return (error);
2383         }
2384
2385         /*
2386          * Check to see if the file system is exported V4 only.
2387          */
2388         if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4))
2389                 return (NFSERR_PROGNOTV4);
2390
2391         /*
2392          * Now, map the user credentials.
2393          * (Note that ND_AUTHNONE will only be set for an NFSv3
2394          *  Fsinfo RPC. If set for anything else, this code might need
2395          *  to change.)
2396          */
2397         if (NFSVNO_EXPORTED(exp) &&
2398             ((!(nd->nd_flag & ND_GSS) && nd->nd_cred->cr_uid == 0) ||
2399              NFSVNO_EXPORTANON(exp) ||
2400              (nd->nd_flag & ND_AUTHNONE))) {
2401                 nd->nd_cred->cr_uid = credanon->cr_uid;
2402                 nd->nd_cred->cr_gid = credanon->cr_gid;
2403                 crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2404                     credanon->cr_groups);
2405         }
2406         return (0);
2407 }
2408
2409 /*
2410  * Check exports.
2411  */
2412 int
2413 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2414     struct ucred **credp)
2415 {
2416         int i, error, *secflavors;
2417
2418         error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2419             &exp->nes_numsecflavor, &secflavors);
2420         if (error) {
2421                 if (nfs_rootfhset) {
2422                         exp->nes_exflag = 0;
2423                         exp->nes_numsecflavor = 0;
2424                         error = 0;
2425                 }
2426         } else {
2427                 /* Copy the security flavors. */
2428                 for (i = 0; i < exp->nes_numsecflavor; i++)
2429                         exp->nes_secflavors[i] = secflavors[i];
2430         }
2431         return (error);
2432 }
2433
2434 /*
2435  * Get a vnode for a file handle and export stuff.
2436  */
2437 int
2438 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2439     struct vnode **vpp, struct nfsexstuff *exp, struct ucred **credp)
2440 {
2441         int i, error, *secflavors;
2442
2443         *credp = NULL;
2444         exp->nes_numsecflavor = 0;
2445         error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp);
2446         if (error != 0)
2447                 /* Make sure the server replies ESTALE to the client. */
2448                 error = ESTALE;
2449         if (nam && !error) {
2450                 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2451                     &exp->nes_numsecflavor, &secflavors);
2452                 if (error) {
2453                         if (nfs_rootfhset) {
2454                                 exp->nes_exflag = 0;
2455                                 exp->nes_numsecflavor = 0;
2456                                 error = 0;
2457                         } else {
2458                                 vput(*vpp);
2459                         }
2460                 } else {
2461                         /* Copy the security flavors. */
2462                         for (i = 0; i < exp->nes_numsecflavor; i++)
2463                                 exp->nes_secflavors[i] = secflavors[i];
2464                 }
2465         }
2466         return (error);
2467 }
2468
2469 /*
2470  * Do the pathconf vnode op.
2471  */
2472 int
2473 nfsvno_pathconf(struct vnode *vp, int flag, register_t *retf,
2474     struct ucred *cred, struct thread *p)
2475 {
2476         int error;
2477
2478         error = VOP_PATHCONF(vp, flag, retf);
2479         return (error);
2480 }
2481
2482 /*
2483  * nfsd_fhtovp() - convert a fh to a vnode ptr
2484  *      - look up fsid in mount list (if not found ret error)
2485  *      - get vp and export rights by calling nfsvno_fhtovp()
2486  *      - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2487  *        for AUTH_SYS
2488  * Also handle getting the Giant lock for the file system,
2489  * as required:
2490  * - if same mount point as *mpp
2491  *       do nothing
2492  *   else if *mpp == NULL
2493  *       if already locked
2494  *           leave it locked
2495  *       else
2496  *           call VFS_LOCK_GIANT()
2497  *   else
2498  *       if already locked
2499  *            unlock Giant
2500  *       call VFS_LOCK_GIANT()
2501  */
2502 void
2503 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp,
2504     struct vnode **vpp, struct nfsexstuff *exp,
2505     struct mount **mpp, int startwrite, struct thread *p)
2506 {
2507         struct mount *mp;
2508         struct ucred *credanon;
2509         fhandle_t *fhp;
2510
2511         fhp = (fhandle_t *)nfp->nfsrvfh_data;
2512         /*
2513          * Check for the special case of the nfsv4root_fh.
2514          */
2515         mp = vfs_getvfs(&fhp->fh_fsid);
2516         if (!mp) {
2517                 *vpp = NULL;
2518                 nd->nd_repstat = ESTALE;
2519                 if (*mpp && exp->nes_vfslocked)
2520                         VFS_UNLOCK_GIANT(*mpp);
2521                 *mpp = NULL;
2522                 exp->nes_vfslocked = 0;
2523                 return;
2524         }
2525
2526         /*
2527          * Now, handle Giant for the file system.
2528          */
2529         if (*mpp != NULL && *mpp != mp && exp->nes_vfslocked) {
2530                 VFS_UNLOCK_GIANT(*mpp);
2531                 exp->nes_vfslocked = 0;
2532         }
2533         if (!exp->nes_vfslocked && *mpp != mp)
2534                 exp->nes_vfslocked = VFS_LOCK_GIANT(mp);
2535
2536         *mpp = mp;
2537         if (startwrite)
2538                 vn_start_write(NULL, mpp, V_WAIT);
2539
2540         nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, vpp, exp,
2541             &credanon);
2542
2543         /*
2544          * For NFSv4 without a pseudo root fs, unexported file handles
2545          * can be returned, so that Lookup works everywhere.
2546          */
2547         if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2548             !(nd->nd_flag & ND_NFSV4)) {
2549                 vput(*vpp);
2550                 nd->nd_repstat = EACCES;
2551         }
2552
2553         /*
2554          * Personally, I've never seen any point in requiring a
2555          * reserved port#, since only in the rare case where the
2556          * clients are all boxes with secure system priviledges,
2557          * does it provide any enhanced security, but... some people
2558          * believe it to be useful and keep putting this code back in.
2559          * (There is also some "security checker" out there that
2560          *  complains if the nfs server doesn't enforce this.)
2561          * However, note the following:
2562          * RFC3530 (NFSv4) specifies that a reserved port# not be
2563          *      required.
2564          * RFC2623 recommends that, if a reserved port# is checked for,
2565          *      that there be a way to turn that off--> ifdef'd.
2566          */
2567 #ifdef NFS_REQRSVPORT
2568         if (!nd->nd_repstat) {
2569                 struct sockaddr_in *saddr;
2570                 struct sockaddr_in6 *saddr6;
2571
2572                 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2573                 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2574                 if (!(nd->nd_flag & ND_NFSV4) &&
2575                     ((saddr->sin_family == AF_INET &&
2576                       ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2577                      (saddr6->sin6_family == AF_INET6 &&
2578                       ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2579                         vput(*vpp);
2580                         nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2581                 }
2582         }
2583 #endif  /* NFS_REQRSVPORT */
2584
2585         /*
2586          * Check/setup credentials.
2587          */
2588         if (!nd->nd_repstat) {
2589                 nd->nd_saveduid = nd->nd_cred->cr_uid;
2590                 nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2591                 if (nd->nd_repstat)
2592                         vput(*vpp);
2593         }
2594         if (credanon != NULL)
2595                 crfree(credanon);
2596         if (nd->nd_repstat) {
2597                 if (startwrite)
2598                         vn_finished_write(mp);
2599                 if (exp->nes_vfslocked) {
2600                         VFS_UNLOCK_GIANT(mp);
2601                         exp->nes_vfslocked = 0;
2602                 }
2603                 vfs_rel(mp);
2604                 *vpp = NULL;
2605                 *mpp = NULL;
2606         } else {
2607                 vfs_rel(mp);
2608         }
2609 }
2610
2611 /*
2612  * glue for fp.
2613  */
2614 int
2615 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2616 {
2617         struct filedesc *fdp;
2618         struct file *fp;
2619
2620         fdp = p->td_proc->p_fd;
2621         if (fd >= fdp->fd_nfiles ||
2622             (fp = fdp->fd_ofiles[fd]) == NULL)
2623                 return (EBADF);
2624         *fpp = fp;
2625         return (0);
2626 }
2627
2628 /*
2629  * Called from nfssvc() to update the exports list. Just call
2630  * vfs_export(). This has to be done, since the v4 root fake fs isn't
2631  * in the mount list.
2632  */
2633 int
2634 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2635 {
2636         struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2637         int error;
2638         struct nameidata nd;
2639         fhandle_t fh;
2640
2641         error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2642         if ((nfsexargp->export.ex_flags & MNT_DELEXPORT)) {
2643                 nfs_rootfhset = 0;
2644                 nfsv4root_set = 0;
2645         } else if (error == 0) {
2646                 if (nfsexargp->fspec == NULL)
2647                         return (EPERM);
2648                 /*
2649                  * If fspec != NULL, this is the v4root path.
2650                  */
2651                 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_USERSPACE,
2652                     nfsexargp->fspec, p);
2653                 if ((error = namei(&nd)) != 0)
2654                         return (error);
2655                 error = nfsvno_getfh(nd.ni_vp, &fh, p);
2656                 vrele(nd.ni_vp);
2657                 if (!error) {
2658                         nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2659                         NFSBCOPY((caddr_t)&fh,
2660                             nfs_rootfh.nfsrvfh_data,
2661                             sizeof (fhandle_t));
2662                         nfs_rootfhset = 1;
2663                 }
2664         }
2665         return (error);
2666 }
2667
2668 /*
2669  * Get the tcp socket sequence numbers we need.
2670  * (Maybe this should be moved to the tcp sources?)
2671  */
2672 int
2673 nfsrv_getsocksndseq(struct socket *so, tcp_seq *maxp, tcp_seq *unap)
2674 {
2675         struct inpcb *inp;
2676         struct tcpcb *tp;
2677
2678         inp = sotoinpcb(so);
2679         KASSERT(inp != NULL, ("nfsrv_getsocksndseq: inp == NULL"));
2680         INP_RLOCK(inp);
2681         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
2682                 INP_RUNLOCK(inp);
2683                 return (EPIPE);
2684         }
2685         tp = intotcpcb(inp);
2686         if (tp->t_state != TCPS_ESTABLISHED) {
2687                 INP_RUNLOCK(inp);
2688                 return (EPIPE);
2689         }
2690         *maxp = tp->snd_max;
2691         *unap = tp->snd_una;
2692         INP_RUNLOCK(inp);
2693         return (0);
2694 }
2695
2696 /*
2697  * This function needs to test to see if the system is near its limit
2698  * for memory allocation via malloc() or mget() and return True iff
2699  * either of these resources are near their limit.
2700  * XXX (For now, this is just a stub.)
2701  */
2702 int nfsrv_testmalloclimit = 0;
2703 int
2704 nfsrv_mallocmget_limit(void)
2705 {
2706         static int printmesg = 0;
2707         static int testval = 1;
2708
2709         if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2710                 if ((printmesg++ % 100) == 0)
2711                         printf("nfsd: malloc/mget near limit\n");
2712                 return (1);
2713         }
2714         return (0);
2715 }
2716
2717 /*
2718  * BSD specific initialization of a mount point.
2719  */
2720 void
2721 nfsd_mntinit(void)
2722 {
2723         static int inited = 0;
2724
2725         if (inited)
2726                 return;
2727         inited = 1;
2728         nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2729         TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2730         nfsv4root_mnt.mnt_export = NULL;
2731         TAILQ_INIT(&nfsv4root_opt);
2732         TAILQ_INIT(&nfsv4root_newopt);
2733         nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2734         nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2735         nfsv4root_mnt.mnt_nvnodelistsize = 0;
2736 }
2737
2738 /*
2739  * Get a vnode for a file handle, without checking exports, etc.
2740  */
2741 struct vnode *
2742 nfsvno_getvp(fhandle_t *fhp)
2743 {
2744         struct mount *mp;
2745         struct vnode *vp;
2746         int error;
2747
2748         mp = vfs_getvfs(&fhp->fh_fsid);
2749         if (mp == NULL)
2750                 return (NULL);
2751         error = VFS_FHTOVP(mp, &fhp->fh_fid, &vp);
2752         if (error)
2753                 return (NULL);
2754         return (vp);
2755 }
2756
2757 /*
2758  * Check to see it a byte range lock held by a process running
2759  * locally on the server conflicts with the new lock.
2760  */
2761 int
2762 nfsvno_localconflict(struct vnode *vp, int ftype, u_int64_t first,
2763     u_int64_t end, struct nfslockconflict *cfp, struct thread *td)
2764 {
2765         int error;
2766         struct flock fl;
2767
2768         if (!nfsrv_dolocallocks)
2769                 return (0);
2770         fl.l_whence = SEEK_SET;
2771         fl.l_type = ftype;
2772         fl.l_start = (off_t)first;
2773         if (end == NFS64BITSSET)
2774                 fl.l_len = 0;
2775         else
2776                 fl.l_len = (off_t)(end - first);
2777         /*
2778          * For FreeBSD8, the l_pid and l_sysid must be set to the same
2779          * values for all calls, so that all locks will be held by the
2780          * nfsd server. (The nfsd server handles conflicts between the
2781          * various clients.)
2782          * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
2783          * bytes, so it can't be put in l_sysid.
2784          */
2785         if (nfsv4_sysid == 0)
2786                 nfsv4_sysid = nlm_acquire_next_sysid();
2787         fl.l_pid = (pid_t)0;
2788         fl.l_sysid = (int)nfsv4_sysid;
2789
2790         NFSVOPUNLOCK(vp, 0, td);
2791         error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_GETLK, &fl,
2792             (F_POSIX | F_REMOTE));
2793         NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, td);
2794         if (error)
2795                 return (error);
2796         if (fl.l_type == F_UNLCK)
2797                 return (0);
2798         if (cfp != NULL) {
2799                 cfp->cl_clientid.lval[0] = cfp->cl_clientid.lval[1] = 0;
2800                 cfp->cl_first = (u_int64_t)fl.l_start;
2801                 if (fl.l_len == 0)
2802                         cfp->cl_end = NFS64BITSSET;
2803                 else
2804                         cfp->cl_end = (u_int64_t)
2805                             (fl.l_start + fl.l_len);
2806                 if (fl.l_type == F_WRLCK)
2807                         cfp->cl_flags = NFSLCK_WRITE;
2808                 else
2809                         cfp->cl_flags = NFSLCK_READ;
2810                 sprintf(cfp->cl_owner, "LOCALID%d", fl.l_pid);
2811                 cfp->cl_ownerlen = strlen(cfp->cl_owner);
2812                 return (NFSERR_DENIED);
2813         }
2814         return (NFSERR_INVAL);
2815 }
2816
2817 /*
2818  * Do a local VOP_ADVLOCK().
2819  */
2820 int
2821 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2822     u_int64_t end, struct thread *td)
2823 {
2824         int error;
2825         struct flock fl;
2826         u_int64_t tlen;
2827
2828         if (!nfsrv_dolocallocks)
2829                 return (0);
2830         fl.l_whence = SEEK_SET;
2831         fl.l_type = ftype;
2832         fl.l_start = (off_t)first;
2833         if (end == NFS64BITSSET) {
2834                 fl.l_len = 0;
2835         } else {
2836                 tlen = end - first;
2837                 fl.l_len = (off_t)tlen;
2838         }
2839         /*
2840          * For FreeBSD8, the l_pid and l_sysid must be set to the same
2841          * values for all calls, so that all locks will be held by the
2842          * nfsd server. (The nfsd server handles conflicts between the
2843          * various clients.)
2844          * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
2845          * bytes, so it can't be put in l_sysid.
2846          */
2847         if (nfsv4_sysid == 0)
2848                 nfsv4_sysid = nlm_acquire_next_sysid();
2849         fl.l_pid = (pid_t)0;
2850         fl.l_sysid = (int)nfsv4_sysid;
2851
2852         NFSVOPUNLOCK(vp, 0, td);
2853         error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
2854             (F_POSIX | F_REMOTE));
2855         NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, td);
2856         return (error);
2857 }
2858
2859 /*
2860  * Unlock an underlying local file system.
2861  */
2862 void
2863 nfsvno_unlockvfs(struct mount *mp)
2864 {
2865
2866         VFS_UNLOCK_GIANT(mp);
2867 }
2868
2869 /*
2870  * Lock an underlying file system, as required, and return
2871  * whether or not it is locked.
2872  */
2873 int
2874 nfsvno_lockvfs(struct mount *mp)
2875 {
2876         int ret;
2877
2878         ret = VFS_LOCK_GIANT(mp);
2879         return (ret);
2880 }
2881
2882 /*
2883  * Check the nfsv4 root exports.
2884  */
2885 int
2886 nfsvno_v4rootexport(struct nfsrv_descript *nd)
2887 {
2888         struct ucred *credanon;
2889         int exflags, error, numsecflavor, *secflavors, i;
2890
2891         error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
2892             &credanon, &numsecflavor, &secflavors);
2893         if (error)
2894                 return (NFSERR_PROGUNAVAIL);
2895         if (credanon != NULL)
2896                 crfree(credanon);
2897         for (i = 0; i < numsecflavor; i++) {
2898                 if (secflavors[i] == AUTH_SYS)
2899                         nd->nd_flag |= ND_EXAUTHSYS;
2900                 else if (secflavors[i] == RPCSEC_GSS_KRB5)
2901                         nd->nd_flag |= ND_EXGSS;
2902                 else if (secflavors[i] == RPCSEC_GSS_KRB5I)
2903                         nd->nd_flag |= ND_EXGSSINTEGRITY;
2904                 else if (secflavors[i] == RPCSEC_GSS_KRB5P)
2905                         nd->nd_flag |= ND_EXGSSPRIVACY;
2906         }
2907         return (0);
2908 }
2909
2910 /*
2911  * Nfs server psuedo system call for the nfsd's
2912  */
2913 /*
2914  * MPSAFE
2915  */
2916 static int
2917 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
2918 {
2919         struct file *fp;
2920         struct nfsd_addsock_args sockarg;
2921         struct nfsd_nfsd_args nfsdarg;
2922         int error;
2923
2924         if (uap->flag & NFSSVC_NFSDADDSOCK) {
2925                 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
2926                 if (error)
2927                         return (error);
2928                 if ((error = fget(td, sockarg.sock, &fp)) != 0) {
2929                         return (error);
2930                 }
2931                 if (fp->f_type != DTYPE_SOCKET) {
2932                         fdrop(fp, td);
2933                         return (EPERM);
2934                 }
2935                 error = nfsrvd_addsock(fp);
2936                 fdrop(fp, td);
2937         } else if (uap->flag & NFSSVC_NFSDNFSD) {
2938                 if (uap->argp == NULL)
2939                         return (EINVAL);
2940                 error = copyin(uap->argp, (caddr_t)&nfsdarg,
2941                     sizeof (nfsdarg));
2942                 if (error)
2943                         return (error);
2944                 error = nfsrvd_nfsd(td, &nfsdarg);
2945         } else {
2946                 error = nfssvc_srvcall(td, uap, td->td_ucred);
2947         }
2948         return (error);
2949 }
2950
2951 static int
2952 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
2953 {
2954         struct nfsex_args export;
2955         struct file *fp = NULL;
2956         int stablefd, len;
2957         struct nfsd_clid adminrevoke;
2958         struct nfsd_dumplist dumplist;
2959         struct nfsd_dumpclients *dumpclients;
2960         struct nfsd_dumplocklist dumplocklist;
2961         struct nfsd_dumplocks *dumplocks;
2962         struct nameidata nd;
2963         vnode_t vp;
2964         int error = EINVAL;
2965
2966         if (uap->flag & NFSSVC_PUBLICFH) {
2967                 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
2968                     sizeof (fhandle_t));
2969                 error = copyin(uap->argp,
2970                     &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
2971                 if (!error)
2972                         nfs_pubfhset = 1;
2973         } else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
2974                 error = copyin(uap->argp,(caddr_t)&export,
2975                     sizeof (struct nfsex_args));
2976                 if (!error)
2977                         error = nfsrv_v4rootexport(&export, cred, p);
2978         } else if (uap->flag & NFSSVC_NOPUBLICFH) {
2979                 nfs_pubfhset = 0;
2980                 error = 0;
2981         } else if (uap->flag & NFSSVC_STABLERESTART) {
2982                 error = copyin(uap->argp, (caddr_t)&stablefd,
2983                     sizeof (int));
2984                 if (!error)
2985                         error = fp_getfvp(p, stablefd, &fp, &vp);
2986                 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
2987                         error = EBADF;
2988                 if (!error && newnfs_numnfsd != 0)
2989                         error = EPERM;
2990                 if (!error) {
2991                         nfsrv_stablefirst.nsf_fp = fp;
2992                         nfsrv_setupstable(p);
2993                 }
2994         } else if (uap->flag & NFSSVC_ADMINREVOKE) {
2995                 error = copyin(uap->argp, (caddr_t)&adminrevoke,
2996                     sizeof (struct nfsd_clid));
2997                 if (!error)
2998                         error = nfsrv_adminrevoke(&adminrevoke, p);
2999         } else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3000                 error = copyin(uap->argp, (caddr_t)&dumplist,
3001                     sizeof (struct nfsd_dumplist));
3002                 if (!error && (dumplist.ndl_size < 1 ||
3003                         dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3004                         error = EPERM;
3005                 if (!error) {
3006                     len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3007                     dumpclients = (struct nfsd_dumpclients *)malloc(len,
3008                         M_TEMP, M_WAITOK);
3009                     nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3010                     error = copyout(dumpclients,
3011                         CAST_USER_ADDR_T(dumplist.ndl_list), len);
3012                     free((caddr_t)dumpclients, M_TEMP);
3013                 }
3014         } else if (uap->flag & NFSSVC_DUMPLOCKS) {
3015                 error = copyin(uap->argp, (caddr_t)&dumplocklist,
3016                     sizeof (struct nfsd_dumplocklist));
3017                 if (!error && (dumplocklist.ndllck_size < 1 ||
3018                         dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3019                         error = EPERM;
3020                 if (!error)
3021                         error = nfsrv_lookupfilename(&nd,
3022                                 dumplocklist.ndllck_fname, p);
3023                 if (!error) {
3024                         len = sizeof (struct nfsd_dumplocks) *
3025                                 dumplocklist.ndllck_size;
3026                         dumplocks = (struct nfsd_dumplocks *)malloc(len,
3027                                 M_TEMP, M_WAITOK);
3028                         nfsrv_dumplocks(nd.ni_vp, dumplocks,
3029                             dumplocklist.ndllck_size, p);
3030                         vput(nd.ni_vp);
3031                         error = copyout(dumplocks,
3032                             CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3033                         free((caddr_t)dumplocks, M_TEMP);
3034                 }
3035         }
3036         return (error);
3037 }
3038
3039 /*
3040  * Check exports.
3041  * Returns 0 if ok, 1 otherwise.
3042  */
3043 int
3044 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3045 {
3046         int i;
3047
3048         /*
3049          * This seems odd, but allow the case where the security flavor
3050          * list is empty. This happens when NFSv4 is traversing non-exported
3051          * file systems. Exported file systems should always have a non-empty
3052          * security flavor list.
3053          */
3054         if (exp->nes_numsecflavor == 0)
3055                 return (0);
3056
3057         for (i = 0; i < exp->nes_numsecflavor; i++) {
3058                 /*
3059                  * The tests for privacy and integrity must be first,
3060                  * since ND_GSS is set for everything but AUTH_SYS.
3061                  */
3062                 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3063                     (nd->nd_flag & ND_GSSPRIVACY))
3064                         return (0);
3065                 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3066                     (nd->nd_flag & ND_GSSINTEGRITY))
3067                         return (0);
3068                 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3069                     (nd->nd_flag & ND_GSS))
3070                         return (0);
3071                 if (exp->nes_secflavors[i] == AUTH_SYS &&
3072                     (nd->nd_flag & ND_GSS) == 0)
3073                         return (0);
3074         }
3075         return (1);
3076 }
3077
3078 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3079
3080 /*
3081  * Called once to initialize data structures...
3082  */
3083 static int
3084 nfsd_modevent(module_t mod, int type, void *data)
3085 {
3086         int error = 0;
3087         static int loaded = 0;
3088
3089         switch (type) {
3090         case MOD_LOAD:
3091                 if (loaded)
3092                         return (0);
3093                 newnfs_portinit();
3094                 mtx_init(&nfs_cache_mutex, "nfs_cache_mutex", NULL, MTX_DEF);
3095                 mtx_init(&nfs_v4root_mutex, "nfs_v4root_mutex", NULL, MTX_DEF);
3096                 mtx_init(&nfsv4root_mnt.mnt_mtx, "struct mount mtx", NULL,
3097                     MTX_DEF);
3098                 lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3099                 nfsrvd_initcache();
3100                 nfsd_init();
3101                 NFSD_LOCK();
3102                 nfsrvd_init(0);
3103                 NFSD_UNLOCK();
3104                 nfsd_mntinit();
3105 #ifdef VV_DISABLEDELEG
3106                 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3107                 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3108 #endif
3109                 nfsd_call_servertimer = nfsrv_servertimer;
3110                 nfsd_call_nfsd = nfssvc_nfsd;
3111                 loaded = 1;
3112                 break;
3113
3114         case MOD_UNLOAD:
3115                 if (newnfs_numnfsd != 0) {
3116                         error = EBUSY;
3117                         break;
3118                 }
3119
3120 #ifdef VV_DISABLEDELEG
3121                 vn_deleg_ops.vndeleg_recall = NULL;
3122                 vn_deleg_ops.vndeleg_disable = NULL;
3123 #endif
3124                 nfsd_call_servertimer = NULL;
3125                 nfsd_call_nfsd = NULL;
3126                 /* and get rid of the locks */
3127                 mtx_destroy(&nfs_cache_mutex);
3128                 mtx_destroy(&nfs_v4root_mutex);
3129                 mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3130                 lockdestroy(&nfsv4root_mnt.mnt_explock);
3131                 loaded = 0;
3132                 break;
3133         default:
3134                 error = EOPNOTSUPP;
3135                 break;
3136         }
3137         return error;
3138 }
3139 static moduledata_t nfsd_mod = {
3140         "nfsd",
3141         nfsd_modevent,
3142         NULL,
3143 };
3144 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3145
3146 /* So that loader and kldload(2) can find us, wherever we are.. */
3147 MODULE_VERSION(nfsd, 1);
3148 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3149 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3150