The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/fs/nfsserver/nfs_nfsdport.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    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: releng/8.0/sys/fs/nfsserver/nfs_nfsdport.c 195699 2009-07-14 22:48:30Z rwatson $");
   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_ACCESS() for two reasons.
  135  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case
  136  * 2 - The owner is to be given access irrespective of mode bits for some
  137  *     operations, so that processes that chmod after opening a file don't
  138  *     break.
  139  */
  140 int
  141 nfsvno_accchk(struct vnode *vp, u_int32_t accessbits, struct ucred *cred,
  142     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked)
  143 {
  144         struct vattr vattr;
  145         int error = 0, getret = 0;
  146         accmode_t accmode;
  147 
  148         /*
  149          * Convert accessbits to Vxxx flags.
  150          */
  151         if (accessbits & (NFSV4ACE_WRITEDATA | NFSV4ACE_APPENDDATA |
  152             NFSV4ACE_ADDFILE | NFSV4ACE_ADDSUBDIRECTORY |
  153             NFSV4ACE_DELETECHILD | NFSV4ACE_WRITEATTRIBUTES |
  154             NFSV4ACE_DELETE | NFSV4ACE_WRITEACL | NFSV4ACE_WRITEOWNER))
  155                 accmode = VWRITE;
  156         else if (accessbits & (NFSV4ACE_EXECUTE | NFSV4ACE_SEARCH))
  157                 accmode = VEXEC;
  158         else
  159                 accmode = VREAD;
  160 
  161         if (accmode & VWRITE) {
  162                 /* Just vn_writechk() changed to check rdonly */
  163                 /*
  164                  * Disallow write attempts on read-only file systems;
  165                  * unless the file is a socket or a block or character
  166                  * device resident on the file system.
  167                  */
  168                 if (NFSVNO_EXRDONLY(exp) ||
  169                         (vp->v_mount->mnt_flag & MNT_RDONLY)) {
  170                         switch (vp->v_type) {
  171                         case VREG:
  172                         case VDIR:
  173                         case VLNK:
  174                                 return (EROFS);
  175                         default:
  176                                 break;
  177                         }
  178                 }
  179                 /*
  180                  * If there's shared text associated with
  181                  * the inode, try to free it up once.  If
  182                  * we fail, we can't allow writing.
  183                  */
  184                 if (vp->v_vflag & VV_TEXT)
  185                         return (ETXTBSY);
  186         }
  187         if (vpislocked == 0)
  188                 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
  189 
  190 #if defined(NFS4_ACL_EXTATTR_NAME) && defined(notyet)
  191         /*
  192          * This function should be called once FFS has NFSv4 ACL support
  193          * in it.
  194          */
  195         /*
  196          * Should the override still be applied when ACLs are enabled?
  197          */
  198         if (nfsrv_useacl != 0 && NFSHASNFS4ACL(vp->v_mount))
  199                 error = nfsrv_aclaccess(vp, accmode, accessbits, cred, p);
  200         else
  201 #endif
  202         if (accessbits == NFSV4ACE_READATTRIBUTES)
  203                 error = 0;
  204         else
  205                 error = VOP_ACCESS(vp, accmode, cred, p);
  206 
  207         /*
  208          * Allow certain operations for the owner (reads and writes
  209          * on files that are already open).
  210          */
  211         if (override != NFSACCCHK_NOOVERRIDE &&
  212             (error == EPERM || error == EACCES)) {
  213                 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
  214                         error = 0;
  215                 else if (override & NFSACCCHK_ALLOWOWNER) {
  216                         getret = VOP_GETATTR(vp, &vattr, cred);
  217                         if (getret == 0 && cred->cr_uid == vattr.va_uid)
  218                                 error = 0;
  219                 }
  220         }
  221         if (vpislocked == 0)
  222                 NFSVOPUNLOCK(vp, 0, p);
  223         return (error);
  224 }
  225 
  226 /*
  227  * Set attribute(s) vnop.
  228  */
  229 int
  230 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
  231     struct thread *p, struct nfsexstuff *exp)
  232 {
  233         int error;
  234 
  235         error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
  236         return (error);
  237 }
  238 
  239 /*
  240  * Set up nameidata for a lookup() call and do it
  241  * For the cases where we are crossing mount points
  242  * (looking up the public fh path or the v4 root path when
  243  *  not using a pseudo-root fs), set/release the Giant lock,
  244  * as required.
  245  */
  246 int
  247 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
  248     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
  249     struct vnode **retdirp)
  250 {
  251         struct componentname *cnp = &ndp->ni_cnd;
  252         int i;
  253         struct iovec aiov;
  254         struct uio auio;
  255         int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
  256         int error = 0, crossmnt;
  257         char *cp;
  258 
  259         *retdirp = NULL;
  260         cnp->cn_nameptr = cnp->cn_pnbuf;
  261         /*
  262          * Extract and set starting directory.
  263          */
  264         if (dp->v_type != VDIR) {
  265                 if (islocked)
  266                         vput(dp);
  267                 else
  268                         vrele(dp);
  269                 nfsvno_relpathbuf(ndp);
  270                 return (ENOTDIR);
  271         }
  272         if (islocked)
  273                 NFSVOPUNLOCK(dp, 0, p);
  274         VREF(dp);
  275         *retdirp = dp;
  276         if (NFSVNO_EXRDONLY(exp))
  277                 cnp->cn_flags |= RDONLY;
  278         ndp->ni_segflg = UIO_SYSSPACE;
  279         crossmnt = 1;
  280 
  281         if (nd->nd_flag & ND_PUBLOOKUP) {
  282                 ndp->ni_loopcnt = 0;
  283                 if (cnp->cn_pnbuf[0] == '/') {
  284                         vrele(dp);
  285                         /*
  286                          * Check for degenerate pathnames here, since lookup()
  287                          * panics on them.
  288                          */
  289                         for (i = 1; i < ndp->ni_pathlen; i++)
  290                                 if (cnp->cn_pnbuf[i] != '/')
  291                                         break;
  292                         if (i == ndp->ni_pathlen) {
  293                                 error = NFSERR_ACCES;
  294                                 goto out;
  295                         }
  296                         dp = rootvnode;
  297                         VREF(dp);
  298                 }
  299         } else if ((enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
  300             (nd->nd_flag & ND_NFSV4) == 0) {
  301                 /*
  302                  * Only cross mount points for NFSv4 when doing a
  303                  * mount while traversing the file system above
  304                  * the mount point, unless enable_crossmntpt is set.
  305                  */
  306                 cnp->cn_flags |= NOCROSSMOUNT;
  307                 crossmnt = 0;
  308         }
  309 
  310         /*
  311          * Initialize for scan, set ni_startdir and bump ref on dp again
  312          * becuase lookup() will dereference ni_startdir.
  313          */
  314 
  315         cnp->cn_thread = p;
  316         ndp->ni_startdir = dp;
  317         ndp->ni_rootdir = rootvnode;
  318 
  319         if (!lockleaf)
  320                 cnp->cn_flags |= LOCKLEAF;
  321         for (;;) {
  322                 cnp->cn_nameptr = cnp->cn_pnbuf;
  323                 /*
  324                  * Call lookup() to do the real work.  If an error occurs,
  325                  * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
  326                  * we do not have to dereference anything before returning.
  327                  * In either case ni_startdir will be dereferenced and NULLed
  328                  * out.
  329                  */
  330                 if (exp->nes_vfslocked)
  331                         ndp->ni_cnd.cn_flags |= GIANTHELD;
  332                 error = lookup(ndp);
  333                 /*
  334                  * The Giant lock should only change when
  335                  * crossing mount points.
  336                  */
  337                 if (crossmnt) {
  338                         exp->nes_vfslocked =
  339                             (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
  340                         ndp->ni_cnd.cn_flags &= ~GIANTHELD;
  341                 }
  342                 if (error)
  343                         break;
  344 
  345                 /*
  346                  * Check for encountering a symbolic link.  Trivial
  347                  * termination occurs if no symlink encountered.
  348                  */
  349                 if ((cnp->cn_flags & ISSYMLINK) == 0) {
  350                         if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
  351                                 nfsvno_relpathbuf(ndp);
  352                         if (ndp->ni_vp && !lockleaf)
  353                                 NFSVOPUNLOCK(ndp->ni_vp, 0, p);
  354                         break;
  355                 }
  356 
  357                 /*
  358                  * Validate symlink
  359                  */
  360                 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
  361                         NFSVOPUNLOCK(ndp->ni_dvp, 0, p);
  362                 if (!(nd->nd_flag & ND_PUBLOOKUP)) {
  363                         error = EINVAL;
  364                         goto badlink2;
  365                 }
  366 
  367                 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
  368                         error = ELOOP;
  369                         goto badlink2;
  370                 }
  371                 if (ndp->ni_pathlen > 1)
  372                         cp = uma_zalloc(namei_zone, M_WAITOK);
  373                 else
  374                         cp = cnp->cn_pnbuf;
  375                 aiov.iov_base = cp;
  376                 aiov.iov_len = MAXPATHLEN;
  377                 auio.uio_iov = &aiov;
  378                 auio.uio_iovcnt = 1;
  379                 auio.uio_offset = 0;
  380                 auio.uio_rw = UIO_READ;
  381                 auio.uio_segflg = UIO_SYSSPACE;
  382                 auio.uio_td = NULL;
  383                 auio.uio_resid = MAXPATHLEN;
  384                 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
  385                 if (error) {
  386                 badlink1:
  387                         if (ndp->ni_pathlen > 1)
  388                                 uma_zfree(namei_zone, cp);
  389                 badlink2:
  390                         vrele(ndp->ni_dvp);
  391                         vput(ndp->ni_vp);
  392                         break;
  393                 }
  394                 linklen = MAXPATHLEN - auio.uio_resid;
  395                 if (linklen == 0) {
  396                         error = ENOENT;
  397                         goto badlink1;
  398                 }
  399                 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
  400                         error = ENAMETOOLONG;
  401                         goto badlink1;
  402                 }
  403 
  404                 /*
  405                  * Adjust or replace path
  406                  */
  407                 if (ndp->ni_pathlen > 1) {
  408                         NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
  409                         uma_zfree(namei_zone, cnp->cn_pnbuf);
  410                         cnp->cn_pnbuf = cp;
  411                 } else
  412                         cnp->cn_pnbuf[linklen] = '\0';
  413                 ndp->ni_pathlen += linklen;
  414 
  415                 /*
  416                  * Cleanup refs for next loop and check if root directory
  417                  * should replace current directory.  Normally ni_dvp
  418                  * becomes the new base directory and is cleaned up when
  419                  * we loop.  Explicitly null pointers after invalidation
  420                  * to clarify operation.
  421                  */
  422                 vput(ndp->ni_vp);
  423                 ndp->ni_vp = NULL;
  424 
  425                 if (cnp->cn_pnbuf[0] == '/') {
  426                         vrele(ndp->ni_dvp);
  427                         ndp->ni_dvp = ndp->ni_rootdir;
  428                         VREF(ndp->ni_dvp);
  429                 }
  430                 ndp->ni_startdir = ndp->ni_dvp;
  431                 ndp->ni_dvp = NULL;
  432         }
  433         if (!lockleaf)
  434                 cnp->cn_flags &= ~LOCKLEAF;
  435 
  436 out:
  437         if (error) {
  438                 uma_zfree(namei_zone, cnp->cn_pnbuf);
  439                 ndp->ni_vp = NULL;
  440                 ndp->ni_dvp = NULL;
  441                 ndp->ni_startdir = NULL;
  442                 cnp->cn_flags &= ~HASBUF;
  443         } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
  444                 ndp->ni_dvp = NULL;
  445         }
  446         return (error);
  447 }
  448 
  449 /*
  450  * Set up a pathname buffer and return a pointer to it and, optionally
  451  * set a hash pointer.
  452  */
  453 void
  454 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
  455 {
  456         struct componentname *cnp = &ndp->ni_cnd;
  457 
  458         cnp->cn_flags |= (NOMACCHECK | HASBUF);
  459         cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
  460         if (hashpp != NULL)
  461                 *hashpp = NULL;
  462         *bufpp = cnp->cn_pnbuf;
  463 }
  464 
  465 /*
  466  * Release the above path buffer, if not released by nfsvno_namei().
  467  */
  468 void
  469 nfsvno_relpathbuf(struct nameidata *ndp)
  470 {
  471 
  472         if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
  473                 panic("nfsrelpath");
  474         uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
  475         ndp->ni_cnd.cn_flags &= ~HASBUF;
  476 }
  477 
  478 /*
  479  * Readlink vnode op into an mbuf list.
  480  */
  481 int
  482 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
  483     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
  484 {
  485         struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
  486         struct iovec *ivp = iv;
  487         struct uio io, *uiop = &io;
  488         struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
  489         int i, len, tlen, error;
  490 
  491         len = 0;
  492         i = 0;
  493         while (len < NFS_MAXPATHLEN) {
  494                 NFSMGET(mp);
  495                 MCLGET(mp, M_WAIT);
  496                 mp->m_len = NFSMSIZ(mp);
  497                 if (len == 0) {
  498                         mp3 = mp2 = mp;
  499                 } else {
  500                         mp2->m_next = mp;
  501                         mp2 = mp;
  502                 }
  503                 if ((len + mp->m_len) > NFS_MAXPATHLEN) {
  504                         mp->m_len = NFS_MAXPATHLEN - len;
  505                         len = NFS_MAXPATHLEN;
  506                 } else {
  507                         len += mp->m_len;
  508                 }
  509                 ivp->iov_base = mtod(mp, caddr_t);
  510                 ivp->iov_len = mp->m_len;
  511                 i++;
  512                 ivp++;
  513         }
  514         uiop->uio_iov = iv;
  515         uiop->uio_iovcnt = i;
  516         uiop->uio_offset = 0;
  517         uiop->uio_resid = len;
  518         uiop->uio_rw = UIO_READ;
  519         uiop->uio_segflg = UIO_SYSSPACE;
  520         uiop->uio_td = NULL;
  521         error = VOP_READLINK(vp, uiop, cred);
  522         if (error) {
  523                 m_freem(mp3);
  524                 *lenp = 0;
  525                 return (error);
  526         }
  527         if (uiop->uio_resid > 0) {
  528                 len -= uiop->uio_resid;
  529                 tlen = NFSM_RNDUP(len);
  530                 nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
  531         }
  532         *lenp = len;
  533         *mpp = mp3;
  534         *mpendp = mp;
  535         return (0);
  536 }
  537 
  538 /*
  539  * Read vnode op call into mbuf list.
  540  */
  541 int
  542 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
  543     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
  544 {
  545         struct mbuf *m;
  546         int i;
  547         struct iovec *iv;
  548         struct iovec *iv2;
  549         int error = 0, len, left, siz, tlen, ioflag = 0, hi, try = 32;
  550         struct mbuf *m2 = NULL, *m3;
  551         struct uio io, *uiop = &io;
  552         struct nfsheur *nh;
  553 
  554         /*
  555          * Calculate seqcount for heuristic
  556          */
  557         /*
  558          * Locate best candidate
  559          */
  560 
  561         hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
  562         nh = &nfsheur[hi];
  563 
  564         while (try--) {
  565                 if (nfsheur[hi].nh_vp == vp) {
  566                         nh = &nfsheur[hi];
  567                         break;
  568                 }
  569                 if (nfsheur[hi].nh_use > 0)
  570                         --nfsheur[hi].nh_use;
  571                 hi = (hi + 1) % NUM_HEURISTIC;
  572                 if (nfsheur[hi].nh_use < nh->nh_use)
  573                         nh = &nfsheur[hi];
  574         }
  575 
  576         if (nh->nh_vp != vp) {
  577                 nh->nh_vp = vp;
  578                 nh->nh_nextr = off;
  579                 nh->nh_use = NHUSE_INIT;
  580                 if (off == 0)
  581                         nh->nh_seqcount = 4;
  582                 else
  583                         nh->nh_seqcount = 1;
  584         }
  585 
  586         /*
  587          * Calculate heuristic
  588          */
  589 
  590         if ((off == 0 && nh->nh_seqcount > 0) || off == nh->nh_nextr) {
  591                 if (++nh->nh_seqcount > IO_SEQMAX)
  592                         nh->nh_seqcount = IO_SEQMAX;
  593         } else if (nh->nh_seqcount > 1) {
  594                 nh->nh_seqcount = 1;
  595         } else {
  596                 nh->nh_seqcount = 0;
  597         }
  598         nh->nh_use += NHUSE_INC;
  599         if (nh->nh_use > NHUSE_MAX)
  600                 nh->nh_use = NHUSE_MAX;
  601         ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
  602 
  603         len = left = NFSM_RNDUP(cnt);
  604         m3 = NULL;
  605         /*
  606          * Generate the mbuf list with the uio_iov ref. to it.
  607          */
  608         i = 0;
  609         while (left > 0) {
  610                 NFSMGET(m);
  611                 MCLGET(m, M_WAIT);
  612                 m->m_len = 0;
  613                 siz = min(M_TRAILINGSPACE(m), left);
  614                 left -= siz;
  615                 i++;
  616                 if (m3)
  617                         m2->m_next = m;
  618                 else
  619                         m3 = m;
  620                 m2 = m;
  621         }
  622         MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
  623             M_TEMP, M_WAITOK);
  624         uiop->uio_iov = iv2 = iv;
  625         m = m3;
  626         left = len;
  627         i = 0;
  628         while (left > 0) {
  629                 if (m == NULL)
  630                         panic("nfsvno_read iov");
  631                 siz = min(M_TRAILINGSPACE(m), left);
  632                 if (siz > 0) {
  633                         iv->iov_base = mtod(m, caddr_t) + m->m_len;
  634                         iv->iov_len = siz;
  635                         m->m_len += siz;
  636                         left -= siz;
  637                         iv++;
  638                         i++;
  639                 }
  640                 m = m->m_next;
  641         }
  642         uiop->uio_iovcnt = i;
  643         uiop->uio_offset = off;
  644         uiop->uio_resid = len;
  645         uiop->uio_rw = UIO_READ;
  646         uiop->uio_segflg = UIO_SYSSPACE;
  647         error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
  648         FREE((caddr_t)iv2, M_TEMP);
  649         if (error) {
  650                 m_freem(m3);
  651                 *mpp = NULL;
  652                 return (error);
  653         }
  654         tlen = len - uiop->uio_resid;
  655         cnt = cnt < tlen ? cnt : tlen;
  656         tlen = NFSM_RNDUP(cnt);
  657         if (tlen == 0) {
  658                 m_freem(m3);
  659                 m3 = NULL;
  660         } else if (len != tlen || tlen != cnt)
  661                 nfsrv_adj(m3, len - tlen, tlen - cnt);
  662         *mpp = m3;
  663         *mpendp = m2;
  664         return (0);
  665 }
  666 
  667 /*
  668  * Write vnode op from an mbuf list.
  669  */
  670 int
  671 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
  672     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
  673 {
  674         struct iovec *ivp;
  675         int i, len;
  676         struct iovec *iv;
  677         int ioflags, error;
  678         struct uio io, *uiop = &io;
  679 
  680         MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
  681             M_WAITOK);
  682         uiop->uio_iov = iv = ivp;
  683         uiop->uio_iovcnt = cnt;
  684         i = mtod(mp, caddr_t) + mp->m_len - cp;
  685         len = retlen;
  686         while (len > 0) {
  687                 if (mp == NULL)
  688                         panic("nfsvno_write");
  689                 if (i > 0) {
  690                         i = min(i, len);
  691                         ivp->iov_base = cp;
  692                         ivp->iov_len = i;
  693                         ivp++;
  694                         len -= i;
  695                 }
  696                 mp = mp->m_next;
  697                 if (mp) {
  698                         i = mp->m_len;
  699                         cp = mtod(mp, caddr_t);
  700                 }
  701         }
  702 
  703         if (stable == NFSWRITE_UNSTABLE)
  704                 ioflags = IO_NODELOCKED;
  705         else
  706                 ioflags = (IO_SYNC | IO_NODELOCKED);
  707         uiop->uio_resid = retlen;
  708         uiop->uio_rw = UIO_WRITE;
  709         uiop->uio_segflg = UIO_SYSSPACE;
  710         NFSUIOPROC(uiop, p);
  711         uiop->uio_offset = off;
  712         error = VOP_WRITE(vp, uiop, ioflags, cred);
  713         FREE((caddr_t)iv, M_TEMP);
  714         return (error);
  715 }
  716 
  717 /*
  718  * Common code for creating a regular file (plus special files for V2).
  719  */
  720 int
  721 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
  722     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
  723     u_char *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
  724 {
  725         u_quad_t tempsize;
  726         int error;
  727 
  728         error = nd->nd_repstat;
  729         if (!error && ndp->ni_vp == NULL) {
  730                 if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
  731                         vrele(ndp->ni_startdir);
  732                         error = VOP_CREATE(ndp->ni_dvp,
  733                             &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
  734                         vput(ndp->ni_dvp);
  735                         nfsvno_relpathbuf(ndp);
  736                         if (!error) {
  737                                 if (*exclusive_flagp) {
  738                                         *exclusive_flagp = 0;
  739                                         NFSVNO_ATTRINIT(nvap);
  740                                         NFSBCOPY(cverf,(caddr_t)&nvap->na_atime,
  741                                             NFSX_VERF);
  742                                         error = VOP_SETATTR(ndp->ni_vp,
  743                                             &nvap->na_vattr, nd->nd_cred);
  744                                 }
  745                         }
  746                 /*
  747                  * NFS V2 Only. nfsrvd_mknod() does this for V3.
  748                  * (This implies, just get out on an error.)
  749                  */
  750                 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
  751                         nvap->na_type == VFIFO) {
  752                         if (nvap->na_type == VCHR && rdev == 0xffffffff)
  753                                 nvap->na_type = VFIFO;
  754                         if (nvap->na_type != VFIFO &&
  755                             (error = priv_check_cred(nd->nd_cred,
  756                              PRIV_VFS_MKNOD_DEV, 0))) {
  757                                 vrele(ndp->ni_startdir);
  758                                 nfsvno_relpathbuf(ndp);
  759                                 vput(ndp->ni_dvp);
  760                                 return (error);
  761                         }
  762                         nvap->na_rdev = rdev;
  763                         error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
  764                             &ndp->ni_cnd, &nvap->na_vattr);
  765                         vput(ndp->ni_dvp);
  766                         nfsvno_relpathbuf(ndp);
  767                         if (error) {
  768                                 vrele(ndp->ni_startdir);
  769                                 return (error);
  770                         }
  771                 } else {
  772                         vrele(ndp->ni_startdir);
  773                         nfsvno_relpathbuf(ndp);
  774                         vput(ndp->ni_dvp);
  775                         return (ENXIO);
  776                 }
  777                 *vpp = ndp->ni_vp;
  778         } else {
  779                 /*
  780                  * Handle cases where error is already set and/or
  781                  * the file exists.
  782                  * 1 - clean up the lookup
  783                  * 2 - iff !error and na_size set, truncate it
  784                  */
  785                 vrele(ndp->ni_startdir);
  786                 nfsvno_relpathbuf(ndp);
  787                 *vpp = ndp->ni_vp;
  788                 if (ndp->ni_dvp == *vpp)
  789                         vrele(ndp->ni_dvp);
  790                 else
  791                         vput(ndp->ni_dvp);
  792                 if (!error && nvap->na_size != VNOVAL) {
  793                         error = nfsvno_accchk(*vpp, NFSV4ACE_ADDFILE,
  794                             nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
  795                             NFSACCCHK_VPISLOCKED);
  796                         if (!error) {
  797                                 tempsize = nvap->na_size;
  798                                 NFSVNO_ATTRINIT(nvap);
  799                                 nvap->na_size = tempsize;
  800                                 error = VOP_SETATTR(*vpp,
  801                                     &nvap->na_vattr, nd->nd_cred);
  802                         }
  803                 }
  804                 if (error)
  805                         vput(*vpp);
  806         }
  807         return (error);
  808 }
  809 
  810 /*
  811  * Do a mknod vnode op.
  812  */
  813 int
  814 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
  815     struct thread *p)
  816 {
  817         int error = 0;
  818         enum vtype vtyp;
  819 
  820         vtyp = nvap->na_type;
  821         /*
  822          * Iff doesn't exist, create it.
  823          */
  824         if (ndp->ni_vp) {
  825                 vrele(ndp->ni_startdir);
  826                 nfsvno_relpathbuf(ndp);
  827                 vput(ndp->ni_dvp);
  828                 vrele(ndp->ni_vp);
  829                 return (EEXIST);
  830         }
  831         if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
  832                 vrele(ndp->ni_startdir);
  833                 nfsvno_relpathbuf(ndp);
  834                 vput(ndp->ni_dvp);
  835                 return (NFSERR_BADTYPE);
  836         }
  837         if (vtyp == VSOCK) {
  838                 vrele(ndp->ni_startdir);
  839                 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
  840                     &ndp->ni_cnd, &nvap->na_vattr);
  841                 vput(ndp->ni_dvp);
  842                 nfsvno_relpathbuf(ndp);
  843         } else {
  844                 if (nvap->na_type != VFIFO &&
  845                     (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
  846                         vrele(ndp->ni_startdir);
  847                         nfsvno_relpathbuf(ndp);
  848                         vput(ndp->ni_dvp);
  849                         return (error);
  850                 }
  851                 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
  852                     &ndp->ni_cnd, &nvap->na_vattr);
  853                 vput(ndp->ni_dvp);
  854                 nfsvno_relpathbuf(ndp);
  855                 if (error)
  856                         vrele(ndp->ni_startdir);
  857                 /*
  858                  * Since VOP_MKNOD returns the ni_vp, I can't
  859                  * see any reason to do the lookup.
  860                  */
  861         }
  862         return (error);
  863 }
  864 
  865 /*
  866  * Mkdir vnode op.
  867  */
  868 int
  869 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
  870     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
  871 {
  872         int error = 0;
  873 
  874         if (ndp->ni_vp != NULL) {
  875                 if (ndp->ni_dvp == ndp->ni_vp)
  876                         vrele(ndp->ni_dvp);
  877                 else
  878                         vput(ndp->ni_dvp);
  879                 vrele(ndp->ni_vp);
  880                 return (EEXIST);
  881         }
  882         error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
  883             &nvap->na_vattr);
  884         vput(ndp->ni_dvp);
  885         nfsvno_relpathbuf(ndp);
  886         return (error);
  887 }
  888 
  889 /*
  890  * symlink vnode op.
  891  */
  892 int
  893 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
  894     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
  895     struct nfsexstuff *exp)
  896 {
  897         int error = 0;
  898 
  899         if (ndp->ni_vp) {
  900                 vrele(ndp->ni_startdir);
  901                 nfsvno_relpathbuf(ndp);
  902                 if (ndp->ni_dvp == ndp->ni_vp)
  903                         vrele(ndp->ni_dvp);
  904                 else
  905                         vput(ndp->ni_dvp);
  906                 vrele(ndp->ni_vp);
  907                 return (EEXIST);
  908         }
  909 
  910         error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
  911             &nvap->na_vattr, pathcp);
  912         vput(ndp->ni_dvp);
  913         vrele(ndp->ni_startdir);
  914         nfsvno_relpathbuf(ndp);
  915         /*
  916          * Although FreeBSD still had the lookup code in
  917          * it for 7/current, there doesn't seem to be any
  918          * point, since VOP_SYMLINK() returns the ni_vp.
  919          * Just vput it for v2.
  920          */
  921         if (!not_v2 && !error)
  922                 vput(ndp->ni_vp);
  923         return (error);
  924 }
  925 
  926 /*
  927  * Parse symbolic link arguments.
  928  * This function has an ugly side effect. It will MALLOC() an area for
  929  * the symlink and set iov_base to point to it, only if it succeeds.
  930  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
  931  * be FREE'd later.
  932  */
  933 int
  934 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
  935     struct thread *p, char **pathcpp, int *lenp)
  936 {
  937         u_int32_t *tl;
  938         char *pathcp = NULL;
  939         int error = 0, len;
  940         struct nfsv2_sattr *sp;
  941 
  942         *pathcpp = NULL;
  943         *lenp = 0;
  944         if ((nd->nd_flag & ND_NFSV3) &&
  945             (error = nfsrv_sattr(nd, nvap, NULL, NULL, p)))
  946                 goto nfsmout;
  947         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
  948         len = fxdr_unsigned(int, *tl);
  949         if (len > NFS_MAXPATHLEN || len <= 0) {
  950                 error = EBADRPC;
  951                 goto nfsmout;
  952         }
  953         MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK);
  954         error = nfsrv_mtostr(nd, pathcp, len);
  955         if (error)
  956                 goto nfsmout;
  957         if (nd->nd_flag & ND_NFSV2) {
  958                 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
  959                 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
  960         }
  961         *pathcpp = pathcp;
  962         *lenp = len;
  963         return (0);
  964 nfsmout:
  965         if (pathcp)
  966                 free(pathcp, M_TEMP);
  967         return (error);
  968 }
  969 
  970 /*
  971  * Remove a non-directory object.
  972  */
  973 int
  974 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
  975     struct thread *p, struct nfsexstuff *exp)
  976 {
  977         struct vnode *vp;
  978         int error = 0;
  979 
  980         vp = ndp->ni_vp;
  981         if (vp->v_type == VDIR)
  982                 error = NFSERR_ISDIR;
  983         else if (is_v4)
  984                 error = nfsrv_checkremove(vp, 1, p);
  985         if (!error)
  986                 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
  987         if (ndp->ni_dvp == vp)
  988                 vrele(ndp->ni_dvp);
  989         else
  990                 vput(ndp->ni_dvp);
  991         vput(vp);
  992         return (error);
  993 }
  994 
  995 /*
  996  * Remove a directory.
  997  */
  998 int
  999 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
 1000     struct thread *p, struct nfsexstuff *exp)
 1001 {
 1002         struct vnode *vp;
 1003         int error = 0;
 1004 
 1005         vp = ndp->ni_vp;
 1006         if (vp->v_type != VDIR) {
 1007                 error = ENOTDIR;
 1008                 goto out;
 1009         }
 1010         /*
 1011          * No rmdir "." please.
 1012          */
 1013         if (ndp->ni_dvp == vp) {
 1014                 error = EINVAL;
 1015                 goto out;
 1016         }
 1017         /*
 1018          * The root of a mounted filesystem cannot be deleted.
 1019          */
 1020         if (vp->v_vflag & VV_ROOT)
 1021                 error = EBUSY;
 1022 out:
 1023         if (!error)
 1024                 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
 1025         if (ndp->ni_dvp == vp)
 1026                 vrele(ndp->ni_dvp);
 1027         else
 1028                 vput(ndp->ni_dvp);
 1029         vput(vp);
 1030         return (error);
 1031 }
 1032 
 1033 /*
 1034  * Rename vnode op.
 1035  */
 1036 int
 1037 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
 1038     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
 1039 {
 1040         struct vnode *fvp, *tvp, *tdvp;
 1041         int error = 0;
 1042 
 1043         fvp = fromndp->ni_vp;
 1044         if (ndstat) {
 1045                 vrele(fromndp->ni_dvp);
 1046                 vrele(fvp);
 1047                 error = ndstat;
 1048                 goto out1;
 1049         }
 1050         tdvp = tondp->ni_dvp;
 1051         tvp = tondp->ni_vp;
 1052         if (tvp != NULL) {
 1053                 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
 1054                         error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
 1055                         goto out;
 1056                 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
 1057                         error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
 1058                         goto out;
 1059                 }
 1060                 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
 1061                         error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
 1062                         goto out;
 1063                 }
 1064 
 1065                 /*
 1066                  * A rename to '.' or '..' results in a prematurely
 1067                  * unlocked vnode on FreeBSD5, so I'm just going to fail that
 1068                  * here.
 1069                  */
 1070                 if ((tondp->ni_cnd.cn_namelen == 1 &&
 1071                      tondp->ni_cnd.cn_nameptr[0] == '.') ||
 1072                     (tondp->ni_cnd.cn_namelen == 2 &&
 1073                      tondp->ni_cnd.cn_nameptr[0] == '.' &&
 1074                      tondp->ni_cnd.cn_nameptr[1] == '.')) {
 1075                         error = EINVAL;
 1076                         goto out;
 1077                 }
 1078         }
 1079         if (fvp->v_type == VDIR && fvp->v_mountedhere) {
 1080                 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
 1081                 goto out;
 1082         }
 1083         if (fvp->v_mount != tdvp->v_mount) {
 1084                 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
 1085                 goto out;
 1086         }
 1087         if (fvp == tdvp) {
 1088                 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
 1089                 goto out;
 1090         }
 1091         if (fvp == tvp) {
 1092                 /*
 1093                  * If source and destination are the same, there is nothing to
 1094                  * do. Set error to -1 to indicate this.
 1095                  */
 1096                 error = -1;
 1097                 goto out;
 1098         }
 1099         if (ndflag & ND_NFSV4) {
 1100                 NFSVOPLOCK(fvp, LK_EXCLUSIVE | LK_RETRY, p);
 1101                 error = nfsrv_checkremove(fvp, 0, p);
 1102                 NFSVOPUNLOCK(fvp, 0, p);
 1103                 if (tvp && !error)
 1104                         error = nfsrv_checkremove(tvp, 1, p);
 1105         } else {
 1106                 /*
 1107                  * For NFSv2 and NFSv3, try to get rid of the delegation, so
 1108                  * that the NFSv4 client won't be confused by the rename.
 1109                  * Since nfsd_recalldelegation() can only be called on an
 1110                  * unlocked vnode at this point and fvp is the file that will
 1111                  * still exist after the rename, just do fvp.
 1112                  */
 1113                 nfsd_recalldelegation(fvp, p);
 1114         }
 1115 out:
 1116         if (!error) {
 1117                 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
 1118                     &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
 1119                     &tondp->ni_cnd);
 1120         } else {
 1121                 if (tdvp == tvp)
 1122                         vrele(tdvp);
 1123                 else
 1124                         vput(tdvp);
 1125                 if (tvp)
 1126                         vput(tvp);
 1127                 vrele(fromndp->ni_dvp);
 1128                 vrele(fvp);
 1129                 if (error == -1)
 1130                         error = 0;
 1131         }
 1132         vrele(tondp->ni_startdir);
 1133         nfsvno_relpathbuf(tondp);
 1134 out1:
 1135         vrele(fromndp->ni_startdir);
 1136         nfsvno_relpathbuf(fromndp);
 1137         return (error);
 1138 }
 1139 
 1140 /*
 1141  * Link vnode op.
 1142  */
 1143 int
 1144 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
 1145     struct thread *p, struct nfsexstuff *exp)
 1146 {
 1147         struct vnode *xp;
 1148         int error = 0;
 1149 
 1150         xp = ndp->ni_vp;
 1151         if (xp != NULL) {
 1152                 error = EEXIST;
 1153         } else {
 1154                 xp = ndp->ni_dvp;
 1155                 if (vp->v_mount != xp->v_mount)
 1156                         error = EXDEV;
 1157         }
 1158         if (!error) {
 1159                 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
 1160                 error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
 1161                 if (ndp->ni_dvp == vp)
 1162                         vrele(ndp->ni_dvp);
 1163                 else
 1164                         vput(ndp->ni_dvp);
 1165                 NFSVOPUNLOCK(vp, 0, p);
 1166         } else {
 1167                 if (ndp->ni_dvp == ndp->ni_vp)
 1168                         vrele(ndp->ni_dvp);
 1169                 else
 1170                         vput(ndp->ni_dvp);
 1171                 if (ndp->ni_vp)
 1172                         vrele(ndp->ni_vp);
 1173         }
 1174         nfsvno_relpathbuf(ndp);
 1175         return (error);
 1176 }
 1177 
 1178 /*
 1179  * Do the fsync() appropriate for the commit.
 1180  */
 1181 int
 1182 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
 1183     struct thread *td)
 1184 {
 1185         int error = 0;
 1186 
 1187         if (cnt > MAX_COMMIT_COUNT) {
 1188                 /*
 1189                  * Give up and do the whole thing
 1190                  */
 1191                 if (vp->v_object &&
 1192                    (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
 1193                         VM_OBJECT_LOCK(vp->v_object);
 1194                         vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
 1195                         VM_OBJECT_UNLOCK(vp->v_object);
 1196                 }
 1197                 error = VOP_FSYNC(vp, MNT_WAIT, td);
 1198         } else {
 1199                 /*
 1200                  * Locate and synchronously write any buffers that fall
 1201                  * into the requested range.  Note:  we are assuming that
 1202                  * f_iosize is a power of 2.
 1203                  */
 1204                 int iosize = vp->v_mount->mnt_stat.f_iosize;
 1205                 int iomask = iosize - 1;
 1206                 struct bufobj *bo;
 1207                 daddr_t lblkno;
 1208 
 1209                 /*
 1210                  * Align to iosize boundry, super-align to page boundry.
 1211                  */
 1212                 if (off & iomask) {
 1213                         cnt += off & iomask;
 1214                         off &= ~(u_quad_t)iomask;
 1215                 }
 1216                 if (off & PAGE_MASK) {
 1217                         cnt += off & PAGE_MASK;
 1218                         off &= ~(u_quad_t)PAGE_MASK;
 1219                 }
 1220                 lblkno = off / iosize;
 1221 
 1222                 if (vp->v_object &&
 1223                    (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
 1224                         VM_OBJECT_LOCK(vp->v_object);
 1225                         vm_object_page_clean(vp->v_object, off / PAGE_SIZE, (cnt + PAGE_MASK) / PAGE_SIZE, OBJPC_SYNC);
 1226                         VM_OBJECT_UNLOCK(vp->v_object);
 1227                 }
 1228 
 1229                 bo = &vp->v_bufobj;
 1230                 BO_LOCK(bo);
 1231                 while (cnt > 0) {
 1232                         struct buf *bp;
 1233 
 1234                         /*
 1235                          * If we have a buffer and it is marked B_DELWRI we
 1236                          * have to lock and write it.  Otherwise the prior
 1237                          * write is assumed to have already been committed.
 1238                          *
 1239                          * gbincore() can return invalid buffers now so we
 1240                          * have to check that bit as well (though B_DELWRI
 1241                          * should not be set if B_INVAL is set there could be
 1242                          * a race here since we haven't locked the buffer).
 1243                          */
 1244                         if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
 1245                                 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
 1246                                     LK_INTERLOCK, BO_MTX(bo)) == ENOLCK) {
 1247                                         BO_LOCK(bo);
 1248                                         continue; /* retry */
 1249                                 }
 1250                                 if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
 1251                                     B_DELWRI) {
 1252                                         bremfree(bp);
 1253                                         bp->b_flags &= ~B_ASYNC;
 1254                                         bwrite(bp);
 1255                                         ++nfs_commit_miss;
 1256                                 } else
 1257                                         BUF_UNLOCK(bp);
 1258                                 BO_LOCK(bo);
 1259                         }
 1260                         ++nfs_commit_blks;
 1261                         if (cnt < iosize)
 1262                                 break;
 1263                         cnt -= iosize;
 1264                         ++lblkno;
 1265                 }
 1266                 BO_UNLOCK(bo);
 1267         }
 1268         return (error);
 1269 }
 1270 
 1271 /*
 1272  * Statfs vnode op.
 1273  */
 1274 int
 1275 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
 1276 {
 1277 
 1278         return (VFS_STATFS(vp->v_mount, sf));
 1279 }
 1280 
 1281 /*
 1282  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
 1283  * must handle nfsrv_opencheck() calls after any other access checks.
 1284  */
 1285 void
 1286 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
 1287     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
 1288     int *exclusive_flagp, struct nfsvattr *nvap, u_char *cverf, int create,
 1289     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
 1290     struct nfsexstuff *exp, struct vnode **vpp)
 1291 {
 1292         struct vnode *vp = NULL;
 1293         u_quad_t tempsize;
 1294         struct nfsexstuff nes;
 1295 
 1296         if (ndp->ni_vp == NULL)
 1297                 nd->nd_repstat = nfsrv_opencheck(clientid,
 1298                     stateidp, stp, NULL, nd, p, nd->nd_repstat);
 1299         if (!nd->nd_repstat) {
 1300                 if (ndp->ni_vp == NULL) {
 1301                         vrele(ndp->ni_startdir);
 1302                         nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
 1303                             &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
 1304                         vput(ndp->ni_dvp);
 1305                         nfsvno_relpathbuf(ndp);
 1306                         if (!nd->nd_repstat) {
 1307                                 if (*exclusive_flagp) {
 1308                                         *exclusive_flagp = 0;
 1309                                         NFSVNO_ATTRINIT(nvap);
 1310                                         NFSBCOPY(cverf,
 1311                                             (caddr_t)&nvap->na_atime,
 1312                                             NFSX_VERF);
 1313                                         nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
 1314                                             &nvap->na_vattr, cred);
 1315                                 } else {
 1316                                         nfsrv_fixattr(nd, ndp->ni_vp, nvap,
 1317                                             aclp, p, attrbitp, exp);
 1318                                 }
 1319                         }
 1320                         vp = ndp->ni_vp;
 1321                 } else {
 1322                         if (ndp->ni_startdir)
 1323                                 vrele(ndp->ni_startdir);
 1324                         nfsvno_relpathbuf(ndp);
 1325                         vp = ndp->ni_vp;
 1326                         if (create == NFSV4OPEN_CREATE) {
 1327                                 if (ndp->ni_dvp == vp)
 1328                                         vrele(ndp->ni_dvp);
 1329                                 else
 1330                                         vput(ndp->ni_dvp);
 1331                         }
 1332                         if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
 1333                                 if (ndp->ni_cnd.cn_flags & RDONLY)
 1334                                         NFSVNO_SETEXRDONLY(&nes);
 1335                                 else
 1336                                         NFSVNO_EXINIT(&nes);
 1337                                 nd->nd_repstat = nfsvno_accchk(vp, 
 1338                                     NFSV4ACE_ADDFILE, cred, &nes, p,
 1339                                     NFSACCCHK_NOOVERRIDE,NFSACCCHK_VPISLOCKED);
 1340                                 nd->nd_repstat = nfsrv_opencheck(clientid,
 1341                                     stateidp, stp, vp, nd, p, nd->nd_repstat);
 1342                                 if (!nd->nd_repstat) {
 1343                                         tempsize = nvap->na_size;
 1344                                         NFSVNO_ATTRINIT(nvap);
 1345                                         nvap->na_size = tempsize;
 1346                                         nd->nd_repstat = VOP_SETATTR(vp,
 1347                                             &nvap->na_vattr, cred);
 1348                                 }
 1349                         } else if (vp->v_type == VREG) {
 1350                                 nd->nd_repstat = nfsrv_opencheck(clientid,
 1351                                     stateidp, stp, vp, nd, p, nd->nd_repstat);
 1352                         }
 1353                 }
 1354         } else {
 1355                 if (ndp->ni_cnd.cn_flags & HASBUF)
 1356                         nfsvno_relpathbuf(ndp);
 1357                 if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
 1358                         vrele(ndp->ni_startdir);
 1359                         if (ndp->ni_dvp == ndp->ni_vp)
 1360                                 vrele(ndp->ni_dvp);
 1361                         else
 1362                                 vput(ndp->ni_dvp);
 1363                         if (ndp->ni_vp)
 1364                                 vput(ndp->ni_vp);
 1365                 }
 1366         }
 1367         *vpp = vp;
 1368 }
 1369 
 1370 /*
 1371  * Updates the file rev and sets the mtime and ctime
 1372  * to the current clock time, returning the va_filerev and va_Xtime
 1373  * values.
 1374  */
 1375 void
 1376 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
 1377     struct ucred *cred, struct thread *p)
 1378 {
 1379         struct vattr va;
 1380 
 1381         VATTR_NULL(&va);
 1382         getnanotime(&va.va_mtime);
 1383         (void) VOP_SETATTR(vp, &va, cred);
 1384         (void) nfsvno_getattr(vp, nvap, cred, p);
 1385 }
 1386 
 1387 /*
 1388  * Glue routine to nfsv4_fillattr().
 1389  */
 1390 int
 1391 nfsvno_fillattr(struct nfsrv_descript *nd, struct vnode *vp,
 1392     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
 1393     struct ucred *cred, struct thread *p, int isdgram, int reterr)
 1394 {
 1395         int error;
 1396 
 1397         error = nfsv4_fillattr(nd, vp, NULL, &nvap->na_vattr, fhp, rderror,
 1398             attrbitp, cred, p, isdgram, reterr);
 1399         return (error);
 1400 }
 1401 
 1402 /* Since the Readdir vnode ops vary, put the entire functions in here. */
 1403 /*
 1404  * nfs readdir service
 1405  * - mallocs what it thinks is enough to read
 1406  *      count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
 1407  * - calls nfsvno_readdir()
 1408  * - loops around building the reply
 1409  *      if the output generated exceeds count break out of loop
 1410  *      The NFSM_CLGET macro is used here so that the reply will be packed
 1411  *      tightly in mbuf clusters.
 1412  * - it only knows that it has encountered eof when the nfsvno_readdir()
 1413  *      reads nothing
 1414  * - as such one readdir rpc will return eof false although you are there
 1415  *      and then the next will return eof
 1416  * - it trims out records with d_fileno == 0
 1417  *      this doesn't matter for Unix clients, but they might confuse clients
 1418  *      for other os'.
 1419  * - it trims out records with d_type == DT_WHT
 1420  *      these cannot be seen through NFS (unless we extend the protocol)
 1421  * NB: It is tempting to set eof to true if the nfsvno_readdir() reads less
 1422  *      than requested, but this may not apply to all filesystems. For
 1423  *      example, client NFS does not { although it is never remote mounted
 1424  *      anyhow }
 1425  *     The alternate call nfsrvd_readdirplus() does lookups as well.
 1426  * PS: The NFS protocol spec. does not clarify what the "count" byte
 1427  *      argument is a count of.. just name strings and file id's or the
 1428  *      entire reply rpc or ...
 1429  *      I tried just file name and id sizes and it confused the Sun client,
 1430  *      so I am using the full rpc size now. The "paranoia.." comment refers
 1431  *      to including the status longwords that are not a part of the dir.
 1432  *      "entry" structures, but are in the rpc.
 1433  */
 1434 int
 1435 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
 1436     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
 1437 {
 1438         struct dirent *dp;
 1439         u_int32_t *tl;
 1440         int dirlen;
 1441         char *cpos, *cend, *rbuf;
 1442         struct nfsvattr at;
 1443         int nlen, error = 0, getret = 1;
 1444         int siz, cnt, fullsiz, eofflag, ncookies;
 1445         u_int64_t off, toff, verf;
 1446         u_long *cookies = NULL, *cookiep;
 1447         struct uio io;
 1448         struct iovec iv;
 1449 
 1450         if (nd->nd_repstat) {
 1451                 nfsrv_postopattr(nd, getret, &at);
 1452                 return (0);
 1453         }
 1454         if (nd->nd_flag & ND_NFSV2) {
 1455                 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
 1456                 off = fxdr_unsigned(u_quad_t, *tl++);
 1457         } else {
 1458                 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
 1459                 off = fxdr_hyper(tl);
 1460                 tl += 2;
 1461                 verf = fxdr_hyper(tl);
 1462                 tl += 2;
 1463         }
 1464         toff = off;
 1465         cnt = fxdr_unsigned(int, *tl);
 1466         if (cnt > NFS_SRVMAXDATA(nd))
 1467                 cnt = NFS_SRVMAXDATA(nd);
 1468         siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
 1469         fullsiz = siz;
 1470         if (nd->nd_flag & ND_NFSV3) {
 1471                 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
 1472                     p);
 1473 #if 0
 1474                 /*
 1475                  * va_filerev is not sufficient as a cookie verifier,
 1476                  * since it is not supposed to change when entries are
 1477                  * removed/added unless that offset cookies returned to
 1478                  * the client are no longer valid.
 1479                  */
 1480                 if (!nd->nd_repstat && toff && verf != at.na_filerev)
 1481                         nd->nd_repstat = NFSERR_BAD_COOKIE;
 1482 #endif
 1483         }
 1484         if (!nd->nd_repstat)
 1485                 nd->nd_repstat = nfsvno_accchk(vp, NFSV4ACE_SEARCH,
 1486                     nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
 1487                     NFSACCCHK_VPISLOCKED);
 1488         if (nd->nd_repstat) {
 1489                 vput(vp);
 1490                 if (nd->nd_flag & ND_NFSV3)
 1491                         nfsrv_postopattr(nd, getret, &at);
 1492                 return (0);
 1493         }
 1494         NFSVOPUNLOCK(vp, 0, p);
 1495         MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
 1496 again:
 1497         eofflag = 0;
 1498         if (cookies) {
 1499                 free((caddr_t)cookies, M_TEMP);
 1500                 cookies = NULL;
 1501         }
 1502 
 1503         iv.iov_base = rbuf;
 1504         iv.iov_len = siz;
 1505         io.uio_iov = &iv;
 1506         io.uio_iovcnt = 1;
 1507         io.uio_offset = (off_t)off;
 1508         io.uio_resid = siz;
 1509         io.uio_segflg = UIO_SYSSPACE;
 1510         io.uio_rw = UIO_READ;
 1511         io.uio_td = NULL;
 1512         NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
 1513         nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
 1514             &cookies);
 1515         NFSVOPUNLOCK(vp, 0, p);
 1516         off = (u_int64_t)io.uio_offset;
 1517         if (io.uio_resid)
 1518                 siz -= io.uio_resid;
 1519 
 1520         if (!cookies && !nd->nd_repstat)
 1521                 nd->nd_repstat = NFSERR_PERM;
 1522         if (nd->nd_flag & ND_NFSV3) {
 1523                 getret = nfsvno_getattr(vp, &at, nd->nd_cred, p);
 1524                 if (!nd->nd_repstat)
 1525                         nd->nd_repstat = getret;
 1526         }
 1527 
 1528         /*
 1529          * Handles the failed cases. nd->nd_repstat == 0 past here.
 1530          */
 1531         if (nd->nd_repstat) {
 1532                 vrele(vp);
 1533                 free((caddr_t)rbuf, M_TEMP);
 1534                 if (cookies)
 1535                         free((caddr_t)cookies, M_TEMP);
 1536                 if (nd->nd_flag & ND_NFSV3)
 1537                         nfsrv_postopattr(nd, getret, &at);
 1538                 return (0);
 1539         }
 1540         /*
 1541          * If nothing read, return eof
 1542          * rpc reply
 1543          */
 1544         if (siz == 0) {
 1545                 vrele(vp);
 1546                 if (nd->nd_flag & ND_NFSV2) {
 1547                         NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
 1548                 } else {
 1549                         nfsrv_postopattr(nd, getret, &at);
 1550                         NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
 1551                         txdr_hyper(at.na_filerev, tl);
 1552                         tl += 2;
 1553                 }
 1554                 *tl++ = newnfs_false;
 1555                 *tl = newnfs_true;
 1556                 FREE((caddr_t)rbuf, M_TEMP);
 1557                 FREE((caddr_t)cookies, M_TEMP);
 1558                 return (0);
 1559         }
 1560 
 1561         /*
 1562          * Check for degenerate cases of nothing useful read.
 1563          * If so go try again
 1564          */
 1565         cpos = rbuf;
 1566         cend = rbuf + siz;
 1567         dp = (struct dirent *)cpos;
 1568         cookiep = cookies;
 1569 
 1570         /*
 1571          * For some reason FreeBSD's ufs_readdir() chooses to back the
 1572          * directory offset up to a block boundary, so it is necessary to
 1573          * skip over the records that precede the requested offset. This
 1574          * requires the assumption that file offset cookies monotonically
 1575          * increase.
 1576          */
 1577         while (cpos < cend && ncookies > 0 &&
 1578             (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
 1579              ((u_quad_t)(*cookiep)) <= toff)) {
 1580                 cpos += dp->d_reclen;
 1581                 dp = (struct dirent *)cpos;
 1582                 cookiep++;
 1583                 ncookies--;
 1584         }
 1585         if (cpos >= cend || ncookies == 0) {
 1586                 siz = fullsiz;
 1587                 toff = off;
 1588                 goto again;
 1589         }
 1590 
 1591         /*
 1592          * dirlen is the size of the reply, including all XDR and must
 1593          * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
 1594          * if the XDR should be included in "count", but to be safe, we do.
 1595          * (Include the two booleans at the end of the reply in dirlen now.)
 1596          */
 1597         if (nd->nd_flag & ND_NFSV3) {
 1598                 nfsrv_postopattr(nd, getret, &at);
 1599                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
 1600                 txdr_hyper(at.na_filerev, tl);
 1601                 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
 1602         } else {
 1603                 dirlen = 2 * NFSX_UNSIGNED;
 1604         }
 1605 
 1606         /* Loop through the records and build reply */
 1607         while (cpos < cend && ncookies > 0) {
 1608                 nlen = dp->d_namlen;
 1609                 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
 1610                         nlen <= NFS_MAXNAMLEN) {
 1611                         if (nd->nd_flag & ND_NFSV3)
 1612                                 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
 1613                         else
 1614                                 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
 1615                         if (dirlen > cnt) {
 1616                                 eofflag = 0;
 1617                                 break;
 1618                         }
 1619 
 1620                         /*
 1621                          * Build the directory record xdr from
 1622                          * the dirent entry.
 1623                          */
 1624                         if (nd->nd_flag & ND_NFSV3) {
 1625                                 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
 1626                                 *tl++ = newnfs_true;
 1627                                 *tl++ = 0;
 1628                         } else {
 1629                                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
 1630                                 *tl++ = newnfs_true;
 1631                         }
 1632                         *tl = txdr_unsigned(dp->d_fileno);
 1633                         (void) nfsm_strtom(nd, dp->d_name, nlen);
 1634                         if (nd->nd_flag & ND_NFSV3) {
 1635                                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
 1636                                 *tl++ = 0;
 1637                         } else
 1638                                 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
 1639                         *tl = txdr_unsigned(*cookiep);
 1640                 }
 1641                 cpos += dp->d_reclen;
 1642                 dp = (struct dirent *)cpos;
 1643                 cookiep++;
 1644                 ncookies--;
 1645         }
 1646         if (cpos < cend)
 1647                 eofflag = 0;
 1648         vrele(vp);
 1649         NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
 1650         *tl++ = newnfs_false;
 1651         if (eofflag)
 1652                 *tl = newnfs_true;
 1653         else
 1654                 *tl = newnfs_false;
 1655         FREE((caddr_t)rbuf, M_TEMP);
 1656         FREE((caddr_t)cookies, M_TEMP);
 1657         return (0);
 1658 nfsmout:
 1659         vput(vp);
 1660         return (error);
 1661 }
 1662 
 1663 /*
 1664  * Readdirplus for V3 and Readdir for V4.
 1665  */
 1666 int
 1667 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
 1668     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
 1669 {
 1670         struct dirent *dp;
 1671         u_int32_t *tl;
 1672         int dirlen;
 1673         char *cpos, *cend, *rbuf;
 1674         struct vnode *nvp;
 1675         fhandle_t nfh;
 1676         struct nfsvattr nva, at, *nvap = &nva;
 1677         struct mbuf *mb0, *mb1;
 1678         struct nfsreferral *refp;
 1679         int nlen, r, error = 0, getret = 1, vgetret;
 1680         int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
 1681         caddr_t bpos0, bpos1;
 1682         u_int64_t off, toff, verf;
 1683         u_long *cookies = NULL, *cookiep;
 1684         nfsattrbit_t attrbits, rderrbits, savbits;
 1685         struct uio io;
 1686         struct iovec iv;
 1687 
 1688         if (nd->nd_repstat) {
 1689                 nfsrv_postopattr(nd, getret, &at);
 1690                 return (0);
 1691         }
 1692         NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
 1693         off = fxdr_hyper(tl);
 1694         toff = off;
 1695         tl += 2;
 1696         verf = fxdr_hyper(tl);
 1697         tl += 2;
 1698         siz = fxdr_unsigned(int, *tl++);
 1699         cnt = fxdr_unsigned(int, *tl);
 1700 
 1701         /*
 1702          * Use the server's maximum data transfer size as the upper bound
 1703          * on reply datalen.
 1704          */
 1705         if (cnt > NFS_SRVMAXDATA(nd))
 1706                 cnt = NFS_SRVMAXDATA(nd);
 1707 
 1708         /*
 1709          * siz is a "hint" of how much directory information (name, fileid,
 1710          * cookie) should be in the reply. At least one client "hints" 0,
 1711          * so I set it to cnt for that case. I also round it up to the
 1712          * next multiple of DIRBLKSIZ.
 1713          */
 1714         if (siz == 0)
 1715                 siz = cnt;
 1716         siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
 1717 
 1718         if (nd->nd_flag & ND_NFSV4) {
 1719                 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
 1720                 if (error)
 1721                         goto nfsmout;
 1722                 NFSSET_ATTRBIT(&savbits, &attrbits);
 1723                 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
 1724                 NFSZERO_ATTRBIT(&rderrbits);
 1725                 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
 1726         } else {
 1727                 NFSZERO_ATTRBIT(&attrbits);
 1728         }
 1729         fullsiz = siz;
 1730         nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p);
 1731         if (!nd->nd_repstat) {
 1732             if (off && verf != at.na_filerev) {
 1733                 /*
 1734                  * va_filerev is not sufficient as a cookie verifier,
 1735                  * since it is not supposed to change when entries are
 1736                  * removed/added unless that offset cookies returned to
 1737                  * the client are no longer valid.
 1738                  */
 1739 #if 0
 1740                 if (nd->nd_flag & ND_NFSV4) {
 1741                         nd->nd_repstat = NFSERR_NOTSAME;
 1742                 } else {
 1743                         nd->nd_repstat = NFSERR_BAD_COOKIE;
 1744                 }
 1745 #endif
 1746             } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
 1747                 nd->nd_repstat = NFSERR_BAD_COOKIE;
 1748             }
 1749         }
 1750         if (!nd->nd_repstat && vp->v_type != VDIR)
 1751                 nd->nd_repstat = NFSERR_NOTDIR;
 1752         if (!nd->nd_repstat && cnt == 0)
 1753                 nd->nd_repstat = NFSERR_TOOSMALL;
 1754         if (!nd->nd_repstat)
 1755                 nd->nd_repstat = nfsvno_accchk(vp, NFSV4ACE_SEARCH,
 1756                     nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
 1757                     NFSACCCHK_VPISLOCKED);
 1758         if (nd->nd_repstat) {
 1759                 vput(vp);
 1760                 if (nd->nd_flag & ND_NFSV3)
 1761                         nfsrv_postopattr(nd, getret, &at);
 1762                 return (0);
 1763         }
 1764 
 1765         NFSVOPUNLOCK(vp, 0, p);
 1766 
 1767         MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
 1768 again:
 1769         eofflag = 0;
 1770         if (cookies) {
 1771                 free((caddr_t)cookies, M_TEMP);
 1772                 cookies = NULL;
 1773         }
 1774 
 1775         iv.iov_base = rbuf;
 1776         iv.iov_len = siz;
 1777         io.uio_iov = &iv;
 1778         io.uio_iovcnt = 1;
 1779         io.uio_offset = (off_t)off;
 1780         io.uio_resid = siz;
 1781         io.uio_segflg = UIO_SYSSPACE;
 1782         io.uio_rw = UIO_READ;
 1783         io.uio_td = NULL;
 1784         NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
 1785         nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
 1786             &cookies);
 1787         NFSVOPUNLOCK(vp, 0, p);
 1788         off = (u_int64_t)io.uio_offset;
 1789         if (io.uio_resid)
 1790                 siz -= io.uio_resid;
 1791 
 1792         getret = nfsvno_getattr(vp, &at, nd->nd_cred, p);
 1793 
 1794         if (!cookies && !nd->nd_repstat)
 1795                 nd->nd_repstat = NFSERR_PERM;
 1796         if (!nd->nd_repstat)
 1797                 nd->nd_repstat = getret;
 1798         if (nd->nd_repstat) {
 1799                 vrele(vp);
 1800                 if (cookies)
 1801                         free((caddr_t)cookies, M_TEMP);
 1802                 free((caddr_t)rbuf, M_TEMP);
 1803                 if (nd->nd_flag & ND_NFSV3)
 1804                         nfsrv_postopattr(nd, getret, &at);
 1805                 return (0);
 1806         }
 1807         /*
 1808          * If nothing read, return eof
 1809          * rpc reply
 1810          */
 1811         if (siz == 0) {
 1812                 vrele(vp);
 1813                 if (nd->nd_flag & ND_NFSV3)
 1814                         nfsrv_postopattr(nd, getret, &at);
 1815                 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
 1816                 txdr_hyper(at.na_filerev, tl);
 1817                 tl += 2;
 1818                 *tl++ = newnfs_false;
 1819                 *tl = newnfs_true;
 1820                 free((caddr_t)cookies, M_TEMP);
 1821                 free((caddr_t)rbuf, M_TEMP);
 1822                 return (0);
 1823         }
 1824 
 1825         /*
 1826          * Check for degenerate cases of nothing useful read.
 1827          * If so go try again
 1828          */
 1829         cpos = rbuf;
 1830         cend = rbuf + siz;
 1831         dp = (struct dirent *)cpos;
 1832         cookiep = cookies;
 1833 
 1834         /*
 1835          * For some reason FreeBSD's ufs_readdir() chooses to back the
 1836          * directory offset up to a block boundary, so it is necessary to
 1837          * skip over the records that precede the requested offset. This
 1838          * requires the assumption that file offset cookies monotonically
 1839          * increase.
 1840          */
 1841         while (cpos < cend && ncookies > 0 &&
 1842           (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
 1843            ((u_quad_t)(*cookiep)) <= toff ||
 1844            ((nd->nd_flag & ND_NFSV4) &&
 1845             ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
 1846              (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
 1847                 cpos += dp->d_reclen;
 1848                 dp = (struct dirent *)cpos;
 1849                 cookiep++;
 1850                 ncookies--;
 1851         }
 1852         if (cpos >= cend || ncookies == 0) {
 1853                 siz = fullsiz;
 1854                 toff = off;
 1855                 goto again;
 1856         }
 1857 
 1858         /*
 1859          * Probe one of the directory entries to see if the filesystem
 1860          * supports VGET for NFSv3. For NFSv4, it will return an
 1861          * error later, if attributes are required.
 1862          * (To be honest, most if not all NFSv4 clients will require
 1863          *  attributes, but??)
 1864          */
 1865         if ((nd->nd_flag & ND_NFSV3)) {
 1866                 vgetret = VFS_VGET(vp->v_mount, dp->d_fileno, LK_EXCLUSIVE,
 1867                     &nvp);
 1868                 if (vgetret != 0) {
 1869                         if (vgetret == EOPNOTSUPP)
 1870                                 nd->nd_repstat = NFSERR_NOTSUPP;
 1871                         else
 1872                                 nd->nd_repstat = NFSERR_SERVERFAULT;
 1873                         vrele(vp);
 1874                         if (cookies)
 1875                                 free((caddr_t)cookies, M_TEMP);
 1876                         free((caddr_t)rbuf, M_TEMP);
 1877                         nfsrv_postopattr(nd, getret, &at);
 1878                         return (0);
 1879                 }
 1880                 if (!vgetret)
 1881                         vput(nvp);
 1882                 nvp = NULL;
 1883         }
 1884 
 1885         /*
 1886          * Save this position, in case there is an error before one entry
 1887          * is created.
 1888          */
 1889         mb0 = nd->nd_mb;
 1890         bpos0 = nd->nd_bpos;
 1891 
 1892         /*
 1893          * Fill in the first part of the reply.
 1894          * dirlen is the reply length in bytes and cannot exceed cnt.
 1895          * (Include the two booleans at the end of the reply in dirlen now,
 1896          *  so we recognize when we have exceeded cnt.)
 1897          */
 1898         if (nd->nd_flag & ND_NFSV3) {
 1899                 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
 1900                 nfsrv_postopattr(nd, getret, &at);
 1901         } else {
 1902                 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
 1903         }
 1904         NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
 1905         txdr_hyper(at.na_filerev, tl);
 1906 
 1907         /*
 1908          * Save this position, in case there is an empty reply needed.
 1909          */
 1910         mb1 = nd->nd_mb;
 1911         bpos1 = nd->nd_bpos;
 1912 
 1913         /* Loop through the records and build reply */
 1914         entrycnt = 0;
 1915         while (cpos < cend && ncookies > 0 && dirlen < cnt) {
 1916                 nlen = dp->d_namlen;
 1917                 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
 1918                     nlen <= NFS_MAXNAMLEN &&
 1919                     ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
 1920                      (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
 1921                       || (nlen == 1 && dp->d_name[0] != '.'))) {
 1922                         /*
 1923                          * Save the current position in the reply, in case
 1924                          * this entry exceeds cnt.
 1925                          */
 1926                         mb1 = nd->nd_mb;
 1927                         bpos1 = nd->nd_bpos;
 1928         
 1929                         /*
 1930                          * For readdir_and_lookup get the vnode using
 1931                          * the file number.
 1932                          */
 1933                         nvp = NULL;
 1934                         refp = NULL;
 1935                         r = 0;
 1936                         if ((nd->nd_flag & ND_NFSV3) ||
 1937                             NFSNONZERO_ATTRBIT(&savbits)) {
 1938                                 if (nd->nd_flag & ND_NFSV4)
 1939                                         refp = nfsv4root_getreferral(NULL,
 1940                                             vp, dp->d_fileno);
 1941                                 if (refp == NULL)
 1942                                         r = VFS_VGET(vp->v_mount, dp->d_fileno,
 1943                                             LK_EXCLUSIVE, &nvp);
 1944                                 if (!r) {
 1945                                     if (refp == NULL &&
 1946                                         ((nd->nd_flag & ND_NFSV3) ||
 1947                                          NFSNONZERO_ATTRBIT(&attrbits))) {
 1948                                         r = nfsvno_getfh(nvp, &nfh, p);
 1949                                         if (!r)
 1950                                             r = nfsvno_getattr(nvp, nvap,
 1951                                                 nd->nd_cred, p);
 1952                                     }
 1953                                 } else {
 1954                                     nvp = NULL;
 1955                                 }
 1956                                 if (r) {
 1957                                         if (!NFSISSET_ATTRBIT(&attrbits,
 1958                                             NFSATTRBIT_RDATTRERROR)) {
 1959                                                 if (nvp != NULL)
 1960                                                         vput(nvp);
 1961                                                 nd->nd_repstat = r;
 1962                                                 break;
 1963                                         }
 1964                                 }
 1965                         }
 1966 
 1967                         /*
 1968                          * Build the directory record xdr
 1969                          */
 1970                         if (nd->nd_flag & ND_NFSV3) {
 1971                                 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
 1972                                 *tl++ = newnfs_true;
 1973                                 *tl++ = 0;
 1974                                 *tl = txdr_unsigned(dp->d_fileno);
 1975                                 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
 1976                                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
 1977                                 *tl++ = 0;
 1978                                 *tl = txdr_unsigned(*cookiep);
 1979                                 nfsrv_postopattr(nd, 0, nvap);
 1980                                 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
 1981                                 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
 1982                                 if (nvp != NULL)
 1983                                         vput(nvp);
 1984                         } else {
 1985                                 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
 1986                                 *tl++ = newnfs_true;
 1987                                 *tl++ = 0;
 1988                                 *tl = txdr_unsigned(*cookiep);
 1989                                 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
 1990                                 if (nvp != NULL)
 1991                                         NFSVOPUNLOCK(nvp, 0, p);
 1992                                 if (refp != NULL) {
 1993                                         dirlen += nfsrv_putreferralattr(nd,
 1994                                             &savbits, refp, 0,
 1995                                             &nd->nd_repstat);
 1996                                         if (nd->nd_repstat) {
 1997                                                 if (nvp != NULL)
 1998                                                         vrele(nvp);
 1999                                                 break;
 2000                                         }
 2001                                 } else if (r) {
 2002                                         dirlen += nfsvno_fillattr(nd, nvp, nvap,
 2003                                             &nfh, r, &rderrbits, nd->nd_cred,
 2004                                             p, isdgram, 0);
 2005                                 } else {
 2006                                         dirlen += nfsvno_fillattr(nd, nvp, nvap,
 2007                                             &nfh, r, &attrbits, nd->nd_cred,
 2008                                             p, isdgram, 0);
 2009                                 }
 2010                                 if (nvp != NULL)
 2011                                         vrele(nvp);
 2012                                 dirlen += (3 * NFSX_UNSIGNED);
 2013                         }
 2014                         if (dirlen <= cnt)
 2015                                 entrycnt++;
 2016                 }
 2017                 cpos += dp->d_reclen;
 2018                 dp = (struct dirent *)cpos;
 2019                 cookiep++;
 2020                 ncookies--;
 2021         }
 2022         vrele(vp);
 2023 
 2024         /*
 2025          * If dirlen > cnt, we must strip off the last entry. If that
 2026          * results in an empty reply, report NFSERR_TOOSMALL.
 2027          */
 2028         if (dirlen > cnt || nd->nd_repstat) {
 2029                 if (!nd->nd_repstat && entrycnt == 0)
 2030                         nd->nd_repstat = NFSERR_TOOSMALL;
 2031                 if (nd->nd_repstat)
 2032                         newnfs_trimtrailing(nd, mb0, bpos0);
 2033                 else
 2034                         newnfs_trimtrailing(nd, mb1, bpos1);
 2035                 eofflag = 0;
 2036         } else if (cpos < cend)
 2037                 eofflag = 0;
 2038         if (!nd->nd_repstat) {
 2039                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
 2040                 *tl++ = newnfs_false;
 2041                 if (eofflag)
 2042                         *tl = newnfs_true;
 2043                 else
 2044                         *tl = newnfs_false;
 2045         }
 2046         FREE((caddr_t)cookies, M_TEMP);
 2047         FREE((caddr_t)rbuf, M_TEMP);
 2048         return (0);
 2049 nfsmout:
 2050         vput(vp);
 2051         return (error);
 2052 }
 2053 
 2054 /*
 2055  * Get the settable attributes out of the mbuf list.
 2056  * (Return 0 or EBADRPC)
 2057  */
 2058 int
 2059 nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
 2060     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
 2061 {
 2062         u_int32_t *tl;
 2063         struct nfsv2_sattr *sp;
 2064         struct timeval curtime;
 2065         int error = 0, toclient = 0;
 2066 
 2067         switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
 2068         case ND_NFSV2:
 2069                 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
 2070                 /*
 2071                  * Some old clients didn't fill in the high order 16bits.
 2072                  * --> check the low order 2 bytes for 0xffff
 2073                  */
 2074                 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
 2075                         nvap->na_mode = nfstov_mode(sp->sa_mode);
 2076                 if (sp->sa_uid != newnfs_xdrneg1)
 2077                         nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
 2078                 if (sp->sa_gid != newnfs_xdrneg1)
 2079                         nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
 2080                 if (sp->sa_size != newnfs_xdrneg1)
 2081                         nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
 2082                 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
 2083 #ifdef notyet
 2084                         fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
 2085 #else
 2086                         nvap->na_atime.tv_sec =
 2087                                 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
 2088                         nvap->na_atime.tv_nsec = 0;
 2089 #endif
 2090                 }
 2091                 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
 2092                         fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
 2093                 break;
 2094         case ND_NFSV3:
 2095                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2096                 if (*tl == newnfs_true) {
 2097                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2098                         nvap->na_mode = nfstov_mode(*tl);
 2099                 }
 2100                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2101                 if (*tl == newnfs_true) {
 2102                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2103                         nvap->na_uid = fxdr_unsigned(uid_t, *tl);
 2104                 }
 2105                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2106                 if (*tl == newnfs_true) {
 2107                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2108                         nvap->na_gid = fxdr_unsigned(gid_t, *tl);
 2109                 }
 2110                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2111                 if (*tl == newnfs_true) {
 2112                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
 2113                         nvap->na_size = fxdr_hyper(tl);
 2114                 }
 2115                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2116                 switch (fxdr_unsigned(int, *tl)) {
 2117                 case NFSV3SATTRTIME_TOCLIENT:
 2118                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
 2119                         fxdr_nfsv3time(tl, &nvap->na_atime);
 2120                         toclient = 1;
 2121                         break;
 2122                 case NFSV3SATTRTIME_TOSERVER:
 2123                         NFSGETTIME(&curtime);
 2124                         nvap->na_atime.tv_sec = curtime.tv_sec;
 2125                         nvap->na_atime.tv_nsec = curtime.tv_usec * 1000;
 2126                         nvap->na_vaflags |= VA_UTIMES_NULL;
 2127                         break;
 2128                 };
 2129                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2130                 switch (fxdr_unsigned(int, *tl)) {
 2131                 case NFSV3SATTRTIME_TOCLIENT:
 2132                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
 2133                         fxdr_nfsv3time(tl, &nvap->na_mtime);
 2134                         nvap->na_vaflags &= ~VA_UTIMES_NULL;
 2135                         break;
 2136                 case NFSV3SATTRTIME_TOSERVER:
 2137                         NFSGETTIME(&curtime);
 2138                         nvap->na_mtime.tv_sec = curtime.tv_sec;
 2139                         nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000;
 2140                         if (!toclient)
 2141                                 nvap->na_vaflags |= VA_UTIMES_NULL;
 2142                         break;
 2143                 };
 2144                 break;
 2145         case ND_NFSV4:
 2146                 error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p);
 2147         };
 2148 nfsmout:
 2149         return (error);
 2150 }
 2151 
 2152 /*
 2153  * Handle the setable attributes for V4.
 2154  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
 2155  */
 2156 int
 2157 nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
 2158     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
 2159 {
 2160         u_int32_t *tl;
 2161         int attrsum = 0;
 2162         int i, j;
 2163         int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
 2164         int toclient = 0;
 2165         u_char *cp, namestr[NFSV4_SMALLSTR + 1];
 2166         uid_t uid;
 2167         gid_t gid;
 2168         struct timeval curtime;
 2169 
 2170         error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
 2171         if (error)
 2172                 return (error);
 2173         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2174         attrsize = fxdr_unsigned(int, *tl);
 2175 
 2176         /*
 2177          * Loop around getting the setable attributes. If an unsupported
 2178          * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
 2179          */
 2180         if (retnotsup) {
 2181                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
 2182                 bitpos = NFSATTRBIT_MAX;
 2183         } else {
 2184                 bitpos = 0;
 2185         }
 2186         for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
 2187             if (attrsum > attrsize) {
 2188                 error = NFSERR_BADXDR;
 2189                 goto nfsmout;
 2190             }
 2191             if (NFSISSET_ATTRBIT(attrbitp, bitpos))
 2192                 switch (bitpos) {
 2193                 case NFSATTRBIT_SIZE:
 2194                         NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
 2195                         nvap->na_size = fxdr_hyper(tl);
 2196                         attrsum += NFSX_HYPER;
 2197                         break;
 2198                 case NFSATTRBIT_ACL:
 2199                         error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
 2200                             p);
 2201                         if (error)
 2202                                 goto nfsmout;
 2203                         if (aceerr && !nd->nd_repstat)
 2204                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
 2205                         attrsum += aclsize;
 2206                         break;
 2207                 case NFSATTRBIT_ARCHIVE:
 2208                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2209                         if (!nd->nd_repstat)
 2210                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
 2211                         attrsum += NFSX_UNSIGNED;
 2212                         break;
 2213                 case NFSATTRBIT_HIDDEN:
 2214                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2215                         if (!nd->nd_repstat)
 2216                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
 2217                         attrsum += NFSX_UNSIGNED;
 2218                         break;
 2219                 case NFSATTRBIT_MIMETYPE:
 2220                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2221                         i = fxdr_unsigned(int, *tl);
 2222                         error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
 2223                         if (error)
 2224                                 goto nfsmout;
 2225                         if (!nd->nd_repstat)
 2226                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
 2227                         attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
 2228                         break;
 2229                 case NFSATTRBIT_MODE:
 2230                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2231                         nvap->na_mode = nfstov_mode(*tl);
 2232                         attrsum += NFSX_UNSIGNED;
 2233                         break;
 2234                 case NFSATTRBIT_OWNER:
 2235                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2236                         j = fxdr_unsigned(int, *tl);
 2237                         if (j < 0)
 2238                                 return (NFSERR_BADXDR);
 2239                         if (j > NFSV4_SMALLSTR)
 2240                                 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
 2241                         else
 2242                                 cp = namestr;
 2243                         error = nfsrv_mtostr(nd, cp, j);
 2244                         if (error) {
 2245                                 if (j > NFSV4_SMALLSTR)
 2246                                         free(cp, M_NFSSTRING);
 2247                                 return (error);
 2248                         }
 2249                         if (!nd->nd_repstat) {
 2250                                 nd->nd_repstat = nfsv4_strtouid(cp,j,&uid,p);
 2251                                 if (!nd->nd_repstat)
 2252                                         nvap->na_uid = uid;
 2253                         }
 2254                         if (j > NFSV4_SMALLSTR)
 2255                                 free(cp, M_NFSSTRING);
 2256                         attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
 2257                         break;
 2258                 case NFSATTRBIT_OWNERGROUP:
 2259                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2260                         j = fxdr_unsigned(int, *tl);
 2261                         if (j < 0)
 2262                                 return (NFSERR_BADXDR);
 2263                         if (j > NFSV4_SMALLSTR)
 2264                                 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
 2265                         else
 2266                                 cp = namestr;
 2267                         error = nfsrv_mtostr(nd, cp, j);
 2268                         if (error) {
 2269                                 if (j > NFSV4_SMALLSTR)
 2270                                         free(cp, M_NFSSTRING);
 2271                                 return (error);
 2272                         }
 2273                         if (!nd->nd_repstat) {
 2274                                 nd->nd_repstat = nfsv4_strtogid(cp,j,&gid,p);
 2275                                 if (!nd->nd_repstat)
 2276                                         nvap->na_gid = gid;
 2277                         }
 2278                         if (j > NFSV4_SMALLSTR)
 2279                                 free(cp, M_NFSSTRING);
 2280                         attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
 2281                         break;
 2282                 case NFSATTRBIT_SYSTEM:
 2283                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2284                         if (!nd->nd_repstat)
 2285                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
 2286                         attrsum += NFSX_UNSIGNED;
 2287                         break;
 2288                 case NFSATTRBIT_TIMEACCESSSET:
 2289                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2290                         attrsum += NFSX_UNSIGNED;
 2291                         if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
 2292                             NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
 2293                             fxdr_nfsv4time(tl, &nvap->na_atime);
 2294                             toclient = 1;
 2295                             attrsum += NFSX_V4TIME;
 2296                         } else {
 2297                             NFSGETTIME(&curtime);
 2298                             nvap->na_atime.tv_sec = curtime.tv_sec;
 2299                             nvap->na_atime.tv_nsec = curtime.tv_usec * 1000;
 2300                             nvap->na_vaflags |= VA_UTIMES_NULL;
 2301                         }
 2302                         break;
 2303                 case NFSATTRBIT_TIMEBACKUP:
 2304                         NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
 2305                         if (!nd->nd_repstat)
 2306                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
 2307                         attrsum += NFSX_V4TIME;
 2308                         break;
 2309                 case NFSATTRBIT_TIMECREATE:
 2310                         NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
 2311                         if (!nd->nd_repstat)
 2312                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
 2313                         attrsum += NFSX_V4TIME;
 2314                         break;
 2315                 case NFSATTRBIT_TIMEMODIFYSET:
 2316                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 2317                         attrsum += NFSX_UNSIGNED;
 2318                         if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
 2319                             NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
 2320                             fxdr_nfsv4time(tl, &nvap->na_mtime);
 2321                             nvap->na_vaflags &= ~VA_UTIMES_NULL;
 2322                             attrsum += NFSX_V4TIME;
 2323                         } else {
 2324                             NFSGETTIME(&curtime);
 2325                             nvap->na_mtime.tv_sec = curtime.tv_sec;
 2326                             nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000;
 2327                             if (!toclient)
 2328                                 nvap->na_vaflags |= VA_UTIMES_NULL;
 2329                         }
 2330                         break;
 2331                 default:
 2332                         nd->nd_repstat = NFSERR_ATTRNOTSUPP;
 2333                         /*
 2334                          * set bitpos so we drop out of the loop.
 2335                          */
 2336                         bitpos = NFSATTRBIT_MAX;
 2337                         break;
 2338                 };
 2339         }
 2340 
 2341         /*
 2342          * some clients pad the attrlist, so we need to skip over the
 2343          * padding.
 2344          */
 2345         if (attrsum > attrsize) {
 2346                 error = NFSERR_BADXDR;
 2347         } else {
 2348                 attrsize = NFSM_RNDUP(attrsize);
 2349                 if (attrsum < attrsize)
 2350                         error = nfsm_advance(nd, attrsize - attrsum, -1);
 2351         }
 2352 nfsmout:
 2353         return (error);
 2354 }
 2355 
 2356 /*
 2357  * Check/setup export credentials.
 2358  */
 2359 int
 2360 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
 2361     struct ucred *credanon)
 2362 {
 2363         int error = 0;
 2364 
 2365         /*
 2366          * Check/setup credentials.
 2367          */
 2368         if (nd->nd_flag & ND_GSS)
 2369                 exp->nes_exflag &= ~MNT_EXPORTANON;
 2370 
 2371         /*
 2372          * Check to see if the operation is allowed for this security flavor.
 2373          * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
 2374          * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
 2375          * Also, allow Secinfo, so that it can acquire the correct flavor(s).
 2376          */
 2377         if (nfsvno_testexp(nd, exp) &&
 2378             nd->nd_procnum != NFSV4OP_SECINFO &&
 2379             nd->nd_procnum != NFSPROC_FSINFO) {
 2380                 if (nd->nd_flag & ND_NFSV4)
 2381                         error = NFSERR_WRONGSEC;
 2382                 else
 2383                         error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
 2384                 return (error);
 2385         }
 2386 
 2387         /*
 2388          * Check to see if the file system is exported V4 only.
 2389          */
 2390         if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4))
 2391                 return (NFSERR_PROGNOTV4);
 2392 
 2393         /*
 2394          * Now, map the user credentials.
 2395          * (Note that ND_AUTHNONE will only be set for an NFSv3
 2396          *  Fsinfo RPC. If set for anything else, this code might need
 2397          *  to change.)
 2398          */
 2399         if (NFSVNO_EXPORTED(exp) &&
 2400             ((!(nd->nd_flag & ND_GSS) && nd->nd_cred->cr_uid == 0) ||
 2401              NFSVNO_EXPORTANON(exp) ||
 2402              (nd->nd_flag & ND_AUTHNONE))) {
 2403                 nd->nd_cred->cr_uid = credanon->cr_uid;
 2404                 nd->nd_cred->cr_gid = credanon->cr_gid;
 2405                 crsetgroups(nd->nd_cred, credanon->cr_ngroups,
 2406                     credanon->cr_groups);
 2407         }
 2408         return (0);
 2409 }
 2410 
 2411 /*
 2412  * Check exports.
 2413  */
 2414 int
 2415 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
 2416     struct ucred **credp)
 2417 {
 2418         int i, error, *secflavors;
 2419 
 2420         error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
 2421             &exp->nes_numsecflavor, &secflavors);
 2422         if (error) {
 2423                 if (nfs_rootfhset) {
 2424                         exp->nes_exflag = 0;
 2425                         exp->nes_numsecflavor = 0;
 2426                         error = 0;
 2427                 }
 2428         } else {
 2429                 /* Copy the security flavors. */
 2430                 for (i = 0; i < exp->nes_numsecflavor; i++)
 2431                         exp->nes_secflavors[i] = secflavors[i];
 2432         }
 2433         return (error);
 2434 }
 2435 
 2436 /*
 2437  * Get a vnode for a file handle and export stuff.
 2438  */
 2439 int
 2440 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
 2441     struct vnode **vpp, struct nfsexstuff *exp, struct ucred **credp)
 2442 {
 2443         int i, error, *secflavors;
 2444 
 2445         *credp = NULL;
 2446         exp->nes_numsecflavor = 0;
 2447         error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp);
 2448         if (nam && !error) {
 2449                 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
 2450                     &exp->nes_numsecflavor, &secflavors);
 2451                 if (error) {
 2452                         if (nfs_rootfhset) {
 2453                                 exp->nes_exflag = 0;
 2454                                 exp->nes_numsecflavor = 0;
 2455                                 error = 0;
 2456                         } else {
 2457                                 vput(*vpp);
 2458                         }
 2459                 } else {
 2460                         /* Copy the security flavors. */
 2461                         for (i = 0; i < exp->nes_numsecflavor; i++)
 2462                                 exp->nes_secflavors[i] = secflavors[i];
 2463                 }
 2464         }
 2465         return (error);
 2466 }
 2467 
 2468 /*
 2469  * Do the pathconf vnode op.
 2470  */
 2471 int
 2472 nfsvno_pathconf(struct vnode *vp, int flag, register_t *retf,
 2473     struct ucred *cred, struct thread *p)
 2474 {
 2475         int error;
 2476 
 2477         error = VOP_PATHCONF(vp, flag, retf);
 2478         return (error);
 2479 }
 2480 
 2481 /*
 2482  * nfsd_fhtovp() - convert a fh to a vnode ptr
 2483  *      - look up fsid in mount list (if not found ret error)
 2484  *      - get vp and export rights by calling nfsvno_fhtovp()
 2485  *      - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
 2486  *        for AUTH_SYS
 2487  * Also handle getting the Giant lock for the file system,
 2488  * as required:
 2489  * - if same mount point as *mpp
 2490  *       do nothing
 2491  *   else if *mpp == NULL
 2492  *       if already locked
 2493  *           leave it locked
 2494  *       else
 2495  *           call VFS_LOCK_GIANT()
 2496  *   else
 2497  *       if already locked
 2498  *            unlock Giant
 2499  *       call VFS_LOCK_GIANT()
 2500  */
 2501 void
 2502 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp,
 2503     struct vnode **vpp, struct nfsexstuff *exp,
 2504     struct mount **mpp, int startwrite, struct thread *p)
 2505 {
 2506         struct mount *mp;
 2507         struct ucred *credanon;
 2508         fhandle_t *fhp;
 2509 
 2510         fhp = (fhandle_t *)nfp->nfsrvfh_data;
 2511         /*
 2512          * Check for the special case of the nfsv4root_fh.
 2513          */
 2514         mp = vfs_getvfs(&fhp->fh_fsid);
 2515         if (!mp) {
 2516                 *vpp = NULL;
 2517                 nd->nd_repstat = ESTALE;
 2518                 if (*mpp && exp->nes_vfslocked)
 2519                         VFS_UNLOCK_GIANT(*mpp);
 2520                 *mpp = NULL;
 2521                 exp->nes_vfslocked = 0;
 2522                 return;
 2523         }
 2524 
 2525         /*
 2526          * Now, handle Giant for the file system.
 2527          */
 2528         if (*mpp != NULL && *mpp != mp && exp->nes_vfslocked) {
 2529                 VFS_UNLOCK_GIANT(*mpp);
 2530                 exp->nes_vfslocked = 0;
 2531         }
 2532         if (!exp->nes_vfslocked && *mpp != mp)
 2533                 exp->nes_vfslocked = VFS_LOCK_GIANT(mp);
 2534 
 2535         *mpp = mp;
 2536         if (startwrite)
 2537                 vn_start_write(NULL, mpp, V_WAIT);
 2538 
 2539         nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, vpp, exp,
 2540             &credanon);
 2541 
 2542         /*
 2543          * For NFSv4 without a pseudo root fs, unexported file handles
 2544          * can be returned, so that Lookup works everywhere.
 2545          */
 2546         if (!nd->nd_repstat && exp->nes_exflag == 0 &&
 2547             !(nd->nd_flag & ND_NFSV4)) {
 2548                 vput(*vpp);
 2549                 nd->nd_repstat = EACCES;
 2550         }
 2551 
 2552         /*
 2553          * Personally, I've never seen any point in requiring a
 2554          * reserved port#, since only in the rare case where the
 2555          * clients are all boxes with secure system priviledges,
 2556          * does it provide any enhanced security, but... some people
 2557          * believe it to be useful and keep putting this code back in.
 2558          * (There is also some "security checker" out there that
 2559          *  complains if the nfs server doesn't enforce this.)
 2560          * However, note the following:
 2561          * RFC3530 (NFSv4) specifies that a reserved port# not be
 2562          *      required.
 2563          * RFC2623 recommends that, if a reserved port# is checked for,
 2564          *      that there be a way to turn that off--> ifdef'd.
 2565          */
 2566 #ifdef NFS_REQRSVPORT
 2567         if (!nd->nd_repstat) {
 2568                 struct sockaddr_in *saddr;
 2569                 struct sockaddr_in6 *saddr6;
 2570 
 2571                 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
 2572                 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
 2573                 if (!(nd->nd_flag & ND_NFSV4) &&
 2574                     ((saddr->sin_family == AF_INET &&
 2575                       ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
 2576                      (saddr6->sin6_family == AF_INET6 &&
 2577                       ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
 2578                         vput(*vpp);
 2579                         nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
 2580                 }
 2581         }
 2582 #endif  /* NFS_REQRSVPORT */
 2583 
 2584         /*
 2585          * Check/setup credentials.
 2586          */
 2587         if (!nd->nd_repstat) {
 2588                 nd->nd_saveduid = nd->nd_cred->cr_uid;
 2589                 nd->nd_repstat = nfsd_excred(nd, exp, credanon);
 2590                 if (nd->nd_repstat)
 2591                         vput(*vpp);
 2592         }
 2593         if (credanon != NULL)
 2594                 crfree(credanon);
 2595         if (nd->nd_repstat) {
 2596                 if (startwrite)
 2597                         vn_finished_write(mp);
 2598                 if (exp->nes_vfslocked) {
 2599                         VFS_UNLOCK_GIANT(mp);
 2600                         exp->nes_vfslocked = 0;
 2601                 }
 2602                 vfs_rel(mp);
 2603                 *vpp = NULL;
 2604                 *mpp = NULL;
 2605         } else {
 2606                 vfs_rel(mp);
 2607         }
 2608 }
 2609 
 2610 /*
 2611  * glue for fp.
 2612  */
 2613 int
 2614 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
 2615 {
 2616         struct filedesc *fdp;
 2617         struct file *fp;
 2618 
 2619         fdp = p->td_proc->p_fd;
 2620         if (fd >= fdp->fd_nfiles ||
 2621             (fp = fdp->fd_ofiles[fd]) == NULL)
 2622                 return (EBADF);
 2623         *fpp = fp;
 2624         return (0);
 2625 }
 2626 
 2627 /*
 2628  * Called from nfssvc() to update the exports list. Just call
 2629  * vfs_export(). This has to be done, since the v4 root fake fs isn't
 2630  * in the mount list.
 2631  */
 2632 int
 2633 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
 2634 {
 2635         struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
 2636         int error;
 2637         struct nameidata nd;
 2638         fhandle_t fh;
 2639 
 2640         error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
 2641         if ((nfsexargp->export.ex_flags & MNT_DELEXPORT)) {
 2642                 nfs_rootfhset = 0;
 2643                 nfsv4root_set = 0;
 2644         } else if (error == 0) {
 2645                 if (nfsexargp->fspec == NULL)
 2646                         return (EPERM);
 2647                 /*
 2648                  * If fspec != NULL, this is the v4root path.
 2649                  */
 2650                 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_USERSPACE,
 2651                     nfsexargp->fspec, p);
 2652                 if ((error = namei(&nd)) != 0)
 2653                         return (error);
 2654                 error = nfsvno_getfh(nd.ni_vp, &fh, p);
 2655                 vrele(nd.ni_vp);
 2656                 if (!error) {
 2657                         nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
 2658                         NFSBCOPY((caddr_t)&fh,
 2659                             nfs_rootfh.nfsrvfh_data,
 2660                             sizeof (fhandle_t));
 2661                         nfs_rootfhset = 1;
 2662                 }
 2663         }
 2664         return (error);
 2665 }
 2666 
 2667 /*
 2668  * Get the tcp socket sequence numbers we need.
 2669  * (Maybe this should be moved to the tcp sources?)
 2670  */
 2671 int
 2672 nfsrv_getsocksndseq(struct socket *so, tcp_seq *maxp, tcp_seq *unap)
 2673 {
 2674         struct inpcb *inp;
 2675         struct tcpcb *tp;
 2676         int error = EPIPE;
 2677 
 2678         INP_INFO_RLOCK(&V_tcbinfo);
 2679         inp = sotoinpcb(so);
 2680         if (inp == NULL) {
 2681                 INP_INFO_RUNLOCK(&V_tcbinfo);
 2682                 return (error);
 2683         }
 2684         INP_RLOCK(inp);
 2685         INP_INFO_RUNLOCK(&V_tcbinfo);
 2686         tp = intotcpcb(inp);
 2687         if (tp != NULL && tp->t_state == TCPS_ESTABLISHED) {
 2688                 *maxp = tp->snd_max;
 2689                 *unap = tp->snd_una;
 2690                 error = 0;
 2691         }
 2692         INP_RUNLOCK(inp);
 2693         return (error);
 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 

Cache object: 3e4c819e1c5faeb6b6eca0fa0524593e


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.