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/nfsclient/nfs_clvnops.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  *      from nfs_vnops.c        8.16 (Berkeley) 5/27/95
   33  */
   34 
   35 #include <sys/cdefs.h>
   36 __FBSDID("$FreeBSD: releng/8.2/sys/fs/nfsclient/nfs_clvnops.c 214821 2010-11-05 02:45:13Z rmacklem $");
   37 
   38 /*
   39  * vnode op calls for Sun NFS version 2, 3 and 4
   40  */
   41 
   42 #include "opt_inet.h"
   43 
   44 #include <sys/param.h>
   45 #include <sys/kernel.h>
   46 #include <sys/systm.h>
   47 #include <sys/resourcevar.h>
   48 #include <sys/proc.h>
   49 #include <sys/mount.h>
   50 #include <sys/bio.h>
   51 #include <sys/buf.h>
   52 #include <sys/jail.h>
   53 #include <sys/malloc.h>
   54 #include <sys/mbuf.h>
   55 #include <sys/namei.h>
   56 #include <sys/socket.h>
   57 #include <sys/vnode.h>
   58 #include <sys/dirent.h>
   59 #include <sys/fcntl.h>
   60 #include <sys/lockf.h>
   61 #include <sys/stat.h>
   62 #include <sys/sysctl.h>
   63 #include <sys/signalvar.h>
   64 
   65 #include <vm/vm.h>
   66 #include <vm/vm_object.h>
   67 #include <vm/vm_extern.h>
   68 #include <vm/vm_object.h>
   69 
   70 #include <fs/nfs/nfsport.h>
   71 #include <fs/nfsclient/nfsnode.h>
   72 #include <fs/nfsclient/nfsmount.h>
   73 #include <fs/nfsclient/nfs.h>
   74 
   75 #include <net/if.h>
   76 #include <netinet/in.h>
   77 #include <netinet/in_var.h>
   78 
   79 #include <nfs/nfs_lock.h>
   80 
   81 /* Defs */
   82 #define TRUE    1
   83 #define FALSE   0
   84 
   85 extern struct nfsstats newnfsstats;
   86 MALLOC_DECLARE(M_NEWNFSREQ);
   87 
   88 /*
   89  * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
   90  * calls are not in getblk() and brelse() so that they would not be necessary
   91  * here.
   92  */
   93 #ifndef B_VMIO
   94 #define vfs_busy_pages(bp, f)
   95 #endif
   96 
   97 static vop_read_t       nfsfifo_read;
   98 static vop_write_t      nfsfifo_write;
   99 static vop_close_t      nfsfifo_close;
  100 static int      nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
  101                     struct thread *);
  102 static vop_lookup_t     nfs_lookup;
  103 static vop_create_t     nfs_create;
  104 static vop_mknod_t      nfs_mknod;
  105 static vop_open_t       nfs_open;
  106 static vop_close_t      nfs_close;
  107 static vop_access_t     nfs_access;
  108 static vop_getattr_t    nfs_getattr;
  109 static vop_setattr_t    nfs_setattr;
  110 static vop_read_t       nfs_read;
  111 static vop_fsync_t      nfs_fsync;
  112 static vop_remove_t     nfs_remove;
  113 static vop_link_t       nfs_link;
  114 static vop_rename_t     nfs_rename;
  115 static vop_mkdir_t      nfs_mkdir;
  116 static vop_rmdir_t      nfs_rmdir;
  117 static vop_symlink_t    nfs_symlink;
  118 static vop_readdir_t    nfs_readdir;
  119 static vop_strategy_t   nfs_strategy;
  120 static vop_lock1_t      nfs_lock1;
  121 static  int     nfs_lookitup(struct vnode *, char *, int,
  122                     struct ucred *, struct thread *, struct nfsnode **);
  123 static  int     nfs_sillyrename(struct vnode *, struct vnode *,
  124                     struct componentname *);
  125 static vop_access_t     nfsspec_access;
  126 static vop_readlink_t   nfs_readlink;
  127 static vop_print_t      nfs_print;
  128 static vop_advlock_t    nfs_advlock;
  129 static vop_advlockasync_t nfs_advlockasync;
  130 #ifdef NFS4_ACL_EXTATTR_NAME
  131 static vop_getacl_t nfs_getacl;
  132 static vop_setacl_t nfs_setacl;
  133 #endif
  134 
  135 /*
  136  * Global vfs data structures for nfs
  137  */
  138 struct vop_vector newnfs_vnodeops = {
  139         .vop_default =          &default_vnodeops,
  140         .vop_access =           nfs_access,
  141         .vop_advlock =          nfs_advlock,
  142         .vop_advlockasync =     nfs_advlockasync,
  143         .vop_close =            nfs_close,
  144         .vop_create =           nfs_create,
  145         .vop_fsync =            nfs_fsync,
  146         .vop_getattr =          nfs_getattr,
  147         .vop_getpages =         ncl_getpages,
  148         .vop_putpages =         ncl_putpages,
  149         .vop_inactive =         ncl_inactive,
  150         .vop_link =             nfs_link,
  151         .vop_lock1 =            nfs_lock1,
  152         .vop_lookup =           nfs_lookup,
  153         .vop_mkdir =            nfs_mkdir,
  154         .vop_mknod =            nfs_mknod,
  155         .vop_open =             nfs_open,
  156         .vop_print =            nfs_print,
  157         .vop_read =             nfs_read,
  158         .vop_readdir =          nfs_readdir,
  159         .vop_readlink =         nfs_readlink,
  160         .vop_reclaim =          ncl_reclaim,
  161         .vop_remove =           nfs_remove,
  162         .vop_rename =           nfs_rename,
  163         .vop_rmdir =            nfs_rmdir,
  164         .vop_setattr =          nfs_setattr,
  165         .vop_strategy =         nfs_strategy,
  166         .vop_symlink =          nfs_symlink,
  167         .vop_write =            ncl_write,
  168 #ifdef NFS4_ACL_EXTATTR_NAME
  169         .vop_getacl =           nfs_getacl,
  170         .vop_setacl =           nfs_setacl,
  171 #endif
  172 };
  173 
  174 struct vop_vector newnfs_fifoops = {
  175         .vop_default =          &fifo_specops,
  176         .vop_access =           nfsspec_access,
  177         .vop_close =            nfsfifo_close,
  178         .vop_fsync =            nfs_fsync,
  179         .vop_getattr =          nfs_getattr,
  180         .vop_inactive =         ncl_inactive,
  181         .vop_print =            nfs_print,
  182         .vop_read =             nfsfifo_read,
  183         .vop_reclaim =          ncl_reclaim,
  184         .vop_setattr =          nfs_setattr,
  185         .vop_write =            nfsfifo_write,
  186 };
  187 
  188 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
  189     struct componentname *cnp, struct vattr *vap);
  190 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
  191     int namelen, struct ucred *cred, struct thread *td);
  192 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
  193     char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
  194     char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
  195 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
  196     struct componentname *scnp, struct sillyrename *sp);
  197 
  198 /*
  199  * Global variables
  200  */
  201 #define DIRHDSIZ        (sizeof (struct dirent) - (MAXNAMLEN + 1))
  202 
  203 SYSCTL_DECL(_vfs_newnfs);
  204 
  205 static int      nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
  206 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
  207            &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
  208 
  209 static int      nfs_prime_access_cache = 0;
  210 SYSCTL_INT(_vfs_newnfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
  211            &nfs_prime_access_cache, 0,
  212            "Prime NFS ACCESS cache when fetching attributes");
  213 
  214 static int      newnfs_commit_on_close = 0;
  215 SYSCTL_INT(_vfs_newnfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
  216     &newnfs_commit_on_close, 0, "write+commit on close, else only write");
  217 
  218 static int      nfs_clean_pages_on_close = 1;
  219 SYSCTL_INT(_vfs_newnfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
  220            &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
  221 
  222 int newnfs_directio_enable = 0;
  223 SYSCTL_INT(_vfs_newnfs, OID_AUTO, directio_enable, CTLFLAG_RW,
  224            &newnfs_directio_enable, 0, "Enable NFS directio");
  225 
  226 /*
  227  * This sysctl allows other processes to mmap a file that has been opened
  228  * O_DIRECT by a process.  In general, having processes mmap the file while
  229  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
  230  * this by default to prevent DoS attacks - to prevent a malicious user from
  231  * opening up files O_DIRECT preventing other users from mmap'ing these
  232  * files.  "Protected" environments where stricter consistency guarantees are
  233  * required can disable this knob.  The process that opened the file O_DIRECT
  234  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
  235  * meaningful.
  236  */
  237 int newnfs_directio_allow_mmap = 1;
  238 SYSCTL_INT(_vfs_newnfs, OID_AUTO, directio_allow_mmap, CTLFLAG_RW,
  239            &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
  240 
  241 #if 0
  242 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
  243            &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
  244 
  245 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
  246            &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
  247 #endif
  248 
  249 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY                \
  250                          | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \
  251                          | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
  252 
  253 /*
  254  * SMP Locking Note :
  255  * The list of locks after the description of the lock is the ordering
  256  * of other locks acquired with the lock held.
  257  * np->n_mtx : Protects the fields in the nfsnode.
  258        VM Object Lock
  259        VI_MTX (acquired indirectly)
  260  * nmp->nm_mtx : Protects the fields in the nfsmount.
  261        rep->r_mtx
  262  * ncl_iod_mutex : Global lock, protects shared nfsiod state.
  263  * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
  264        nmp->nm_mtx
  265        rep->r_mtx
  266  * rep->r_mtx : Protects the fields in an nfsreq.
  267  */
  268 
  269 static int
  270 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
  271     struct ucred *cred, u_int32_t *retmode)
  272 {
  273         int error = 0, attrflag, i, lrupos;
  274         u_int32_t rmode;
  275         struct nfsnode *np = VTONFS(vp);
  276         struct nfsvattr nfsva;
  277 
  278         error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
  279             &rmode, NULL);
  280         if (attrflag)
  281                 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
  282         if (!error) {
  283                 lrupos = 0;
  284                 mtx_lock(&np->n_mtx);
  285                 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
  286                         if (np->n_accesscache[i].uid == cred->cr_uid) {
  287                                 np->n_accesscache[i].mode = rmode;
  288                                 np->n_accesscache[i].stamp = time_second;
  289                                 break;
  290                         }
  291                         if (i > 0 && np->n_accesscache[i].stamp <
  292                             np->n_accesscache[lrupos].stamp)
  293                                 lrupos = i;
  294                 }
  295                 if (i == NFS_ACCESSCACHESIZE) {
  296                         np->n_accesscache[lrupos].uid = cred->cr_uid;
  297                         np->n_accesscache[lrupos].mode = rmode;
  298                         np->n_accesscache[lrupos].stamp = time_second;
  299                 }
  300                 mtx_unlock(&np->n_mtx);
  301                 if (retmode != NULL)
  302                         *retmode = rmode;
  303         } else if (NFS_ISV4(vp)) {
  304                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
  305         }
  306         return (error);
  307 }
  308 
  309 /*
  310  * nfs access vnode op.
  311  * For nfs version 2, just return ok. File accesses may fail later.
  312  * For nfs version 3, use the access rpc to check accessibility. If file modes
  313  * are changed on the server, accesses might still fail later.
  314  */
  315 static int
  316 nfs_access(struct vop_access_args *ap)
  317 {
  318         struct vnode *vp = ap->a_vp;
  319         int error = 0, i, gotahit;
  320         u_int32_t mode, wmode, rmode;
  321         int v34 = NFS_ISV34(vp);
  322         struct nfsnode *np = VTONFS(vp);
  323 
  324         /*
  325          * Disallow write attempts on filesystems mounted read-only;
  326          * unless the file is a socket, fifo, or a block or character
  327          * device resident on the filesystem.
  328          */
  329         if ((ap->a_accmode & (VWRITE | VAPPEND
  330 #ifdef NFS4_ACL_EXTATTR_NAME
  331             | VWRITE_NAMED_ATTRS | VDELETE_CHILD | VWRITE_ATTRIBUTES |
  332             VDELETE | VWRITE_ACL | VWRITE_OWNER
  333 #endif
  334             )) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
  335                 switch (vp->v_type) {
  336                 case VREG:
  337                 case VDIR:
  338                 case VLNK:
  339                         return (EROFS);
  340                 default:
  341                         break;
  342                 }
  343         }
  344         /*
  345          * For nfs v3 or v4, check to see if we have done this recently, and if
  346          * so return our cached result instead of making an ACCESS call.
  347          * If not, do an access rpc, otherwise you are stuck emulating
  348          * ufs_access() locally using the vattr. This may not be correct,
  349          * since the server may apply other access criteria such as
  350          * client uid-->server uid mapping that we do not know about.
  351          */
  352         if (v34) {
  353                 if (ap->a_accmode & VREAD)
  354                         mode = NFSACCESS_READ;
  355                 else
  356                         mode = 0;
  357                 if (vp->v_type != VDIR) {
  358                         if (ap->a_accmode & VWRITE)
  359                                 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
  360                         if (ap->a_accmode & VAPPEND)
  361                                 mode |= NFSACCESS_EXTEND;
  362                         if (ap->a_accmode & VEXEC)
  363                                 mode |= NFSACCESS_EXECUTE;
  364 #ifdef NFS4_ACL_EXTATTR_NAME
  365                         if (ap->a_accmode & VDELETE)
  366                                 mode |= NFSACCESS_DELETE;
  367 #endif
  368                 } else {
  369                         if (ap->a_accmode & VWRITE)
  370                                 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
  371                         if (ap->a_accmode & VAPPEND)
  372                                 mode |= NFSACCESS_EXTEND;
  373                         if (ap->a_accmode & VEXEC)
  374                                 mode |= NFSACCESS_LOOKUP;
  375 #ifdef NFS4_ACL_EXTATTR_NAME
  376                         if (ap->a_accmode & VDELETE)
  377                                 mode |= NFSACCESS_DELETE;
  378                         if (ap->a_accmode & VDELETE_CHILD)
  379                                 mode |= NFSACCESS_MODIFY;
  380 #endif
  381                 }
  382                 /* XXX safety belt, only make blanket request if caching */
  383                 if (nfsaccess_cache_timeout > 0) {
  384                         wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
  385                                 NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
  386                                 NFSACCESS_DELETE | NFSACCESS_LOOKUP;
  387                 } else {
  388                         wmode = mode;
  389                 }
  390 
  391                 /*
  392                  * Does our cached result allow us to give a definite yes to
  393                  * this request?
  394                  */
  395                 gotahit = 0;
  396                 mtx_lock(&np->n_mtx);
  397                 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
  398                         if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
  399                             if (time_second < (np->n_accesscache[i].stamp
  400                                 + nfsaccess_cache_timeout) &&
  401                                 (np->n_accesscache[i].mode & mode) == mode) {
  402                                 NFSINCRGLOBAL(newnfsstats.accesscache_hits);
  403                                 gotahit = 1;
  404                             }
  405                             break;
  406                         }
  407                 }
  408                 mtx_unlock(&np->n_mtx);
  409                 if (gotahit == 0) {
  410                         /*
  411                          * Either a no, or a don't know.  Go to the wire.
  412                          */
  413                         NFSINCRGLOBAL(newnfsstats.accesscache_misses);
  414                         error = nfs34_access_otw(vp, wmode, ap->a_td,
  415                             ap->a_cred, &rmode);
  416                         if (!error &&
  417                             (rmode & mode) != mode)
  418                                 error = EACCES;
  419                 }
  420                 return (error);
  421         } else {
  422                 if ((error = nfsspec_access(ap)) != 0) {
  423                         return (error);
  424                 }
  425                 /*
  426                  * Attempt to prevent a mapped root from accessing a file
  427                  * which it shouldn't.  We try to read a byte from the file
  428                  * if the user is root and the file is not zero length.
  429                  * After calling nfsspec_access, we should have the correct
  430                  * file size cached.
  431                  */
  432                 mtx_lock(&np->n_mtx);
  433                 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
  434                     && VTONFS(vp)->n_size > 0) {
  435                         struct iovec aiov;
  436                         struct uio auio;
  437                         char buf[1];
  438 
  439                         mtx_unlock(&np->n_mtx);
  440                         aiov.iov_base = buf;
  441                         aiov.iov_len = 1;
  442                         auio.uio_iov = &aiov;
  443                         auio.uio_iovcnt = 1;
  444                         auio.uio_offset = 0;
  445                         auio.uio_resid = 1;
  446                         auio.uio_segflg = UIO_SYSSPACE;
  447                         auio.uio_rw = UIO_READ;
  448                         auio.uio_td = ap->a_td;
  449 
  450                         if (vp->v_type == VREG)
  451                                 error = ncl_readrpc(vp, &auio, ap->a_cred);
  452                         else if (vp->v_type == VDIR) {
  453                                 char* bp;
  454                                 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
  455                                 aiov.iov_base = bp;
  456                                 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
  457                                 error = ncl_readdirrpc(vp, &auio, ap->a_cred,
  458                                     ap->a_td);
  459                                 free(bp, M_TEMP);
  460                         } else if (vp->v_type == VLNK)
  461                                 error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
  462                         else
  463                                 error = EACCES;
  464                 } else
  465                         mtx_unlock(&np->n_mtx);
  466                 return (error);
  467         }
  468 }
  469 
  470 
  471 /*
  472  * nfs open vnode op
  473  * Check to see if the type is ok
  474  * and that deletion is not in progress.
  475  * For paged in text files, you will need to flush the page cache
  476  * if consistency is lost.
  477  */
  478 /* ARGSUSED */
  479 static int
  480 nfs_open(struct vop_open_args *ap)
  481 {
  482         struct vnode *vp = ap->a_vp;
  483         struct nfsnode *np = VTONFS(vp);
  484         struct vattr vattr;
  485         int error;
  486         int fmode = ap->a_mode;
  487 
  488         if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
  489                 return (EOPNOTSUPP);
  490 
  491         /*
  492          * For NFSv4, we need to do the Open Op before cache validation,
  493          * so that we conform to RFC3530 Sec. 9.3.1.
  494          */
  495         if (NFS_ISV4(vp)) {
  496                 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
  497                 if (error) {
  498                         error = nfscl_maperr(ap->a_td, error, (uid_t)0,
  499                             (gid_t)0);
  500                         return (error);
  501                 }
  502         }
  503 
  504         /*
  505          * Now, if this Open will be doing reading, re-validate/flush the
  506          * cache, so that Close/Open coherency is maintained.
  507          */
  508         mtx_lock(&np->n_mtx);
  509         if (np->n_flag & NMODIFIED) {
  510                 mtx_unlock(&np->n_mtx);
  511                 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  512                 if (error == EINTR || error == EIO) {
  513                         if (NFS_ISV4(vp))
  514                                 (void) nfsrpc_close(vp, 0, ap->a_td);
  515                         return (error);
  516                 }
  517                 mtx_lock(&np->n_mtx);
  518                 np->n_attrstamp = 0;
  519                 if (vp->v_type == VDIR)
  520                         np->n_direofoffset = 0;
  521                 mtx_unlock(&np->n_mtx);
  522                 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
  523                 if (error) {
  524                         if (NFS_ISV4(vp))
  525                                 (void) nfsrpc_close(vp, 0, ap->a_td);
  526                         return (error);
  527                 }
  528                 mtx_lock(&np->n_mtx);
  529                 np->n_mtime = vattr.va_mtime;
  530                 if (NFS_ISV4(vp))
  531                         np->n_change = vattr.va_filerev;
  532         } else {
  533                 mtx_unlock(&np->n_mtx);
  534                 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
  535                 if (error) {
  536                         if (NFS_ISV4(vp))
  537                                 (void) nfsrpc_close(vp, 0, ap->a_td);
  538                         return (error);
  539                 }
  540                 mtx_lock(&np->n_mtx);
  541                 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
  542                     NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
  543                         if (vp->v_type == VDIR)
  544                                 np->n_direofoffset = 0;
  545                         mtx_unlock(&np->n_mtx);
  546                         error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  547                         if (error == EINTR || error == EIO) {
  548                                 if (NFS_ISV4(vp))
  549                                         (void) nfsrpc_close(vp, 0, ap->a_td);
  550                                 return (error);
  551                         }
  552                         mtx_lock(&np->n_mtx);
  553                         np->n_mtime = vattr.va_mtime;
  554                         if (NFS_ISV4(vp))
  555                                 np->n_change = vattr.va_filerev;
  556                 }
  557         }
  558 
  559         /*
  560          * If the object has >= 1 O_DIRECT active opens, we disable caching.
  561          */
  562         if (newnfs_directio_enable && (fmode & O_DIRECT) &&
  563             (vp->v_type == VREG)) {
  564                 if (np->n_directio_opens == 0) {
  565                         mtx_unlock(&np->n_mtx);
  566                         error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  567                         if (error) {
  568                                 if (NFS_ISV4(vp))
  569                                         (void) nfsrpc_close(vp, 0, ap->a_td);
  570                                 return (error);
  571                         }
  572                         mtx_lock(&np->n_mtx);
  573                         np->n_flag |= NNONCACHE;
  574                 }
  575                 np->n_directio_opens++;
  576         }
  577         mtx_unlock(&np->n_mtx);
  578         vnode_create_vobject(vp, vattr.va_size, ap->a_td);
  579         return (0);
  580 }
  581 
  582 /*
  583  * nfs close vnode op
  584  * What an NFS client should do upon close after writing is a debatable issue.
  585  * Most NFS clients push delayed writes to the server upon close, basically for
  586  * two reasons:
  587  * 1 - So that any write errors may be reported back to the client process
  588  *     doing the close system call. By far the two most likely errors are
  589  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
  590  * 2 - To put a worst case upper bound on cache inconsistency between
  591  *     multiple clients for the file.
  592  * There is also a consistency problem for Version 2 of the protocol w.r.t.
  593  * not being able to tell if other clients are writing a file concurrently,
  594  * since there is no way of knowing if the changed modify time in the reply
  595  * is only due to the write for this client.
  596  * (NFS Version 3 provides weak cache consistency data in the reply that
  597  *  should be sufficient to detect and handle this case.)
  598  *
  599  * The current code does the following:
  600  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
  601  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
  602  *                     or commit them (this satisfies 1 and 2 except for the
  603  *                     case where the server crashes after this close but
  604  *                     before the commit RPC, which is felt to be "good
  605  *                     enough". Changing the last argument to ncl_flush() to
  606  *                     a 1 would force a commit operation, if it is felt a
  607  *                     commit is necessary now.
  608  * for NFS Version 4 - flush the dirty buffers and commit them, if
  609  *                     nfscl_mustflush() says this is necessary.
  610  *                     It is necessary if there is no write delegation held,
  611  *                     in order to satisfy open/close coherency.
  612  *                     If the file isn't cached on local stable storage,
  613  *                     it may be necessary in order to detect "out of space"
  614  *                     errors from the server, if the write delegation
  615  *                     issued by the server doesn't allow the file to grow.
  616  */
  617 /* ARGSUSED */
  618 static int
  619 nfs_close(struct vop_close_args *ap)
  620 {
  621         struct vnode *vp = ap->a_vp;
  622         struct nfsnode *np = VTONFS(vp);
  623         struct nfsvattr nfsva;
  624         struct ucred *cred;
  625         int error = 0, ret, localcred = 0;
  626         int fmode = ap->a_fflag;
  627 
  628         if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
  629                 return (0);
  630         /*
  631          * During shutdown, a_cred isn't valid, so just use root.
  632          */
  633         if (ap->a_cred == NOCRED) {
  634                 cred = newnfs_getcred();
  635                 localcred = 1;
  636         } else {
  637                 cred = ap->a_cred;
  638         }
  639         if (vp->v_type == VREG) {
  640             /*
  641              * Examine and clean dirty pages, regardless of NMODIFIED.
  642              * This closes a major hole in close-to-open consistency.
  643              * We want to push out all dirty pages (and buffers) on
  644              * close, regardless of whether they were dirtied by
  645              * mmap'ed writes or via write().
  646              */
  647             if (nfs_clean_pages_on_close && vp->v_object) {
  648                 VM_OBJECT_LOCK(vp->v_object);
  649                 vm_object_page_clean(vp->v_object, 0, 0, 0);
  650                 VM_OBJECT_UNLOCK(vp->v_object);
  651             }
  652             mtx_lock(&np->n_mtx);
  653             if (np->n_flag & NMODIFIED) {
  654                 mtx_unlock(&np->n_mtx);
  655                 if (NFS_ISV3(vp)) {
  656                     /*
  657                      * Under NFSv3 we have dirty buffers to dispose of.  We
  658                      * must flush them to the NFS server.  We have the option
  659                      * of waiting all the way through the commit rpc or just
  660                      * waiting for the initial write.  The default is to only
  661                      * wait through the initial write so the data is in the
  662                      * server's cache, which is roughly similar to the state
  663                      * a standard disk subsystem leaves the file in on close().
  664                      *
  665                      * We cannot clear the NMODIFIED bit in np->n_flag due to
  666                      * potential races with other processes, and certainly
  667                      * cannot clear it if we don't commit.
  668                      * These races occur when there is no longer the old
  669                      * traditional vnode locking implemented for Vnode Ops.
  670                      */
  671                     int cm = newnfs_commit_on_close ? 1 : 0;
  672                     error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0);
  673                     /* np->n_flag &= ~NMODIFIED; */
  674                 } else if (NFS_ISV4(vp)) { 
  675                         if (nfscl_mustflush(vp) != 0) {
  676                                 int cm = newnfs_commit_on_close ? 1 : 0;
  677                                 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
  678                                     cm, 0);
  679                                 /*
  680                                  * as above w.r.t races when clearing
  681                                  * NMODIFIED.
  682                                  * np->n_flag &= ~NMODIFIED;
  683                                  */
  684                         }
  685                 } else
  686                     error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  687                 mtx_lock(&np->n_mtx);
  688             }
  689             /* 
  690              * Invalidate the attribute cache in all cases.
  691              * An open is going to fetch fresh attrs any way, other procs
  692              * on this node that have file open will be forced to do an 
  693              * otw attr fetch, but this is safe.
  694              * --> A user found that their RPC count dropped by 20% when
  695              *     this was commented out and I can't see any requirement
  696              *     for it, so I've disabled it when negative lookups are
  697              *     enabled. (What does this have to do with negative lookup
  698              *     caching? Well nothing, except it was reported by the
  699              *     same user that needed negative lookup caching and I wanted
  700              *     there to be a way to disable it to see if it
  701              *     is the cause of some caching/coherency issue that might
  702              *     crop up.)
  703              */
  704             if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0)
  705                     np->n_attrstamp = 0;
  706             if (np->n_flag & NWRITEERR) {
  707                 np->n_flag &= ~NWRITEERR;
  708                 error = np->n_error;
  709             }
  710             mtx_unlock(&np->n_mtx);
  711         }
  712 
  713         if (NFS_ISV4(vp)) {
  714                 /*
  715                  * Get attributes so "change" is up to date.
  716                  */
  717                 if (error == 0 && nfscl_mustflush(vp) != 0) {
  718                         ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
  719                             NULL);
  720                         if (!ret) {
  721                                 np->n_change = nfsva.na_filerev;
  722                                 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
  723                                     NULL, 0, 0);
  724                         }
  725                 }
  726 
  727                 /*
  728                  * and do the close.
  729                  */
  730                 ret = nfsrpc_close(vp, 0, ap->a_td);
  731                 if (!error && ret)
  732                         error = ret;
  733                 if (error)
  734                         error = nfscl_maperr(ap->a_td, error, (uid_t)0,
  735                             (gid_t)0);
  736         }
  737         if (newnfs_directio_enable)
  738                 KASSERT((np->n_directio_asyncwr == 0),
  739                         ("nfs_close: dirty unflushed (%d) directio buffers\n",
  740                          np->n_directio_asyncwr));
  741         if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
  742                 mtx_lock(&np->n_mtx);
  743                 KASSERT((np->n_directio_opens > 0), 
  744                         ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
  745                 np->n_directio_opens--;
  746                 if (np->n_directio_opens == 0)
  747                         np->n_flag &= ~NNONCACHE;
  748                 mtx_unlock(&np->n_mtx);
  749         }
  750         if (localcred)
  751                 NFSFREECRED(cred);
  752         return (error);
  753 }
  754 
  755 /*
  756  * nfs getattr call from vfs.
  757  */
  758 static int
  759 nfs_getattr(struct vop_getattr_args *ap)
  760 {
  761         struct vnode *vp = ap->a_vp;
  762         struct thread *td = curthread;  /* XXX */
  763         struct nfsnode *np = VTONFS(vp);
  764         int error = 0;
  765         struct nfsvattr nfsva;
  766         struct vattr *vap = ap->a_vap;
  767         struct vattr vattr;
  768 
  769         /*
  770          * Update local times for special files.
  771          */
  772         mtx_lock(&np->n_mtx);
  773         if (np->n_flag & (NACC | NUPD))
  774                 np->n_flag |= NCHG;
  775         mtx_unlock(&np->n_mtx);
  776         /*
  777          * First look in the cache.
  778          */
  779         if (ncl_getattrcache(vp, &vattr) == 0) {
  780                 vap->va_type = vattr.va_type;
  781                 vap->va_mode = vattr.va_mode;
  782                 vap->va_nlink = vattr.va_nlink;
  783                 vap->va_uid = vattr.va_uid;
  784                 vap->va_gid = vattr.va_gid;
  785                 vap->va_fsid = vattr.va_fsid;
  786                 vap->va_fileid = vattr.va_fileid;
  787                 vap->va_size = vattr.va_size;
  788                 vap->va_blocksize = vattr.va_blocksize;
  789                 vap->va_atime = vattr.va_atime;
  790                 vap->va_mtime = vattr.va_mtime;
  791                 vap->va_ctime = vattr.va_ctime;
  792                 vap->va_gen = vattr.va_gen;
  793                 vap->va_flags = vattr.va_flags;
  794                 vap->va_rdev = vattr.va_rdev;
  795                 vap->va_bytes = vattr.va_bytes;
  796                 vap->va_filerev = vattr.va_filerev;
  797                 /*
  798                  * Get the local modify time for the case of a write
  799                  * delegation.
  800                  */
  801                 nfscl_deleggetmodtime(vp, &vap->va_mtime);
  802                 return (0);
  803         }
  804 
  805         if (NFS_ISV34(vp) && nfs_prime_access_cache &&
  806             nfsaccess_cache_timeout > 0) {
  807                 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
  808                 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
  809                 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
  810                         nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
  811                         return (0);
  812                 }
  813         }
  814         error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
  815         if (!error)
  816                 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
  817         if (!error) {
  818                 /*
  819                  * Get the local modify time for the case of a write
  820                  * delegation.
  821                  */
  822                 nfscl_deleggetmodtime(vp, &vap->va_mtime);
  823         } else if (NFS_ISV4(vp)) {
  824                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
  825         }
  826         return (error);
  827 }
  828 
  829 /*
  830  * nfs setattr call.
  831  */
  832 static int
  833 nfs_setattr(struct vop_setattr_args *ap)
  834 {
  835         struct vnode *vp = ap->a_vp;
  836         struct nfsnode *np = VTONFS(vp);
  837         struct thread *td = curthread;  /* XXX */
  838         struct vattr *vap = ap->a_vap;
  839         int error = 0;
  840         u_quad_t tsize;
  841 
  842 #ifndef nolint
  843         tsize = (u_quad_t)0;
  844 #endif
  845 
  846         /*
  847          * Setting of flags and marking of atimes are not supported.
  848          */
  849         if (vap->va_flags != VNOVAL)
  850                 return (EOPNOTSUPP);
  851 
  852         /*
  853          * Disallow write attempts if the filesystem is mounted read-only.
  854          */
  855         if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
  856             vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
  857             vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
  858             (vp->v_mount->mnt_flag & MNT_RDONLY))
  859                 return (EROFS);
  860         if (vap->va_size != VNOVAL) {
  861                 switch (vp->v_type) {
  862                 case VDIR:
  863                         return (EISDIR);
  864                 case VCHR:
  865                 case VBLK:
  866                 case VSOCK:
  867                 case VFIFO:
  868                         if (vap->va_mtime.tv_sec == VNOVAL &&
  869                             vap->va_atime.tv_sec == VNOVAL &&
  870                             vap->va_mode == (mode_t)VNOVAL &&
  871                             vap->va_uid == (uid_t)VNOVAL &&
  872                             vap->va_gid == (gid_t)VNOVAL)
  873                                 return (0);             
  874                         vap->va_size = VNOVAL;
  875                         break;
  876                 default:
  877                         /*
  878                          * Disallow write attempts if the filesystem is
  879                          * mounted read-only.
  880                          */
  881                         if (vp->v_mount->mnt_flag & MNT_RDONLY)
  882                                 return (EROFS);
  883                         /*
  884                          *  We run vnode_pager_setsize() early (why?),
  885                          * we must set np->n_size now to avoid vinvalbuf
  886                          * V_SAVE races that might setsize a lower
  887                          * value.
  888                          */
  889                         mtx_lock(&np->n_mtx);
  890                         tsize = np->n_size;
  891                         mtx_unlock(&np->n_mtx);
  892                         error = ncl_meta_setsize(vp, ap->a_cred, td,
  893                             vap->va_size);
  894                         mtx_lock(&np->n_mtx);
  895                         if (np->n_flag & NMODIFIED) {
  896                             tsize = np->n_size;
  897                             mtx_unlock(&np->n_mtx);
  898                             if (vap->va_size == 0)
  899                                 error = ncl_vinvalbuf(vp, 0, td, 1);
  900                             else
  901                                 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
  902                             if (error) {
  903                                 vnode_pager_setsize(vp, tsize);
  904                                 return (error);
  905                             }
  906                             /*
  907                              * Call nfscl_delegmodtime() to set the modify time
  908                              * locally, as required.
  909                              */
  910                             nfscl_delegmodtime(vp);
  911                         } else
  912                             mtx_unlock(&np->n_mtx);
  913                         /*
  914                          * np->n_size has already been set to vap->va_size
  915                          * in ncl_meta_setsize(). We must set it again since
  916                          * nfs_loadattrcache() could be called through
  917                          * ncl_meta_setsize() and could modify np->n_size.
  918                          */
  919                         mtx_lock(&np->n_mtx);
  920                         np->n_vattr.na_size = np->n_size = vap->va_size;
  921                         mtx_unlock(&np->n_mtx);
  922                 };
  923         } else {
  924                 mtx_lock(&np->n_mtx);
  925                 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 
  926                     (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
  927                         mtx_unlock(&np->n_mtx);
  928                         if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
  929                             (error == EINTR || error == EIO))
  930                                 return (error);
  931                 } else
  932                         mtx_unlock(&np->n_mtx);
  933         }
  934         error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
  935         if (error && vap->va_size != VNOVAL) {
  936                 mtx_lock(&np->n_mtx);
  937                 np->n_size = np->n_vattr.na_size = tsize;
  938                 vnode_pager_setsize(vp, tsize);
  939                 mtx_unlock(&np->n_mtx);
  940         }
  941         return (error);
  942 }
  943 
  944 /*
  945  * Do an nfs setattr rpc.
  946  */
  947 static int
  948 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
  949     struct thread *td)
  950 {
  951         struct nfsnode *np = VTONFS(vp);
  952         int error, ret, attrflag, i;
  953         struct nfsvattr nfsva;
  954 
  955         if (NFS_ISV34(vp)) {
  956                 mtx_lock(&np->n_mtx);
  957                 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
  958                         np->n_accesscache[i].stamp = 0;
  959                 np->n_flag |= NDELEGMOD;
  960                 mtx_unlock(&np->n_mtx);
  961         }
  962         error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
  963             NULL);
  964         if (attrflag) {
  965                 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
  966                 if (ret && !error)
  967                         error = ret;
  968         }
  969         if (error && NFS_ISV4(vp))
  970                 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
  971         return (error);
  972 }
  973 
  974 /*
  975  * nfs lookup call, one step at a time...
  976  * First look in cache
  977  * If not found, unlock the directory nfsnode and do the rpc
  978  */
  979 static int
  980 nfs_lookup(struct vop_lookup_args *ap)
  981 {
  982         struct componentname *cnp = ap->a_cnp;
  983         struct vnode *dvp = ap->a_dvp;
  984         struct vnode **vpp = ap->a_vpp;
  985         struct mount *mp = dvp->v_mount;
  986         int flags = cnp->cn_flags;
  987         struct vnode *newvp;
  988         struct nfsmount *nmp;
  989         struct nfsnode *np, *newnp;
  990         int error = 0, attrflag, dattrflag, ltype;
  991         struct thread *td = cnp->cn_thread;
  992         struct nfsfh *nfhp;
  993         struct nfsvattr dnfsva, nfsva;
  994         struct vattr vattr;
  995         struct timespec dmtime;
  996         
  997         *vpp = NULLVP;
  998         if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
  999             (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
 1000                 return (EROFS);
 1001         if (dvp->v_type != VDIR)
 1002                 return (ENOTDIR);
 1003         nmp = VFSTONFS(mp);
 1004         np = VTONFS(dvp);
 1005 
 1006         /* For NFSv4, wait until any remove is done. */
 1007         mtx_lock(&np->n_mtx);
 1008         while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
 1009                 np->n_flag |= NREMOVEWANT;
 1010                 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
 1011         }
 1012         mtx_unlock(&np->n_mtx);
 1013 
 1014         if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
 1015                 return (error);
 1016         error = cache_lookup(dvp, vpp, cnp);
 1017         if (error > 0 && error != ENOENT)
 1018                 return (error);
 1019         if (error == -1) {
 1020                 /*
 1021                  * We only accept a positive hit in the cache if the
 1022                  * change time of the file matches our cached copy.
 1023                  * Otherwise, we discard the cache entry and fallback
 1024                  * to doing a lookup RPC.
 1025                  *
 1026                  * To better handle stale file handles and attributes,
 1027                  * clear the attribute cache of this node if it is a
 1028                  * leaf component, part of an open() call, and not
 1029                  * locally modified before fetching the attributes.
 1030                  * This should allow stale file handles to be detected
 1031                  * here where we can fall back to a LOOKUP RPC to
 1032                  * recover rather than having nfs_open() detect the
 1033                  * stale file handle and failing open(2) with ESTALE.
 1034                  */
 1035                 newvp = *vpp;
 1036                 newnp = VTONFS(newvp);
 1037                 if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
 1038                     !(newnp->n_flag & NMODIFIED)) {
 1039                         mtx_lock(&newnp->n_mtx);
 1040                         newnp->n_attrstamp = 0;
 1041                         mtx_unlock(&newnp->n_mtx);
 1042                 }
 1043                 if (nfscl_nodeleg(newvp, 0) == 0 ||
 1044                     (VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
 1045                     timespeccmp(&vattr.va_ctime, &newnp->n_ctime, ==))) {
 1046                         NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
 1047                         if (cnp->cn_nameiop != LOOKUP &&
 1048                             (flags & ISLASTCN))
 1049                                 cnp->cn_flags |= SAVENAME;
 1050                         return (0);
 1051                 }
 1052                 cache_purge(newvp);
 1053                 if (dvp != newvp)
 1054                         vput(newvp);
 1055                 else 
 1056                         vrele(newvp);
 1057                 *vpp = NULLVP;
 1058         } else if (error == ENOENT) {
 1059                 if (dvp->v_iflag & VI_DOOMED)
 1060                         return (ENOENT);
 1061                 /*
 1062                  * We only accept a negative hit in the cache if the
 1063                  * modification time of the parent directory matches
 1064                  * our cached copy.  Otherwise, we discard all of the
 1065                  * negative cache entries for this directory. We also
 1066                  * only trust -ve cache entries for less than
 1067                  * nm_negative_namecache_timeout seconds.
 1068                  */
 1069                 if ((u_int)(ticks - np->n_dmtime_ticks) <
 1070                     (nmp->nm_negnametimeo * hz) &&
 1071                     VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
 1072                     timespeccmp(&vattr.va_mtime, &np->n_dmtime, ==)) {
 1073                         NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
 1074                         return (ENOENT);
 1075                 }
 1076                 cache_purge_negative(dvp);
 1077                 mtx_lock(&np->n_mtx);
 1078                 timespecclear(&np->n_dmtime);
 1079                 mtx_unlock(&np->n_mtx);
 1080         }
 1081 
 1082         /*
 1083          * Cache the modification time of the parent directory in case
 1084          * the lookup fails and results in adding the first negative
 1085          * name cache entry for the directory.  Since this is reading
 1086          * a single time_t, don't bother with locking.  The
 1087          * modification time may be a bit stale, but it must be read
 1088          * before performing the lookup RPC to prevent a race where
 1089          * another lookup updates the timestamp on the directory after
 1090          * the lookup RPC has been performed on the server but before
 1091          * n_dmtime is set at the end of this function.
 1092          */
 1093         dmtime = np->n_vattr.na_mtime;
 1094         error = 0;
 1095         newvp = NULLVP;
 1096         NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
 1097         error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 1098             cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
 1099             NULL);
 1100         if (dattrflag)
 1101                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 1102         if (error) {
 1103                 if (newvp != NULLVP) {
 1104                         vput(newvp);
 1105                         *vpp = NULLVP;
 1106                 }
 1107 
 1108                 if (error != ENOENT) {
 1109                         if (NFS_ISV4(dvp))
 1110                                 error = nfscl_maperr(td, error, (uid_t)0,
 1111                                     (gid_t)0);
 1112                         return (error);
 1113                 }
 1114 
 1115                 /* The requested file was not found. */
 1116                 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
 1117                     (flags & ISLASTCN)) {
 1118                         /*
 1119                          * XXX: UFS does a full VOP_ACCESS(dvp,
 1120                          * VWRITE) here instead of just checking
 1121                          * MNT_RDONLY.
 1122                          */
 1123                         if (mp->mnt_flag & MNT_RDONLY)
 1124                                 return (EROFS);
 1125                         cnp->cn_flags |= SAVENAME;
 1126                         return (EJUSTRETURN);
 1127                 }
 1128 
 1129                 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) {
 1130                         /*
 1131                          * Maintain n_dmtime as the modification time
 1132                          * of the parent directory when the oldest -ve
 1133                          * name cache entry for this directory was
 1134                          * added.  If a -ve cache entry has already
 1135                          * been added with a newer modification time
 1136                          * by a concurrent lookup, then don't bother
 1137                          * adding a cache entry.  The modification
 1138                          * time of the directory might have changed
 1139                          * due to the file this lookup failed to find
 1140                          * being created.  In that case a subsequent
 1141                          * lookup would incorrectly use the entry
 1142                          * added here instead of doing an extra
 1143                          * lookup.
 1144                          */
 1145                         mtx_lock(&np->n_mtx);
 1146                         if (timespeccmp(&np->n_dmtime, &dmtime, <=)) {
 1147                                 if (!timespecisset(&np->n_dmtime)) {
 1148                                         np->n_dmtime = dmtime;
 1149                                         np->n_dmtime_ticks = ticks;
 1150                                 }
 1151                                 mtx_unlock(&np->n_mtx);
 1152                                 cache_enter(dvp, NULL, cnp);
 1153                         } else
 1154                                 mtx_unlock(&np->n_mtx);
 1155                 }
 1156                 return (ENOENT);
 1157         }
 1158 
 1159         /*
 1160          * Handle RENAME case...
 1161          */
 1162         if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
 1163                 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
 1164                         FREE((caddr_t)nfhp, M_NFSFH);
 1165                         return (EISDIR);
 1166                 }
 1167                 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL);
 1168                 if (error)
 1169                         return (error);
 1170                 newvp = NFSTOV(np);
 1171                 if (attrflag)
 1172                         (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1173                             0, 1);
 1174                 *vpp = newvp;
 1175                 cnp->cn_flags |= SAVENAME;
 1176                 return (0);
 1177         }
 1178 
 1179         if (flags & ISDOTDOT) {
 1180                 ltype = VOP_ISLOCKED(dvp);
 1181                 error = vfs_busy(mp, MBF_NOWAIT);
 1182                 if (error != 0) {
 1183                         vfs_ref(mp);
 1184                         VOP_UNLOCK(dvp, 0);
 1185                         error = vfs_busy(mp, 0);
 1186                         vn_lock(dvp, ltype | LK_RETRY);
 1187                         vfs_rel(mp);
 1188                         if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
 1189                                 vfs_unbusy(mp);
 1190                                 error = ENOENT;
 1191                         }
 1192                         if (error != 0)
 1193                                 return (error);
 1194                 }
 1195                 VOP_UNLOCK(dvp, 0);
 1196                 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL);
 1197                 if (error == 0)
 1198                         newvp = NFSTOV(np);
 1199                 vfs_unbusy(mp);
 1200                 if (newvp != dvp)
 1201                         vn_lock(dvp, ltype | LK_RETRY);
 1202                 if (dvp->v_iflag & VI_DOOMED) {
 1203                         if (error == 0) {
 1204                                 if (newvp == dvp)
 1205                                         vrele(newvp);
 1206                                 else
 1207                                         vput(newvp);
 1208                         }
 1209                         error = ENOENT;
 1210                 }
 1211                 if (error != 0)
 1212                         return (error);
 1213                 if (attrflag)
 1214                         (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1215                             0, 1);
 1216         } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
 1217                 FREE((caddr_t)nfhp, M_NFSFH);
 1218                 VREF(dvp);
 1219                 newvp = dvp;
 1220                 if (attrflag)
 1221                         (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1222                             0, 1);
 1223         } else {
 1224                 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL);
 1225                 if (error)
 1226                         return (error);
 1227                 newvp = NFSTOV(np);
 1228                 if (attrflag)
 1229                         (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1230                             0, 1);
 1231                 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
 1232                     !(np->n_flag & NMODIFIED)) {                        
 1233                         /*
 1234                          * Flush the attribute cache when opening a
 1235                          * leaf node to ensure that fresh attributes
 1236                          * are fetched in nfs_open() since we did not
 1237                          * fetch attributes from the LOOKUP reply.
 1238                          */
 1239                         mtx_lock(&np->n_mtx);
 1240                         np->n_attrstamp = 0;
 1241                         mtx_unlock(&np->n_mtx);
 1242                 }
 1243         }
 1244         if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
 1245                 cnp->cn_flags |= SAVENAME;
 1246         if ((cnp->cn_flags & MAKEENTRY) &&
 1247             (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
 1248                 np->n_ctime = np->n_vattr.na_vattr.va_ctime;
 1249                 cache_enter(dvp, newvp, cnp);
 1250         }
 1251         *vpp = newvp;
 1252         return (0);
 1253 }
 1254 
 1255 /*
 1256  * nfs read call.
 1257  * Just call ncl_bioread() to do the work.
 1258  */
 1259 static int
 1260 nfs_read(struct vop_read_args *ap)
 1261 {
 1262         struct vnode *vp = ap->a_vp;
 1263 
 1264         switch (vp->v_type) {
 1265         case VREG:
 1266                 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
 1267         case VDIR:
 1268                 return (EISDIR);
 1269         default:
 1270                 return (EOPNOTSUPP);
 1271         }
 1272 }
 1273 
 1274 /*
 1275  * nfs readlink call
 1276  */
 1277 static int
 1278 nfs_readlink(struct vop_readlink_args *ap)
 1279 {
 1280         struct vnode *vp = ap->a_vp;
 1281 
 1282         if (vp->v_type != VLNK)
 1283                 return (EINVAL);
 1284         return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
 1285 }
 1286 
 1287 /*
 1288  * Do a readlink rpc.
 1289  * Called by ncl_doio() from below the buffer cache.
 1290  */
 1291 int
 1292 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 1293 {
 1294         int error, ret, attrflag;
 1295         struct nfsvattr nfsva;
 1296 
 1297         error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
 1298             &attrflag, NULL);
 1299         if (attrflag) {
 1300                 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
 1301                 if (ret && !error)
 1302                         error = ret;
 1303         }
 1304         if (error && NFS_ISV4(vp))
 1305                 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
 1306         return (error);
 1307 }
 1308 
 1309 /*
 1310  * nfs read rpc call
 1311  * Ditto above
 1312  */
 1313 int
 1314 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 1315 {
 1316         int error, ret, attrflag;
 1317         struct nfsvattr nfsva;
 1318 
 1319         error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag,
 1320             NULL);
 1321         if (attrflag) {
 1322                 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
 1323                 if (ret && !error)
 1324                         error = ret;
 1325         }
 1326         if (error && NFS_ISV4(vp))
 1327                 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
 1328         return (error);
 1329 }
 1330 
 1331 /*
 1332  * nfs write call
 1333  */
 1334 int
 1335 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
 1336     int *iomode, int *must_commit, int called_from_strategy)
 1337 {
 1338         struct nfsvattr nfsva;
 1339         int error = 0, attrflag, ret;
 1340         u_char verf[NFSX_VERF];
 1341         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 1342 
 1343         *must_commit = 0;
 1344         error = nfsrpc_write(vp, uiop, iomode, verf, cred,
 1345             uiop->uio_td, &nfsva, &attrflag, NULL, called_from_strategy);
 1346         NFSLOCKMNT(nmp);
 1347         if (!error && NFSHASWRITEVERF(nmp) &&
 1348             NFSBCMP(verf, nmp->nm_verf, NFSX_VERF)) {
 1349                 *must_commit = 1;
 1350                 NFSBCOPY(verf, nmp->nm_verf, NFSX_VERF);
 1351         }
 1352         NFSUNLOCKMNT(nmp);
 1353         if (attrflag) {
 1354                 if (VTONFS(vp)->n_flag & ND_NFSV4)
 1355                         ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
 1356                             1);
 1357                 else
 1358                         ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
 1359                             1);
 1360                 if (ret && !error)
 1361                         error = ret;
 1362         }
 1363         if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
 1364                 *iomode = NFSWRITE_FILESYNC;
 1365         if (error && NFS_ISV4(vp))
 1366                 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
 1367         return (error);
 1368 }
 1369 
 1370 /*
 1371  * nfs mknod rpc
 1372  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
 1373  * mode set to specify the file type and the size field for rdev.
 1374  */
 1375 static int
 1376 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
 1377     struct vattr *vap)
 1378 {
 1379         struct nfsvattr nfsva, dnfsva;
 1380         struct vnode *newvp = NULL;
 1381         struct nfsnode *np = NULL, *dnp;
 1382         struct nfsfh *nfhp;
 1383         struct vattr vattr;
 1384         int error = 0, attrflag, dattrflag;
 1385         u_int32_t rdev;
 1386 
 1387         if (vap->va_type == VCHR || vap->va_type == VBLK)
 1388                 rdev = vap->va_rdev;
 1389         else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
 1390                 rdev = 0xffffffff;
 1391         else
 1392                 return (EOPNOTSUPP);
 1393         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
 1394                 return (error);
 1395         error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
 1396             rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
 1397             &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
 1398         if (!error) {
 1399                 if (!nfhp)
 1400                         (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
 1401                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
 1402                             &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
 1403                             NULL);
 1404                 if (nfhp)
 1405                         error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
 1406                             cnp->cn_thread, &np, NULL);
 1407         }
 1408         if (dattrflag)
 1409                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 1410         if (!error) {
 1411                 newvp = NFSTOV(np);
 1412                 if (attrflag)
 1413                         error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1414                             0, 1);
 1415         }
 1416         if (!error) {
 1417                 if ((cnp->cn_flags & MAKEENTRY))
 1418                         cache_enter(dvp, newvp, cnp);
 1419                 *vpp = newvp;
 1420         } else if (NFS_ISV4(dvp)) {
 1421                 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
 1422                     vap->va_gid);
 1423         }
 1424         dnp = VTONFS(dvp);
 1425         mtx_lock(&dnp->n_mtx);
 1426         dnp->n_flag |= NMODIFIED;
 1427         if (!dattrflag)
 1428                 dnp->n_attrstamp = 0;
 1429         mtx_unlock(&dnp->n_mtx);
 1430         return (error);
 1431 }
 1432 
 1433 /*
 1434  * nfs mknod vop
 1435  * just call nfs_mknodrpc() to do the work.
 1436  */
 1437 /* ARGSUSED */
 1438 static int
 1439 nfs_mknod(struct vop_mknod_args *ap)
 1440 {
 1441         return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
 1442 }
 1443 
 1444 static u_long create_verf;
 1445 /*
 1446  * nfs file create call
 1447  */
 1448 static int
 1449 nfs_create(struct vop_create_args *ap)
 1450 {
 1451         struct vnode *dvp = ap->a_dvp;
 1452         struct vattr *vap = ap->a_vap;
 1453         struct componentname *cnp = ap->a_cnp;
 1454         struct nfsnode *np = NULL, *dnp;
 1455         struct vnode *newvp = NULL;
 1456         struct nfsmount *nmp;
 1457         struct nfsvattr dnfsva, nfsva;
 1458         struct nfsfh *nfhp;
 1459         nfsquad_t cverf;
 1460         int error = 0, attrflag, dattrflag, fmode = 0;
 1461         struct vattr vattr;
 1462 
 1463         /*
 1464          * Oops, not for me..
 1465          */
 1466         if (vap->va_type == VSOCK)
 1467                 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
 1468 
 1469         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
 1470                 return (error);
 1471         if (vap->va_vaflags & VA_EXCLUSIVE)
 1472                 fmode |= O_EXCL;
 1473         dnp = VTONFS(dvp);
 1474         nmp = VFSTONFS(vnode_mount(dvp));
 1475 again:
 1476         /* For NFSv4, wait until any remove is done. */
 1477         mtx_lock(&dnp->n_mtx);
 1478         while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
 1479                 dnp->n_flag |= NREMOVEWANT;
 1480                 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
 1481         }
 1482         mtx_unlock(&dnp->n_mtx);
 1483 
 1484 #ifdef INET
 1485         CURVNET_SET(CRED_TO_VNET(cnp->cn_cred));
 1486         IN_IFADDR_RLOCK();
 1487         if (!TAILQ_EMPTY(&V_in_ifaddrhead))
 1488                 cverf.lval[0] = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
 1489         else
 1490 #endif
 1491                 cverf.lval[0] = create_verf;
 1492 #ifdef INET
 1493         IN_IFADDR_RUNLOCK();
 1494         CURVNET_RESTORE();
 1495 #endif
 1496         cverf.lval[1] = ++create_verf;
 1497         error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 1498             vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
 1499             &nfhp, &attrflag, &dattrflag, NULL);
 1500         if (!error) {
 1501                 if (nfhp == NULL)
 1502                         (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
 1503                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
 1504                             &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
 1505                             NULL);
 1506                 if (nfhp != NULL)
 1507                         error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
 1508                             cnp->cn_thread, &np, NULL);
 1509         }
 1510         if (dattrflag)
 1511                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 1512         if (!error) {
 1513                 newvp = NFSTOV(np);
 1514                 if (attrflag)
 1515                         error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1516                             0, 1);
 1517         }
 1518         if (error) {
 1519                 if (newvp != NULL) {
 1520                         vrele(newvp);
 1521                         newvp = NULL;
 1522                 }
 1523                 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
 1524                     error == NFSERR_NOTSUPP) {
 1525                         fmode &= ~O_EXCL;
 1526                         goto again;
 1527                 }
 1528         } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
 1529                 if (nfscl_checksattr(vap, &nfsva)) {
 1530                         error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
 1531                             cnp->cn_thread, &nfsva, &attrflag, NULL);
 1532                         if (error && (vap->va_uid != (uid_t)VNOVAL ||
 1533                             vap->va_gid != (gid_t)VNOVAL)) {
 1534                                 /* try again without setting uid/gid */
 1535                                 vap->va_uid = (uid_t)VNOVAL;
 1536                                 vap->va_gid = (uid_t)VNOVAL;
 1537                                 error = nfsrpc_setattr(newvp, vap, NULL, 
 1538                                     cnp->cn_cred, cnp->cn_thread, &nfsva,
 1539                                     &attrflag, NULL);
 1540                         }
 1541                         if (attrflag)
 1542                                 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
 1543                                     NULL, 0, 1);
 1544                 }
 1545         }
 1546         if (!error) {
 1547                 if (cnp->cn_flags & MAKEENTRY)
 1548                         cache_enter(dvp, newvp, cnp);
 1549                 *ap->a_vpp = newvp;
 1550         } else if (NFS_ISV4(dvp)) {
 1551                 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
 1552                     vap->va_gid);
 1553         }
 1554         mtx_lock(&dnp->n_mtx);
 1555         dnp->n_flag |= NMODIFIED;
 1556         if (!dattrflag)
 1557                 dnp->n_attrstamp = 0;
 1558         mtx_unlock(&dnp->n_mtx);
 1559         return (error);
 1560 }
 1561 
 1562 /*
 1563  * nfs file remove call
 1564  * To try and make nfs semantics closer to ufs semantics, a file that has
 1565  * other processes using the vnode is renamed instead of removed and then
 1566  * removed later on the last close.
 1567  * - If v_usecount > 1
 1568  *        If a rename is not already in the works
 1569  *           call nfs_sillyrename() to set it up
 1570  *     else
 1571  *        do the remove rpc
 1572  */
 1573 static int
 1574 nfs_remove(struct vop_remove_args *ap)
 1575 {
 1576         struct vnode *vp = ap->a_vp;
 1577         struct vnode *dvp = ap->a_dvp;
 1578         struct componentname *cnp = ap->a_cnp;
 1579         struct nfsnode *np = VTONFS(vp);
 1580         int error = 0;
 1581         struct vattr vattr;
 1582 
 1583         KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
 1584         KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
 1585         if (vp->v_type == VDIR)
 1586                 error = EPERM;
 1587         else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
 1588             VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
 1589             vattr.va_nlink > 1)) {
 1590                 /*
 1591                  * Purge the name cache so that the chance of a lookup for
 1592                  * the name succeeding while the remove is in progress is
 1593                  * minimized. Without node locking it can still happen, such
 1594                  * that an I/O op returns ESTALE, but since you get this if
 1595                  * another host removes the file..
 1596                  */
 1597                 cache_purge(vp);
 1598                 /*
 1599                  * throw away biocache buffers, mainly to avoid
 1600                  * unnecessary delayed writes later.
 1601                  */
 1602                 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
 1603                 /* Do the rpc */
 1604                 if (error != EINTR && error != EIO)
 1605                         error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
 1606                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
 1607                 /*
 1608                  * Kludge City: If the first reply to the remove rpc is lost..
 1609                  *   the reply to the retransmitted request will be ENOENT
 1610                  *   since the file was in fact removed
 1611                  *   Therefore, we cheat and return success.
 1612                  */
 1613                 if (error == ENOENT)
 1614                         error = 0;
 1615         } else if (!np->n_sillyrename)
 1616                 error = nfs_sillyrename(dvp, vp, cnp);
 1617         np->n_attrstamp = 0;
 1618         return (error);
 1619 }
 1620 
 1621 /*
 1622  * nfs file remove rpc called from nfs_inactive
 1623  */
 1624 int
 1625 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
 1626 {
 1627         /*
 1628          * Make sure that the directory vnode is still valid.
 1629          * XXX we should lock sp->s_dvp here.
 1630          */
 1631         if (sp->s_dvp->v_type == VBAD)
 1632                 return (0);
 1633         return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
 1634             sp->s_cred, NULL));
 1635 }
 1636 
 1637 /*
 1638  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
 1639  */
 1640 static int
 1641 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
 1642     int namelen, struct ucred *cred, struct thread *td)
 1643 {
 1644         struct nfsvattr dnfsva;
 1645         struct nfsnode *dnp = VTONFS(dvp);
 1646         int error = 0, dattrflag;
 1647 
 1648         mtx_lock(&dnp->n_mtx);
 1649         dnp->n_flag |= NREMOVEINPROG;
 1650         mtx_unlock(&dnp->n_mtx);
 1651         error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
 1652             &dattrflag, NULL);
 1653         mtx_lock(&dnp->n_mtx);
 1654         if ((dnp->n_flag & NREMOVEWANT)) {
 1655                 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
 1656                 mtx_unlock(&dnp->n_mtx);
 1657                 wakeup((caddr_t)dnp);
 1658         } else {
 1659                 dnp->n_flag &= ~NREMOVEINPROG;
 1660                 mtx_unlock(&dnp->n_mtx);
 1661         }
 1662         if (dattrflag)
 1663                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 1664         mtx_lock(&dnp->n_mtx);
 1665         dnp->n_flag |= NMODIFIED;
 1666         if (!dattrflag)
 1667                 dnp->n_attrstamp = 0;
 1668         mtx_unlock(&dnp->n_mtx);
 1669         if (error && NFS_ISV4(dvp))
 1670                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
 1671         return (error);
 1672 }
 1673 
 1674 /*
 1675  * nfs file rename call
 1676  */
 1677 static int
 1678 nfs_rename(struct vop_rename_args *ap)
 1679 {
 1680         struct vnode *fvp = ap->a_fvp;
 1681         struct vnode *tvp = ap->a_tvp;
 1682         struct vnode *fdvp = ap->a_fdvp;
 1683         struct vnode *tdvp = ap->a_tdvp;
 1684         struct componentname *tcnp = ap->a_tcnp;
 1685         struct componentname *fcnp = ap->a_fcnp;
 1686         struct nfsnode *fnp = VTONFS(ap->a_fvp);
 1687         struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
 1688         struct nfsv4node *newv4 = NULL;
 1689         int error;
 1690 
 1691         KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
 1692             (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
 1693         /* Check for cross-device rename */
 1694         if ((fvp->v_mount != tdvp->v_mount) ||
 1695             (tvp && (fvp->v_mount != tvp->v_mount))) {
 1696                 error = EXDEV;
 1697                 goto out;
 1698         }
 1699 
 1700         if (fvp == tvp) {
 1701                 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
 1702                 error = 0;
 1703                 goto out;
 1704         }
 1705         if ((error = vn_lock(fvp, LK_EXCLUSIVE)))
 1706                 goto out;
 1707 
 1708         /*
 1709          * We have to flush B_DELWRI data prior to renaming
 1710          * the file.  If we don't, the delayed-write buffers
 1711          * can be flushed out later after the file has gone stale
 1712          * under NFSV3.  NFSV2 does not have this problem because
 1713          * ( as far as I can tell ) it flushes dirty buffers more
 1714          * often.
 1715          * 
 1716          * Skip the rename operation if the fsync fails, this can happen
 1717          * due to the server's volume being full, when we pushed out data
 1718          * that was written back to our cache earlier. Not checking for
 1719          * this condition can result in potential (silent) data loss.
 1720          */
 1721         error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
 1722         VOP_UNLOCK(fvp, 0);
 1723         if (!error && tvp)
 1724                 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
 1725         if (error)
 1726                 goto out;
 1727 
 1728         /*
 1729          * If the tvp exists and is in use, sillyrename it before doing the
 1730          * rename of the new file over it.
 1731          * XXX Can't sillyrename a directory.
 1732          */
 1733         if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
 1734                 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
 1735                 vput(tvp);
 1736                 tvp = NULL;
 1737         }
 1738 
 1739         error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
 1740             tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
 1741             tcnp->cn_thread);
 1742 
 1743         if (!error) {
 1744                 /*
 1745                  * For NFSv4, check to see if it is the same name and
 1746                  * replace the name, if it is different.
 1747                  */
 1748                 MALLOC(newv4, struct nfsv4node *,
 1749                     sizeof (struct nfsv4node) +
 1750                     tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
 1751                     M_NFSV4NODE, M_WAITOK);
 1752                 mtx_lock(&tdnp->n_mtx);
 1753                 mtx_lock(&fnp->n_mtx);
 1754                 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
 1755                     (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
 1756                       NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
 1757                       tcnp->cn_namelen) ||
 1758                       tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
 1759                       NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
 1760                         tdnp->n_fhp->nfh_len))) {
 1761 #ifdef notdef
 1762 { char nnn[100]; int nnnl;
 1763 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
 1764 bcopy(tcnp->cn_nameptr, nnn, nnnl);
 1765 nnn[nnnl] = '\0';
 1766 printf("ren replace=%s\n",nnn);
 1767 }
 1768 #endif
 1769                         FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
 1770                         fnp->n_v4 = newv4;
 1771                         newv4 = NULL;
 1772                         fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
 1773                         fnp->n_v4->n4_namelen = tcnp->cn_namelen;
 1774                         NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
 1775                             tdnp->n_fhp->nfh_len);
 1776                         NFSBCOPY(tcnp->cn_nameptr,
 1777                             NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
 1778                 }
 1779                 mtx_unlock(&tdnp->n_mtx);
 1780                 mtx_unlock(&fnp->n_mtx);
 1781                 if (newv4 != NULL)
 1782                         FREE((caddr_t)newv4, M_NFSV4NODE);
 1783         }
 1784 
 1785         if (fvp->v_type == VDIR) {
 1786                 if (tvp != NULL && tvp->v_type == VDIR)
 1787                         cache_purge(tdvp);
 1788                 cache_purge(fdvp);
 1789         }
 1790 
 1791 out:
 1792         if (tdvp == tvp)
 1793                 vrele(tdvp);
 1794         else
 1795                 vput(tdvp);
 1796         if (tvp)
 1797                 vput(tvp);
 1798         vrele(fdvp);
 1799         vrele(fvp);
 1800         /*
 1801          * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
 1802          */
 1803         if (error == ENOENT)
 1804                 error = 0;
 1805         return (error);
 1806 }
 1807 
 1808 /*
 1809  * nfs file rename rpc called from nfs_remove() above
 1810  */
 1811 static int
 1812 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
 1813     struct sillyrename *sp)
 1814 {
 1815 
 1816         return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
 1817             sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
 1818             scnp->cn_thread));
 1819 }
 1820 
 1821 /*
 1822  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
 1823  */
 1824 static int
 1825 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
 1826     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
 1827     int tnamelen, struct ucred *cred, struct thread *td)
 1828 {
 1829         struct nfsvattr fnfsva, tnfsva;
 1830         struct nfsnode *fdnp = VTONFS(fdvp);
 1831         struct nfsnode *tdnp = VTONFS(tdvp);
 1832         int error = 0, fattrflag, tattrflag;
 1833 
 1834         error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
 1835             tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
 1836             &tattrflag, NULL, NULL);
 1837         mtx_lock(&fdnp->n_mtx);
 1838         fdnp->n_flag |= NMODIFIED;
 1839         mtx_unlock(&fdnp->n_mtx);
 1840         mtx_lock(&tdnp->n_mtx);
 1841         tdnp->n_flag |= NMODIFIED;
 1842         mtx_unlock(&tdnp->n_mtx);
 1843         if (fattrflag)
 1844                 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
 1845         else
 1846                 fdnp->n_attrstamp = 0;
 1847         if (tattrflag)
 1848                 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
 1849         else
 1850                 tdnp->n_attrstamp = 0;
 1851         if (error && NFS_ISV4(fdvp))
 1852                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
 1853         return (error);
 1854 }
 1855 
 1856 /*
 1857  * nfs hard link create call
 1858  */
 1859 static int
 1860 nfs_link(struct vop_link_args *ap)
 1861 {
 1862         struct vnode *vp = ap->a_vp;
 1863         struct vnode *tdvp = ap->a_tdvp;
 1864         struct componentname *cnp = ap->a_cnp;
 1865         struct nfsnode *tdnp;
 1866         struct nfsvattr nfsva, dnfsva;
 1867         int error = 0, attrflag, dattrflag;
 1868 
 1869         if (vp->v_mount != tdvp->v_mount) {
 1870                 return (EXDEV);
 1871         }
 1872 
 1873         /*
 1874          * Push all writes to the server, so that the attribute cache
 1875          * doesn't get "out of sync" with the server.
 1876          * XXX There should be a better way!
 1877          */
 1878         VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
 1879 
 1880         error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
 1881             cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
 1882             &dattrflag, NULL);
 1883         tdnp = VTONFS(tdvp);
 1884         mtx_lock(&tdnp->n_mtx);
 1885         tdnp->n_flag |= NMODIFIED;
 1886         mtx_unlock(&tdnp->n_mtx);
 1887         if (attrflag)
 1888                 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
 1889         else
 1890                 VTONFS(vp)->n_attrstamp = 0;
 1891         if (dattrflag)
 1892                 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
 1893         else
 1894                 tdnp->n_attrstamp = 0;
 1895         /*
 1896          * If negative lookup caching is enabled, I might as well
 1897          * add an entry for this node. Not necessary for correctness,
 1898          * but if negative caching is enabled, then the system
 1899          * must care about lookup caching hit rate, so...
 1900          */
 1901         if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
 1902             (cnp->cn_flags & MAKEENTRY))
 1903                 cache_enter(tdvp, vp, cnp);
 1904         if (error && NFS_ISV4(vp))
 1905                 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
 1906                     (gid_t)0);
 1907         return (error);
 1908 }
 1909 
 1910 /*
 1911  * nfs symbolic link create call
 1912  */
 1913 static int
 1914 nfs_symlink(struct vop_symlink_args *ap)
 1915 {
 1916         struct vnode *dvp = ap->a_dvp;
 1917         struct vattr *vap = ap->a_vap;
 1918         struct componentname *cnp = ap->a_cnp;
 1919         struct nfsvattr nfsva, dnfsva;
 1920         struct nfsfh *nfhp;
 1921         struct nfsnode *np = NULL, *dnp;
 1922         struct vnode *newvp = NULL;
 1923         int error = 0, attrflag, dattrflag, ret;
 1924 
 1925         vap->va_type = VLNK;
 1926         error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 1927             ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
 1928             &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
 1929         if (nfhp) {
 1930                 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
 1931                     &np, NULL);
 1932                 if (!ret)
 1933                         newvp = NFSTOV(np);
 1934                 else if (!error)
 1935                         error = ret;
 1936         }
 1937         if (newvp != NULL) {
 1938                 if (attrflag)
 1939                         (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1940                             0, 1);
 1941         } else if (!error) {
 1942                 /*
 1943                  * If we do not have an error and we could not extract the
 1944                  * newvp from the response due to the request being NFSv2, we
 1945                  * have to do a lookup in order to obtain a newvp to return.
 1946                  */
 1947                 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 1948                     cnp->cn_cred, cnp->cn_thread, &np);
 1949                 if (!error)
 1950                         newvp = NFSTOV(np);
 1951         }
 1952         if (error) {
 1953                 if (newvp)
 1954                         vput(newvp);
 1955                 if (NFS_ISV4(dvp))
 1956                         error = nfscl_maperr(cnp->cn_thread, error,
 1957                             vap->va_uid, vap->va_gid);
 1958         } else {
 1959                 /*
 1960                  * If negative lookup caching is enabled, I might as well
 1961                  * add an entry for this node. Not necessary for correctness,
 1962                  * but if negative caching is enabled, then the system
 1963                  * must care about lookup caching hit rate, so...
 1964                  */
 1965                 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
 1966                     (cnp->cn_flags & MAKEENTRY))
 1967                         cache_enter(dvp, newvp, cnp);
 1968                 *ap->a_vpp = newvp;
 1969         }
 1970 
 1971         dnp = VTONFS(dvp);
 1972         mtx_lock(&dnp->n_mtx);
 1973         dnp->n_flag |= NMODIFIED;
 1974         mtx_unlock(&dnp->n_mtx);
 1975         if (dattrflag)
 1976                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 1977         else
 1978                 dnp->n_attrstamp = 0;
 1979         return (error);
 1980 }
 1981 
 1982 /*
 1983  * nfs make dir call
 1984  */
 1985 static int
 1986 nfs_mkdir(struct vop_mkdir_args *ap)
 1987 {
 1988         struct vnode *dvp = ap->a_dvp;
 1989         struct vattr *vap = ap->a_vap;
 1990         struct componentname *cnp = ap->a_cnp;
 1991         struct nfsnode *np = NULL, *dnp;
 1992         struct vnode *newvp = NULL;
 1993         struct vattr vattr;
 1994         struct nfsfh *nfhp;
 1995         struct nfsvattr nfsva, dnfsva;
 1996         int error = 0, attrflag, dattrflag, ret;
 1997 
 1998         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
 1999                 return (error);
 2000         vap->va_type = VDIR;
 2001         error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 2002             vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
 2003             &attrflag, &dattrflag, NULL);
 2004         dnp = VTONFS(dvp);
 2005         mtx_lock(&dnp->n_mtx);
 2006         dnp->n_flag |= NMODIFIED;
 2007         mtx_unlock(&dnp->n_mtx);
 2008         if (dattrflag)
 2009                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 2010         else
 2011                 dnp->n_attrstamp = 0;
 2012         if (nfhp) {
 2013                 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
 2014                     &np, NULL);
 2015                 if (!ret) {
 2016                         newvp = NFSTOV(np);
 2017                         if (attrflag)
 2018                            (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
 2019                                 NULL, 0, 1);
 2020                 } else if (!error)
 2021                         error = ret;
 2022         }
 2023         if (!error && newvp == NULL) {
 2024                 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 2025                     cnp->cn_cred, cnp->cn_thread, &np);
 2026                 if (!error) {
 2027                         newvp = NFSTOV(np);
 2028                         if (newvp->v_type != VDIR)
 2029                                 error = EEXIST;
 2030                 }
 2031         }
 2032         if (error) {
 2033                 if (newvp)
 2034                         vput(newvp);
 2035                 if (NFS_ISV4(dvp))
 2036                         error = nfscl_maperr(cnp->cn_thread, error,
 2037                             vap->va_uid, vap->va_gid);
 2038         } else {
 2039                 /*
 2040                  * If negative lookup caching is enabled, I might as well
 2041                  * add an entry for this node. Not necessary for correctness,
 2042                  * but if negative caching is enabled, then the system
 2043                  * must care about lookup caching hit rate, so...
 2044                  */
 2045                 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
 2046                     (cnp->cn_flags & MAKEENTRY))
 2047                         cache_enter(dvp, newvp, cnp);
 2048                 *ap->a_vpp = newvp;
 2049         }
 2050         return (error);
 2051 }
 2052 
 2053 /*
 2054  * nfs remove directory call
 2055  */
 2056 static int
 2057 nfs_rmdir(struct vop_rmdir_args *ap)
 2058 {
 2059         struct vnode *vp = ap->a_vp;
 2060         struct vnode *dvp = ap->a_dvp;
 2061         struct componentname *cnp = ap->a_cnp;
 2062         struct nfsnode *dnp;
 2063         struct nfsvattr dnfsva;
 2064         int error, dattrflag;
 2065 
 2066         if (dvp == vp)
 2067                 return (EINVAL);
 2068         error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 2069             cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
 2070         dnp = VTONFS(dvp);
 2071         mtx_lock(&dnp->n_mtx);
 2072         dnp->n_flag |= NMODIFIED;
 2073         mtx_unlock(&dnp->n_mtx);
 2074         if (dattrflag)
 2075                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 2076         else
 2077                 dnp->n_attrstamp = 0;
 2078 
 2079         cache_purge(dvp);
 2080         cache_purge(vp);
 2081         if (error && NFS_ISV4(dvp))
 2082                 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
 2083                     (gid_t)0);
 2084         /*
 2085          * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
 2086          */
 2087         if (error == ENOENT)
 2088                 error = 0;
 2089         return (error);
 2090 }
 2091 
 2092 /*
 2093  * nfs readdir call
 2094  */
 2095 static int
 2096 nfs_readdir(struct vop_readdir_args *ap)
 2097 {
 2098         struct vnode *vp = ap->a_vp;
 2099         struct nfsnode *np = VTONFS(vp);
 2100         struct uio *uio = ap->a_uio;
 2101         int tresid, error = 0;
 2102         struct vattr vattr;
 2103         
 2104         if (vp->v_type != VDIR) 
 2105                 return(EPERM);
 2106 
 2107         /*
 2108          * First, check for hit on the EOF offset cache
 2109          */
 2110         if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
 2111             (np->n_flag & NMODIFIED) == 0) {
 2112                 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
 2113                         mtx_lock(&np->n_mtx);
 2114                         if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
 2115                             !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
 2116                                 mtx_unlock(&np->n_mtx);
 2117                                 NFSINCRGLOBAL(newnfsstats.direofcache_hits);
 2118                                 return (0);
 2119                         } else
 2120                                 mtx_unlock(&np->n_mtx);
 2121                 }
 2122         }
 2123 
 2124         /*
 2125          * Call ncl_bioread() to do the real work.
 2126          */
 2127         tresid = uio->uio_resid;
 2128         error = ncl_bioread(vp, uio, 0, ap->a_cred);
 2129 
 2130         if (!error && uio->uio_resid == tresid)
 2131                 NFSINCRGLOBAL(newnfsstats.direofcache_misses);
 2132         return (error);
 2133 }
 2134 
 2135 /*
 2136  * Readdir rpc call.
 2137  * Called from below the buffer cache by ncl_doio().
 2138  */
 2139 int
 2140 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
 2141     struct thread *td)
 2142 {
 2143         struct nfsvattr nfsva;
 2144         nfsuint64 *cookiep, cookie;
 2145         struct nfsnode *dnp = VTONFS(vp);
 2146         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2147         int error = 0, eof, attrflag;
 2148 
 2149         KASSERT(uiop->uio_iovcnt == 1 &&
 2150             (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
 2151             (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
 2152             ("nfs readdirrpc bad uio"));
 2153 
 2154         /*
 2155          * If there is no cookie, assume directory was stale.
 2156          */
 2157         ncl_dircookie_lock(dnp);
 2158         cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
 2159         if (cookiep) {
 2160                 cookie = *cookiep;
 2161                 ncl_dircookie_unlock(dnp);
 2162         } else {
 2163                 ncl_dircookie_unlock(dnp);              
 2164                 return (NFSERR_BAD_COOKIE);
 2165         }
 2166 
 2167         if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
 2168                 (void)ncl_fsinfo(nmp, vp, cred, td);
 2169 
 2170         error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
 2171             &attrflag, &eof, NULL);
 2172         if (attrflag)
 2173                 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
 2174 
 2175         if (!error) {
 2176                 /*
 2177                  * We are now either at the end of the directory or have filled
 2178                  * the block.
 2179                  */
 2180                 if (eof)
 2181                         dnp->n_direofoffset = uiop->uio_offset;
 2182                 else {
 2183                         if (uiop->uio_resid > 0)
 2184                                 ncl_printf("EEK! readdirrpc resid > 0\n");
 2185                         ncl_dircookie_lock(dnp);
 2186                         cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
 2187                         *cookiep = cookie;
 2188                         ncl_dircookie_unlock(dnp);
 2189                 }
 2190         } else if (NFS_ISV4(vp)) {
 2191                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
 2192         }
 2193         return (error);
 2194 }
 2195 
 2196 /*
 2197  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
 2198  */
 2199 int
 2200 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
 2201     struct thread *td)
 2202 {
 2203         struct nfsvattr nfsva;
 2204         nfsuint64 *cookiep, cookie;
 2205         struct nfsnode *dnp = VTONFS(vp);
 2206         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2207         int error = 0, attrflag, eof;
 2208 
 2209         KASSERT(uiop->uio_iovcnt == 1 &&
 2210             (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
 2211             (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
 2212             ("nfs readdirplusrpc bad uio"));
 2213 
 2214         /*
 2215          * If there is no cookie, assume directory was stale.
 2216          */
 2217         ncl_dircookie_lock(dnp);
 2218         cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
 2219         if (cookiep) {
 2220                 cookie = *cookiep;
 2221                 ncl_dircookie_unlock(dnp);
 2222         } else {
 2223                 ncl_dircookie_unlock(dnp);
 2224                 return (NFSERR_BAD_COOKIE);
 2225         }
 2226 
 2227         if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
 2228                 (void)ncl_fsinfo(nmp, vp, cred, td);
 2229         error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
 2230             &attrflag, &eof, NULL);
 2231         if (attrflag)
 2232                 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
 2233 
 2234         if (!error) {
 2235                 /*
 2236                  * We are now either at end of the directory or have filled the
 2237                  * the block.
 2238                  */
 2239                 if (eof)
 2240                         dnp->n_direofoffset = uiop->uio_offset;
 2241                 else {
 2242                         if (uiop->uio_resid > 0)
 2243                                 ncl_printf("EEK! readdirplusrpc resid > 0\n");
 2244                         ncl_dircookie_lock(dnp);
 2245                         cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
 2246                         *cookiep = cookie;
 2247                         ncl_dircookie_unlock(dnp);
 2248                 }
 2249         } else if (NFS_ISV4(vp)) {
 2250                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
 2251         }
 2252         return (error);
 2253 }
 2254 
 2255 /*
 2256  * Silly rename. To make the NFS filesystem that is stateless look a little
 2257  * more like the "ufs" a remove of an active vnode is translated to a rename
 2258  * to a funny looking filename that is removed by nfs_inactive on the
 2259  * nfsnode. There is the potential for another process on a different client
 2260  * to create the same funny name between the nfs_lookitup() fails and the
 2261  * nfs_rename() completes, but...
 2262  */
 2263 static int
 2264 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
 2265 {
 2266         struct sillyrename *sp;
 2267         struct nfsnode *np;
 2268         int error;
 2269         short pid;
 2270         unsigned int lticks;
 2271 
 2272         cache_purge(dvp);
 2273         np = VTONFS(vp);
 2274         KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
 2275         MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
 2276             M_NEWNFSREQ, M_WAITOK);
 2277         sp->s_cred = crhold(cnp->cn_cred);
 2278         sp->s_dvp = dvp;
 2279         VREF(dvp);
 2280 
 2281         /* 
 2282          * Fudge together a funny name.
 2283          * Changing the format of the funny name to accomodate more 
 2284          * sillynames per directory.
 2285          * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 
 2286          * CPU ticks since boot.
 2287          */
 2288         pid = cnp->cn_thread->td_proc->p_pid;
 2289         lticks = (unsigned int)ticks;
 2290         for ( ; ; ) {
 2291                 sp->s_namlen = sprintf(sp->s_name, 
 2292                                        ".nfs.%08x.%04x4.4", lticks, 
 2293                                        pid);
 2294                 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2295                                  cnp->cn_thread, NULL))
 2296                         break;
 2297                 lticks++;
 2298         }
 2299         error = nfs_renameit(dvp, vp, cnp, sp);
 2300         if (error)
 2301                 goto bad;
 2302         error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2303                 cnp->cn_thread, &np);
 2304         np->n_sillyrename = sp;
 2305         return (0);
 2306 bad:
 2307         vrele(sp->s_dvp);
 2308         crfree(sp->s_cred);
 2309         free((caddr_t)sp, M_NEWNFSREQ);
 2310         return (error);
 2311 }
 2312 
 2313 /*
 2314  * Look up a file name and optionally either update the file handle or
 2315  * allocate an nfsnode, depending on the value of npp.
 2316  * npp == NULL  --> just do the lookup
 2317  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
 2318  *                      handled too
 2319  * *npp != NULL --> update the file handle in the vnode
 2320  */
 2321 static int
 2322 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
 2323     struct thread *td, struct nfsnode **npp)
 2324 {
 2325         struct vnode *newvp = NULL, *vp;
 2326         struct nfsnode *np, *dnp = VTONFS(dvp);
 2327         struct nfsfh *nfhp, *onfhp;
 2328         struct nfsvattr nfsva, dnfsva;
 2329         struct componentname cn;
 2330         int error = 0, attrflag, dattrflag;
 2331         u_int hash;
 2332 
 2333         error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
 2334             &nfhp, &attrflag, &dattrflag, NULL);
 2335         if (dattrflag)
 2336                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 2337         if (npp && !error) {
 2338                 if (*npp != NULL) {
 2339                     np = *npp;
 2340                     vp = NFSTOV(np);
 2341                     /*
 2342                      * For NFSv4, check to see if it is the same name and
 2343                      * replace the name, if it is different.
 2344                      */
 2345                     if (np->n_v4 != NULL && nfsva.na_type == VREG &&
 2346                         (np->n_v4->n4_namelen != len ||
 2347                          NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
 2348                          dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
 2349                          NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
 2350                          dnp->n_fhp->nfh_len))) {
 2351 #ifdef notdef
 2352 { char nnn[100]; int nnnl;
 2353 nnnl = (len < 100) ? len : 99;
 2354 bcopy(name, nnn, nnnl);
 2355 nnn[nnnl] = '\0';
 2356 printf("replace=%s\n",nnn);
 2357 }
 2358 #endif
 2359                             FREE((caddr_t)np->n_v4, M_NFSV4NODE);
 2360                             MALLOC(np->n_v4, struct nfsv4node *,
 2361                                 sizeof (struct nfsv4node) +
 2362                                 dnp->n_fhp->nfh_len + len - 1,
 2363                                 M_NFSV4NODE, M_WAITOK);
 2364                             np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
 2365                             np->n_v4->n4_namelen = len;
 2366                             NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
 2367                                 dnp->n_fhp->nfh_len);
 2368                             NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
 2369                     }
 2370                     hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
 2371                         FNV1_32_INIT);
 2372                     onfhp = np->n_fhp;
 2373                     /*
 2374                      * Rehash node for new file handle.
 2375                      */
 2376                     vfs_hash_rehash(vp, hash);
 2377                     np->n_fhp = nfhp;
 2378                     if (onfhp != NULL)
 2379                         FREE((caddr_t)onfhp, M_NFSFH);
 2380                     newvp = NFSTOV(np);
 2381                 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
 2382                     FREE((caddr_t)nfhp, M_NFSFH);
 2383                     VREF(dvp);
 2384                     newvp = dvp;
 2385                 } else {
 2386                     cn.cn_nameptr = name;
 2387                     cn.cn_namelen = len;
 2388                     error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
 2389                         &np, NULL);
 2390                     if (error)
 2391                         return (error);
 2392                     newvp = NFSTOV(np);
 2393                 }
 2394                 if (!attrflag && *npp == NULL) {
 2395                         vrele(newvp);
 2396                         return (ENOENT);
 2397                 }
 2398                 if (attrflag)
 2399                         (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 2400                             0, 1);
 2401         }
 2402         if (npp && *npp == NULL) {
 2403                 if (error) {
 2404                         if (newvp) {
 2405                                 if (newvp == dvp)
 2406                                         vrele(newvp);
 2407                                 else
 2408                                         vput(newvp);
 2409                         }
 2410                 } else
 2411                         *npp = np;
 2412         }
 2413         if (error && NFS_ISV4(dvp))
 2414                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
 2415         return (error);
 2416 }
 2417 
 2418 /*
 2419  * Nfs Version 3 and 4 commit rpc
 2420  */
 2421 int
 2422 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
 2423    struct thread *td)
 2424 {
 2425         struct nfsvattr nfsva;
 2426         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2427         int error, attrflag;
 2428         u_char verf[NFSX_VERF];
 2429 
 2430         mtx_lock(&nmp->nm_mtx);
 2431         if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
 2432                 mtx_unlock(&nmp->nm_mtx);
 2433                 return (0);
 2434         }
 2435         mtx_unlock(&nmp->nm_mtx);
 2436         error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva,
 2437             &attrflag, NULL);
 2438         if (!error) {
 2439                 if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) {
 2440                         NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF);
 2441                         error = NFSERR_STALEWRITEVERF;
 2442                 }
 2443                 if (!error && attrflag)
 2444                         (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
 2445                             0, 1);
 2446         } else if (NFS_ISV4(vp)) {
 2447                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
 2448         }
 2449         return (error);
 2450 }
 2451 
 2452 /*
 2453  * Strategy routine.
 2454  * For async requests when nfsiod(s) are running, queue the request by
 2455  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
 2456  * request.
 2457  */
 2458 static int
 2459 nfs_strategy(struct vop_strategy_args *ap)
 2460 {
 2461         struct buf *bp = ap->a_bp;
 2462         struct ucred *cr;
 2463 
 2464         KASSERT(!(bp->b_flags & B_DONE),
 2465             ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
 2466         BUF_ASSERT_HELD(bp);
 2467 
 2468         if (bp->b_iocmd == BIO_READ)
 2469                 cr = bp->b_rcred;
 2470         else
 2471                 cr = bp->b_wcred;
 2472 
 2473         /*
 2474          * If the op is asynchronous and an i/o daemon is waiting
 2475          * queue the request, wake it up and wait for completion
 2476          * otherwise just do it ourselves.
 2477          */
 2478         if ((bp->b_flags & B_ASYNC) == 0 ||
 2479             ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
 2480                 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1);
 2481         return (0);
 2482 }
 2483 
 2484 /*
 2485  * fsync vnode op. Just call ncl_flush() with commit == 1.
 2486  */
 2487 /* ARGSUSED */
 2488 static int
 2489 nfs_fsync(struct vop_fsync_args *ap)
 2490 {
 2491         return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0));
 2492 }
 2493 
 2494 /*
 2495  * Flush all the blocks associated with a vnode.
 2496  *      Walk through the buffer pool and push any dirty pages
 2497  *      associated with the vnode.
 2498  * If the called_from_renewthread argument is TRUE, it has been called
 2499  * from the NFSv4 renew thread and, as such, cannot block indefinitely
 2500  * waiting for a buffer write to complete.
 2501  */
 2502 int
 2503 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
 2504     int commit, int called_from_renewthread)
 2505 {
 2506         struct nfsnode *np = VTONFS(vp);
 2507         struct buf *bp;
 2508         int i;
 2509         struct buf *nbp;
 2510         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2511         int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
 2512         int passone = 1, trycnt = 0;
 2513         u_quad_t off, endoff, toff;
 2514         struct ucred* wcred = NULL;
 2515         struct buf **bvec = NULL;
 2516         struct bufobj *bo;
 2517 #ifndef NFS_COMMITBVECSIZ
 2518 #define NFS_COMMITBVECSIZ       20
 2519 #endif
 2520         struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
 2521         int bvecsize = 0, bveccount;
 2522 
 2523         if (called_from_renewthread != 0)
 2524                 slptimeo = hz;
 2525         if (nmp->nm_flag & NFSMNT_INT)
 2526                 slpflag = NFS_PCATCH;
 2527         if (!commit)
 2528                 passone = 0;
 2529         bo = &vp->v_bufobj;
 2530         /*
 2531          * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
 2532          * server, but has not been committed to stable storage on the server
 2533          * yet. On the first pass, the byte range is worked out and the commit
 2534          * rpc is done. On the second pass, ncl_writebp() is called to do the
 2535          * job.
 2536          */
 2537 again:
 2538         off = (u_quad_t)-1;
 2539         endoff = 0;
 2540         bvecpos = 0;
 2541         if (NFS_ISV34(vp) && commit) {
 2542                 if (bvec != NULL && bvec != bvec_on_stack)
 2543                         free(bvec, M_TEMP);
 2544                 /*
 2545                  * Count up how many buffers waiting for a commit.
 2546                  */
 2547                 bveccount = 0;
 2548                 BO_LOCK(bo);
 2549                 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 2550                         if (!BUF_ISLOCKED(bp) &&
 2551                             (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
 2552                                 == (B_DELWRI | B_NEEDCOMMIT))
 2553                                 bveccount++;
 2554                 }
 2555                 /*
 2556                  * Allocate space to remember the list of bufs to commit.  It is
 2557                  * important to use M_NOWAIT here to avoid a race with nfs_write.
 2558                  * If we can't get memory (for whatever reason), we will end up
 2559                  * committing the buffers one-by-one in the loop below.
 2560                  */
 2561                 if (bveccount > NFS_COMMITBVECSIZ) {
 2562                         /*
 2563                          * Release the vnode interlock to avoid a lock
 2564                          * order reversal.
 2565                          */
 2566                         BO_UNLOCK(bo);
 2567                         bvec = (struct buf **)
 2568                                 malloc(bveccount * sizeof(struct buf *),
 2569                                        M_TEMP, M_NOWAIT);
 2570                         BO_LOCK(bo);
 2571                         if (bvec == NULL) {
 2572                                 bvec = bvec_on_stack;
 2573                                 bvecsize = NFS_COMMITBVECSIZ;
 2574                         } else
 2575                                 bvecsize = bveccount;
 2576                 } else {
 2577                         bvec = bvec_on_stack;
 2578                         bvecsize = NFS_COMMITBVECSIZ;
 2579                 }
 2580                 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 2581                         if (bvecpos >= bvecsize)
 2582                                 break;
 2583                         if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
 2584                                 nbp = TAILQ_NEXT(bp, b_bobufs);
 2585                                 continue;
 2586                         }
 2587                         if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
 2588                             (B_DELWRI | B_NEEDCOMMIT)) {
 2589                                 BUF_UNLOCK(bp);
 2590                                 nbp = TAILQ_NEXT(bp, b_bobufs);
 2591                                 continue;
 2592                         }
 2593                         BO_UNLOCK(bo);
 2594                         bremfree(bp);
 2595                         /*
 2596                          * Work out if all buffers are using the same cred
 2597                          * so we can deal with them all with one commit.
 2598                          *
 2599                          * NOTE: we are not clearing B_DONE here, so we have
 2600                          * to do it later on in this routine if we intend to
 2601                          * initiate I/O on the bp.
 2602                          *
 2603                          * Note: to avoid loopback deadlocks, we do not
 2604                          * assign b_runningbufspace.
 2605                          */
 2606                         if (wcred == NULL)
 2607                                 wcred = bp->b_wcred;
 2608                         else if (wcred != bp->b_wcred)
 2609                                 wcred = NOCRED;
 2610                         vfs_busy_pages(bp, 1);
 2611 
 2612                         BO_LOCK(bo);
 2613                         /*
 2614                          * bp is protected by being locked, but nbp is not
 2615                          * and vfs_busy_pages() may sleep.  We have to
 2616                          * recalculate nbp.
 2617                          */
 2618                         nbp = TAILQ_NEXT(bp, b_bobufs);
 2619 
 2620                         /*
 2621                          * A list of these buffers is kept so that the
 2622                          * second loop knows which buffers have actually
 2623                          * been committed. This is necessary, since there
 2624                          * may be a race between the commit rpc and new
 2625                          * uncommitted writes on the file.
 2626                          */
 2627                         bvec[bvecpos++] = bp;
 2628                         toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2629                                 bp->b_dirtyoff;
 2630                         if (toff < off)
 2631                                 off = toff;
 2632                         toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
 2633                         if (toff > endoff)
 2634                                 endoff = toff;
 2635                 }
 2636                 BO_UNLOCK(bo);
 2637         }
 2638         if (bvecpos > 0) {
 2639                 /*
 2640                  * Commit data on the server, as required.
 2641                  * If all bufs are using the same wcred, then use that with
 2642                  * one call for all of them, otherwise commit each one
 2643                  * separately.
 2644                  */
 2645                 if (wcred != NOCRED)
 2646                         retv = ncl_commit(vp, off, (int)(endoff - off),
 2647                                           wcred, td);
 2648                 else {
 2649                         retv = 0;
 2650                         for (i = 0; i < bvecpos; i++) {
 2651                                 off_t off, size;
 2652                                 bp = bvec[i];
 2653                                 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2654                                         bp->b_dirtyoff;
 2655                                 size = (u_quad_t)(bp->b_dirtyend
 2656                                                   - bp->b_dirtyoff);
 2657                                 retv = ncl_commit(vp, off, (int)size,
 2658                                                   bp->b_wcred, td);
 2659                                 if (retv) break;
 2660                         }
 2661                 }
 2662 
 2663                 if (retv == NFSERR_STALEWRITEVERF)
 2664                         ncl_clearcommit(vp->v_mount);
 2665 
 2666                 /*
 2667                  * Now, either mark the blocks I/O done or mark the
 2668                  * blocks dirty, depending on whether the commit
 2669                  * succeeded.
 2670                  */
 2671                 for (i = 0; i < bvecpos; i++) {
 2672                         bp = bvec[i];
 2673                         bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
 2674                         if (retv) {
 2675                                 /*
 2676                                  * Error, leave B_DELWRI intact
 2677                                  */
 2678                                 vfs_unbusy_pages(bp);
 2679                                 brelse(bp);
 2680                         } else {
 2681                                 /*
 2682                                  * Success, remove B_DELWRI ( bundirty() ).
 2683                                  *
 2684                                  * b_dirtyoff/b_dirtyend seem to be NFS
 2685                                  * specific.  We should probably move that
 2686                                  * into bundirty(). XXX
 2687                                  */
 2688                                 bufobj_wref(bo);
 2689                                 bp->b_flags |= B_ASYNC;
 2690                                 bundirty(bp);
 2691                                 bp->b_flags &= ~B_DONE;
 2692                                 bp->b_ioflags &= ~BIO_ERROR;
 2693                                 bp->b_dirtyoff = bp->b_dirtyend = 0;
 2694                                 bufdone(bp);
 2695                         }
 2696                 }
 2697         }
 2698 
 2699         /*
 2700          * Start/do any write(s) that are required.
 2701          */
 2702 loop:
 2703         BO_LOCK(bo);
 2704         TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 2705                 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
 2706                         if (waitfor != MNT_WAIT || passone)
 2707                                 continue;
 2708 
 2709                         error = BUF_TIMELOCK(bp,
 2710                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 2711                             BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
 2712                         if (error == 0) {
 2713                                 BUF_UNLOCK(bp);
 2714                                 goto loop;
 2715                         }
 2716                         if (error == ENOLCK) {
 2717                                 error = 0;
 2718                                 goto loop;
 2719                         }
 2720                         if (called_from_renewthread != 0) {
 2721                                 /*
 2722                                  * Return EIO so the flush will be retried
 2723                                  * later.
 2724                                  */
 2725                                 error = EIO;
 2726                                 goto done;
 2727                         }
 2728                         if (newnfs_sigintr(nmp, td)) {
 2729                                 error = EINTR;
 2730                                 goto done;
 2731                         }
 2732                         if (slpflag & PCATCH) {
 2733                                 slpflag = 0;
 2734                                 slptimeo = 2 * hz;
 2735                         }
 2736                         goto loop;
 2737                 }
 2738                 if ((bp->b_flags & B_DELWRI) == 0)
 2739                         panic("nfs_fsync: not dirty");
 2740                 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
 2741                         BUF_UNLOCK(bp);
 2742                         continue;
 2743                 }
 2744                 BO_UNLOCK(bo);
 2745                 bremfree(bp);
 2746                 if (passone || !commit)
 2747                     bp->b_flags |= B_ASYNC;
 2748                 else
 2749                     bp->b_flags |= B_ASYNC;
 2750                 bwrite(bp);
 2751                 if (newnfs_sigintr(nmp, td)) {
 2752                         error = EINTR;
 2753                         goto done;
 2754                 }
 2755                 goto loop;
 2756         }
 2757         if (passone) {
 2758                 passone = 0;
 2759                 BO_UNLOCK(bo);
 2760                 goto again;
 2761         }
 2762         if (waitfor == MNT_WAIT) {
 2763                 while (bo->bo_numoutput) {
 2764                         error = bufobj_wwait(bo, slpflag, slptimeo);
 2765                         if (error) {
 2766                             BO_UNLOCK(bo);
 2767                             if (called_from_renewthread != 0) {
 2768                                 /*
 2769                                  * Return EIO so that the flush will be
 2770                                  * retried later.
 2771                                  */
 2772                                 error = EIO;
 2773                                 goto done;
 2774                             }
 2775                             error = newnfs_sigintr(nmp, td);
 2776                             if (error)
 2777                                 goto done;
 2778                             if (slpflag & PCATCH) {
 2779                                 slpflag = 0;
 2780                                 slptimeo = 2 * hz;
 2781                             }
 2782                             BO_LOCK(bo);
 2783                         }
 2784                 }
 2785                 if (bo->bo_dirty.bv_cnt != 0 && commit) {
 2786                         BO_UNLOCK(bo);
 2787                         goto loop;
 2788                 }
 2789                 /*
 2790                  * Wait for all the async IO requests to drain
 2791                  */
 2792                 BO_UNLOCK(bo);
 2793                 mtx_lock(&np->n_mtx);
 2794                 while (np->n_directio_asyncwr > 0) {
 2795                         np->n_flag |= NFSYNCWAIT;
 2796                         error = newnfs_msleep(td, &np->n_directio_asyncwr,
 2797                             &np->n_mtx, slpflag | (PRIBIO + 1), 
 2798                             "nfsfsync", 0);
 2799                         if (error) {
 2800                                 if (newnfs_sigintr(nmp, td)) {
 2801                                         mtx_unlock(&np->n_mtx);
 2802                                         error = EINTR;  
 2803                                         goto done;
 2804                                 }
 2805                         }
 2806                 }
 2807                 mtx_unlock(&np->n_mtx);
 2808         } else
 2809                 BO_UNLOCK(bo);
 2810         mtx_lock(&np->n_mtx);
 2811         if (np->n_flag & NWRITEERR) {
 2812                 error = np->n_error;
 2813                 np->n_flag &= ~NWRITEERR;
 2814         }
 2815         if (commit && bo->bo_dirty.bv_cnt == 0 &&
 2816             bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
 2817                 np->n_flag &= ~NMODIFIED;
 2818         mtx_unlock(&np->n_mtx);
 2819 done:
 2820         if (bvec != NULL && bvec != bvec_on_stack)
 2821                 free(bvec, M_TEMP);
 2822         if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
 2823             (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
 2824              np->n_directio_asyncwr != 0) && trycnt++ < 5) {
 2825                 /* try, try again... */
 2826                 passone = 1;
 2827                 wcred = NULL;
 2828                 bvec = NULL;
 2829                 bvecsize = 0;
 2830 printf("try%d\n", trycnt);
 2831                 goto again;
 2832         }
 2833         return (error);
 2834 }
 2835 
 2836 /*
 2837  * NFS advisory byte-level locks.
 2838  */
 2839 static int
 2840 nfs_advlock(struct vop_advlock_args *ap)
 2841 {
 2842         struct vnode *vp = ap->a_vp;
 2843         struct ucred *cred;
 2844         struct nfsnode *np = VTONFS(ap->a_vp);
 2845         struct proc *p = (struct proc *)ap->a_id;
 2846         struct thread *td = curthread;  /* XXX */
 2847         struct vattr va;
 2848         int ret, error = EOPNOTSUPP;
 2849         u_quad_t size;
 2850         
 2851         if (NFS_ISV4(vp) && (ap->a_flags & F_POSIX)) {
 2852                 cred = p->p_ucred;
 2853                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 2854                 if (vp->v_iflag & VI_DOOMED) {
 2855                         VOP_UNLOCK(vp, 0);
 2856                         return (EBADF);
 2857                 }
 2858 
 2859                 /*
 2860                  * If this is unlocking a write locked region, flush and
 2861                  * commit them before unlocking. This is required by
 2862                  * RFC3530 Sec. 9.3.2.
 2863                  */
 2864                 if (ap->a_op == F_UNLCK &&
 2865                     nfscl_checkwritelocked(vp, ap->a_fl, cred, td))
 2866                         (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0);
 2867 
 2868                 /*
 2869                  * Loop around doing the lock op, while a blocking lock
 2870                  * must wait for the lock op to succeed.
 2871                  */
 2872                 do {
 2873                         ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
 2874                             ap->a_fl, 0, cred, td);
 2875                         if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
 2876                             ap->a_op == F_SETLK) {
 2877                                 VOP_UNLOCK(vp, 0);
 2878                                 error = nfs_catnap(PZERO | PCATCH, ret,
 2879                                     "ncladvl");
 2880                                 if (error)
 2881                                         return (EINTR);
 2882                                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 2883                                 if (vp->v_iflag & VI_DOOMED) {
 2884                                         VOP_UNLOCK(vp, 0);
 2885                                         return (EBADF);
 2886                                 }
 2887                         }
 2888                 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
 2889                      ap->a_op == F_SETLK);
 2890                 if (ret == NFSERR_DENIED) {
 2891                         VOP_UNLOCK(vp, 0);
 2892                         return (EAGAIN);
 2893                 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
 2894                         VOP_UNLOCK(vp, 0);
 2895                         return (ret);
 2896                 } else if (ret != 0) {
 2897                         VOP_UNLOCK(vp, 0);
 2898                         return (EACCES);
 2899                 }
 2900 
 2901                 /*
 2902                  * Now, if we just got a lock, invalidate data in the buffer
 2903                  * cache, as required, so that the coherency conforms with
 2904                  * RFC3530 Sec. 9.3.2.
 2905                  */
 2906                 if (ap->a_op == F_SETLK) {
 2907                         if ((np->n_flag & NMODIFIED) == 0) {
 2908                                 np->n_attrstamp = 0;
 2909                                 ret = VOP_GETATTR(vp, &va, cred);
 2910                         }
 2911                         if ((np->n_flag & NMODIFIED) || ret ||
 2912                             np->n_change != va.va_filerev) {
 2913                                 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
 2914                                 np->n_attrstamp = 0;
 2915                                 ret = VOP_GETATTR(vp, &va, cred);
 2916                                 if (!ret) {
 2917                                         np->n_mtime = va.va_mtime;
 2918                                         np->n_change = va.va_filerev;
 2919                                 }
 2920                         }
 2921                 }
 2922                 VOP_UNLOCK(vp, 0);
 2923                 return (0);
 2924         } else if (!NFS_ISV4(vp)) {
 2925                 error = vn_lock(vp, LK_SHARED);
 2926                 if (error)
 2927                         return (error);
 2928                 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
 2929                         size = VTONFS(vp)->n_size;
 2930                         VOP_UNLOCK(vp, 0);
 2931                         error = lf_advlock(ap, &(vp->v_lockf), size);
 2932                 } else {
 2933                         if (nfs_advlock_p != NULL)
 2934                                 error = nfs_advlock_p(ap);
 2935                         else {
 2936                                 VOP_UNLOCK(vp, 0);
 2937                                 error = ENOLCK;
 2938                         }
 2939                 }
 2940         }
 2941         return (error);
 2942 }
 2943 
 2944 /*
 2945  * NFS advisory byte-level locks.
 2946  */
 2947 static int
 2948 nfs_advlockasync(struct vop_advlockasync_args *ap)
 2949 {
 2950         struct vnode *vp = ap->a_vp;
 2951         u_quad_t size;
 2952         int error;
 2953         
 2954         if (NFS_ISV4(vp))
 2955                 return (EOPNOTSUPP);
 2956         error = vn_lock(vp, LK_SHARED);
 2957         if (error)
 2958                 return (error);
 2959         if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
 2960                 size = VTONFS(vp)->n_size;
 2961                 VOP_UNLOCK(vp, 0);
 2962                 error = lf_advlockasync(ap, &(vp->v_lockf), size);
 2963         } else {
 2964                 VOP_UNLOCK(vp, 0);
 2965                 error = EOPNOTSUPP;
 2966         }
 2967         return (error);
 2968 }
 2969 
 2970 /*
 2971  * Print out the contents of an nfsnode.
 2972  */
 2973 static int
 2974 nfs_print(struct vop_print_args *ap)
 2975 {
 2976         struct vnode *vp = ap->a_vp;
 2977         struct nfsnode *np = VTONFS(vp);
 2978 
 2979         ncl_printf("\tfileid %ld fsid 0x%x",
 2980            np->n_vattr.na_fileid, np->n_vattr.na_fsid);
 2981         if (vp->v_type == VFIFO)
 2982                 fifo_printinfo(vp);
 2983         printf("\n");
 2984         return (0);
 2985 }
 2986 
 2987 /*
 2988  * This is the "real" nfs::bwrite(struct buf*).
 2989  * We set B_CACHE if this is a VMIO buffer.
 2990  */
 2991 int
 2992 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
 2993 {
 2994         int s;
 2995         int oldflags = bp->b_flags;
 2996 #if 0
 2997         int retv = 1;
 2998         off_t off;
 2999 #endif
 3000 
 3001         BUF_ASSERT_HELD(bp);
 3002 
 3003         if (bp->b_flags & B_INVAL) {
 3004                 brelse(bp);
 3005                 return(0);
 3006         }
 3007 
 3008         bp->b_flags |= B_CACHE;
 3009 
 3010         /*
 3011          * Undirty the bp.  We will redirty it later if the I/O fails.
 3012          */
 3013 
 3014         s = splbio();
 3015         bundirty(bp);
 3016         bp->b_flags &= ~B_DONE;
 3017         bp->b_ioflags &= ~BIO_ERROR;
 3018         bp->b_iocmd = BIO_WRITE;
 3019 
 3020         bufobj_wref(bp->b_bufobj);
 3021         curthread->td_ru.ru_oublock++;
 3022         splx(s);
 3023 
 3024         /*
 3025          * Note: to avoid loopback deadlocks, we do not
 3026          * assign b_runningbufspace.
 3027          */
 3028         vfs_busy_pages(bp, 1);
 3029 
 3030         BUF_KERNPROC(bp);
 3031         bp->b_iooffset = dbtob(bp->b_blkno);
 3032         bstrategy(bp);
 3033 
 3034         if( (oldflags & B_ASYNC) == 0) {
 3035                 int rtval = bufwait(bp);
 3036 
 3037                 if (oldflags & B_DELWRI) {
 3038                         s = splbio();
 3039                         reassignbuf(bp);
 3040                         splx(s);
 3041                 }
 3042                 brelse(bp);
 3043                 return (rtval);
 3044         }
 3045 
 3046         return (0);
 3047 }
 3048 
 3049 /*
 3050  * nfs special file access vnode op.
 3051  * Essentially just get vattr and then imitate iaccess() since the device is
 3052  * local to the client.
 3053  */
 3054 static int
 3055 nfsspec_access(struct vop_access_args *ap)
 3056 {
 3057         struct vattr *vap;
 3058         struct ucred *cred = ap->a_cred;
 3059         struct vnode *vp = ap->a_vp;
 3060         accmode_t accmode = ap->a_accmode;
 3061         struct vattr vattr;
 3062         int error;
 3063 
 3064         /*
 3065          * Disallow write attempts on filesystems mounted read-only;
 3066          * unless the file is a socket, fifo, or a block or character
 3067          * device resident on the filesystem.
 3068          */
 3069         if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
 3070                 switch (vp->v_type) {
 3071                 case VREG:
 3072                 case VDIR:
 3073                 case VLNK:
 3074                         return (EROFS);
 3075                 default:
 3076                         break;
 3077                 }
 3078         }
 3079         vap = &vattr;
 3080         error = VOP_GETATTR(vp, vap, cred);
 3081         if (error)
 3082                 goto out;
 3083         error  = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
 3084             accmode, cred, NULL);
 3085 out:
 3086         return error;
 3087 }
 3088 
 3089 /*
 3090  * Read wrapper for fifos.
 3091  */
 3092 static int
 3093 nfsfifo_read(struct vop_read_args *ap)
 3094 {
 3095         struct nfsnode *np = VTONFS(ap->a_vp);
 3096         int error;
 3097 
 3098         /*
 3099          * Set access flag.
 3100          */
 3101         mtx_lock(&np->n_mtx);
 3102         np->n_flag |= NACC;
 3103         getnanotime(&np->n_atim);
 3104         mtx_unlock(&np->n_mtx);
 3105         error = fifo_specops.vop_read(ap);
 3106         return error;   
 3107 }
 3108 
 3109 /*
 3110  * Write wrapper for fifos.
 3111  */
 3112 static int
 3113 nfsfifo_write(struct vop_write_args *ap)
 3114 {
 3115         struct nfsnode *np = VTONFS(ap->a_vp);
 3116 
 3117         /*
 3118          * Set update flag.
 3119          */
 3120         mtx_lock(&np->n_mtx);
 3121         np->n_flag |= NUPD;
 3122         getnanotime(&np->n_mtim);
 3123         mtx_unlock(&np->n_mtx);
 3124         return(fifo_specops.vop_write(ap));
 3125 }
 3126 
 3127 /*
 3128  * Close wrapper for fifos.
 3129  *
 3130  * Update the times on the nfsnode then do fifo close.
 3131  */
 3132 static int
 3133 nfsfifo_close(struct vop_close_args *ap)
 3134 {
 3135         struct vnode *vp = ap->a_vp;
 3136         struct nfsnode *np = VTONFS(vp);
 3137         struct vattr vattr;
 3138         struct timespec ts;
 3139 
 3140         mtx_lock(&np->n_mtx);
 3141         if (np->n_flag & (NACC | NUPD)) {
 3142                 getnanotime(&ts);
 3143                 if (np->n_flag & NACC)
 3144                         np->n_atim = ts;
 3145                 if (np->n_flag & NUPD)
 3146                         np->n_mtim = ts;
 3147                 np->n_flag |= NCHG;
 3148                 if (vrefcnt(vp) == 1 &&
 3149                     (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
 3150                         VATTR_NULL(&vattr);
 3151                         if (np->n_flag & NACC)
 3152                                 vattr.va_atime = np->n_atim;
 3153                         if (np->n_flag & NUPD)
 3154                                 vattr.va_mtime = np->n_mtim;
 3155                         mtx_unlock(&np->n_mtx);
 3156                         (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
 3157                         goto out;
 3158                 }
 3159         }
 3160         mtx_unlock(&np->n_mtx);
 3161 out:
 3162         return (fifo_specops.vop_close(ap));
 3163 }
 3164 
 3165 /*
 3166  * Just call ncl_writebp() with the force argument set to 1.
 3167  *
 3168  * NOTE: B_DONE may or may not be set in a_bp on call.
 3169  */
 3170 static int
 3171 nfs_bwrite(struct buf *bp)
 3172 {
 3173 
 3174         return (ncl_writebp(bp, 1, curthread));
 3175 }
 3176 
 3177 struct buf_ops buf_ops_newnfs = {
 3178         .bop_name       =       "buf_ops_nfs",
 3179         .bop_write      =       nfs_bwrite,
 3180         .bop_strategy   =       bufstrategy,
 3181         .bop_sync       =       bufsync,
 3182         .bop_bdflush    =       bufbdflush,
 3183 };
 3184 
 3185 /*
 3186  * Cloned from vop_stdlock(), and then the ugly hack added.
 3187  */
 3188 static int
 3189 nfs_lock1(struct vop_lock1_args *ap)
 3190 {
 3191         struct vnode *vp = ap->a_vp;
 3192         int error = 0;
 3193 
 3194         /*
 3195          * Since vfs_hash_get() calls vget() and it will no longer work
 3196          * for FreeBSD8 with flags == 0, I can only think of this horrible
 3197          * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
 3198          * and then handle it here. All I want for this case is a v_usecount
 3199          * on the vnode to use for recovery, while another thread might
 3200          * hold a lock on the vnode. I have the other threads blocked, so
 3201          * there isn't any race problem.
 3202          */
 3203         if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
 3204                 if ((ap->a_flags & LK_INTERLOCK) == 0)
 3205                         panic("ncllock1");
 3206                 if ((vp->v_iflag & VI_DOOMED))
 3207                         error = ENOENT;
 3208                 VI_UNLOCK(vp);
 3209                 return (error);
 3210         }
 3211         return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
 3212             LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
 3213             ap->a_line));
 3214 }
 3215 
 3216 #ifdef NFS4_ACL_EXTATTR_NAME
 3217 static int
 3218 nfs_getacl(struct vop_getacl_args *ap)
 3219 {
 3220         int error;
 3221 
 3222         if (ap->a_type != ACL_TYPE_NFS4)
 3223                 return (EOPNOTSUPP);
 3224         error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
 3225             NULL);
 3226         if (error > NFSERR_STALE) {
 3227                 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
 3228                 error = EPERM;
 3229         }
 3230         return (error);
 3231 }
 3232 
 3233 static int
 3234 nfs_setacl(struct vop_setacl_args *ap)
 3235 {
 3236         int error;
 3237 
 3238         if (ap->a_type != ACL_TYPE_NFS4)
 3239                 return (EOPNOTSUPP);
 3240         error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
 3241             NULL);
 3242         if (error > NFSERR_STALE) {
 3243                 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
 3244                 error = EPERM;
 3245         }
 3246         return (error);
 3247 }
 3248 
 3249 #endif  /* NFS4_ACL_EXTATTR_NAME */

Cache object: ad2d15b98a4f873fb3258aac48fc749c


[ 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.