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-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-2  -  FREEBSD-11-1  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-4  -  FREEBSD-10-3  -  FREEBSD-10-2  -  FREEBSD-10-1  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-3  -  FREEBSD-9-2  -  FREEBSD-9-1  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-4  -  FREEBSD-8-3  -  FREEBSD-8-2  -  FREEBSD-8-1  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-4  -  FREEBSD-7-3  -  FREEBSD-7-2  -  FREEBSD-7-1  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-4  -  FREEBSD-6-3  -  FREEBSD-6-2  -  FREEBSD-6-1  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-5  -  FREEBSD-5-4  -  FREEBSD-5-3  -  FREEBSD-5-2  -  FREEBSD-5-1  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  linux-2.6  -  linux-2.4.22  -  MK83  -  MK84  -  PLAN9  -  DFBSD  -  NETBSD  -  NETBSD5  -  NETBSD4  -  NETBSD3  -  NETBSD20  -  OPENBSD  -  xnu-517  -  xnu-792  -  xnu-792.6.70  -  xnu-1228  -  xnu-1456.1.26  -  xnu-1699.24.8  -  xnu-2050.18.24  -  OPENSOLARIS  -  minix-3-1-1 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * SPDX-License-Identifier: BSD-3-Clause
    3  *
    4  * Copyright (c) 1989, 1993
    5  *      The Regents of the University of California.  All rights reserved.
    6  *
    7  * This code is derived from software contributed to Berkeley by
    8  * Rick Macklem at The University of Guelph.
    9  *
   10  * Redistribution and use in source and binary forms, with or without
   11  * modification, are permitted provided that the following conditions
   12  * are met:
   13  * 1. Redistributions of source code must retain the above copyright
   14  *    notice, this list of conditions and the following disclaimer.
   15  * 2. Redistributions in binary form must reproduce the above copyright
   16  *    notice, this list of conditions and the following disclaimer in the
   17  *    documentation and/or other materials provided with the distribution.
   18  * 3. Neither the name of the University nor the names of its contributors
   19  *    may be used to endorse or promote products derived from this software
   20  *    without specific prior written permission.
   21  *
   22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   32  * SUCH DAMAGE.
   33  *
   34  *      from nfs_vnops.c        8.16 (Berkeley) 5/27/95
   35  */
   36 
   37 #include <sys/cdefs.h>
   38 __FBSDID("$FreeBSD: stable/12/sys/fs/nfsclient/nfs_clvnops.c 371029 2021-11-18 01:03:32Z rmacklem $");
   39 
   40 /*
   41  * vnode op calls for Sun NFS version 2, 3 and 4
   42  */
   43 
   44 #include "opt_inet.h"
   45 
   46 #include <sys/param.h>
   47 #include <sys/kernel.h>
   48 #include <sys/systm.h>
   49 #include <sys/resourcevar.h>
   50 #include <sys/proc.h>
   51 #include <sys/mount.h>
   52 #include <sys/bio.h>
   53 #include <sys/buf.h>
   54 #include <sys/jail.h>
   55 #include <sys/malloc.h>
   56 #include <sys/mbuf.h>
   57 #include <sys/namei.h>
   58 #include <sys/socket.h>
   59 #include <sys/vnode.h>
   60 #include <sys/dirent.h>
   61 #include <sys/fcntl.h>
   62 #include <sys/lockf.h>
   63 #include <sys/stat.h>
   64 #include <sys/sysctl.h>
   65 #include <sys/signalvar.h>
   66 
   67 #include <vm/vm.h>
   68 #include <vm/vm_extern.h>
   69 #include <vm/vm_object.h>
   70 
   71 #include <fs/nfs/nfsport.h>
   72 #include <fs/nfsclient/nfsnode.h>
   73 #include <fs/nfsclient/nfsmount.h>
   74 #include <fs/nfsclient/nfs.h>
   75 #include <fs/nfsclient/nfs_kdtrace.h>
   76 
   77 #include <net/if.h>
   78 #include <netinet/in.h>
   79 #include <netinet/in_var.h>
   80 
   81 #include <nfs/nfs_lock.h>
   82 
   83 #ifdef KDTRACE_HOOKS
   84 #include <sys/dtrace_bsd.h>
   85 
   86 dtrace_nfsclient_accesscache_flush_probe_func_t
   87                 dtrace_nfscl_accesscache_flush_done_probe;
   88 uint32_t        nfscl_accesscache_flush_done_id;
   89 
   90 dtrace_nfsclient_accesscache_get_probe_func_t
   91                 dtrace_nfscl_accesscache_get_hit_probe,
   92                 dtrace_nfscl_accesscache_get_miss_probe;
   93 uint32_t        nfscl_accesscache_get_hit_id;
   94 uint32_t        nfscl_accesscache_get_miss_id;
   95 
   96 dtrace_nfsclient_accesscache_load_probe_func_t
   97                 dtrace_nfscl_accesscache_load_done_probe;
   98 uint32_t        nfscl_accesscache_load_done_id;
   99 #endif /* !KDTRACE_HOOKS */
  100 
  101 /* Defs */
  102 #define TRUE    1
  103 #define FALSE   0
  104 
  105 extern struct nfsstatsv1 nfsstatsv1;
  106 extern int nfsrv_useacl;
  107 extern int nfscl_debuglevel;
  108 MALLOC_DECLARE(M_NEWNFSREQ);
  109 
  110 static vop_read_t       nfsfifo_read;
  111 static vop_write_t      nfsfifo_write;
  112 static vop_close_t      nfsfifo_close;
  113 static int      nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
  114                     struct thread *);
  115 static vop_lookup_t     nfs_lookup;
  116 static vop_create_t     nfs_create;
  117 static vop_mknod_t      nfs_mknod;
  118 static vop_open_t       nfs_open;
  119 static vop_pathconf_t   nfs_pathconf;
  120 static vop_close_t      nfs_close;
  121 static vop_access_t     nfs_access;
  122 static vop_getattr_t    nfs_getattr;
  123 static vop_setattr_t    nfs_setattr;
  124 static vop_read_t       nfs_read;
  125 static vop_fsync_t      nfs_fsync;
  126 static vop_remove_t     nfs_remove;
  127 static vop_link_t       nfs_link;
  128 static vop_rename_t     nfs_rename;
  129 static vop_mkdir_t      nfs_mkdir;
  130 static vop_rmdir_t      nfs_rmdir;
  131 static vop_symlink_t    nfs_symlink;
  132 static vop_readdir_t    nfs_readdir;
  133 static vop_strategy_t   nfs_strategy;
  134 static  int     nfs_lookitup(struct vnode *, char *, int,
  135                     struct ucred *, struct thread *, struct nfsnode **);
  136 static  int     nfs_sillyrename(struct vnode *, struct vnode *,
  137                     struct componentname *);
  138 static vop_access_t     nfsspec_access;
  139 static vop_readlink_t   nfs_readlink;
  140 static vop_print_t      nfs_print;
  141 static vop_advlock_t    nfs_advlock;
  142 static vop_advlockasync_t nfs_advlockasync;
  143 static vop_getacl_t nfs_getacl;
  144 static vop_setacl_t nfs_setacl;
  145 static vop_lock1_t      nfs_lock;
  146 
  147 /*
  148  * Global vfs data structures for nfs
  149  */
  150 struct vop_vector newnfs_vnodeops = {
  151         .vop_default =          &default_vnodeops,
  152         .vop_access =           nfs_access,
  153         .vop_advlock =          nfs_advlock,
  154         .vop_advlockasync =     nfs_advlockasync,
  155         .vop_close =            nfs_close,
  156         .vop_create =           nfs_create,
  157         .vop_fsync =            nfs_fsync,
  158         .vop_getattr =          nfs_getattr,
  159         .vop_getpages =         ncl_getpages,
  160         .vop_putpages =         ncl_putpages,
  161         .vop_inactive =         ncl_inactive,
  162         .vop_link =             nfs_link,
  163         .vop_lock1 =            nfs_lock,
  164         .vop_lookup =           nfs_lookup,
  165         .vop_mkdir =            nfs_mkdir,
  166         .vop_mknod =            nfs_mknod,
  167         .vop_open =             nfs_open,
  168         .vop_pathconf =         nfs_pathconf,
  169         .vop_print =            nfs_print,
  170         .vop_read =             nfs_read,
  171         .vop_readdir =          nfs_readdir,
  172         .vop_readlink =         nfs_readlink,
  173         .vop_reclaim =          ncl_reclaim,
  174         .vop_remove =           nfs_remove,
  175         .vop_rename =           nfs_rename,
  176         .vop_rmdir =            nfs_rmdir,
  177         .vop_setattr =          nfs_setattr,
  178         .vop_strategy =         nfs_strategy,
  179         .vop_symlink =          nfs_symlink,
  180         .vop_write =            ncl_write,
  181         .vop_getacl =           nfs_getacl,
  182         .vop_setacl =           nfs_setacl,
  183 };
  184 
  185 struct vop_vector newnfs_fifoops = {
  186         .vop_default =          &fifo_specops,
  187         .vop_access =           nfsspec_access,
  188         .vop_close =            nfsfifo_close,
  189         .vop_fsync =            nfs_fsync,
  190         .vop_getattr =          nfs_getattr,
  191         .vop_inactive =         ncl_inactive,
  192         .vop_pathconf =         nfs_pathconf,
  193         .vop_print =            nfs_print,
  194         .vop_read =             nfsfifo_read,
  195         .vop_reclaim =          ncl_reclaim,
  196         .vop_setattr =          nfs_setattr,
  197         .vop_write =            nfsfifo_write,
  198 };
  199 
  200 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
  201     struct componentname *cnp, struct vattr *vap);
  202 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
  203     int namelen, struct ucred *cred, struct thread *td);
  204 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
  205     char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
  206     char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
  207 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
  208     struct componentname *scnp, struct sillyrename *sp);
  209 
  210 /*
  211  * Global variables
  212  */
  213 SYSCTL_DECL(_vfs_nfs);
  214 
  215 static int      nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
  216 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
  217            &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
  218 
  219 static int      nfs_prime_access_cache = 0;
  220 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
  221            &nfs_prime_access_cache, 0,
  222            "Prime NFS ACCESS cache when fetching attributes");
  223 
  224 static int      newnfs_commit_on_close = 0;
  225 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
  226     &newnfs_commit_on_close, 0, "write+commit on close, else only write");
  227 
  228 static int      nfs_clean_pages_on_close = 1;
  229 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
  230            &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
  231 
  232 int newnfs_directio_enable = 0;
  233 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
  234            &newnfs_directio_enable, 0, "Enable NFS directio");
  235 
  236 int nfs_keep_dirty_on_error;
  237 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
  238     &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
  239 
  240 /*
  241  * This sysctl allows other processes to mmap a file that has been opened
  242  * O_DIRECT by a process.  In general, having processes mmap the file while
  243  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
  244  * this by default to prevent DoS attacks - to prevent a malicious user from
  245  * opening up files O_DIRECT preventing other users from mmap'ing these
  246  * files.  "Protected" environments where stricter consistency guarantees are
  247  * required can disable this knob.  The process that opened the file O_DIRECT
  248  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
  249  * meaningful.
  250  */
  251 int newnfs_directio_allow_mmap = 1;
  252 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
  253            &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
  254 
  255 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY                \
  256                          | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \
  257                          | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
  258 
  259 /*
  260  * SMP Locking Note :
  261  * The list of locks after the description of the lock is the ordering
  262  * of other locks acquired with the lock held.
  263  * np->n_mtx : Protects the fields in the nfsnode.
  264        VM Object Lock
  265        VI_MTX (acquired indirectly)
  266  * nmp->nm_mtx : Protects the fields in the nfsmount.
  267        rep->r_mtx
  268  * ncl_iod_mutex : Global lock, protects shared nfsiod state.
  269  * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
  270        nmp->nm_mtx
  271        rep->r_mtx
  272  * rep->r_mtx : Protects the fields in an nfsreq.
  273  */
  274 
  275 static int
  276 nfs_lock(struct vop_lock1_args *ap)
  277 {
  278         struct vnode *vp;
  279         struct nfsnode *np;
  280         u_quad_t nsize;
  281         int error, lktype;
  282         bool onfault;
  283 
  284         vp = ap->a_vp;
  285         lktype = ap->a_flags & LK_TYPE_MASK;
  286         error = VOP_LOCK1_APV(&default_vnodeops, ap);
  287         if (error != 0 || vp->v_op != &newnfs_vnodeops)
  288                 return (error);
  289         np = VTONFS(vp);
  290         if (np == NULL)
  291                 return (0);
  292         NFSLOCKNODE(np);
  293         if ((np->n_flag & NVNSETSZSKIP) == 0 || (lktype != LK_SHARED &&
  294             lktype != LK_EXCLUSIVE && lktype != LK_UPGRADE &&
  295             lktype != LK_TRYUPGRADE)) {
  296                 NFSUNLOCKNODE(np);
  297                 return (0);
  298         }
  299         onfault = (ap->a_flags & LK_EATTR_MASK) == LK_NOWAIT &&
  300             (ap->a_flags & LK_INIT_MASK) == LK_CANRECURSE &&
  301             (lktype == LK_SHARED || lktype == LK_EXCLUSIVE);
  302         if (onfault && vp->v_vnlock->lk_recurse == 0) {
  303                 /*
  304                  * Force retry in vm_fault(), to make the lock request
  305                  * sleepable, which allows us to piggy-back the
  306                  * sleepable call to vnode_pager_setsize().
  307                  */
  308                 NFSUNLOCKNODE(np);
  309                 VOP_UNLOCK(vp, 0);
  310                 return (EBUSY);
  311         }
  312         if ((ap->a_flags & LK_NOWAIT) != 0 ||
  313             (lktype == LK_SHARED && vp->v_vnlock->lk_recurse > 0)) {
  314                 NFSUNLOCKNODE(np);
  315                 return (0);
  316         }
  317         if (lktype == LK_SHARED) {
  318                 NFSUNLOCKNODE(np);
  319                 VOP_UNLOCK(vp, 0);
  320                 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
  321                 ap->a_flags |= LK_EXCLUSIVE;
  322                 error = VOP_LOCK1_APV(&default_vnodeops, ap);
  323                 if (error != 0 || vp->v_op != &newnfs_vnodeops)
  324                         return (error);
  325                 if (vp->v_data == NULL)
  326                         goto downgrade;
  327                 MPASS(vp->v_data == np);
  328                 NFSLOCKNODE(np);
  329                 if ((np->n_flag & NVNSETSZSKIP) == 0) {
  330                         NFSUNLOCKNODE(np);
  331                         goto downgrade;
  332                 }
  333         }
  334         np->n_flag &= ~NVNSETSZSKIP;
  335         nsize = np->n_size;
  336         NFSUNLOCKNODE(np);
  337         vnode_pager_setsize(vp, nsize);
  338 downgrade:
  339         if (lktype == LK_SHARED) {
  340                 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
  341                 ap->a_flags |= LK_DOWNGRADE;
  342                 (void)VOP_LOCK1_APV(&default_vnodeops, ap);
  343         }
  344         return (0);
  345 }
  346 
  347 static int
  348 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
  349     struct ucred *cred, u_int32_t *retmode)
  350 {
  351         int error = 0, attrflag, i, lrupos;
  352         u_int32_t rmode;
  353         struct nfsnode *np = VTONFS(vp);
  354         struct nfsvattr nfsva;
  355 
  356         error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
  357             &rmode, NULL);
  358         if (attrflag)
  359                 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
  360         if (!error) {
  361                 lrupos = 0;
  362                 NFSLOCKNODE(np);
  363                 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
  364                         if (np->n_accesscache[i].uid == cred->cr_uid) {
  365                                 np->n_accesscache[i].mode = rmode;
  366                                 np->n_accesscache[i].stamp = time_second;
  367                                 break;
  368                         }
  369                         if (i > 0 && np->n_accesscache[i].stamp <
  370                             np->n_accesscache[lrupos].stamp)
  371                                 lrupos = i;
  372                 }
  373                 if (i == NFS_ACCESSCACHESIZE) {
  374                         np->n_accesscache[lrupos].uid = cred->cr_uid;
  375                         np->n_accesscache[lrupos].mode = rmode;
  376                         np->n_accesscache[lrupos].stamp = time_second;
  377                 }
  378                 NFSUNLOCKNODE(np);
  379                 if (retmode != NULL)
  380                         *retmode = rmode;
  381                 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
  382         } else if (NFS_ISV4(vp)) {
  383                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
  384         }
  385 #ifdef KDTRACE_HOOKS
  386         if (error != 0)
  387                 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
  388                     error);
  389 #endif
  390         return (error);
  391 }
  392 
  393 /*
  394  * nfs access vnode op.
  395  * For nfs version 2, just return ok. File accesses may fail later.
  396  * For nfs version 3, use the access rpc to check accessibility. If file modes
  397  * are changed on the server, accesses might still fail later.
  398  */
  399 static int
  400 nfs_access(struct vop_access_args *ap)
  401 {
  402         struct vnode *vp = ap->a_vp;
  403         int error = 0, i, gotahit;
  404         u_int32_t mode, wmode, rmode;
  405         int v34 = NFS_ISV34(vp);
  406         struct nfsnode *np = VTONFS(vp);
  407 
  408         /*
  409          * Disallow write attempts on filesystems mounted read-only;
  410          * unless the file is a socket, fifo, or a block or character
  411          * device resident on the filesystem.
  412          */
  413         if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
  414             VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
  415             VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
  416                 switch (vp->v_type) {
  417                 case VREG:
  418                 case VDIR:
  419                 case VLNK:
  420                         return (EROFS);
  421                 default:
  422                         break;
  423                 }
  424         }
  425         /*
  426          * For nfs v3 or v4, check to see if we have done this recently, and if
  427          * so return our cached result instead of making an ACCESS call.
  428          * If not, do an access rpc, otherwise you are stuck emulating
  429          * ufs_access() locally using the vattr. This may not be correct,
  430          * since the server may apply other access criteria such as
  431          * client uid-->server uid mapping that we do not know about.
  432          */
  433         if (v34) {
  434                 if (ap->a_accmode & VREAD)
  435                         mode = NFSACCESS_READ;
  436                 else
  437                         mode = 0;
  438                 if (vp->v_type != VDIR) {
  439                         if (ap->a_accmode & VWRITE)
  440                                 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
  441                         if (ap->a_accmode & VAPPEND)
  442                                 mode |= NFSACCESS_EXTEND;
  443                         if (ap->a_accmode & VEXEC)
  444                                 mode |= NFSACCESS_EXECUTE;
  445                         if (ap->a_accmode & VDELETE)
  446                                 mode |= NFSACCESS_DELETE;
  447                 } else {
  448                         if (ap->a_accmode & VWRITE)
  449                                 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
  450                         if (ap->a_accmode & VAPPEND)
  451                                 mode |= NFSACCESS_EXTEND;
  452                         if (ap->a_accmode & VEXEC)
  453                                 mode |= NFSACCESS_LOOKUP;
  454                         if (ap->a_accmode & VDELETE)
  455                                 mode |= NFSACCESS_DELETE;
  456                         if (ap->a_accmode & VDELETE_CHILD)
  457                                 mode |= NFSACCESS_MODIFY;
  458                 }
  459                 /* XXX safety belt, only make blanket request if caching */
  460                 if (nfsaccess_cache_timeout > 0) {
  461                         wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
  462                                 NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
  463                                 NFSACCESS_DELETE | NFSACCESS_LOOKUP;
  464                 } else {
  465                         wmode = mode;
  466                 }
  467 
  468                 /*
  469                  * Does our cached result allow us to give a definite yes to
  470                  * this request?
  471                  */
  472                 gotahit = 0;
  473                 NFSLOCKNODE(np);
  474                 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
  475                         if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
  476                             if (time_second < (np->n_accesscache[i].stamp
  477                                 + nfsaccess_cache_timeout) &&
  478                                 (np->n_accesscache[i].mode & mode) == mode) {
  479                                 NFSINCRGLOBAL(nfsstatsv1.accesscache_hits);
  480                                 gotahit = 1;
  481                             }
  482                             break;
  483                         }
  484                 }
  485                 NFSUNLOCKNODE(np);
  486 #ifdef KDTRACE_HOOKS
  487                 if (gotahit != 0)
  488                         KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
  489                             ap->a_cred->cr_uid, mode);
  490                 else
  491                         KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
  492                             ap->a_cred->cr_uid, mode);
  493 #endif
  494                 if (gotahit == 0) {
  495                         /*
  496                          * Either a no, or a don't know.  Go to the wire.
  497                          */
  498                         NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
  499                         error = nfs34_access_otw(vp, wmode, ap->a_td,
  500                             ap->a_cred, &rmode);
  501                         if (!error &&
  502                             (rmode & mode) != mode)
  503                                 error = EACCES;
  504                 }
  505                 return (error);
  506         } else {
  507                 if ((error = nfsspec_access(ap)) != 0) {
  508                         return (error);
  509                 }
  510                 /*
  511                  * Attempt to prevent a mapped root from accessing a file
  512                  * which it shouldn't.  We try to read a byte from the file
  513                  * if the user is root and the file is not zero length.
  514                  * After calling nfsspec_access, we should have the correct
  515                  * file size cached.
  516                  */
  517                 NFSLOCKNODE(np);
  518                 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
  519                     && VTONFS(vp)->n_size > 0) {
  520                         struct iovec aiov;
  521                         struct uio auio;
  522                         char buf[1];
  523 
  524                         NFSUNLOCKNODE(np);
  525                         aiov.iov_base = buf;
  526                         aiov.iov_len = 1;
  527                         auio.uio_iov = &aiov;
  528                         auio.uio_iovcnt = 1;
  529                         auio.uio_offset = 0;
  530                         auio.uio_resid = 1;
  531                         auio.uio_segflg = UIO_SYSSPACE;
  532                         auio.uio_rw = UIO_READ;
  533                         auio.uio_td = ap->a_td;
  534 
  535                         if (vp->v_type == VREG)
  536                                 error = ncl_readrpc(vp, &auio, ap->a_cred);
  537                         else if (vp->v_type == VDIR) {
  538                                 char* bp;
  539                                 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
  540                                 aiov.iov_base = bp;
  541                                 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
  542                                 error = ncl_readdirrpc(vp, &auio, ap->a_cred,
  543                                     ap->a_td);
  544                                 free(bp, M_TEMP);
  545                         } else if (vp->v_type == VLNK)
  546                                 error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
  547                         else
  548                                 error = EACCES;
  549                 } else
  550                         NFSUNLOCKNODE(np);
  551                 return (error);
  552         }
  553 }
  554 
  555 
  556 /*
  557  * nfs open vnode op
  558  * Check to see if the type is ok
  559  * and that deletion is not in progress.
  560  * For paged in text files, you will need to flush the page cache
  561  * if consistency is lost.
  562  */
  563 /* ARGSUSED */
  564 static int
  565 nfs_open(struct vop_open_args *ap)
  566 {
  567         struct vnode *vp = ap->a_vp;
  568         struct nfsnode *np = VTONFS(vp);
  569         struct vattr vattr;
  570         int error;
  571         int fmode = ap->a_mode;
  572         struct ucred *cred;
  573         vm_object_t obj;
  574 
  575         if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
  576                 return (EOPNOTSUPP);
  577 
  578         /*
  579          * For NFSv4, we need to do the Open Op before cache validation,
  580          * so that we conform to RFC3530 Sec. 9.3.1.
  581          */
  582         if (NFS_ISV4(vp)) {
  583                 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
  584                 if (error) {
  585                         error = nfscl_maperr(ap->a_td, error, (uid_t)0,
  586                             (gid_t)0);
  587                         return (error);
  588                 }
  589         }
  590 
  591         /*
  592          * Now, if this Open will be doing reading, re-validate/flush the
  593          * cache, so that Close/Open coherency is maintained.
  594          */
  595         NFSLOCKNODE(np);
  596         if (np->n_flag & NMODIFIED) {
  597                 NFSUNLOCKNODE(np);
  598                 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  599                 if (error == EINTR || error == EIO) {
  600                         if (NFS_ISV4(vp))
  601                                 (void) nfsrpc_close(vp, 0, ap->a_td);
  602                         return (error);
  603                 }
  604                 NFSLOCKNODE(np);
  605                 np->n_attrstamp = 0;
  606                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
  607                 if (vp->v_type == VDIR)
  608                         np->n_direofoffset = 0;
  609                 NFSUNLOCKNODE(np);
  610                 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
  611                 if (error) {
  612                         if (NFS_ISV4(vp))
  613                                 (void) nfsrpc_close(vp, 0, ap->a_td);
  614                         return (error);
  615                 }
  616                 NFSLOCKNODE(np);
  617                 np->n_mtime = vattr.va_mtime;
  618                 if (NFS_ISV4(vp))
  619                         np->n_change = vattr.va_filerev;
  620         } else {
  621                 NFSUNLOCKNODE(np);
  622                 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
  623                 if (error) {
  624                         if (NFS_ISV4(vp))
  625                                 (void) nfsrpc_close(vp, 0, ap->a_td);
  626                         return (error);
  627                 }
  628                 NFSLOCKNODE(np);
  629                 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
  630                     NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
  631                         if (vp->v_type == VDIR)
  632                                 np->n_direofoffset = 0;
  633                         NFSUNLOCKNODE(np);
  634                         error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  635                         if (error == EINTR || error == EIO) {
  636                                 if (NFS_ISV4(vp))
  637                                         (void) nfsrpc_close(vp, 0, ap->a_td);
  638                                 return (error);
  639                         }
  640                         NFSLOCKNODE(np);
  641                         np->n_mtime = vattr.va_mtime;
  642                         if (NFS_ISV4(vp))
  643                                 np->n_change = vattr.va_filerev;
  644                 }
  645         }
  646 
  647         /*
  648          * If the object has >= 1 O_DIRECT active opens, we disable caching.
  649          */
  650         if (newnfs_directio_enable && (fmode & O_DIRECT) &&
  651             (vp->v_type == VREG)) {
  652                 if (np->n_directio_opens == 0) {
  653                         NFSUNLOCKNODE(np);
  654                         error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  655                         if (error) {
  656                                 if (NFS_ISV4(vp))
  657                                         (void) nfsrpc_close(vp, 0, ap->a_td);
  658                                 return (error);
  659                         }
  660                         NFSLOCKNODE(np);
  661                         np->n_flag |= NNONCACHE;
  662                 }
  663                 np->n_directio_opens++;
  664         }
  665 
  666         /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
  667         if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
  668                 np->n_flag |= NWRITEOPENED;
  669 
  670         /*
  671          * If this is an open for writing, capture a reference to the
  672          * credentials, so they can be used by ncl_putpages(). Using
  673          * these write credentials is preferable to the credentials of
  674          * whatever thread happens to be doing the VOP_PUTPAGES() since
  675          * the write RPCs are less likely to fail with EACCES.
  676          */
  677         if ((fmode & FWRITE) != 0) {
  678                 cred = np->n_writecred;
  679                 np->n_writecred = crhold(ap->a_cred);
  680         } else
  681                 cred = NULL;
  682         NFSUNLOCKNODE(np);
  683 
  684         if (cred != NULL)
  685                 crfree(cred);
  686         vnode_create_vobject(vp, vattr.va_size, ap->a_td);
  687 
  688         /*
  689          * If the text file has been mmap'd, flush any dirty pages to the
  690          * buffer cache and then...
  691          * Make sure all writes are pushed to the NFS server.  If this is not
  692          * done, the modify time of the file can change while the text
  693          * file is being executed.  This will cause the process that is
  694          * executing the text file to be terminated.
  695          */
  696         if (vp->v_writecount <= -1) {
  697                 if ((obj = vp->v_object) != NULL &&
  698                     (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
  699                         VM_OBJECT_WLOCK(obj);
  700                         vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
  701                         VM_OBJECT_WUNLOCK(obj);
  702                 }
  703 
  704                 /* Now, flush the buffer cache. */
  705                 ncl_flush(vp, MNT_WAIT, curthread, 0, 0);
  706 
  707                 /* And, finally, make sure that n_mtime is up to date. */
  708                 np = VTONFS(vp);
  709                 NFSLOCKNODE(np);
  710                 np->n_mtime = np->n_vattr.na_mtime;
  711                 NFSUNLOCKNODE(np);
  712         }
  713         return (0);
  714 }
  715 
  716 /*
  717  * nfs close vnode op
  718  * What an NFS client should do upon close after writing is a debatable issue.
  719  * Most NFS clients push delayed writes to the server upon close, basically for
  720  * two reasons:
  721  * 1 - So that any write errors may be reported back to the client process
  722  *     doing the close system call. By far the two most likely errors are
  723  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
  724  * 2 - To put a worst case upper bound on cache inconsistency between
  725  *     multiple clients for the file.
  726  * There is also a consistency problem for Version 2 of the protocol w.r.t.
  727  * not being able to tell if other clients are writing a file concurrently,
  728  * since there is no way of knowing if the changed modify time in the reply
  729  * is only due to the write for this client.
  730  * (NFS Version 3 provides weak cache consistency data in the reply that
  731  *  should be sufficient to detect and handle this case.)
  732  *
  733  * The current code does the following:
  734  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
  735  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
  736  *                     or commit them (this satisfies 1 and 2 except for the
  737  *                     case where the server crashes after this close but
  738  *                     before the commit RPC, which is felt to be "good
  739  *                     enough". Changing the last argument to ncl_flush() to
  740  *                     a 1 would force a commit operation, if it is felt a
  741  *                     commit is necessary now.
  742  * for NFS Version 4 - flush the dirty buffers and commit them, if
  743  *                     nfscl_mustflush() says this is necessary.
  744  *                     It is necessary if there is no write delegation held,
  745  *                     in order to satisfy open/close coherency.
  746  *                     If the file isn't cached on local stable storage,
  747  *                     it may be necessary in order to detect "out of space"
  748  *                     errors from the server, if the write delegation
  749  *                     issued by the server doesn't allow the file to grow.
  750  */
  751 /* ARGSUSED */
  752 static int
  753 nfs_close(struct vop_close_args *ap)
  754 {
  755         struct vnode *vp = ap->a_vp;
  756         struct nfsnode *np = VTONFS(vp);
  757         struct nfsvattr nfsva;
  758         struct ucred *cred;
  759         int error = 0, ret, localcred = 0;
  760         int fmode = ap->a_fflag;
  761 
  762         if (NFSCL_FORCEDISM(vp->v_mount))
  763                 return (0);
  764         /*
  765          * During shutdown, a_cred isn't valid, so just use root.
  766          */
  767         if (ap->a_cred == NOCRED) {
  768                 cred = newnfs_getcred();
  769                 localcred = 1;
  770         } else {
  771                 cred = ap->a_cred;
  772         }
  773         if (vp->v_type == VREG) {
  774             /*
  775              * Examine and clean dirty pages, regardless of NMODIFIED.
  776              * This closes a major hole in close-to-open consistency.
  777              * We want to push out all dirty pages (and buffers) on
  778              * close, regardless of whether they were dirtied by
  779              * mmap'ed writes or via write().
  780              */
  781             if (nfs_clean_pages_on_close && vp->v_object) {
  782                 VM_OBJECT_WLOCK(vp->v_object);
  783                 vm_object_page_clean(vp->v_object, 0, 0, 0);
  784                 VM_OBJECT_WUNLOCK(vp->v_object);
  785             }
  786             NFSLOCKNODE(np);
  787             if (np->n_flag & NMODIFIED) {
  788                 NFSUNLOCKNODE(np);
  789                 if (NFS_ISV3(vp)) {
  790                     /*
  791                      * Under NFSv3 we have dirty buffers to dispose of.  We
  792                      * must flush them to the NFS server.  We have the option
  793                      * of waiting all the way through the commit rpc or just
  794                      * waiting for the initial write.  The default is to only
  795                      * wait through the initial write so the data is in the
  796                      * server's cache, which is roughly similar to the state
  797                      * a standard disk subsystem leaves the file in on close().
  798                      *
  799                      * We cannot clear the NMODIFIED bit in np->n_flag due to
  800                      * potential races with other processes, and certainly
  801                      * cannot clear it if we don't commit.
  802                      * These races occur when there is no longer the old
  803                      * traditional vnode locking implemented for Vnode Ops.
  804                      */
  805                     int cm = newnfs_commit_on_close ? 1 : 0;
  806                     error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0);
  807                     /* np->n_flag &= ~NMODIFIED; */
  808                 } else if (NFS_ISV4(vp)) { 
  809                         if (nfscl_mustflush(vp) != 0) {
  810                                 int cm = newnfs_commit_on_close ? 1 : 0;
  811                                 error = ncl_flush(vp, MNT_WAIT, ap->a_td,
  812                                     cm, 0);
  813                                 /*
  814                                  * as above w.r.t races when clearing
  815                                  * NMODIFIED.
  816                                  * np->n_flag &= ~NMODIFIED;
  817                                  */
  818                         }
  819                 } else {
  820                         error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  821                 }
  822                 NFSLOCKNODE(np);
  823             }
  824             /* 
  825              * Invalidate the attribute cache in all cases.
  826              * An open is going to fetch fresh attrs any way, other procs
  827              * on this node that have file open will be forced to do an 
  828              * otw attr fetch, but this is safe.
  829              * --> A user found that their RPC count dropped by 20% when
  830              *     this was commented out and I can't see any requirement
  831              *     for it, so I've disabled it when negative lookups are
  832              *     enabled. (What does this have to do with negative lookup
  833              *     caching? Well nothing, except it was reported by the
  834              *     same user that needed negative lookup caching and I wanted
  835              *     there to be a way to disable it to see if it
  836              *     is the cause of some caching/coherency issue that might
  837              *     crop up.)
  838              */
  839             if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
  840                     np->n_attrstamp = 0;
  841                     KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
  842             }
  843             if (np->n_flag & NWRITEERR) {
  844                 np->n_flag &= ~NWRITEERR;
  845                 error = np->n_error;
  846             }
  847             NFSUNLOCKNODE(np);
  848         }
  849 
  850         if (NFS_ISV4(vp)) {
  851                 /*
  852                  * Get attributes so "change" is up to date.
  853                  */
  854                 if (error == 0 && nfscl_mustflush(vp) != 0 &&
  855                     vp->v_type == VREG &&
  856                     (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
  857                         ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
  858                             NULL);
  859                         if (!ret) {
  860                                 np->n_change = nfsva.na_filerev;
  861                                 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
  862                                     NULL, 0, 0);
  863                         }
  864                 }
  865 
  866                 /*
  867                  * and do the close.
  868                  */
  869                 ret = nfsrpc_close(vp, 0, ap->a_td);
  870                 if (!error && ret)
  871                         error = ret;
  872                 if (error)
  873                         error = nfscl_maperr(ap->a_td, error, (uid_t)0,
  874                             (gid_t)0);
  875         }
  876         if (newnfs_directio_enable)
  877                 KASSERT((np->n_directio_asyncwr == 0),
  878                         ("nfs_close: dirty unflushed (%d) directio buffers\n",
  879                          np->n_directio_asyncwr));
  880         if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
  881                 NFSLOCKNODE(np);
  882                 KASSERT((np->n_directio_opens > 0), 
  883                         ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
  884                 np->n_directio_opens--;
  885                 if (np->n_directio_opens == 0)
  886                         np->n_flag &= ~NNONCACHE;
  887                 NFSUNLOCKNODE(np);
  888         }
  889         if (localcred)
  890                 NFSFREECRED(cred);
  891         return (error);
  892 }
  893 
  894 /*
  895  * nfs getattr call from vfs.
  896  */
  897 static int
  898 nfs_getattr(struct vop_getattr_args *ap)
  899 {
  900         struct vnode *vp = ap->a_vp;
  901         struct thread *td = curthread;  /* XXX */
  902         struct nfsnode *np = VTONFS(vp);
  903         int error = 0;
  904         struct nfsvattr nfsva;
  905         struct vattr *vap = ap->a_vap;
  906         struct vattr vattr;
  907 
  908         /*
  909          * Update local times for special files.
  910          */
  911         NFSLOCKNODE(np);
  912         if (np->n_flag & (NACC | NUPD))
  913                 np->n_flag |= NCHG;
  914         NFSUNLOCKNODE(np);
  915         /*
  916          * First look in the cache.
  917          */
  918         if (ncl_getattrcache(vp, &vattr) == 0) {
  919                 ncl_copy_vattr(vap, &vattr);
  920 
  921                 /*
  922                  * Get the local modify time for the case of a write
  923                  * delegation.
  924                  */
  925                 nfscl_deleggetmodtime(vp, &vap->va_mtime);
  926                 return (0);
  927         }
  928 
  929         if (NFS_ISV34(vp) && nfs_prime_access_cache &&
  930             nfsaccess_cache_timeout > 0) {
  931                 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
  932                 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
  933                 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
  934                         nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
  935                         return (0);
  936                 }
  937         }
  938         error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
  939         if (!error)
  940                 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
  941         if (!error) {
  942                 /*
  943                  * Get the local modify time for the case of a write
  944                  * delegation.
  945                  */
  946                 nfscl_deleggetmodtime(vp, &vap->va_mtime);
  947         } else if (NFS_ISV4(vp)) {
  948                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
  949         }
  950         return (error);
  951 }
  952 
  953 /*
  954  * nfs setattr call.
  955  */
  956 static int
  957 nfs_setattr(struct vop_setattr_args *ap)
  958 {
  959         struct vnode *vp = ap->a_vp;
  960         struct nfsnode *np = VTONFS(vp);
  961         struct thread *td = curthread;  /* XXX */
  962         struct vattr *vap = ap->a_vap;
  963         int error = 0;
  964         u_quad_t tsize;
  965         struct timespec ts;
  966 
  967 #ifndef nolint
  968         tsize = (u_quad_t)0;
  969 #endif
  970 
  971         /*
  972          * Setting of flags and marking of atimes are not supported.
  973          */
  974         if (vap->va_flags != VNOVAL)
  975                 return (EOPNOTSUPP);
  976 
  977         /*
  978          * Disallow write attempts if the filesystem is mounted read-only.
  979          */
  980         if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
  981             vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
  982             vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
  983             (vp->v_mount->mnt_flag & MNT_RDONLY))
  984                 return (EROFS);
  985         if (vap->va_size != VNOVAL) {
  986                 switch (vp->v_type) {
  987                 case VDIR:
  988                         return (EISDIR);
  989                 case VCHR:
  990                 case VBLK:
  991                 case VSOCK:
  992                 case VFIFO:
  993                         if (vap->va_mtime.tv_sec == VNOVAL &&
  994                             vap->va_atime.tv_sec == VNOVAL &&
  995                             vap->va_mode == (mode_t)VNOVAL &&
  996                             vap->va_uid == (uid_t)VNOVAL &&
  997                             vap->va_gid == (gid_t)VNOVAL)
  998                                 return (0);             
  999                         vap->va_size = VNOVAL;
 1000                         break;
 1001                 default:
 1002                         /*
 1003                          * Disallow write attempts if the filesystem is
 1004                          * mounted read-only.
 1005                          */
 1006                         if (vp->v_mount->mnt_flag & MNT_RDONLY)
 1007                                 return (EROFS);
 1008                         /*
 1009                          *  We run vnode_pager_setsize() early (why?),
 1010                          * we must set np->n_size now to avoid vinvalbuf
 1011                          * V_SAVE races that might setsize a lower
 1012                          * value.
 1013                          */
 1014                         NFSLOCKNODE(np);
 1015                         tsize = np->n_size;
 1016                         NFSUNLOCKNODE(np);
 1017                         error = ncl_meta_setsize(vp, td, vap->va_size);
 1018                         NFSLOCKNODE(np);
 1019                         if (np->n_flag & NMODIFIED) {
 1020                             tsize = np->n_size;
 1021                             NFSUNLOCKNODE(np);
 1022                             error = ncl_vinvalbuf(vp, vap->va_size == 0 ?
 1023                                 0 : V_SAVE, td, 1);
 1024                             if (error != 0) {
 1025                                     vnode_pager_setsize(vp, tsize);
 1026                                     return (error);
 1027                             }
 1028                             /*
 1029                              * Call nfscl_delegmodtime() to set the modify time
 1030                              * locally, as required.
 1031                              */
 1032                             nfscl_delegmodtime(vp);
 1033                         } else
 1034                             NFSUNLOCKNODE(np);
 1035                         /*
 1036                          * np->n_size has already been set to vap->va_size
 1037                          * in ncl_meta_setsize(). We must set it again since
 1038                          * nfs_loadattrcache() could be called through
 1039                          * ncl_meta_setsize() and could modify np->n_size.
 1040                          */
 1041                         NFSLOCKNODE(np);
 1042                         np->n_vattr.na_size = np->n_size = vap->va_size;
 1043                         NFSUNLOCKNODE(np);
 1044                 }
 1045         } else {
 1046                 NFSLOCKNODE(np);
 1047                 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 
 1048                     (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
 1049                         NFSUNLOCKNODE(np);
 1050                         error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
 1051                         if (error == EINTR || error == EIO)
 1052                                 return (error);
 1053                 } else
 1054                         NFSUNLOCKNODE(np);
 1055         }
 1056         error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
 1057         if (vap->va_size != VNOVAL) {
 1058                 if (error == 0) {
 1059                         nanouptime(&ts);
 1060                         NFSLOCKNODE(np);
 1061                         np->n_localmodtime = ts;
 1062                         NFSUNLOCKNODE(np);
 1063                 } else {
 1064                         NFSLOCKNODE(np);
 1065                         np->n_size = np->n_vattr.na_size = tsize;
 1066                         vnode_pager_setsize(vp, tsize);
 1067                         NFSUNLOCKNODE(np);
 1068                 }
 1069         }
 1070         return (error);
 1071 }
 1072 
 1073 /*
 1074  * Do an nfs setattr rpc.
 1075  */
 1076 static int
 1077 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
 1078     struct thread *td)
 1079 {
 1080         struct nfsnode *np = VTONFS(vp);
 1081         int error, ret, attrflag, i;
 1082         struct nfsvattr nfsva;
 1083 
 1084         if (NFS_ISV34(vp)) {
 1085                 NFSLOCKNODE(np);
 1086                 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
 1087                         np->n_accesscache[i].stamp = 0;
 1088                 np->n_flag |= NDELEGMOD;
 1089                 NFSUNLOCKNODE(np);
 1090                 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
 1091         }
 1092         error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
 1093             NULL);
 1094         if (attrflag) {
 1095                 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
 1096                 if (ret && !error)
 1097                         error = ret;
 1098         }
 1099         if (error && NFS_ISV4(vp))
 1100                 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
 1101         return (error);
 1102 }
 1103 
 1104 /*
 1105  * nfs lookup call, one step at a time...
 1106  * First look in cache
 1107  * If not found, unlock the directory nfsnode and do the rpc
 1108  */
 1109 static int
 1110 nfs_lookup(struct vop_lookup_args *ap)
 1111 {
 1112         struct componentname *cnp = ap->a_cnp;
 1113         struct vnode *dvp = ap->a_dvp;
 1114         struct vnode **vpp = ap->a_vpp;
 1115         struct mount *mp = dvp->v_mount;
 1116         int flags = cnp->cn_flags;
 1117         struct vnode *newvp;
 1118         struct nfsmount *nmp;
 1119         struct nfsnode *np, *newnp;
 1120         int error = 0, attrflag, dattrflag, ltype, ncticks;
 1121         struct thread *td = cnp->cn_thread;
 1122         struct nfsfh *nfhp;
 1123         struct nfsvattr dnfsva, nfsva;
 1124         struct vattr vattr;
 1125         struct timespec nctime, ts;
 1126 
 1127         *vpp = NULLVP;
 1128         if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
 1129             (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
 1130                 return (EROFS);
 1131         if (dvp->v_type != VDIR)
 1132                 return (ENOTDIR);
 1133         nmp = VFSTONFS(mp);
 1134         np = VTONFS(dvp);
 1135 
 1136         /* For NFSv4, wait until any remove is done. */
 1137         NFSLOCKNODE(np);
 1138         while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
 1139                 np->n_flag |= NREMOVEWANT;
 1140                 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
 1141         }
 1142         NFSUNLOCKNODE(np);
 1143 
 1144         error = vn_dir_check_exec(dvp, cnp);
 1145         if (error != 0)
 1146                 return (error);
 1147         error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
 1148         if (error > 0 && error != ENOENT)
 1149                 return (error);
 1150         if (error == -1) {
 1151                 /*
 1152                  * Lookups of "." are special and always return the
 1153                  * current directory.  cache_lookup() already handles
 1154                  * associated locking bookkeeping, etc.
 1155                  */
 1156                 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
 1157                         /* XXX: Is this really correct? */
 1158                         if (cnp->cn_nameiop != LOOKUP &&
 1159                             (flags & ISLASTCN))
 1160                                 cnp->cn_flags |= SAVENAME;
 1161                         return (0);
 1162                 }
 1163 
 1164                 /*
 1165                  * We only accept a positive hit in the cache if the
 1166                  * change time of the file matches our cached copy.
 1167                  * Otherwise, we discard the cache entry and fallback
 1168                  * to doing a lookup RPC.  We also only trust cache
 1169                  * entries for less than nm_nametimeo seconds.
 1170                  *
 1171                  * To better handle stale file handles and attributes,
 1172                  * clear the attribute cache of this node if it is a
 1173                  * leaf component, part of an open() call, and not
 1174                  * locally modified before fetching the attributes.
 1175                  * This should allow stale file handles to be detected
 1176                  * here where we can fall back to a LOOKUP RPC to
 1177                  * recover rather than having nfs_open() detect the
 1178                  * stale file handle and failing open(2) with ESTALE.
 1179                  */
 1180                 newvp = *vpp;
 1181                 newnp = VTONFS(newvp);
 1182                 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
 1183                     (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
 1184                     !(newnp->n_flag & NMODIFIED)) {
 1185                         NFSLOCKNODE(newnp);
 1186                         newnp->n_attrstamp = 0;
 1187                         KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
 1188                         NFSUNLOCKNODE(newnp);
 1189                 }
 1190                 if (nfscl_nodeleg(newvp, 0) == 0 ||
 1191                     ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
 1192                     VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
 1193                     timespeccmp(&vattr.va_ctime, &nctime, ==))) {
 1194                         NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
 1195                         if (cnp->cn_nameiop != LOOKUP &&
 1196                             (flags & ISLASTCN))
 1197                                 cnp->cn_flags |= SAVENAME;
 1198                         return (0);
 1199                 }
 1200                 cache_purge(newvp);
 1201                 if (dvp != newvp)
 1202                         vput(newvp);
 1203                 else 
 1204                         vrele(newvp);
 1205                 *vpp = NULLVP;
 1206         } else if (error == ENOENT) {
 1207                 if (dvp->v_iflag & VI_DOOMED)
 1208                         return (ENOENT);
 1209                 /*
 1210                  * We only accept a negative hit in the cache if the
 1211                  * modification time of the parent directory matches
 1212                  * the cached copy in the name cache entry.
 1213                  * Otherwise, we discard all of the negative cache
 1214                  * entries for this directory.  We also only trust
 1215                  * negative cache entries for up to nm_negnametimeo
 1216                  * seconds.
 1217                  */
 1218                 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
 1219                     VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
 1220                     timespeccmp(&vattr.va_mtime, &nctime, ==)) {
 1221                         NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
 1222                         return (ENOENT);
 1223                 }
 1224                 cache_purge_negative(dvp);
 1225         }
 1226 
 1227         error = 0;
 1228         newvp = NULLVP;
 1229         NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses);
 1230         nanouptime(&ts);
 1231         error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 1232             cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
 1233             NULL);
 1234         if (dattrflag)
 1235                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 1236         if (error) {
 1237                 if (newvp != NULLVP) {
 1238                         vput(newvp);
 1239                         *vpp = NULLVP;
 1240                 }
 1241 
 1242                 if (error != ENOENT) {
 1243                         if (NFS_ISV4(dvp))
 1244                                 error = nfscl_maperr(td, error, (uid_t)0,
 1245                                     (gid_t)0);
 1246                         return (error);
 1247                 }
 1248 
 1249                 /* The requested file was not found. */
 1250                 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
 1251                     (flags & ISLASTCN)) {
 1252                         /*
 1253                          * XXX: UFS does a full VOP_ACCESS(dvp,
 1254                          * VWRITE) here instead of just checking
 1255                          * MNT_RDONLY.
 1256                          */
 1257                         if (mp->mnt_flag & MNT_RDONLY)
 1258                                 return (EROFS);
 1259                         cnp->cn_flags |= SAVENAME;
 1260                         return (EJUSTRETURN);
 1261                 }
 1262 
 1263                 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
 1264                         /*
 1265                          * Cache the modification time of the parent
 1266                          * directory from the post-op attributes in
 1267                          * the name cache entry.  The negative cache
 1268                          * entry will be ignored once the directory
 1269                          * has changed.  Don't bother adding the entry
 1270                          * if the directory has already changed.
 1271                          */
 1272                         NFSLOCKNODE(np);
 1273                         if (timespeccmp(&np->n_vattr.na_mtime,
 1274                             &dnfsva.na_mtime, ==)) {
 1275                                 NFSUNLOCKNODE(np);
 1276                                 cache_enter_time(dvp, NULL, cnp,
 1277                                     &dnfsva.na_mtime, NULL);
 1278                         } else
 1279                                 NFSUNLOCKNODE(np);
 1280                 }
 1281                 return (ENOENT);
 1282         }
 1283 
 1284         /*
 1285          * Handle RENAME case...
 1286          */
 1287         if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
 1288                 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
 1289                         free(nfhp, M_NFSFH);
 1290                         return (EISDIR);
 1291                 }
 1292                 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
 1293                     LK_EXCLUSIVE);
 1294                 if (error)
 1295                         return (error);
 1296                 newvp = NFSTOV(np);
 1297                 /*
 1298                  * If n_localmodtime >= time before RPC, then
 1299                  * a file modification operation, such as
 1300                  * VOP_SETATTR() of size, has occurred while
 1301                  * the Lookup RPC and acquisition of the vnode
 1302                  * happened.  As such, the attributes might
 1303                  * be stale, with possibly an incorrect size.
 1304                  */
 1305                 NFSLOCKNODE(np);
 1306                 if (timespecisset(&np->n_localmodtime) &&
 1307                     timespeccmp(&np->n_localmodtime, &ts, >=)) {
 1308                         NFSCL_DEBUG(4, "nfs_lookup: rename localmod "
 1309                             "stale attributes\n");
 1310                         attrflag = 0;
 1311                 }
 1312                 NFSUNLOCKNODE(np);
 1313                 if (attrflag)
 1314                         (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1315                             0, 1);
 1316                 *vpp = newvp;
 1317                 cnp->cn_flags |= SAVENAME;
 1318                 return (0);
 1319         }
 1320 
 1321         if (flags & ISDOTDOT) {
 1322                 ltype = NFSVOPISLOCKED(dvp);
 1323                 error = vfs_busy(mp, MBF_NOWAIT);
 1324                 if (error != 0) {
 1325                         vfs_ref(mp);
 1326                         NFSVOPUNLOCK(dvp, 0);
 1327                         error = vfs_busy(mp, 0);
 1328                         NFSVOPLOCK(dvp, ltype | LK_RETRY);
 1329                         vfs_rel(mp);
 1330                         if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
 1331                                 vfs_unbusy(mp);
 1332                                 error = ENOENT;
 1333                         }
 1334                         if (error != 0)
 1335                                 return (error);
 1336                 }
 1337                 NFSVOPUNLOCK(dvp, 0);
 1338                 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
 1339                     cnp->cn_lkflags);
 1340                 if (error == 0)
 1341                         newvp = NFSTOV(np);
 1342                 vfs_unbusy(mp);
 1343                 if (newvp != dvp)
 1344                         NFSVOPLOCK(dvp, ltype | LK_RETRY);
 1345                 if (dvp->v_iflag & VI_DOOMED) {
 1346                         if (error == 0) {
 1347                                 if (newvp == dvp)
 1348                                         vrele(newvp);
 1349                                 else
 1350                                         vput(newvp);
 1351                         }
 1352                         error = ENOENT;
 1353                 }
 1354                 if (error != 0)
 1355                         return (error);
 1356                 if (attrflag)
 1357                         (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1358                             0, 1);
 1359         } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
 1360                 free(nfhp, M_NFSFH);
 1361                 VREF(dvp);
 1362                 newvp = dvp;
 1363                 if (attrflag)
 1364                         (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1365                             0, 1);
 1366         } else {
 1367                 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
 1368                     cnp->cn_lkflags);
 1369                 if (error)
 1370                         return (error);
 1371                 newvp = NFSTOV(np);
 1372                 /*
 1373                  * If n_localmodtime >= time before RPC, then
 1374                  * a file modification operation, such as
 1375                  * VOP_SETATTR() of size, has occurred while
 1376                  * the Lookup RPC and acquisition of the vnode
 1377                  * happened.  As such, the attributes might
 1378                  * be stale, with possibly an incorrect size.
 1379                  */
 1380                 NFSLOCKNODE(np);
 1381                 if (timespecisset(&np->n_localmodtime) &&
 1382                     timespeccmp(&np->n_localmodtime, &ts, >=)) {
 1383                         NFSCL_DEBUG(4, "nfs_lookup: localmod "
 1384                             "stale attributes\n");
 1385                         attrflag = 0;
 1386                 }
 1387                 NFSUNLOCKNODE(np);
 1388                 if (attrflag)
 1389                         (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1390                             0, 1);
 1391                 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
 1392                     !(np->n_flag & NMODIFIED)) {                        
 1393                         /*
 1394                          * Flush the attribute cache when opening a
 1395                          * leaf node to ensure that fresh attributes
 1396                          * are fetched in nfs_open() since we did not
 1397                          * fetch attributes from the LOOKUP reply.
 1398                          */
 1399                         NFSLOCKNODE(np);
 1400                         np->n_attrstamp = 0;
 1401                         KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
 1402                         NFSUNLOCKNODE(np);
 1403                 }
 1404         }
 1405         if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
 1406                 cnp->cn_flags |= SAVENAME;
 1407         if ((cnp->cn_flags & MAKEENTRY) && dvp != newvp &&
 1408             (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
 1409             attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
 1410                 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
 1411                     newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
 1412         *vpp = newvp;
 1413         return (0);
 1414 }
 1415 
 1416 /*
 1417  * nfs read call.
 1418  * Just call ncl_bioread() to do the work.
 1419  */
 1420 static int
 1421 nfs_read(struct vop_read_args *ap)
 1422 {
 1423         struct vnode *vp = ap->a_vp;
 1424 
 1425         switch (vp->v_type) {
 1426         case VREG:
 1427                 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
 1428         case VDIR:
 1429                 return (EISDIR);
 1430         default:
 1431                 return (EOPNOTSUPP);
 1432         }
 1433 }
 1434 
 1435 /*
 1436  * nfs readlink call
 1437  */
 1438 static int
 1439 nfs_readlink(struct vop_readlink_args *ap)
 1440 {
 1441         struct vnode *vp = ap->a_vp;
 1442 
 1443         if (vp->v_type != VLNK)
 1444                 return (EINVAL);
 1445         return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
 1446 }
 1447 
 1448 /*
 1449  * Do a readlink rpc.
 1450  * Called by ncl_doio() from below the buffer cache.
 1451  */
 1452 int
 1453 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 1454 {
 1455         int error, ret, attrflag;
 1456         struct nfsvattr nfsva;
 1457 
 1458         error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
 1459             &attrflag, NULL);
 1460         if (attrflag) {
 1461                 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
 1462                 if (ret && !error)
 1463                         error = ret;
 1464         }
 1465         if (error && NFS_ISV4(vp))
 1466                 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
 1467         return (error);
 1468 }
 1469 
 1470 /*
 1471  * nfs read rpc call
 1472  * Ditto above
 1473  */
 1474 int
 1475 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 1476 {
 1477         int error, ret, attrflag;
 1478         struct nfsvattr nfsva;
 1479         struct nfsmount *nmp;
 1480 
 1481         nmp = VFSTONFS(vnode_mount(vp));
 1482         error = EIO;
 1483         attrflag = 0;
 1484         if (NFSHASPNFS(nmp))
 1485                 error = nfscl_doiods(vp, uiop, NULL, NULL,
 1486                     NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td);
 1487         NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
 1488         if (error != 0)
 1489                 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
 1490                     &attrflag, NULL);
 1491         if (attrflag) {
 1492                 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
 1493                 if (ret && !error)
 1494                         error = ret;
 1495         }
 1496         if (error && NFS_ISV4(vp))
 1497                 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
 1498         return (error);
 1499 }
 1500 
 1501 /*
 1502  * nfs write call
 1503  */
 1504 int
 1505 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
 1506     int *iomode, int *must_commit, int called_from_strategy)
 1507 {
 1508         struct nfsvattr nfsva;
 1509         int error, attrflag, ret;
 1510         struct nfsmount *nmp;
 1511 
 1512         nmp = VFSTONFS(vnode_mount(vp));
 1513         error = EIO;
 1514         attrflag = 0;
 1515         if (NFSHASPNFS(nmp))
 1516                 error = nfscl_doiods(vp, uiop, iomode, must_commit,
 1517                     NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td);
 1518         NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
 1519         if (error != 0)
 1520                 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
 1521                     uiop->uio_td, &nfsva, &attrflag, NULL,
 1522                     called_from_strategy);
 1523         if (attrflag) {
 1524                 if (VTONFS(vp)->n_flag & ND_NFSV4)
 1525                         ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
 1526                             1);
 1527                 else
 1528                         ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
 1529                             1);
 1530                 if (ret && !error)
 1531                         error = ret;
 1532         }
 1533         if (DOINGASYNC(vp))
 1534                 *iomode = NFSWRITE_FILESYNC;
 1535         if (error && NFS_ISV4(vp))
 1536                 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
 1537         return (error);
 1538 }
 1539 
 1540 /*
 1541  * nfs mknod rpc
 1542  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
 1543  * mode set to specify the file type and the size field for rdev.
 1544  */
 1545 static int
 1546 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
 1547     struct vattr *vap)
 1548 {
 1549         struct nfsvattr nfsva, dnfsva;
 1550         struct vnode *newvp = NULL;
 1551         struct nfsnode *np = NULL, *dnp;
 1552         struct nfsfh *nfhp;
 1553         struct vattr vattr;
 1554         int error = 0, attrflag, dattrflag;
 1555         u_int32_t rdev;
 1556 
 1557         if (vap->va_type == VCHR || vap->va_type == VBLK)
 1558                 rdev = vap->va_rdev;
 1559         else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
 1560                 rdev = 0xffffffff;
 1561         else
 1562                 return (EOPNOTSUPP);
 1563         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
 1564                 return (error);
 1565         error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
 1566             rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
 1567             &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
 1568         if (!error) {
 1569                 if (!nfhp)
 1570                         (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
 1571                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
 1572                             &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
 1573                             NULL);
 1574                 if (nfhp)
 1575                         error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
 1576                             cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
 1577         }
 1578         if (dattrflag)
 1579                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 1580         if (!error) {
 1581                 newvp = NFSTOV(np);
 1582                 if (attrflag != 0) {
 1583                         error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1584                             0, 1);
 1585                         if (error != 0)
 1586                                 vput(newvp);
 1587                 }
 1588         }
 1589         if (!error) {
 1590                 *vpp = newvp;
 1591         } else if (NFS_ISV4(dvp)) {
 1592                 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
 1593                     vap->va_gid);
 1594         }
 1595         dnp = VTONFS(dvp);
 1596         NFSLOCKNODE(dnp);
 1597         dnp->n_flag |= NMODIFIED;
 1598         if (!dattrflag) {
 1599                 dnp->n_attrstamp = 0;
 1600                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
 1601         }
 1602         NFSUNLOCKNODE(dnp);
 1603         return (error);
 1604 }
 1605 
 1606 /*
 1607  * nfs mknod vop
 1608  * just call nfs_mknodrpc() to do the work.
 1609  */
 1610 /* ARGSUSED */
 1611 static int
 1612 nfs_mknod(struct vop_mknod_args *ap)
 1613 {
 1614         return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
 1615 }
 1616 
 1617 static struct mtx nfs_cverf_mtx;
 1618 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
 1619     MTX_DEF);
 1620 
 1621 static nfsquad_t
 1622 nfs_get_cverf(void)
 1623 {
 1624         static nfsquad_t cverf;
 1625         nfsquad_t ret;
 1626         static int cverf_initialized = 0;
 1627 
 1628         mtx_lock(&nfs_cverf_mtx);
 1629         if (cverf_initialized == 0) {
 1630                 cverf.lval[0] = arc4random();
 1631                 cverf.lval[1] = arc4random();
 1632                 cverf_initialized = 1;
 1633         } else
 1634                 cverf.qval++;
 1635         ret = cverf;
 1636         mtx_unlock(&nfs_cverf_mtx);
 1637 
 1638         return (ret);
 1639 }
 1640 
 1641 /*
 1642  * nfs file create call
 1643  */
 1644 static int
 1645 nfs_create(struct vop_create_args *ap)
 1646 {
 1647         struct vnode *dvp = ap->a_dvp;
 1648         struct vattr *vap = ap->a_vap;
 1649         struct componentname *cnp = ap->a_cnp;
 1650         struct nfsnode *np = NULL, *dnp;
 1651         struct vnode *newvp = NULL;
 1652         struct nfsmount *nmp;
 1653         struct nfsvattr dnfsva, nfsva;
 1654         struct nfsfh *nfhp;
 1655         nfsquad_t cverf;
 1656         int error = 0, attrflag, dattrflag, fmode = 0;
 1657         struct vattr vattr;
 1658 
 1659         /*
 1660          * Oops, not for me..
 1661          */
 1662         if (vap->va_type == VSOCK)
 1663                 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
 1664 
 1665         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
 1666                 return (error);
 1667         if (vap->va_vaflags & VA_EXCLUSIVE)
 1668                 fmode |= O_EXCL;
 1669         dnp = VTONFS(dvp);
 1670         nmp = VFSTONFS(vnode_mount(dvp));
 1671 again:
 1672         /* For NFSv4, wait until any remove is done. */
 1673         NFSLOCKNODE(dnp);
 1674         while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
 1675                 dnp->n_flag |= NREMOVEWANT;
 1676                 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
 1677         }
 1678         NFSUNLOCKNODE(dnp);
 1679 
 1680         cverf = nfs_get_cverf();
 1681         error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 1682             vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
 1683             &nfhp, &attrflag, &dattrflag, NULL);
 1684         if (!error) {
 1685                 if (nfhp == NULL)
 1686                         (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
 1687                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
 1688                             &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
 1689                             NULL);
 1690                 if (nfhp != NULL)
 1691                         error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
 1692                             cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
 1693         }
 1694         if (dattrflag)
 1695                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 1696         if (!error) {
 1697                 newvp = NFSTOV(np);
 1698                 if (attrflag == 0)
 1699                         error = nfsrpc_getattr(newvp, cnp->cn_cred,
 1700                             cnp->cn_thread, &nfsva, NULL);
 1701                 if (error == 0)
 1702                         error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1703                             0, 1);
 1704         }
 1705         if (error) {
 1706                 if (newvp != NULL) {
 1707                         vput(newvp);
 1708                         newvp = NULL;
 1709                 }
 1710                 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
 1711                     error == NFSERR_NOTSUPP) {
 1712                         fmode &= ~O_EXCL;
 1713                         goto again;
 1714                 }
 1715         } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
 1716                 if (nfscl_checksattr(vap, &nfsva)) {
 1717                         error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
 1718                             cnp->cn_thread, &nfsva, &attrflag, NULL);
 1719                         if (error && (vap->va_uid != (uid_t)VNOVAL ||
 1720                             vap->va_gid != (gid_t)VNOVAL)) {
 1721                                 /* try again without setting uid/gid */
 1722                                 vap->va_uid = (uid_t)VNOVAL;
 1723                                 vap->va_gid = (uid_t)VNOVAL;
 1724                                 error = nfsrpc_setattr(newvp, vap, NULL, 
 1725                                     cnp->cn_cred, cnp->cn_thread, &nfsva,
 1726                                     &attrflag, NULL);
 1727                         }
 1728                         if (attrflag)
 1729                                 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
 1730                                     NULL, 0, 1);
 1731                         if (error != 0)
 1732                                 vput(newvp);
 1733                 }
 1734         }
 1735         if (!error) {
 1736                 if ((cnp->cn_flags & MAKEENTRY) && attrflag) {
 1737                         if (dvp != newvp)
 1738                                 cache_enter_time(dvp, newvp, cnp,
 1739                                     &nfsva.na_ctime, NULL);
 1740                         else
 1741                                 printf("nfs_create: bogus NFS server returned "
 1742                                     "the directory as the new file object\n");
 1743                 }
 1744                 *ap->a_vpp = newvp;
 1745         } else if (NFS_ISV4(dvp)) {
 1746                 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
 1747                     vap->va_gid);
 1748         }
 1749         NFSLOCKNODE(dnp);
 1750         dnp->n_flag |= NMODIFIED;
 1751         if (!dattrflag) {
 1752                 dnp->n_attrstamp = 0;
 1753                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
 1754         }
 1755         NFSUNLOCKNODE(dnp);
 1756         return (error);
 1757 }
 1758 
 1759 /*
 1760  * nfs file remove call
 1761  * To try and make nfs semantics closer to ufs semantics, a file that has
 1762  * other processes using the vnode is renamed instead of removed and then
 1763  * removed later on the last close.
 1764  * - If v_usecount > 1
 1765  *        If a rename is not already in the works
 1766  *           call nfs_sillyrename() to set it up
 1767  *     else
 1768  *        do the remove rpc
 1769  */
 1770 static int
 1771 nfs_remove(struct vop_remove_args *ap)
 1772 {
 1773         struct vnode *vp = ap->a_vp;
 1774         struct vnode *dvp = ap->a_dvp;
 1775         struct componentname *cnp = ap->a_cnp;
 1776         struct nfsnode *np = VTONFS(vp);
 1777         int error = 0;
 1778         struct vattr vattr;
 1779 
 1780         KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
 1781         KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
 1782         if (vp->v_type == VDIR)
 1783                 error = EPERM;
 1784         else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
 1785             VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
 1786             vattr.va_nlink > 1)) {
 1787                 /*
 1788                  * Purge the name cache so that the chance of a lookup for
 1789                  * the name succeeding while the remove is in progress is
 1790                  * minimized. Without node locking it can still happen, such
 1791                  * that an I/O op returns ESTALE, but since you get this if
 1792                  * another host removes the file..
 1793                  */
 1794                 cache_purge(vp);
 1795                 /*
 1796                  * throw away biocache buffers, mainly to avoid
 1797                  * unnecessary delayed writes later.
 1798                  */
 1799                 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
 1800                 if (error != EINTR && error != EIO)
 1801                         /* Do the rpc */
 1802                         error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
 1803                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
 1804                 /*
 1805                  * Kludge City: If the first reply to the remove rpc is lost..
 1806                  *   the reply to the retransmitted request will be ENOENT
 1807                  *   since the file was in fact removed
 1808                  *   Therefore, we cheat and return success.
 1809                  */
 1810                 if (error == ENOENT)
 1811                         error = 0;
 1812         } else if (!np->n_sillyrename)
 1813                 error = nfs_sillyrename(dvp, vp, cnp);
 1814         NFSLOCKNODE(np);
 1815         np->n_attrstamp = 0;
 1816         NFSUNLOCKNODE(np);
 1817         KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
 1818         return (error);
 1819 }
 1820 
 1821 /*
 1822  * nfs file remove rpc called from nfs_inactive
 1823  */
 1824 int
 1825 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
 1826 {
 1827         /*
 1828          * Make sure that the directory vnode is still valid.
 1829          * XXX we should lock sp->s_dvp here.
 1830          */
 1831         if (sp->s_dvp->v_type == VBAD)
 1832                 return (0);
 1833         return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
 1834             sp->s_cred, NULL));
 1835 }
 1836 
 1837 /*
 1838  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
 1839  */
 1840 static int
 1841 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
 1842     int namelen, struct ucred *cred, struct thread *td)
 1843 {
 1844         struct nfsvattr dnfsva;
 1845         struct nfsnode *dnp = VTONFS(dvp);
 1846         int error = 0, dattrflag;
 1847 
 1848         NFSLOCKNODE(dnp);
 1849         dnp->n_flag |= NREMOVEINPROG;
 1850         NFSUNLOCKNODE(dnp);
 1851         error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
 1852             &dattrflag, NULL);
 1853         NFSLOCKNODE(dnp);
 1854         if ((dnp->n_flag & NREMOVEWANT)) {
 1855                 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
 1856                 NFSUNLOCKNODE(dnp);
 1857                 wakeup((caddr_t)dnp);
 1858         } else {
 1859                 dnp->n_flag &= ~NREMOVEINPROG;
 1860                 NFSUNLOCKNODE(dnp);
 1861         }
 1862         if (dattrflag)
 1863                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 1864         NFSLOCKNODE(dnp);
 1865         dnp->n_flag |= NMODIFIED;
 1866         if (!dattrflag) {
 1867                 dnp->n_attrstamp = 0;
 1868                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
 1869         }
 1870         NFSUNLOCKNODE(dnp);
 1871         if (error && NFS_ISV4(dvp))
 1872                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
 1873         return (error);
 1874 }
 1875 
 1876 /*
 1877  * nfs file rename call
 1878  */
 1879 static int
 1880 nfs_rename(struct vop_rename_args *ap)
 1881 {
 1882         struct vnode *fvp = ap->a_fvp;
 1883         struct vnode *tvp = ap->a_tvp;
 1884         struct vnode *fdvp = ap->a_fdvp;
 1885         struct vnode *tdvp = ap->a_tdvp;
 1886         struct componentname *tcnp = ap->a_tcnp;
 1887         struct componentname *fcnp = ap->a_fcnp;
 1888         struct nfsnode *fnp = VTONFS(ap->a_fvp);
 1889         struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
 1890         struct nfsv4node *newv4 = NULL;
 1891         int error;
 1892 
 1893         KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
 1894             (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
 1895         /* Check for cross-device rename */
 1896         if ((fvp->v_mount != tdvp->v_mount) ||
 1897             (tvp && (fvp->v_mount != tvp->v_mount))) {
 1898                 error = EXDEV;
 1899                 goto out;
 1900         }
 1901 
 1902         if (fvp == tvp) {
 1903                 printf("nfs_rename: fvp == tvp (can't happen)\n");
 1904                 error = 0;
 1905                 goto out;
 1906         }
 1907         if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
 1908                 goto out;
 1909 
 1910         /*
 1911          * We have to flush B_DELWRI data prior to renaming
 1912          * the file.  If we don't, the delayed-write buffers
 1913          * can be flushed out later after the file has gone stale
 1914          * under NFSV3.  NFSV2 does not have this problem because
 1915          * ( as far as I can tell ) it flushes dirty buffers more
 1916          * often.
 1917          * 
 1918          * Skip the rename operation if the fsync fails, this can happen
 1919          * due to the server's volume being full, when we pushed out data
 1920          * that was written back to our cache earlier. Not checking for
 1921          * this condition can result in potential (silent) data loss.
 1922          */
 1923         error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
 1924         NFSVOPUNLOCK(fvp, 0);
 1925         if (!error && tvp)
 1926                 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
 1927         if (error)
 1928                 goto out;
 1929 
 1930         /*
 1931          * If the tvp exists and is in use, sillyrename it before doing the
 1932          * rename of the new file over it.
 1933          * XXX Can't sillyrename a directory.
 1934          */
 1935         if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
 1936                 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
 1937                 vput(tvp);
 1938                 tvp = NULL;
 1939         }
 1940 
 1941         error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
 1942             tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
 1943             tcnp->cn_thread);
 1944 
 1945         if (error == 0 && NFS_ISV4(tdvp)) {
 1946                 /*
 1947                  * For NFSv4, check to see if it is the same name and
 1948                  * replace the name, if it is different.
 1949                  */
 1950                 newv4 = malloc(
 1951                     sizeof (struct nfsv4node) +
 1952                     tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
 1953                     M_NFSV4NODE, M_WAITOK);
 1954                 NFSLOCKNODE(tdnp);
 1955                 NFSLOCKNODE(fnp);
 1956                 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
 1957                     (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
 1958                       NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
 1959                       tcnp->cn_namelen) ||
 1960                       tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
 1961                       NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
 1962                         tdnp->n_fhp->nfh_len))) {
 1963 #ifdef notdef
 1964 { char nnn[100]; int nnnl;
 1965 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
 1966 bcopy(tcnp->cn_nameptr, nnn, nnnl);
 1967 nnn[nnnl] = '\0';
 1968 printf("ren replace=%s\n",nnn);
 1969 }
 1970 #endif
 1971                         free(fnp->n_v4, M_NFSV4NODE);
 1972                         fnp->n_v4 = newv4;
 1973                         newv4 = NULL;
 1974                         fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
 1975                         fnp->n_v4->n4_namelen = tcnp->cn_namelen;
 1976                         NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
 1977                             tdnp->n_fhp->nfh_len);
 1978                         NFSBCOPY(tcnp->cn_nameptr,
 1979                             NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
 1980                 }
 1981                 NFSUNLOCKNODE(tdnp);
 1982                 NFSUNLOCKNODE(fnp);
 1983                 if (newv4 != NULL)
 1984                         free(newv4, M_NFSV4NODE);
 1985         }
 1986 
 1987         if (fvp->v_type == VDIR) {
 1988                 if (tvp != NULL && tvp->v_type == VDIR)
 1989                         cache_purge(tdvp);
 1990                 cache_purge(fdvp);
 1991         }
 1992 
 1993 out:
 1994         if (tdvp == tvp)
 1995                 vrele(tdvp);
 1996         else
 1997                 vput(tdvp);
 1998         if (tvp)
 1999                 vput(tvp);
 2000         vrele(fdvp);
 2001         vrele(fvp);
 2002         /*
 2003          * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
 2004          */
 2005         if (error == ENOENT)
 2006                 error = 0;
 2007         return (error);
 2008 }
 2009 
 2010 /*
 2011  * nfs file rename rpc called from nfs_remove() above
 2012  */
 2013 static int
 2014 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
 2015     struct sillyrename *sp)
 2016 {
 2017 
 2018         return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
 2019             sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
 2020             scnp->cn_thread));
 2021 }
 2022 
 2023 /*
 2024  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
 2025  */
 2026 static int
 2027 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
 2028     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
 2029     int tnamelen, struct ucred *cred, struct thread *td)
 2030 {
 2031         struct nfsvattr fnfsva, tnfsva;
 2032         struct nfsnode *fdnp = VTONFS(fdvp);
 2033         struct nfsnode *tdnp = VTONFS(tdvp);
 2034         int error = 0, fattrflag, tattrflag;
 2035 
 2036         error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
 2037             tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
 2038             &tattrflag, NULL, NULL);
 2039         NFSLOCKNODE(fdnp);
 2040         fdnp->n_flag |= NMODIFIED;
 2041         if (fattrflag != 0) {
 2042                 NFSUNLOCKNODE(fdnp);
 2043                 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
 2044         } else {
 2045                 fdnp->n_attrstamp = 0;
 2046                 NFSUNLOCKNODE(fdnp);
 2047                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
 2048         }
 2049         NFSLOCKNODE(tdnp);
 2050         tdnp->n_flag |= NMODIFIED;
 2051         if (tattrflag != 0) {
 2052                 NFSUNLOCKNODE(tdnp);
 2053                 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
 2054         } else {
 2055                 tdnp->n_attrstamp = 0;
 2056                 NFSUNLOCKNODE(tdnp);
 2057                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
 2058         }
 2059         if (error && NFS_ISV4(fdvp))
 2060                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
 2061         return (error);
 2062 }
 2063 
 2064 /*
 2065  * nfs hard link create call
 2066  */
 2067 static int
 2068 nfs_link(struct vop_link_args *ap)
 2069 {
 2070         struct vnode *vp = ap->a_vp;
 2071         struct vnode *tdvp = ap->a_tdvp;
 2072         struct componentname *cnp = ap->a_cnp;
 2073         struct nfsnode *np, *tdnp;
 2074         struct nfsvattr nfsva, dnfsva;
 2075         int error = 0, attrflag, dattrflag;
 2076 
 2077         /*
 2078          * Push all writes to the server, so that the attribute cache
 2079          * doesn't get "out of sync" with the server.
 2080          * XXX There should be a better way!
 2081          */
 2082         VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
 2083 
 2084         error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
 2085             cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
 2086             &dattrflag, NULL);
 2087         tdnp = VTONFS(tdvp);
 2088         NFSLOCKNODE(tdnp);
 2089         tdnp->n_flag |= NMODIFIED;
 2090         if (dattrflag != 0) {
 2091                 NFSUNLOCKNODE(tdnp);
 2092                 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
 2093         } else {
 2094                 tdnp->n_attrstamp = 0;
 2095                 NFSUNLOCKNODE(tdnp);
 2096                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
 2097         }
 2098         if (attrflag)
 2099                 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
 2100         else {
 2101                 np = VTONFS(vp);
 2102                 NFSLOCKNODE(np);
 2103                 np->n_attrstamp = 0;
 2104                 NFSUNLOCKNODE(np);
 2105                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
 2106         }
 2107         /*
 2108          * If negative lookup caching is enabled, I might as well
 2109          * add an entry for this node. Not necessary for correctness,
 2110          * but if negative caching is enabled, then the system
 2111          * must care about lookup caching hit rate, so...
 2112          */
 2113         if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
 2114             (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
 2115                 if (tdvp != vp)
 2116                         cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
 2117                 else
 2118                         printf("nfs_link: bogus NFS server returned "
 2119                             "the directory as the new link\n");
 2120         }
 2121         if (error && NFS_ISV4(vp))
 2122                 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
 2123                     (gid_t)0);
 2124         return (error);
 2125 }
 2126 
 2127 /*
 2128  * nfs symbolic link create call
 2129  */
 2130 static int
 2131 nfs_symlink(struct vop_symlink_args *ap)
 2132 {
 2133         struct vnode *dvp = ap->a_dvp;
 2134         struct vattr *vap = ap->a_vap;
 2135         struct componentname *cnp = ap->a_cnp;
 2136         struct nfsvattr nfsva, dnfsva;
 2137         struct nfsfh *nfhp;
 2138         struct nfsnode *np = NULL, *dnp;
 2139         struct vnode *newvp = NULL;
 2140         int error = 0, attrflag, dattrflag, ret;
 2141 
 2142         vap->va_type = VLNK;
 2143         error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 2144             ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
 2145             &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
 2146         if (nfhp) {
 2147                 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
 2148                     &np, NULL, LK_EXCLUSIVE);
 2149                 if (!ret)
 2150                         newvp = NFSTOV(np);
 2151                 else if (!error)
 2152                         error = ret;
 2153         }
 2154         if (newvp != NULL) {
 2155                 if (attrflag)
 2156                         (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 2157                             0, 1);
 2158         } else if (!error) {
 2159                 /*
 2160                  * If we do not have an error and we could not extract the
 2161                  * newvp from the response due to the request being NFSv2, we
 2162                  * have to do a lookup in order to obtain a newvp to return.
 2163                  */
 2164                 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 2165                     cnp->cn_cred, cnp->cn_thread, &np);
 2166                 if (!error)
 2167                         newvp = NFSTOV(np);
 2168         }
 2169         if (error) {
 2170                 if (newvp)
 2171                         vput(newvp);
 2172                 if (NFS_ISV4(dvp))
 2173                         error = nfscl_maperr(cnp->cn_thread, error,
 2174                             vap->va_uid, vap->va_gid);
 2175         } else {
 2176                 *ap->a_vpp = newvp;
 2177         }
 2178 
 2179         dnp = VTONFS(dvp);
 2180         NFSLOCKNODE(dnp);
 2181         dnp->n_flag |= NMODIFIED;
 2182         if (dattrflag != 0) {
 2183                 NFSUNLOCKNODE(dnp);
 2184                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 2185         } else {
 2186                 dnp->n_attrstamp = 0;
 2187                 NFSUNLOCKNODE(dnp);
 2188                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
 2189         }
 2190         /*
 2191          * If negative lookup caching is enabled, I might as well
 2192          * add an entry for this node. Not necessary for correctness,
 2193          * but if negative caching is enabled, then the system
 2194          * must care about lookup caching hit rate, so...
 2195          */
 2196         if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
 2197             (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
 2198                 if (dvp != newvp)
 2199                         cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
 2200                             NULL);
 2201                 else
 2202                         printf("nfs_symlink: bogus NFS server returned "
 2203                             "the directory as the new file object\n");
 2204         }
 2205         return (error);
 2206 }
 2207 
 2208 /*
 2209  * nfs make dir call
 2210  */
 2211 static int
 2212 nfs_mkdir(struct vop_mkdir_args *ap)
 2213 {
 2214         struct vnode *dvp = ap->a_dvp;
 2215         struct vattr *vap = ap->a_vap;
 2216         struct componentname *cnp = ap->a_cnp;
 2217         struct nfsnode *np = NULL, *dnp;
 2218         struct vnode *newvp = NULL;
 2219         struct vattr vattr;
 2220         struct nfsfh *nfhp;
 2221         struct nfsvattr nfsva, dnfsva;
 2222         int error = 0, attrflag, dattrflag, ret;
 2223 
 2224         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
 2225                 return (error);
 2226         vap->va_type = VDIR;
 2227         error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 2228             vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
 2229             &attrflag, &dattrflag, NULL);
 2230         dnp = VTONFS(dvp);
 2231         NFSLOCKNODE(dnp);
 2232         dnp->n_flag |= NMODIFIED;
 2233         if (dattrflag != 0) {
 2234                 NFSUNLOCKNODE(dnp);
 2235                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 2236         } else {
 2237                 dnp->n_attrstamp = 0;
 2238                 NFSUNLOCKNODE(dnp);
 2239                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
 2240         }
 2241         if (nfhp) {
 2242                 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
 2243                     &np, NULL, LK_EXCLUSIVE);
 2244                 if (!ret) {
 2245                         newvp = NFSTOV(np);
 2246                         if (attrflag)
 2247                            (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
 2248                                 NULL, 0, 1);
 2249                 } else if (!error)
 2250                         error = ret;
 2251         }
 2252         if (!error && newvp == NULL) {
 2253                 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 2254                     cnp->cn_cred, cnp->cn_thread, &np);
 2255                 if (!error) {
 2256                         newvp = NFSTOV(np);
 2257                         if (newvp->v_type != VDIR)
 2258                                 error = EEXIST;
 2259                 }
 2260         }
 2261         if (error) {
 2262                 if (newvp)
 2263                         vput(newvp);
 2264                 if (NFS_ISV4(dvp))
 2265                         error = nfscl_maperr(cnp->cn_thread, error,
 2266                             vap->va_uid, vap->va_gid);
 2267         } else {
 2268                 /*
 2269                  * If negative lookup caching is enabled, I might as well
 2270                  * add an entry for this node. Not necessary for correctness,
 2271                  * but if negative caching is enabled, then the system
 2272                  * must care about lookup caching hit rate, so...
 2273                  */
 2274                 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
 2275                     (cnp->cn_flags & MAKEENTRY) &&
 2276                     attrflag != 0 && dattrflag != 0) {
 2277                         if (dvp != newvp)
 2278                                 cache_enter_time(dvp, newvp, cnp,
 2279                                     &nfsva.na_ctime, &dnfsva.na_ctime);
 2280                         else
 2281                                 printf("nfs_mkdir: bogus NFS server returned "
 2282                                     "the directory that the directory was "
 2283                                     "created in as the new file object\n");
 2284                 }
 2285                 *ap->a_vpp = newvp;
 2286         }
 2287         return (error);
 2288 }
 2289 
 2290 /*
 2291  * nfs remove directory call
 2292  */
 2293 static int
 2294 nfs_rmdir(struct vop_rmdir_args *ap)
 2295 {
 2296         struct vnode *vp = ap->a_vp;
 2297         struct vnode *dvp = ap->a_dvp;
 2298         struct componentname *cnp = ap->a_cnp;
 2299         struct nfsnode *dnp;
 2300         struct nfsvattr dnfsva;
 2301         int error, dattrflag;
 2302 
 2303         if (dvp == vp)
 2304                 return (EINVAL);
 2305         error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 2306             cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
 2307         dnp = VTONFS(dvp);
 2308         NFSLOCKNODE(dnp);
 2309         dnp->n_flag |= NMODIFIED;
 2310         if (dattrflag != 0) {
 2311                 NFSUNLOCKNODE(dnp);
 2312                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 2313         } else {
 2314                 dnp->n_attrstamp = 0;
 2315                 NFSUNLOCKNODE(dnp);
 2316                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
 2317         }
 2318 
 2319         cache_purge(dvp);
 2320         cache_purge(vp);
 2321         if (error && NFS_ISV4(dvp))
 2322                 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
 2323                     (gid_t)0);
 2324         /*
 2325          * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
 2326          */
 2327         if (error == ENOENT)
 2328                 error = 0;
 2329         return (error);
 2330 }
 2331 
 2332 /*
 2333  * nfs readdir call
 2334  */
 2335 static int
 2336 nfs_readdir(struct vop_readdir_args *ap)
 2337 {
 2338         struct vnode *vp = ap->a_vp;
 2339         struct nfsnode *np = VTONFS(vp);
 2340         struct uio *uio = ap->a_uio;
 2341         ssize_t tresid, left;
 2342         int error = 0;
 2343         struct vattr vattr;
 2344         
 2345         if (ap->a_eofflag != NULL)
 2346                 *ap->a_eofflag = 0;
 2347         if (vp->v_type != VDIR) 
 2348                 return(EPERM);
 2349 
 2350         /*
 2351          * First, check for hit on the EOF offset cache
 2352          */
 2353         NFSLOCKNODE(np);
 2354         if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
 2355             (np->n_flag & NMODIFIED) == 0) {
 2356                 NFSUNLOCKNODE(np);
 2357                 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
 2358                         NFSLOCKNODE(np);
 2359                         if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
 2360                             !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
 2361                                 NFSUNLOCKNODE(np);
 2362                                 NFSINCRGLOBAL(nfsstatsv1.direofcache_hits);
 2363                                 if (ap->a_eofflag != NULL)
 2364                                         *ap->a_eofflag = 1;
 2365                                 return (0);
 2366                         } else
 2367                                 NFSUNLOCKNODE(np);
 2368                 }
 2369         } else
 2370                 NFSUNLOCKNODE(np);
 2371 
 2372         /*
 2373          * NFS always guarantees that directory entries don't straddle
 2374          * DIRBLKSIZ boundaries.  As such, we need to limit the size
 2375          * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
 2376          * directory entry.
 2377          */
 2378         left = uio->uio_resid % DIRBLKSIZ;
 2379         if (left == uio->uio_resid)
 2380                 return (EINVAL);
 2381         uio->uio_resid -= left;
 2382 
 2383         /*
 2384          * Call ncl_bioread() to do the real work.
 2385          */
 2386         tresid = uio->uio_resid;
 2387         error = ncl_bioread(vp, uio, 0, ap->a_cred);
 2388 
 2389         if (!error && uio->uio_resid == tresid) {
 2390                 NFSINCRGLOBAL(nfsstatsv1.direofcache_misses);
 2391                 if (ap->a_eofflag != NULL)
 2392                         *ap->a_eofflag = 1;
 2393         }
 2394         
 2395         /* Add the partial DIRBLKSIZ (left) back in. */
 2396         uio->uio_resid += left;
 2397         return (error);
 2398 }
 2399 
 2400 /*
 2401  * Readdir rpc call.
 2402  * Called from below the buffer cache by ncl_doio().
 2403  */
 2404 int
 2405 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
 2406     struct thread *td)
 2407 {
 2408         struct nfsvattr nfsva;
 2409         nfsuint64 *cookiep, cookie;
 2410         struct nfsnode *dnp = VTONFS(vp);
 2411         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2412         int error = 0, eof, attrflag;
 2413 
 2414         KASSERT(uiop->uio_iovcnt == 1 &&
 2415             (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
 2416             (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
 2417             ("nfs readdirrpc bad uio"));
 2418 
 2419         /*
 2420          * If there is no cookie, assume directory was stale.
 2421          */
 2422         ncl_dircookie_lock(dnp);
 2423         NFSUNLOCKNODE(dnp);
 2424         cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
 2425         if (cookiep) {
 2426                 cookie = *cookiep;
 2427                 ncl_dircookie_unlock(dnp);
 2428         } else {
 2429                 ncl_dircookie_unlock(dnp);              
 2430                 return (NFSERR_BAD_COOKIE);
 2431         }
 2432 
 2433         if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
 2434                 (void)ncl_fsinfo(nmp, vp, cred, td);
 2435 
 2436         error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
 2437             &attrflag, &eof, NULL);
 2438         if (attrflag)
 2439                 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
 2440 
 2441         if (!error) {
 2442                 /*
 2443                  * We are now either at the end of the directory or have filled
 2444                  * the block.
 2445                  */
 2446                 if (eof) {
 2447                         NFSLOCKNODE(dnp);
 2448                         dnp->n_direofoffset = uiop->uio_offset;
 2449                         NFSUNLOCKNODE(dnp);
 2450                 } else {
 2451                         if (uiop->uio_resid > 0)
 2452                                 printf("EEK! readdirrpc resid > 0\n");
 2453                         ncl_dircookie_lock(dnp);
 2454                         NFSUNLOCKNODE(dnp);
 2455                         cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
 2456                         *cookiep = cookie;
 2457                         ncl_dircookie_unlock(dnp);
 2458                 }
 2459         } else if (NFS_ISV4(vp)) {
 2460                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
 2461         }
 2462         return (error);
 2463 }
 2464 
 2465 /*
 2466  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
 2467  */
 2468 int
 2469 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
 2470     struct thread *td)
 2471 {
 2472         struct nfsvattr nfsva;
 2473         nfsuint64 *cookiep, cookie;
 2474         struct nfsnode *dnp = VTONFS(vp);
 2475         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2476         int error = 0, attrflag, eof;
 2477 
 2478         KASSERT(uiop->uio_iovcnt == 1 &&
 2479             (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
 2480             (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
 2481             ("nfs readdirplusrpc bad uio"));
 2482 
 2483         /*
 2484          * If there is no cookie, assume directory was stale.
 2485          */
 2486         ncl_dircookie_lock(dnp);
 2487         NFSUNLOCKNODE(dnp);
 2488         cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
 2489         if (cookiep) {
 2490                 cookie = *cookiep;
 2491                 ncl_dircookie_unlock(dnp);
 2492         } else {
 2493                 ncl_dircookie_unlock(dnp);
 2494                 return (NFSERR_BAD_COOKIE);
 2495         }
 2496 
 2497         if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
 2498                 (void)ncl_fsinfo(nmp, vp, cred, td);
 2499         error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
 2500             &attrflag, &eof, NULL);
 2501         if (attrflag)
 2502                 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
 2503 
 2504         if (!error) {
 2505                 /*
 2506                  * We are now either at end of the directory or have filled the
 2507                  * the block.
 2508                  */
 2509                 if (eof) {
 2510                         NFSLOCKNODE(dnp);
 2511                         dnp->n_direofoffset = uiop->uio_offset;
 2512                         NFSUNLOCKNODE(dnp);
 2513                 } else {
 2514                         if (uiop->uio_resid > 0)
 2515                                 printf("EEK! readdirplusrpc resid > 0\n");
 2516                         ncl_dircookie_lock(dnp);
 2517                         NFSUNLOCKNODE(dnp);
 2518                         cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
 2519                         *cookiep = cookie;
 2520                         ncl_dircookie_unlock(dnp);
 2521                 }
 2522         } else if (NFS_ISV4(vp)) {
 2523                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
 2524         }
 2525         return (error);
 2526 }
 2527 
 2528 /*
 2529  * Silly rename. To make the NFS filesystem that is stateless look a little
 2530  * more like the "ufs" a remove of an active vnode is translated to a rename
 2531  * to a funny looking filename that is removed by nfs_inactive on the
 2532  * nfsnode. There is the potential for another process on a different client
 2533  * to create the same funny name between the nfs_lookitup() fails and the
 2534  * nfs_rename() completes, but...
 2535  */
 2536 static int
 2537 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
 2538 {
 2539         struct sillyrename *sp;
 2540         struct nfsnode *np;
 2541         int error;
 2542         short pid;
 2543         unsigned int lticks;
 2544 
 2545         cache_purge(dvp);
 2546         np = VTONFS(vp);
 2547         KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
 2548         sp = malloc(sizeof (struct sillyrename),
 2549             M_NEWNFSREQ, M_WAITOK);
 2550         sp->s_cred = crhold(cnp->cn_cred);
 2551         sp->s_dvp = dvp;
 2552         VREF(dvp);
 2553 
 2554         /* 
 2555          * Fudge together a funny name.
 2556          * Changing the format of the funny name to accommodate more 
 2557          * sillynames per directory.
 2558          * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 
 2559          * CPU ticks since boot.
 2560          */
 2561         pid = cnp->cn_thread->td_proc->p_pid;
 2562         lticks = (unsigned int)ticks;
 2563         for ( ; ; ) {
 2564                 sp->s_namlen = sprintf(sp->s_name, 
 2565                                        ".nfs.%08x.%04x4.4", lticks, 
 2566                                        pid);
 2567                 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2568                                  cnp->cn_thread, NULL))
 2569                         break;
 2570                 lticks++;
 2571         }
 2572         error = nfs_renameit(dvp, vp, cnp, sp);
 2573         if (error)
 2574                 goto bad;
 2575         error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2576                 cnp->cn_thread, &np);
 2577         np->n_sillyrename = sp;
 2578         return (0);
 2579 bad:
 2580         vrele(sp->s_dvp);
 2581         crfree(sp->s_cred);
 2582         free(sp, M_NEWNFSREQ);
 2583         return (error);
 2584 }
 2585 
 2586 /*
 2587  * Look up a file name and optionally either update the file handle or
 2588  * allocate an nfsnode, depending on the value of npp.
 2589  * npp == NULL  --> just do the lookup
 2590  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
 2591  *                      handled too
 2592  * *npp != NULL --> update the file handle in the vnode
 2593  */
 2594 static int
 2595 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
 2596     struct thread *td, struct nfsnode **npp)
 2597 {
 2598         struct vnode *newvp = NULL, *vp;
 2599         struct nfsnode *np, *dnp = VTONFS(dvp);
 2600         struct nfsfh *nfhp, *onfhp;
 2601         struct nfsvattr nfsva, dnfsva;
 2602         struct componentname cn;
 2603         int error = 0, attrflag, dattrflag;
 2604         u_int hash;
 2605         struct timespec ts;
 2606 
 2607         nanouptime(&ts);
 2608         error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
 2609             &nfhp, &attrflag, &dattrflag, NULL);
 2610         if (dattrflag)
 2611                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 2612         if (npp && !error) {
 2613                 if (*npp != NULL) {
 2614                     np = *npp;
 2615                     vp = NFSTOV(np);
 2616                     /*
 2617                      * For NFSv4, check to see if it is the same name and
 2618                      * replace the name, if it is different.
 2619                      */
 2620                     if (np->n_v4 != NULL && nfsva.na_type == VREG &&
 2621                         (np->n_v4->n4_namelen != len ||
 2622                          NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
 2623                          dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
 2624                          NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
 2625                          dnp->n_fhp->nfh_len))) {
 2626 #ifdef notdef
 2627 { char nnn[100]; int nnnl;
 2628 nnnl = (len < 100) ? len : 99;
 2629 bcopy(name, nnn, nnnl);
 2630 nnn[nnnl] = '\0';
 2631 printf("replace=%s\n",nnn);
 2632 }
 2633 #endif
 2634                             free(np->n_v4, M_NFSV4NODE);
 2635                             np->n_v4 = malloc(
 2636                                 sizeof (struct nfsv4node) +
 2637                                 dnp->n_fhp->nfh_len + len - 1,
 2638                                 M_NFSV4NODE, M_WAITOK);
 2639                             np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
 2640                             np->n_v4->n4_namelen = len;
 2641                             NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
 2642                                 dnp->n_fhp->nfh_len);
 2643                             NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
 2644                     }
 2645                     hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
 2646                         FNV1_32_INIT);
 2647                     onfhp = np->n_fhp;
 2648                     /*
 2649                      * Rehash node for new file handle.
 2650                      */
 2651                     vfs_hash_rehash(vp, hash);
 2652                     np->n_fhp = nfhp;
 2653                     if (onfhp != NULL)
 2654                         free(onfhp, M_NFSFH);
 2655                     newvp = NFSTOV(np);
 2656                 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
 2657                     free(nfhp, M_NFSFH);
 2658                     VREF(dvp);
 2659                     newvp = dvp;
 2660                 } else {
 2661                     cn.cn_nameptr = name;
 2662                     cn.cn_namelen = len;
 2663                     error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
 2664                         &np, NULL, LK_EXCLUSIVE);
 2665                     if (error)
 2666                         return (error);
 2667                     newvp = NFSTOV(np);
 2668                     /*
 2669                      * If n_localmodtime >= time before RPC, then
 2670                      * a file modification operation, such as
 2671                      * VOP_SETATTR() of size, has occurred while
 2672                      * the Lookup RPC and acquisition of the vnode
 2673                      * happened.  As such, the attributes might
 2674                      * be stale, with possibly an incorrect size.
 2675                      */
 2676                     NFSLOCKNODE(np);
 2677                     if (timespecisset(&np->n_localmodtime) &&
 2678                         timespeccmp(&np->n_localmodtime, &ts, >=)) {
 2679                         NFSCL_DEBUG(4, "nfs_lookitup: localmod "
 2680                             "stale attributes\n");
 2681                         attrflag = 0;
 2682                     }
 2683                     NFSUNLOCKNODE(np);
 2684                 }
 2685                 if (!attrflag && *npp == NULL) {
 2686                         if (newvp == dvp)
 2687                                 vrele(newvp);
 2688                         else
 2689                                 vput(newvp);
 2690                         return (ENOENT);
 2691                 }
 2692                 if (attrflag)
 2693                         (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 2694                             0, 1);
 2695         }
 2696         if (npp && *npp == NULL) {
 2697                 if (error) {
 2698                         if (newvp) {
 2699                                 if (newvp == dvp)
 2700                                         vrele(newvp);
 2701                                 else
 2702                                         vput(newvp);
 2703                         }
 2704                 } else
 2705                         *npp = np;
 2706         }
 2707         if (error && NFS_ISV4(dvp))
 2708                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
 2709         return (error);
 2710 }
 2711 
 2712 /*
 2713  * Nfs Version 3 and 4 commit rpc
 2714  */
 2715 int
 2716 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
 2717    struct thread *td)
 2718 {
 2719         struct nfsvattr nfsva;
 2720         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2721         struct nfsnode *np;
 2722         struct uio uio;
 2723         int error, attrflag;
 2724 
 2725         np = VTONFS(vp);
 2726         error = EIO;
 2727         attrflag = 0;
 2728         if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) {
 2729                 uio.uio_offset = offset;
 2730                 uio.uio_resid = cnt;
 2731                 error = nfscl_doiods(vp, &uio, NULL, NULL,
 2732                     NFSV4OPEN_ACCESSWRITE, 1, cred, td);
 2733                 if (error != 0) {
 2734                         NFSLOCKNODE(np);
 2735                         np->n_flag &= ~NDSCOMMIT;
 2736                         NFSUNLOCKNODE(np);
 2737                 }
 2738         }
 2739         if (error != 0) {
 2740                 mtx_lock(&nmp->nm_mtx);
 2741                 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
 2742                         mtx_unlock(&nmp->nm_mtx);
 2743                         return (0);
 2744                 }
 2745                 mtx_unlock(&nmp->nm_mtx);
 2746                 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
 2747                     &attrflag, NULL);
 2748         }
 2749         if (attrflag != 0)
 2750                 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
 2751                     0, 1);
 2752         if (error != 0 && NFS_ISV4(vp))
 2753                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
 2754         return (error);
 2755 }
 2756 
 2757 /*
 2758  * Strategy routine.
 2759  * For async requests when nfsiod(s) are running, queue the request by
 2760  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
 2761  * request.
 2762  */
 2763 static int
 2764 nfs_strategy(struct vop_strategy_args *ap)
 2765 {
 2766         struct buf *bp;
 2767         struct vnode *vp;
 2768         struct ucred *cr;
 2769 
 2770         bp = ap->a_bp;
 2771         vp = ap->a_vp;
 2772         KASSERT(bp->b_vp == vp, ("missing b_getvp"));
 2773         KASSERT(!(bp->b_flags & B_DONE),
 2774             ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
 2775         BUF_ASSERT_HELD(bp);
 2776 
 2777         if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno)
 2778                 bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize /
 2779                     DEV_BSIZE);
 2780         if (bp->b_iocmd == BIO_READ)
 2781                 cr = bp->b_rcred;
 2782         else
 2783                 cr = bp->b_wcred;
 2784 
 2785         /*
 2786          * If the op is asynchronous and an i/o daemon is waiting
 2787          * queue the request, wake it up and wait for completion
 2788          * otherwise just do it ourselves.
 2789          */
 2790         if ((bp->b_flags & B_ASYNC) == 0 ||
 2791             ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread))
 2792                 (void) ncl_doio(vp, bp, cr, curthread, 1);
 2793         return (0);
 2794 }
 2795 
 2796 /*
 2797  * fsync vnode op. Just call ncl_flush() with commit == 1.
 2798  */
 2799 /* ARGSUSED */
 2800 static int
 2801 nfs_fsync(struct vop_fsync_args *ap)
 2802 {
 2803 
 2804         if (ap->a_vp->v_type != VREG) {
 2805                 /*
 2806                  * For NFS, metadata is changed synchronously on the server,
 2807                  * so there is nothing to flush. Also, ncl_flush() clears
 2808                  * the NMODIFIED flag and that shouldn't be done here for
 2809                  * directories.
 2810                  */
 2811                 return (0);
 2812         }
 2813         return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0));
 2814 }
 2815 
 2816 /*
 2817  * Flush all the blocks associated with a vnode.
 2818  *      Walk through the buffer pool and push any dirty pages
 2819  *      associated with the vnode.
 2820  * If the called_from_renewthread argument is TRUE, it has been called
 2821  * from the NFSv4 renew thread and, as such, cannot block indefinitely
 2822  * waiting for a buffer write to complete.
 2823  */
 2824 int
 2825 ncl_flush(struct vnode *vp, int waitfor, struct thread *td,
 2826     int commit, int called_from_renewthread)
 2827 {
 2828         struct nfsnode *np = VTONFS(vp);
 2829         struct buf *bp;
 2830         int i;
 2831         struct buf *nbp;
 2832         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2833         int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
 2834         int passone = 1, trycnt = 0;
 2835         u_quad_t off, endoff, toff;
 2836         struct ucred* wcred = NULL;
 2837         struct buf **bvec = NULL;
 2838         struct bufobj *bo;
 2839 #ifndef NFS_COMMITBVECSIZ
 2840 #define NFS_COMMITBVECSIZ       20
 2841 #endif
 2842         struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
 2843         u_int bvecsize = 0, bveccount;
 2844         struct timespec ts;
 2845 
 2846         if (called_from_renewthread != 0)
 2847                 slptimeo = hz;
 2848         if (nmp->nm_flag & NFSMNT_INT)
 2849                 slpflag = PCATCH;
 2850         if (!commit)
 2851                 passone = 0;
 2852         bo = &vp->v_bufobj;
 2853         /*
 2854          * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
 2855          * server, but has not been committed to stable storage on the server
 2856          * yet. On the first pass, the byte range is worked out and the commit
 2857          * rpc is done. On the second pass, ncl_writebp() is called to do the
 2858          * job.
 2859          */
 2860 again:
 2861         off = (u_quad_t)-1;
 2862         endoff = 0;
 2863         bvecpos = 0;
 2864         if (NFS_ISV34(vp) && commit) {
 2865                 if (bvec != NULL && bvec != bvec_on_stack)
 2866                         free(bvec, M_TEMP);
 2867                 /*
 2868                  * Count up how many buffers waiting for a commit.
 2869                  */
 2870                 bveccount = 0;
 2871                 BO_LOCK(bo);
 2872                 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 2873                         if (!BUF_ISLOCKED(bp) &&
 2874                             (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
 2875                                 == (B_DELWRI | B_NEEDCOMMIT))
 2876                                 bveccount++;
 2877                 }
 2878                 /*
 2879                  * Allocate space to remember the list of bufs to commit.  It is
 2880                  * important to use M_NOWAIT here to avoid a race with nfs_write.
 2881                  * If we can't get memory (for whatever reason), we will end up
 2882                  * committing the buffers one-by-one in the loop below.
 2883                  */
 2884                 if (bveccount > NFS_COMMITBVECSIZ) {
 2885                         /*
 2886                          * Release the vnode interlock to avoid a lock
 2887                          * order reversal.
 2888                          */
 2889                         BO_UNLOCK(bo);
 2890                         bvec = (struct buf **)
 2891                                 malloc(bveccount * sizeof(struct buf *),
 2892                                        M_TEMP, M_NOWAIT);
 2893                         BO_LOCK(bo);
 2894                         if (bvec == NULL) {
 2895                                 bvec = bvec_on_stack;
 2896                                 bvecsize = NFS_COMMITBVECSIZ;
 2897                         } else
 2898                                 bvecsize = bveccount;
 2899                 } else {
 2900                         bvec = bvec_on_stack;
 2901                         bvecsize = NFS_COMMITBVECSIZ;
 2902                 }
 2903                 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 2904                         if (bvecpos >= bvecsize)
 2905                                 break;
 2906                         if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
 2907                                 nbp = TAILQ_NEXT(bp, b_bobufs);
 2908                                 continue;
 2909                         }
 2910                         if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
 2911                             (B_DELWRI | B_NEEDCOMMIT)) {
 2912                                 BUF_UNLOCK(bp);
 2913                                 nbp = TAILQ_NEXT(bp, b_bobufs);
 2914                                 continue;
 2915                         }
 2916                         BO_UNLOCK(bo);
 2917                         bremfree(bp);
 2918                         /*
 2919                          * Work out if all buffers are using the same cred
 2920                          * so we can deal with them all with one commit.
 2921                          *
 2922                          * NOTE: we are not clearing B_DONE here, so we have
 2923                          * to do it later on in this routine if we intend to
 2924                          * initiate I/O on the bp.
 2925                          *
 2926                          * Note: to avoid loopback deadlocks, we do not
 2927                          * assign b_runningbufspace.
 2928                          */
 2929                         if (wcred == NULL)
 2930                                 wcred = bp->b_wcred;
 2931                         else if (wcred != bp->b_wcred)
 2932                                 wcred = NOCRED;
 2933                         vfs_busy_pages(bp, 1);
 2934 
 2935                         BO_LOCK(bo);
 2936                         /*
 2937                          * bp is protected by being locked, but nbp is not
 2938                          * and vfs_busy_pages() may sleep.  We have to
 2939                          * recalculate nbp.
 2940                          */
 2941                         nbp = TAILQ_NEXT(bp, b_bobufs);
 2942 
 2943                         /*
 2944                          * A list of these buffers is kept so that the
 2945                          * second loop knows which buffers have actually
 2946                          * been committed. This is necessary, since there
 2947                          * may be a race between the commit rpc and new
 2948                          * uncommitted writes on the file.
 2949                          */
 2950                         bvec[bvecpos++] = bp;
 2951                         toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2952                                 bp->b_dirtyoff;
 2953                         if (toff < off)
 2954                                 off = toff;
 2955                         toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
 2956                         if (toff > endoff)
 2957                                 endoff = toff;
 2958                 }
 2959                 BO_UNLOCK(bo);
 2960         }
 2961         if (bvecpos > 0) {
 2962                 /*
 2963                  * Commit data on the server, as required.
 2964                  * If all bufs are using the same wcred, then use that with
 2965                  * one call for all of them, otherwise commit each one
 2966                  * separately.
 2967                  */
 2968                 if (wcred != NOCRED)
 2969                         retv = ncl_commit(vp, off, (int)(endoff - off),
 2970                                           wcred, td);
 2971                 else {
 2972                         retv = 0;
 2973                         for (i = 0; i < bvecpos; i++) {
 2974                                 off_t off, size;
 2975                                 bp = bvec[i];
 2976                                 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2977                                         bp->b_dirtyoff;
 2978                                 size = (u_quad_t)(bp->b_dirtyend
 2979                                                   - bp->b_dirtyoff);
 2980                                 retv = ncl_commit(vp, off, (int)size,
 2981                                                   bp->b_wcred, td);
 2982                                 if (retv) break;
 2983                         }
 2984                 }
 2985 
 2986                 if (retv == NFSERR_STALEWRITEVERF)
 2987                         ncl_clearcommit(vp->v_mount);
 2988 
 2989                 /*
 2990                  * Now, either mark the blocks I/O done or mark the
 2991                  * blocks dirty, depending on whether the commit
 2992                  * succeeded.
 2993                  */
 2994                 for (i = 0; i < bvecpos; i++) {
 2995                         bp = bvec[i];
 2996                         bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
 2997                         if (!NFSCL_FORCEDISM(vp->v_mount) && retv) {
 2998                                 /*
 2999                                  * Error, leave B_DELWRI intact
 3000                                  */
 3001                                 vfs_unbusy_pages(bp);
 3002                                 brelse(bp);
 3003                         } else {
 3004                                 /*
 3005                                  * Success, remove B_DELWRI ( bundirty() ).
 3006                                  *
 3007                                  * b_dirtyoff/b_dirtyend seem to be NFS
 3008                                  * specific.  We should probably move that
 3009                                  * into bundirty(). XXX
 3010                                  */
 3011                                 bufobj_wref(bo);
 3012                                 bp->b_flags |= B_ASYNC;
 3013                                 bundirty(bp);
 3014                                 bp->b_flags &= ~B_DONE;
 3015                                 bp->b_ioflags &= ~BIO_ERROR;
 3016                                 bp->b_dirtyoff = bp->b_dirtyend = 0;
 3017                                 bufdone(bp);
 3018                         }
 3019                 }
 3020         }
 3021 
 3022         /*
 3023          * Start/do any write(s) that are required.
 3024          */
 3025 loop:
 3026         BO_LOCK(bo);
 3027         TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 3028                 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
 3029                         if (waitfor != MNT_WAIT || passone)
 3030                                 continue;
 3031 
 3032                         error = BUF_TIMELOCK(bp,
 3033                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 3034                             BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
 3035                         if (error == 0) {
 3036                                 BUF_UNLOCK(bp);
 3037                                 goto loop;
 3038                         }
 3039                         if (error == ENOLCK) {
 3040                                 error = 0;
 3041                                 goto loop;
 3042                         }
 3043                         if (called_from_renewthread != 0) {
 3044                                 /*
 3045                                  * Return EIO so the flush will be retried
 3046                                  * later.
 3047                                  */
 3048                                 error = EIO;
 3049                                 goto done;
 3050                         }
 3051                         if (newnfs_sigintr(nmp, td)) {
 3052                                 error = EINTR;
 3053                                 goto done;
 3054                         }
 3055                         if (slpflag == PCATCH) {
 3056                                 slpflag = 0;
 3057                                 slptimeo = 2 * hz;
 3058                         }
 3059                         goto loop;
 3060                 }
 3061                 if ((bp->b_flags & B_DELWRI) == 0)
 3062                         panic("nfs_fsync: not dirty");
 3063                 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
 3064                         BUF_UNLOCK(bp);
 3065                         continue;
 3066                 }
 3067                 BO_UNLOCK(bo);
 3068                 bremfree(bp);
 3069                 if (passone || !commit)
 3070                     bp->b_flags |= B_ASYNC;
 3071                 else
 3072                     bp->b_flags |= B_ASYNC;
 3073                 bwrite(bp);
 3074                 if (newnfs_sigintr(nmp, td)) {
 3075                         error = EINTR;
 3076                         goto done;
 3077                 }
 3078                 goto loop;
 3079         }
 3080         if (passone) {
 3081                 passone = 0;
 3082                 BO_UNLOCK(bo);
 3083                 goto again;
 3084         }
 3085         if (waitfor == MNT_WAIT) {
 3086                 while (bo->bo_numoutput) {
 3087                         error = bufobj_wwait(bo, slpflag, slptimeo);
 3088                         if (error) {
 3089                             BO_UNLOCK(bo);
 3090                             if (called_from_renewthread != 0) {
 3091                                 /*
 3092                                  * Return EIO so that the flush will be
 3093                                  * retried later.
 3094                                  */
 3095                                 error = EIO;
 3096                                 goto done;
 3097                             }
 3098                             error = newnfs_sigintr(nmp, td);
 3099                             if (error)
 3100                                 goto done;
 3101                             if (slpflag == PCATCH) {
 3102                                 slpflag = 0;
 3103                                 slptimeo = 2 * hz;
 3104                             }
 3105                             BO_LOCK(bo);
 3106                         }
 3107                 }
 3108                 if (bo->bo_dirty.bv_cnt != 0 && commit) {
 3109                         BO_UNLOCK(bo);
 3110                         goto loop;
 3111                 }
 3112                 /*
 3113                  * Wait for all the async IO requests to drain
 3114                  */
 3115                 BO_UNLOCK(bo);
 3116                 NFSLOCKNODE(np);
 3117                 while (np->n_directio_asyncwr > 0) {
 3118                         np->n_flag |= NFSYNCWAIT;
 3119                         error = newnfs_msleep(td, &np->n_directio_asyncwr,
 3120                             &np->n_mtx, slpflag | (PRIBIO + 1), 
 3121                             "nfsfsync", 0);
 3122                         if (error) {
 3123                                 if (newnfs_sigintr(nmp, td)) {
 3124                                         NFSUNLOCKNODE(np);
 3125                                         error = EINTR;  
 3126                                         goto done;
 3127                                 }
 3128                         }
 3129                 }
 3130                 NFSUNLOCKNODE(np);
 3131         } else
 3132                 BO_UNLOCK(bo);
 3133         if (NFSHASPNFS(nmp)) {
 3134                 nfscl_layoutcommit(vp, td);
 3135                 /*
 3136                  * Invalidate the attribute cache, since writes to a DS
 3137                  * won't update the size attribute.
 3138                  */
 3139                 NFSLOCKNODE(np);
 3140                 np->n_attrstamp = 0;
 3141         } else
 3142                 NFSLOCKNODE(np);
 3143         if (np->n_flag & NWRITEERR) {
 3144                 error = np->n_error;
 3145                 np->n_flag &= ~NWRITEERR;
 3146         }
 3147         if (commit && bo->bo_dirty.bv_cnt == 0 &&
 3148             bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
 3149                 np->n_flag &= ~NMODIFIED;
 3150         NFSUNLOCKNODE(np);
 3151 done:
 3152         if (bvec != NULL && bvec != bvec_on_stack)
 3153                 free(bvec, M_TEMP);
 3154         if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
 3155             (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
 3156             np->n_directio_asyncwr != 0)) {
 3157                 if (trycnt++ < 5) {
 3158                         /* try, try again... */
 3159                         passone = 1;
 3160                         wcred = NULL;
 3161                         bvec = NULL;
 3162                         bvecsize = 0;
 3163                         goto again;
 3164                 }
 3165                 vn_printf(vp, "ncl_flush failed");
 3166                 error = called_from_renewthread != 0 ? EIO : EBUSY;
 3167         }
 3168         if (error == 0) {
 3169                 nanouptime(&ts);
 3170                 NFSLOCKNODE(np);
 3171                 np->n_localmodtime = ts;
 3172                 NFSUNLOCKNODE(np);
 3173         }
 3174         return (error);
 3175 }
 3176 
 3177 /*
 3178  * NFS advisory byte-level locks.
 3179  */
 3180 static int
 3181 nfs_advlock(struct vop_advlock_args *ap)
 3182 {
 3183         struct vnode *vp = ap->a_vp;
 3184         struct ucred *cred;
 3185         struct nfsnode *np = VTONFS(ap->a_vp);
 3186         struct proc *p = (struct proc *)ap->a_id;
 3187         struct thread *td = curthread;  /* XXX */
 3188         struct vattr va;
 3189         int ret, error = EOPNOTSUPP;
 3190         u_quad_t size;
 3191         
 3192         ret = NFSVOPLOCK(vp, LK_SHARED);
 3193         if (ret != 0)
 3194                 return (EBADF);
 3195         if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
 3196                 if (vp->v_type != VREG) {
 3197                         NFSVOPUNLOCK(vp, 0);
 3198                         return (EINVAL);
 3199                 }
 3200                 if ((ap->a_flags & F_POSIX) != 0)
 3201                         cred = p->p_ucred;
 3202                 else
 3203                         cred = td->td_ucred;
 3204                 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
 3205                 if (vp->v_iflag & VI_DOOMED) {
 3206                         NFSVOPUNLOCK(vp, 0);
 3207                         return (EBADF);
 3208                 }
 3209 
 3210                 /*
 3211                  * If this is unlocking a write locked region, flush and
 3212                  * commit them before unlocking. This is required by
 3213                  * RFC3530 Sec. 9.3.2.
 3214                  */
 3215                 if (ap->a_op == F_UNLCK &&
 3216                     nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
 3217                     ap->a_flags))
 3218                         (void) ncl_flush(vp, MNT_WAIT, td, 1, 0);
 3219 
 3220                 /*
 3221                  * Loop around doing the lock op, while a blocking lock
 3222                  * must wait for the lock op to succeed.
 3223                  */
 3224                 do {
 3225                         ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
 3226                             ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
 3227                         if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
 3228                             ap->a_op == F_SETLK) {
 3229                                 NFSVOPUNLOCK(vp, 0);
 3230                                 error = nfs_catnap(PZERO | PCATCH, ret,
 3231                                     "ncladvl");
 3232                                 if (error)
 3233                                         return (EINTR);
 3234                                 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
 3235                                 if (vp->v_iflag & VI_DOOMED) {
 3236                                         NFSVOPUNLOCK(vp, 0);
 3237                                         return (EBADF);
 3238                                 }
 3239                         }
 3240                 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
 3241                      ap->a_op == F_SETLK);
 3242                 if (ret == NFSERR_DENIED) {
 3243                         NFSVOPUNLOCK(vp, 0);
 3244                         return (EAGAIN);
 3245                 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
 3246                         NFSVOPUNLOCK(vp, 0);
 3247                         return (ret);
 3248                 } else if (ret != 0) {
 3249                         NFSVOPUNLOCK(vp, 0);
 3250                         return (EACCES);
 3251                 }
 3252 
 3253                 /*
 3254                  * Now, if we just got a lock, invalidate data in the buffer
 3255                  * cache, as required, so that the coherency conforms with
 3256                  * RFC3530 Sec. 9.3.2.
 3257                  */
 3258                 if (ap->a_op == F_SETLK) {
 3259                         if ((np->n_flag & NMODIFIED) == 0) {
 3260                                 np->n_attrstamp = 0;
 3261                                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
 3262                                 ret = VOP_GETATTR(vp, &va, cred);
 3263                         }
 3264                         if ((np->n_flag & NMODIFIED) || ret ||
 3265                             np->n_change != va.va_filerev) {
 3266                                 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
 3267                                 np->n_attrstamp = 0;
 3268                                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
 3269                                 ret = VOP_GETATTR(vp, &va, cred);
 3270                                 if (!ret) {
 3271                                         np->n_mtime = va.va_mtime;
 3272                                         np->n_change = va.va_filerev;
 3273                                 }
 3274                         }
 3275                         /* Mark that a file lock has been acquired. */
 3276                         NFSLOCKNODE(np);
 3277                         np->n_flag |= NHASBEENLOCKED;
 3278                         NFSUNLOCKNODE(np);
 3279                 }
 3280                 NFSVOPUNLOCK(vp, 0);
 3281                 return (0);
 3282         } else if (!NFS_ISV4(vp)) {
 3283                 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
 3284                         size = VTONFS(vp)->n_size;
 3285                         NFSVOPUNLOCK(vp, 0);
 3286                         error = lf_advlock(ap, &(vp->v_lockf), size);
 3287                 } else {
 3288                         if (nfs_advlock_p != NULL)
 3289                                 error = nfs_advlock_p(ap);
 3290                         else {
 3291                                 NFSVOPUNLOCK(vp, 0);
 3292                                 error = ENOLCK;
 3293                         }
 3294                 }
 3295                 if (error == 0 && ap->a_op == F_SETLK) {
 3296                         error = NFSVOPLOCK(vp, LK_SHARED);
 3297                         if (error == 0) {
 3298                                 /* Mark that a file lock has been acquired. */
 3299                                 NFSLOCKNODE(np);
 3300                                 np->n_flag |= NHASBEENLOCKED;
 3301                                 NFSUNLOCKNODE(np);
 3302                                 NFSVOPUNLOCK(vp, 0);
 3303                         }
 3304                 }
 3305         } else
 3306                 NFSVOPUNLOCK(vp, 0);
 3307         return (error);
 3308 }
 3309 
 3310 /*
 3311  * NFS advisory byte-level locks.
 3312  */
 3313 static int
 3314 nfs_advlockasync(struct vop_advlockasync_args *ap)
 3315 {
 3316         struct vnode *vp = ap->a_vp;
 3317         u_quad_t size;
 3318         int error;
 3319         
 3320         if (NFS_ISV4(vp))
 3321                 return (EOPNOTSUPP);
 3322         error = NFSVOPLOCK(vp, LK_SHARED);
 3323         if (error)
 3324                 return (error);
 3325         if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
 3326                 size = VTONFS(vp)->n_size;
 3327                 NFSVOPUNLOCK(vp, 0);
 3328                 error = lf_advlockasync(ap, &(vp->v_lockf), size);
 3329         } else {
 3330                 NFSVOPUNLOCK(vp, 0);
 3331                 error = EOPNOTSUPP;
 3332         }
 3333         return (error);
 3334 }
 3335 
 3336 /*
 3337  * Print out the contents of an nfsnode.
 3338  */
 3339 static int
 3340 nfs_print(struct vop_print_args *ap)
 3341 {
 3342         struct vnode *vp = ap->a_vp;
 3343         struct nfsnode *np = VTONFS(vp);
 3344 
 3345         printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid,
 3346             (uintmax_t)np->n_vattr.na_fsid);
 3347         if (vp->v_type == VFIFO)
 3348                 fifo_printinfo(vp);
 3349         printf("\n");
 3350         return (0);
 3351 }
 3352 
 3353 /*
 3354  * This is the "real" nfs::bwrite(struct buf*).
 3355  * We set B_CACHE if this is a VMIO buffer.
 3356  */
 3357 int
 3358 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
 3359 {
 3360         int oldflags, rtval;
 3361 
 3362         BUF_ASSERT_HELD(bp);
 3363 
 3364         if (bp->b_flags & B_INVAL) {
 3365                 brelse(bp);
 3366                 return (0);
 3367         }
 3368 
 3369         oldflags = bp->b_flags;
 3370         bp->b_flags |= B_CACHE;
 3371 
 3372         /*
 3373          * Undirty the bp.  We will redirty it later if the I/O fails.
 3374          */
 3375         bundirty(bp);
 3376         bp->b_flags &= ~B_DONE;
 3377         bp->b_ioflags &= ~BIO_ERROR;
 3378         bp->b_iocmd = BIO_WRITE;
 3379 
 3380         bufobj_wref(bp->b_bufobj);
 3381         curthread->td_ru.ru_oublock++;
 3382 
 3383         /*
 3384          * Note: to avoid loopback deadlocks, we do not
 3385          * assign b_runningbufspace.
 3386          */
 3387         vfs_busy_pages(bp, 1);
 3388 
 3389         BUF_KERNPROC(bp);
 3390         bp->b_iooffset = dbtob(bp->b_blkno);
 3391         bstrategy(bp);
 3392 
 3393         if ((oldflags & B_ASYNC) != 0)
 3394                 return (0);
 3395 
 3396         rtval = bufwait(bp);
 3397         if (oldflags & B_DELWRI)
 3398                 reassignbuf(bp);
 3399         brelse(bp);
 3400         return (rtval);
 3401 }
 3402 
 3403 /*
 3404  * nfs special file access vnode op.
 3405  * Essentially just get vattr and then imitate iaccess() since the device is
 3406  * local to the client.
 3407  */
 3408 static int
 3409 nfsspec_access(struct vop_access_args *ap)
 3410 {
 3411         struct vattr *vap;
 3412         struct ucred *cred = ap->a_cred;
 3413         struct vnode *vp = ap->a_vp;
 3414         accmode_t accmode = ap->a_accmode;
 3415         struct vattr vattr;
 3416         int error;
 3417 
 3418         /*
 3419          * Disallow write attempts on filesystems mounted read-only;
 3420          * unless the file is a socket, fifo, or a block or character
 3421          * device resident on the filesystem.
 3422          */
 3423         if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
 3424                 switch (vp->v_type) {
 3425                 case VREG:
 3426                 case VDIR:
 3427                 case VLNK:
 3428                         return (EROFS);
 3429                 default:
 3430                         break;
 3431                 }
 3432         }
 3433         vap = &vattr;
 3434         error = VOP_GETATTR(vp, vap, cred);
 3435         if (error)
 3436                 goto out;
 3437         error  = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
 3438             accmode, cred, NULL);
 3439 out:
 3440         return error;
 3441 }
 3442 
 3443 /*
 3444  * Read wrapper for fifos.
 3445  */
 3446 static int
 3447 nfsfifo_read(struct vop_read_args *ap)
 3448 {
 3449         struct nfsnode *np = VTONFS(ap->a_vp);
 3450         int error;
 3451 
 3452         /*
 3453          * Set access flag.
 3454          */
 3455         NFSLOCKNODE(np);
 3456         np->n_flag |= NACC;
 3457         vfs_timestamp(&np->n_atim);
 3458         NFSUNLOCKNODE(np);
 3459         error = fifo_specops.vop_read(ap);
 3460         return error;   
 3461 }
 3462 
 3463 /*
 3464  * Write wrapper for fifos.
 3465  */
 3466 static int
 3467 nfsfifo_write(struct vop_write_args *ap)
 3468 {
 3469         struct nfsnode *np = VTONFS(ap->a_vp);
 3470 
 3471         /*
 3472          * Set update flag.
 3473          */
 3474         NFSLOCKNODE(np);
 3475         np->n_flag |= NUPD;
 3476         vfs_timestamp(&np->n_mtim);
 3477         NFSUNLOCKNODE(np);
 3478         return(fifo_specops.vop_write(ap));
 3479 }
 3480 
 3481 /*
 3482  * Close wrapper for fifos.
 3483  *
 3484  * Update the times on the nfsnode then do fifo close.
 3485  */
 3486 static int
 3487 nfsfifo_close(struct vop_close_args *ap)
 3488 {
 3489         struct vnode *vp = ap->a_vp;
 3490         struct nfsnode *np = VTONFS(vp);
 3491         struct vattr vattr;
 3492         struct timespec ts;
 3493 
 3494         NFSLOCKNODE(np);
 3495         if (np->n_flag & (NACC | NUPD)) {
 3496                 vfs_timestamp(&ts);
 3497                 if (np->n_flag & NACC)
 3498                         np->n_atim = ts;
 3499                 if (np->n_flag & NUPD)
 3500                         np->n_mtim = ts;
 3501                 np->n_flag |= NCHG;
 3502                 if (vrefcnt(vp) == 1 &&
 3503                     (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
 3504                         VATTR_NULL(&vattr);
 3505                         if (np->n_flag & NACC)
 3506                                 vattr.va_atime = np->n_atim;
 3507                         if (np->n_flag & NUPD)
 3508                                 vattr.va_mtime = np->n_mtim;
 3509                         NFSUNLOCKNODE(np);
 3510                         (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
 3511                         goto out;
 3512                 }
 3513         }
 3514         NFSUNLOCKNODE(np);
 3515 out:
 3516         return (fifo_specops.vop_close(ap));
 3517 }
 3518 
 3519 /*
 3520  * Just call ncl_writebp() with the force argument set to 1.
 3521  *
 3522  * NOTE: B_DONE may or may not be set in a_bp on call.
 3523  */
 3524 static int
 3525 nfs_bwrite(struct buf *bp)
 3526 {
 3527 
 3528         return (ncl_writebp(bp, 1, curthread));
 3529 }
 3530 
 3531 struct buf_ops buf_ops_newnfs = {
 3532         .bop_name       =       "buf_ops_nfs",
 3533         .bop_write      =       nfs_bwrite,
 3534         .bop_strategy   =       bufstrategy,
 3535         .bop_sync       =       bufsync,
 3536         .bop_bdflush    =       bufbdflush,
 3537 };
 3538 
 3539 static int
 3540 nfs_getacl(struct vop_getacl_args *ap)
 3541 {
 3542         int error;
 3543 
 3544         if (ap->a_type != ACL_TYPE_NFS4)
 3545                 return (EOPNOTSUPP);
 3546         error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
 3547             NULL);
 3548         if (error > NFSERR_STALE) {
 3549                 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
 3550                 error = EPERM;
 3551         }
 3552         return (error);
 3553 }
 3554 
 3555 static int
 3556 nfs_setacl(struct vop_setacl_args *ap)
 3557 {
 3558         int error;
 3559 
 3560         if (ap->a_type != ACL_TYPE_NFS4)
 3561                 return (EOPNOTSUPP);
 3562         error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
 3563             NULL);
 3564         if (error > NFSERR_STALE) {
 3565                 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
 3566                 error = EPERM;
 3567         }
 3568         return (error);
 3569 }
 3570 
 3571 /*
 3572  * Return POSIX pathconf information applicable to nfs filesystems.
 3573  */
 3574 static int
 3575 nfs_pathconf(struct vop_pathconf_args *ap)
 3576 {
 3577         struct nfsv3_pathconf pc;
 3578         struct nfsvattr nfsva;
 3579         struct vnode *vp = ap->a_vp;
 3580         struct thread *td = curthread;
 3581         int attrflag, error;
 3582 
 3583         if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
 3584             ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
 3585             ap->a_name == _PC_NO_TRUNC)) ||
 3586             (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
 3587                 /*
 3588                  * Since only the above 4 a_names are returned by the NFSv3
 3589                  * Pathconf RPC, there is no point in doing it for others.
 3590                  * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
 3591                  * be used for _PC_NFS4_ACL as well.
 3592                  */
 3593                 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
 3594                     &attrflag, NULL);
 3595                 if (attrflag != 0)
 3596                         (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
 3597                             1);
 3598                 if (error != 0)
 3599                         return (error);
 3600         } else {
 3601                 /*
 3602                  * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
 3603                  * just fake them.
 3604                  */
 3605                 pc.pc_linkmax = NFS_LINK_MAX;
 3606                 pc.pc_namemax = NFS_MAXNAMLEN;
 3607                 pc.pc_notrunc = 1;
 3608                 pc.pc_chownrestricted = 1;
 3609                 pc.pc_caseinsensitive = 0;
 3610                 pc.pc_casepreserving = 1;
 3611                 error = 0;
 3612         }
 3613         switch (ap->a_name) {
 3614         case _PC_LINK_MAX:
 3615 #ifdef _LP64
 3616                 *ap->a_retval = pc.pc_linkmax;
 3617 #else
 3618                 *ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax);
 3619 #endif
 3620                 break;
 3621         case _PC_NAME_MAX:
 3622                 *ap->a_retval = pc.pc_namemax;
 3623                 break;
 3624         case _PC_PIPE_BUF:
 3625                 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
 3626                         *ap->a_retval = PIPE_BUF;
 3627                 else
 3628                         error = EINVAL;
 3629                 break;
 3630         case _PC_CHOWN_RESTRICTED:
 3631                 *ap->a_retval = pc.pc_chownrestricted;
 3632                 break;
 3633         case _PC_NO_TRUNC:
 3634                 *ap->a_retval = pc.pc_notrunc;
 3635                 break;
 3636         case _PC_ACL_NFS4:
 3637                 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
 3638                     NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
 3639                         *ap->a_retval = 1;
 3640                 else
 3641                         *ap->a_retval = 0;
 3642                 break;
 3643         case _PC_ACL_PATH_MAX:
 3644                 if (NFS_ISV4(vp))
 3645                         *ap->a_retval = ACL_MAX_ENTRIES;
 3646                 else
 3647                         *ap->a_retval = 3;
 3648                 break;
 3649         case _PC_PRIO_IO:
 3650                 *ap->a_retval = 0;
 3651                 break;
 3652         case _PC_SYNC_IO:
 3653                 *ap->a_retval = 0;
 3654                 break;
 3655         case _PC_ALLOC_SIZE_MIN:
 3656                 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
 3657                 break;
 3658         case _PC_FILESIZEBITS:
 3659                 if (NFS_ISV34(vp))
 3660                         *ap->a_retval = 64;
 3661                 else
 3662                         *ap->a_retval = 32;
 3663                 break;
 3664         case _PC_REC_INCR_XFER_SIZE:
 3665                 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
 3666                 break;
 3667         case _PC_REC_MAX_XFER_SIZE:
 3668                 *ap->a_retval = -1; /* means ``unlimited'' */
 3669                 break;
 3670         case _PC_REC_MIN_XFER_SIZE:
 3671                 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
 3672                 break;
 3673         case _PC_REC_XFER_ALIGN:
 3674                 *ap->a_retval = PAGE_SIZE;
 3675                 break;
 3676         case _PC_SYMLINK_MAX:
 3677                 *ap->a_retval = NFS_MAXPATHLEN;
 3678                 break;
 3679 
 3680         default:
 3681                 error = vop_stdpathconf(ap);
 3682                 break;
 3683         }
 3684         return (error);
 3685 }
 3686 

Cache object: 31ecd22b2ce0de45acd4a22225d7910d


[ 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.