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/nfsclient/nfs_vnops.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  * Copyright (c) 1989, 1993
    3  *      The Regents of the University of California.  All rights reserved.
    4  *
    5  * This code is derived from software contributed to Berkeley by
    6  * Rick Macklem at The University of Guelph.
    7  *
    8  * Redistribution and use in source and binary forms, with or without
    9  * modification, are permitted provided that the following conditions
   10  * are met:
   11  * 1. Redistributions of source code must retain the above copyright
   12  *    notice, this list of conditions and the following disclaimer.
   13  * 2. Redistributions in binary form must reproduce the above copyright
   14  *    notice, this list of conditions and the following disclaimer in the
   15  *    documentation and/or other materials provided with the distribution.
   16  * 4. Neither the name of the University nor the names of its contributors
   17  *    may be used to endorse or promote products derived from this software
   18  *    without specific prior written permission.
   19  *
   20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   30  * SUCH DAMAGE.
   31  *
   32  *      @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
   33  */
   34 
   35 #include <sys/cdefs.h>
   36 __FBSDID("$FreeBSD: releng/5.3/sys/nfsclient/nfs_vnops.c 132640 2004-07-25 21:24:23Z phk $");
   37 
   38 /*
   39  * vnode op calls for Sun NFS version 2 and 3
   40  */
   41 
   42 #include "opt_inet.h"
   43 
   44 #include <sys/param.h>
   45 #include <sys/kernel.h>
   46 #include <sys/systm.h>
   47 #include <sys/resourcevar.h>
   48 #include <sys/proc.h>
   49 #include <sys/mount.h>
   50 #include <sys/bio.h>
   51 #include <sys/buf.h>
   52 #include <sys/malloc.h>
   53 #include <sys/mbuf.h>
   54 #include <sys/namei.h>
   55 #include <sys/socket.h>
   56 #include <sys/vnode.h>
   57 #include <sys/dirent.h>
   58 #include <sys/fcntl.h>
   59 #include <sys/lockf.h>
   60 #include <sys/stat.h>
   61 #include <sys/sysctl.h>
   62 
   63 #include <vm/vm.h>
   64 #include <vm/vm_extern.h>
   65 
   66 #include <fs/fifofs/fifo.h>
   67 
   68 #include <rpc/rpcclnt.h>
   69 
   70 #include <nfs/rpcv2.h>
   71 #include <nfs/nfsproto.h>
   72 #include <nfsclient/nfs.h>
   73 #include <nfsclient/nfsnode.h>
   74 #include <nfsclient/nfsmount.h>
   75 #include <nfsclient/nfs_lock.h>
   76 #include <nfs/xdr_subs.h>
   77 #include <nfsclient/nfsm_subs.h>
   78 
   79 #include <net/if.h>
   80 #include <netinet/in.h>
   81 #include <netinet/in_var.h>
   82 
   83 /* Defs */
   84 #define TRUE    1
   85 #define FALSE   0
   86 
   87 /*
   88  * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
   89  * calls are not in getblk() and brelse() so that they would not be necessary
   90  * here.
   91  */
   92 #ifndef B_VMIO
   93 #define vfs_busy_pages(bp, f)
   94 #endif
   95 
   96 static int      nfsspec_read(struct vop_read_args *);
   97 static int      nfsspec_write(struct vop_write_args *);
   98 static int      nfsfifo_read(struct vop_read_args *);
   99 static int      nfsfifo_write(struct vop_write_args *);
  100 static int      nfsspec_close(struct vop_close_args *);
  101 static int      nfsfifo_close(struct vop_close_args *);
  102 static int      nfs_flush(struct vnode *, struct ucred *, int, struct thread *,
  103                     int);
  104 static int      nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
  105                     struct thread *);
  106 static  int     nfs_lookup(struct vop_lookup_args *);
  107 static  int     nfs_create(struct vop_create_args *);
  108 static  int     nfs_mknod(struct vop_mknod_args *);
  109 static  int     nfs_open(struct vop_open_args *);
  110 static  int     nfs_close(struct vop_close_args *);
  111 static  int     nfs_access(struct vop_access_args *);
  112 static  int     nfs_getattr(struct vop_getattr_args *);
  113 static  int     nfs_setattr(struct vop_setattr_args *);
  114 static  int     nfs_read(struct vop_read_args *);
  115 static  int     nfs_fsync(struct vop_fsync_args *);
  116 static  int     nfs_remove(struct vop_remove_args *);
  117 static  int     nfs_link(struct vop_link_args *);
  118 static  int     nfs_rename(struct vop_rename_args *);
  119 static  int     nfs_mkdir(struct vop_mkdir_args *);
  120 static  int     nfs_rmdir(struct vop_rmdir_args *);
  121 static  int     nfs_symlink(struct vop_symlink_args *);
  122 static  int     nfs_readdir(struct vop_readdir_args *);
  123 static  int     nfs_strategy(struct vop_strategy_args *);
  124 static  int     nfs_lookitup(struct vnode *, const char *, int,
  125                     struct ucred *, struct thread *, struct nfsnode **);
  126 static  int     nfs_sillyrename(struct vnode *, struct vnode *,
  127                     struct componentname *);
  128 static int      nfsspec_access(struct vop_access_args *);
  129 static int      nfs_readlink(struct vop_readlink_args *);
  130 static int      nfs_print(struct vop_print_args *);
  131 static int      nfs_advlock(struct vop_advlock_args *);
  132 
  133 /*
  134  * Global vfs data structures for nfs
  135  */
  136 vop_t **nfs_vnodeop_p;
  137 static struct vnodeopv_entry_desc nfs_vnodeop_entries[] = {
  138         { &vop_default_desc,            (vop_t *) vop_defaultop },
  139         { &vop_access_desc,             (vop_t *) nfs_access },
  140         { &vop_advlock_desc,            (vop_t *) nfs_advlock },
  141         { &vop_close_desc,              (vop_t *) nfs_close },
  142         { &vop_create_desc,             (vop_t *) nfs_create },
  143         { &vop_fsync_desc,              (vop_t *) nfs_fsync },
  144         { &vop_getattr_desc,            (vop_t *) nfs_getattr },
  145         { &vop_getpages_desc,           (vop_t *) nfs_getpages },
  146         { &vop_putpages_desc,           (vop_t *) nfs_putpages },
  147         { &vop_inactive_desc,           (vop_t *) nfs_inactive },
  148         { &vop_lease_desc,              (vop_t *) vop_null },
  149         { &vop_link_desc,               (vop_t *) nfs_link },
  150         { &vop_lookup_desc,             (vop_t *) nfs_lookup },
  151         { &vop_mkdir_desc,              (vop_t *) nfs_mkdir },
  152         { &vop_mknod_desc,              (vop_t *) nfs_mknod },
  153         { &vop_open_desc,               (vop_t *) nfs_open },
  154         { &vop_print_desc,              (vop_t *) nfs_print },
  155         { &vop_read_desc,               (vop_t *) nfs_read },
  156         { &vop_readdir_desc,            (vop_t *) nfs_readdir },
  157         { &vop_readlink_desc,           (vop_t *) nfs_readlink },
  158         { &vop_reclaim_desc,            (vop_t *) nfs_reclaim },
  159         { &vop_remove_desc,             (vop_t *) nfs_remove },
  160         { &vop_rename_desc,             (vop_t *) nfs_rename },
  161         { &vop_rmdir_desc,              (vop_t *) nfs_rmdir },
  162         { &vop_setattr_desc,            (vop_t *) nfs_setattr },
  163         { &vop_strategy_desc,           (vop_t *) nfs_strategy },
  164         { &vop_symlink_desc,            (vop_t *) nfs_symlink },
  165         { &vop_write_desc,              (vop_t *) nfs_write },
  166         { NULL, NULL }
  167 };
  168 static struct vnodeopv_desc nfs_vnodeop_opv_desc =
  169         { &nfs_vnodeop_p, nfs_vnodeop_entries };
  170 VNODEOP_SET(nfs_vnodeop_opv_desc);
  171 
  172 /*
  173  * Special device vnode ops
  174  */
  175 vop_t **spec_nfsnodeop_p;
  176 static struct vnodeopv_entry_desc nfs_specop_entries[] = {
  177         { &vop_default_desc,            (vop_t *) spec_vnoperate },
  178         { &vop_access_desc,             (vop_t *) nfsspec_access },
  179         { &vop_close_desc,              (vop_t *) nfsspec_close },
  180         { &vop_fsync_desc,              (vop_t *) nfs_fsync },
  181         { &vop_getattr_desc,            (vop_t *) nfs_getattr },
  182         { &vop_inactive_desc,           (vop_t *) nfs_inactive },
  183         { &vop_print_desc,              (vop_t *) nfs_print },
  184         { &vop_read_desc,               (vop_t *) nfsspec_read },
  185         { &vop_reclaim_desc,            (vop_t *) nfs_reclaim },
  186         { &vop_setattr_desc,            (vop_t *) nfs_setattr },
  187         { &vop_write_desc,              (vop_t *) nfsspec_write },
  188         { NULL, NULL }
  189 };
  190 static struct vnodeopv_desc spec_nfsnodeop_opv_desc =
  191         { &spec_nfsnodeop_p, nfs_specop_entries };
  192 VNODEOP_SET(spec_nfsnodeop_opv_desc);
  193 
  194 vop_t **fifo_nfsnodeop_p;
  195 static struct vnodeopv_entry_desc nfs_fifoop_entries[] = {
  196         { &vop_default_desc,            (vop_t *) fifo_vnoperate },
  197         { &vop_access_desc,             (vop_t *) nfsspec_access },
  198         { &vop_close_desc,              (vop_t *) nfsfifo_close },
  199         { &vop_fsync_desc,              (vop_t *) nfs_fsync },
  200         { &vop_getattr_desc,            (vop_t *) nfs_getattr },
  201         { &vop_inactive_desc,           (vop_t *) nfs_inactive },
  202         { &vop_print_desc,              (vop_t *) nfs_print },
  203         { &vop_read_desc,               (vop_t *) nfsfifo_read },
  204         { &vop_reclaim_desc,            (vop_t *) nfs_reclaim },
  205         { &vop_setattr_desc,            (vop_t *) nfs_setattr },
  206         { &vop_write_desc,              (vop_t *) nfsfifo_write },
  207         { NULL, NULL }
  208 };
  209 static struct vnodeopv_desc fifo_nfsnodeop_opv_desc =
  210         { &fifo_nfsnodeop_p, nfs_fifoop_entries };
  211 VNODEOP_SET(fifo_nfsnodeop_opv_desc);
  212 
  213 static int      nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
  214                              struct componentname *cnp, struct vattr *vap);
  215 static int      nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
  216                               struct ucred *cred, struct thread *td);
  217 static int      nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
  218                               int fnamelen, struct vnode *tdvp,
  219                               const char *tnameptr, int tnamelen,
  220                               struct ucred *cred, struct thread *td);
  221 static int      nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
  222                              struct sillyrename *sp);
  223 
  224 /*
  225  * Global variables
  226  */
  227 struct proc     *nfs_iodwant[NFS_MAXASYNCDAEMON];
  228 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
  229 int              nfs_numasync = 0;
  230 #define DIRHDSIZ        (sizeof (struct dirent) - (MAXNAMLEN + 1))
  231 
  232 SYSCTL_DECL(_vfs_nfs);
  233 
  234 static int      nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
  235 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
  236            &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
  237 
  238 static int      nfsv3_commit_on_close = 0;
  239 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
  240            &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
  241 #if 0
  242 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
  243            &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
  244 
  245 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
  246            &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
  247 #endif
  248 
  249 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY          \
  250                          | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE     \
  251                          | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
  252 static int
  253 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
  254     struct ucred *cred)
  255 {
  256         const int v3 = 1;
  257         u_int32_t *tl;
  258         int error = 0, attrflag;
  259 
  260         struct mbuf *mreq, *mrep, *md, *mb;
  261         caddr_t bpos, dpos;
  262         u_int32_t rmode;
  263         struct nfsnode *np = VTONFS(vp);
  264 
  265         nfsstats.rpccnt[NFSPROC_ACCESS]++;
  266         mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
  267         mb = mreq;
  268         bpos = mtod(mb, caddr_t);
  269         nfsm_fhtom(vp, v3);
  270         tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
  271         *tl = txdr_unsigned(wmode);
  272         nfsm_request(vp, NFSPROC_ACCESS, td, cred);
  273         nfsm_postop_attr(vp, attrflag);
  274         if (!error) {
  275                 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
  276                 rmode = fxdr_unsigned(u_int32_t, *tl);
  277                 np->n_mode = rmode;
  278                 np->n_modeuid = cred->cr_uid;
  279                 np->n_modestamp = time_second;
  280         }
  281         m_freem(mrep);
  282 nfsmout:
  283         return (error);
  284 }
  285 
  286 /*
  287  * nfs access vnode op.
  288  * For nfs version 2, just return ok. File accesses may fail later.
  289  * For nfs version 3, use the access rpc to check accessibility. If file modes
  290  * are changed on the server, accesses might still fail later.
  291  */
  292 static int
  293 nfs_access(struct vop_access_args *ap)
  294 {
  295         struct vnode *vp = ap->a_vp;
  296         int error = 0;
  297         u_int32_t mode, wmode;
  298         int v3 = NFS_ISV3(vp);
  299         struct nfsnode *np = VTONFS(vp);
  300 
  301         /*
  302          * Disallow write attempts on filesystems mounted read-only;
  303          * unless the file is a socket, fifo, or a block or character
  304          * device resident on the filesystem.
  305          */
  306         if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
  307                 switch (vp->v_type) {
  308                 case VREG:
  309                 case VDIR:
  310                 case VLNK:
  311                         return (EROFS);
  312                 default:
  313                         break;
  314                 }
  315         }
  316         /*
  317          * For nfs v3, check to see if we have done this recently, and if
  318          * so return our cached result instead of making an ACCESS call.
  319          * If not, do an access rpc, otherwise you are stuck emulating
  320          * ufs_access() locally using the vattr. This may not be correct,
  321          * since the server may apply other access criteria such as
  322          * client uid-->server uid mapping that we do not know about.
  323          */
  324         if (v3) {
  325                 if (ap->a_mode & VREAD)
  326                         mode = NFSV3ACCESS_READ;
  327                 else
  328                         mode = 0;
  329                 if (vp->v_type != VDIR) {
  330                         if (ap->a_mode & VWRITE)
  331                                 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
  332                         if (ap->a_mode & VEXEC)
  333                                 mode |= NFSV3ACCESS_EXECUTE;
  334                 } else {
  335                         if (ap->a_mode & VWRITE)
  336                                 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
  337                                          NFSV3ACCESS_DELETE);
  338                         if (ap->a_mode & VEXEC)
  339                                 mode |= NFSV3ACCESS_LOOKUP;
  340                 }
  341                 /* XXX safety belt, only make blanket request if caching */
  342                 if (nfsaccess_cache_timeout > 0) {
  343                         wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
  344                                 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
  345                                 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
  346                 } else {
  347                         wmode = mode;
  348                 }
  349 
  350                 /*
  351                  * Does our cached result allow us to give a definite yes to
  352                  * this request?
  353                  */
  354                 if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
  355                     (ap->a_cred->cr_uid == np->n_modeuid) &&
  356                     ((np->n_mode & mode) == mode)) {
  357                         nfsstats.accesscache_hits++;
  358                 } else {
  359                         /*
  360                          * Either a no, or a don't know.  Go to the wire.
  361                          */
  362                         nfsstats.accesscache_misses++;
  363                         error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
  364                         if (!error) {
  365                                 if ((np->n_mode & mode) != mode) {
  366                                         error = EACCES;
  367                                 }
  368                         }
  369                 }
  370                 return (error);
  371         } else {
  372                 if ((error = nfsspec_access(ap)) != 0)
  373                         return (error);
  374 
  375                 /*
  376                  * Attempt to prevent a mapped root from accessing a file
  377                  * which it shouldn't.  We try to read a byte from the file
  378                  * if the user is root and the file is not zero length.
  379                  * After calling nfsspec_access, we should have the correct
  380                  * file size cached.
  381                  */
  382                 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
  383                     && VTONFS(vp)->n_size > 0) {
  384                         struct iovec aiov;
  385                         struct uio auio;
  386                         char buf[1];
  387 
  388                         aiov.iov_base = buf;
  389                         aiov.iov_len = 1;
  390                         auio.uio_iov = &aiov;
  391                         auio.uio_iovcnt = 1;
  392                         auio.uio_offset = 0;
  393                         auio.uio_resid = 1;
  394                         auio.uio_segflg = UIO_SYSSPACE;
  395                         auio.uio_rw = UIO_READ;
  396                         auio.uio_td = ap->a_td;
  397 
  398                         if (vp->v_type == VREG)
  399                                 error = nfs_readrpc(vp, &auio, ap->a_cred);
  400                         else if (vp->v_type == VDIR) {
  401                                 char* bp;
  402                                 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
  403                                 aiov.iov_base = bp;
  404                                 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
  405                                 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
  406                                 free(bp, M_TEMP);
  407                         } else if (vp->v_type == VLNK)
  408                                 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
  409                         else
  410                                 error = EACCES;
  411                 }
  412                 return (error);
  413         }
  414 }
  415 
  416 /*
  417  * nfs open vnode op
  418  * Check to see if the type is ok
  419  * and that deletion is not in progress.
  420  * For paged in text files, you will need to flush the page cache
  421  * if consistency is lost.
  422  */
  423 /* ARGSUSED */
  424 static int
  425 nfs_open(struct vop_open_args *ap)
  426 {
  427         struct vnode *vp = ap->a_vp;
  428         struct nfsnode *np = VTONFS(vp);
  429         struct vattr vattr;
  430         int error;
  431 
  432         if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
  433 #ifdef DIAGNOSTIC
  434                 printf("open eacces vtyp=%d\n", vp->v_type);
  435 #endif
  436                 return (EOPNOTSUPP);
  437         }
  438         /*
  439          * Get a valid lease. If cached data is stale, flush it.
  440          */
  441         if (np->n_flag & NMODIFIED) {
  442                 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_td, 1);
  443                 if (error == EINTR || error == EIO)
  444                         return (error);
  445                 np->n_attrstamp = 0;
  446                 if (vp->v_type == VDIR)
  447                         np->n_direofoffset = 0;
  448                 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
  449                 if (error)
  450                         return (error);
  451                 np->n_mtime = vattr.va_mtime.tv_sec;
  452         } else {
  453                 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
  454                 if (error)
  455                         return (error);
  456                 if (np->n_mtime != vattr.va_mtime.tv_sec) {
  457                         if (vp->v_type == VDIR)
  458                                 np->n_direofoffset = 0;
  459                         error = nfs_vinvalbuf(vp, V_SAVE,
  460                                 ap->a_cred, ap->a_td, 1);
  461                         if (error == EINTR || error == EIO)
  462                                 return (error);
  463                         np->n_mtime = vattr.va_mtime.tv_sec;
  464                 }
  465         }
  466         np->n_attrstamp = 0; /* For Open/Close consistency */
  467         return (0);
  468 }
  469 
  470 /*
  471  * nfs close vnode op
  472  * What an NFS client should do upon close after writing is a debatable issue.
  473  * Most NFS clients push delayed writes to the server upon close, basically for
  474  * two reasons:
  475  * 1 - So that any write errors may be reported back to the client process
  476  *     doing the close system call. By far the two most likely errors are
  477  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
  478  * 2 - To put a worst case upper bound on cache inconsistency between
  479  *     multiple clients for the file.
  480  * There is also a consistency problem for Version 2 of the protocol w.r.t.
  481  * not being able to tell if other clients are writing a file concurrently,
  482  * since there is no way of knowing if the changed modify time in the reply
  483  * is only due to the write for this client.
  484  * (NFS Version 3 provides weak cache consistency data in the reply that
  485  *  should be sufficient to detect and handle this case.)
  486  *
  487  * The current code does the following:
  488  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
  489  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
  490  *                     or commit them (this satisfies 1 and 2 except for the
  491  *                     case where the server crashes after this close but
  492  *                     before the commit RPC, which is felt to be "good
  493  *                     enough". Changing the last argument to nfs_flush() to
  494  *                     a 1 would force a commit operation, if it is felt a
  495  *                     commit is necessary now.
  496  */
  497 /* ARGSUSED */
  498 static int
  499 nfs_close(struct vop_close_args *ap)
  500 {
  501         struct vnode *vp = ap->a_vp;
  502         struct nfsnode *np = VTONFS(vp);
  503         int error = 0;
  504 
  505         if (vp->v_type == VREG) {
  506             if (np->n_flag & NMODIFIED) {
  507                 if (NFS_ISV3(vp)) {
  508                     /*
  509                      * Under NFSv3 we have dirty buffers to dispose of.  We
  510                      * must flush them to the NFS server.  We have the option
  511                      * of waiting all the way through the commit rpc or just
  512                      * waiting for the initial write.  The default is to only
  513                      * wait through the initial write so the data is in the
  514                      * server's cache, which is roughly similar to the state
  515                      * a standard disk subsystem leaves the file in on close().
  516                      *
  517                      * We cannot clear the NMODIFIED bit in np->n_flag due to
  518                      * potential races with other processes, and certainly
  519                      * cannot clear it if we don't commit.
  520                      */
  521                     int cm = nfsv3_commit_on_close ? 1 : 0;
  522                     error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_td, cm);
  523                     /* np->n_flag &= ~NMODIFIED; */
  524                 } else {
  525                     vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, ap->a_td);
  526                     error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_td, 1);
  527                     VOP_UNLOCK(vp, 0, ap->a_td);
  528                 }
  529                 np->n_attrstamp = 0;
  530             }
  531             if (np->n_flag & NWRITEERR) {
  532                 np->n_flag &= ~NWRITEERR;
  533                 error = np->n_error;
  534             }
  535         }
  536         return (error);
  537 }
  538 
  539 /*
  540  * nfs getattr call from vfs.
  541  */
  542 static int
  543 nfs_getattr(struct vop_getattr_args *ap)
  544 {
  545         struct vnode *vp = ap->a_vp;
  546         struct nfsnode *np = VTONFS(vp);
  547         caddr_t bpos, dpos;
  548         int error = 0;
  549         struct mbuf *mreq, *mrep, *md, *mb;
  550         int v3 = NFS_ISV3(vp);
  551 
  552         /*
  553          * Update local times for special files.
  554          */
  555         if (np->n_flag & (NACC | NUPD))
  556                 np->n_flag |= NCHG;
  557         /*
  558          * First look in the cache.
  559          */
  560         if (nfs_getattrcache(vp, ap->a_vap) == 0)
  561                 return (0);
  562 
  563         if (v3 && nfsaccess_cache_timeout > 0) {
  564                 nfsstats.accesscache_misses++;
  565                 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, ap->a_cred);
  566                 if (nfs_getattrcache(vp, ap->a_vap) == 0)
  567                         return (0);
  568         }
  569 
  570         nfsstats.rpccnt[NFSPROC_GETATTR]++;
  571         mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
  572         mb = mreq;
  573         bpos = mtod(mb, caddr_t);
  574         nfsm_fhtom(vp, v3);
  575         nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, ap->a_cred);
  576         if (!error) {
  577                 nfsm_loadattr(vp, ap->a_vap);
  578         }
  579         m_freem(mrep);
  580 nfsmout:
  581         return (error);
  582 }
  583 
  584 /*
  585  * nfs setattr call.
  586  */
  587 static int
  588 nfs_setattr(struct vop_setattr_args *ap)
  589 {
  590         struct vnode *vp = ap->a_vp;
  591         struct nfsnode *np = VTONFS(vp);
  592         struct vattr *vap = ap->a_vap;
  593         int error = 0;
  594         u_quad_t tsize;
  595 
  596 #ifndef nolint
  597         tsize = (u_quad_t)0;
  598 #endif
  599 
  600         /*
  601          * Setting of flags is not supported.
  602          */
  603         if (vap->va_flags != VNOVAL)
  604                 return (EOPNOTSUPP);
  605 
  606         /*
  607          * Disallow write attempts if the filesystem is mounted read-only.
  608          */
  609         if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
  610             vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
  611             vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
  612             (vp->v_mount->mnt_flag & MNT_RDONLY))
  613                 return (EROFS);
  614         if (vap->va_size != VNOVAL) {
  615                 switch (vp->v_type) {
  616                 case VDIR:
  617                         return (EISDIR);
  618                 case VCHR:
  619                 case VBLK:
  620                 case VSOCK:
  621                 case VFIFO:
  622                         if (vap->va_mtime.tv_sec == VNOVAL &&
  623                             vap->va_atime.tv_sec == VNOVAL &&
  624                             vap->va_mode == (mode_t)VNOVAL &&
  625                             vap->va_uid == (uid_t)VNOVAL &&
  626                             vap->va_gid == (gid_t)VNOVAL)
  627                                 return (0);
  628                         vap->va_size = VNOVAL;
  629                         break;
  630                 default:
  631                         /*
  632                          * Disallow write attempts if the filesystem is
  633                          * mounted read-only.
  634                          */
  635                         if (vp->v_mount->mnt_flag & MNT_RDONLY)
  636                                 return (EROFS);
  637 
  638                         /*
  639                          *  We run vnode_pager_setsize() early (why?),
  640                          * we must set np->n_size now to avoid vinvalbuf
  641                          * V_SAVE races that might setsize a lower
  642                          * value.
  643                          */
  644 
  645                         tsize = np->n_size;
  646                         error = nfs_meta_setsize(vp, ap->a_cred, 
  647                                                 ap->a_td, vap->va_size);
  648 
  649                         if (np->n_flag & NMODIFIED) {
  650                             if (vap->va_size == 0)
  651                                 error = nfs_vinvalbuf(vp, 0,
  652                                         ap->a_cred, ap->a_td, 1);
  653                             else
  654                                 error = nfs_vinvalbuf(vp, V_SAVE,
  655                                         ap->a_cred, ap->a_td, 1);
  656                             if (error) {
  657                                 vnode_pager_setsize(vp, np->n_size);
  658                                 return (error);
  659                             }
  660                         }
  661                         /*
  662                          * np->n_size has already been set to vap->va_size
  663                          * in nfs_meta_setsize(). We must set it again since
  664                          * nfs_loadattrcache() could be called through
  665                          * nfs_meta_setsize() and could modify np->n_size.
  666                          */
  667                         np->n_vattr.va_size = np->n_size = vap->va_size;
  668                 };
  669         } else if ((vap->va_mtime.tv_sec != VNOVAL ||
  670                 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) &&
  671                 vp->v_type == VREG &&
  672                 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
  673                  ap->a_td, 1)) != 0 && (error == EINTR || error == EIO))
  674                 return (error);
  675         error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
  676         if (error && vap->va_size != VNOVAL) {
  677                 np->n_size = np->n_vattr.va_size = tsize;
  678                 vnode_pager_setsize(vp, np->n_size);
  679         }
  680         return (error);
  681 }
  682 
  683 /*
  684  * Do an nfs setattr rpc.
  685  */
  686 static int
  687 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
  688     struct thread *td)
  689 {
  690         struct nfsv2_sattr *sp;
  691         struct nfsnode *np = VTONFS(vp);
  692         caddr_t bpos, dpos;
  693         u_int32_t *tl;
  694         int error = 0, wccflag = NFSV3_WCCRATTR;
  695         struct mbuf *mreq, *mrep, *md, *mb;
  696         int v3 = NFS_ISV3(vp);
  697 
  698         nfsstats.rpccnt[NFSPROC_SETATTR]++;
  699         mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
  700         mb = mreq;
  701         bpos = mtod(mb, caddr_t);
  702         nfsm_fhtom(vp, v3);
  703         if (v3) {
  704                 nfsm_v3attrbuild(vap, TRUE);
  705                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
  706                 *tl = nfs_false;
  707         } else {
  708                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
  709                 if (vap->va_mode == (mode_t)VNOVAL)
  710                         sp->sa_mode = nfs_xdrneg1;
  711                 else
  712                         sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
  713                 if (vap->va_uid == (uid_t)VNOVAL)
  714                         sp->sa_uid = nfs_xdrneg1;
  715                 else
  716                         sp->sa_uid = txdr_unsigned(vap->va_uid);
  717                 if (vap->va_gid == (gid_t)VNOVAL)
  718                         sp->sa_gid = nfs_xdrneg1;
  719                 else
  720                         sp->sa_gid = txdr_unsigned(vap->va_gid);
  721                 sp->sa_size = txdr_unsigned(vap->va_size);
  722                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
  723                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
  724         }
  725         nfsm_request(vp, NFSPROC_SETATTR, td, cred);
  726         if (v3) {
  727                 np->n_modestamp = 0;
  728                 nfsm_wcc_data(vp, wccflag);
  729         } else
  730                 nfsm_loadattr(vp, NULL);
  731         m_freem(mrep);
  732 nfsmout:
  733         return (error);
  734 }
  735 
  736 /*
  737  * nfs lookup call, one step at a time...
  738  * First look in cache
  739  * If not found, unlock the directory nfsnode and do the rpc
  740  */
  741 static int
  742 nfs_lookup(struct vop_lookup_args *ap)
  743 {
  744         struct componentname *cnp = ap->a_cnp;
  745         struct vnode *dvp = ap->a_dvp;
  746         struct vnode **vpp = ap->a_vpp;
  747         int flags = cnp->cn_flags;
  748         struct vnode *newvp;
  749         struct nfsmount *nmp;
  750         caddr_t bpos, dpos;
  751         struct mbuf *mreq, *mrep, *md, *mb;
  752         long len;
  753         nfsfh_t *fhp;
  754         struct nfsnode *np;
  755         int lockparent, wantparent, error = 0, attrflag, fhsize;
  756         int v3 = NFS_ISV3(dvp);
  757         struct thread *td = cnp->cn_thread;
  758 
  759         *vpp = NULLVP;
  760         cnp->cn_flags &= ~PDIRUNLOCK;
  761         if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
  762             (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
  763                 return (EROFS);
  764         if (dvp->v_type != VDIR)
  765                 return (ENOTDIR);
  766         lockparent = flags & LOCKPARENT;
  767         wantparent = flags & (LOCKPARENT|WANTPARENT);
  768         nmp = VFSTONFS(dvp->v_mount);
  769         np = VTONFS(dvp);
  770         if ((error = cache_lookup(dvp, vpp, cnp)) && error != ENOENT) {
  771                 struct vattr vattr;
  772                 int vpid;
  773 
  774                 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
  775                         *vpp = NULLVP;
  776                         return (error);
  777                 }
  778 
  779                 vhold(*vpp);
  780                 newvp = *vpp;
  781                 vpid = newvp->v_id;
  782                 /*
  783                  * See the comment starting `Step through' in ufs/ufs_lookup.c
  784                  * for an explanation of the locking protocol
  785                  */
  786                 if (dvp == newvp) {
  787                         VREF(newvp);
  788                         error = 0;
  789                 } else if (flags & ISDOTDOT) {
  790                         VOP_UNLOCK(dvp, 0, td);
  791                         cnp->cn_flags |= PDIRUNLOCK;
  792                         error = vget(newvp, LK_EXCLUSIVE, td);
  793                         if (!error && lockparent && (flags & ISLASTCN)) {
  794                                 error = vn_lock(dvp, LK_EXCLUSIVE, td);
  795                                 if (error == 0)
  796                                         cnp->cn_flags &= ~PDIRUNLOCK;
  797                         }
  798                 } else {
  799                         error = vget(newvp, LK_EXCLUSIVE, td);
  800                         if (!lockparent || error || !(flags & ISLASTCN)) {
  801                                 VOP_UNLOCK(dvp, 0, td);
  802                                 cnp->cn_flags |= PDIRUNLOCK;
  803                         }
  804                 }
  805                 if (!error) {
  806                         if (vpid == newvp->v_id) {
  807                            if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, td)
  808                             && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
  809                                 nfsstats.lookupcache_hits++;
  810                                 if (cnp->cn_nameiop != LOOKUP &&
  811                                     (flags & ISLASTCN))
  812                                         cnp->cn_flags |= SAVENAME;
  813                                 vdrop(newvp);
  814                                 return (0);
  815                            }
  816                            cache_purge(newvp);
  817                         }
  818                         vput(newvp);
  819                         if (lockparent && dvp != newvp && (flags & ISLASTCN))
  820                                 VOP_UNLOCK(dvp, 0, td);
  821                 }
  822                 vdrop(newvp);
  823                 error = vn_lock(dvp, LK_EXCLUSIVE, td);
  824                 *vpp = NULLVP;
  825                 if (error) {
  826                         cnp->cn_flags |= PDIRUNLOCK;
  827                         return (error);
  828                 }
  829                 cnp->cn_flags &= ~PDIRUNLOCK;
  830         }
  831         error = 0;
  832         newvp = NULLVP;
  833         nfsstats.lookupcache_misses++;
  834         nfsstats.rpccnt[NFSPROC_LOOKUP]++;
  835         len = cnp->cn_namelen;
  836         mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
  837                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
  838         mb = mreq;
  839         bpos = mtod(mb, caddr_t);
  840         nfsm_fhtom(dvp, v3);
  841         nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
  842         nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
  843         if (error) {
  844                 nfsm_postop_attr(dvp, attrflag);
  845                 m_freem(mrep);
  846                 goto nfsmout;
  847         }
  848         nfsm_getfh(fhp, fhsize, v3);
  849 
  850         /*
  851          * Handle RENAME case...
  852          */
  853         if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
  854                 if (NFS_CMPFH(np, fhp, fhsize)) {
  855                         m_freem(mrep);
  856                         return (EISDIR);
  857                 }
  858                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
  859                 if (error) {
  860                         m_freem(mrep);
  861                         return (error);
  862                 }
  863                 newvp = NFSTOV(np);
  864                 if (v3) {
  865                         nfsm_postop_attr(newvp, attrflag);
  866                         nfsm_postop_attr(dvp, attrflag);
  867                 } else
  868                         nfsm_loadattr(newvp, NULL);
  869                 *vpp = newvp;
  870                 m_freem(mrep);
  871                 cnp->cn_flags |= SAVENAME;
  872                 if (!lockparent) {
  873                         VOP_UNLOCK(dvp, 0, td);
  874                         cnp->cn_flags |= PDIRUNLOCK;
  875                 }
  876                 return (0);
  877         }
  878 
  879         if (flags & ISDOTDOT) {
  880                 VOP_UNLOCK(dvp, 0, td);
  881                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
  882                 if (error) {
  883                         vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
  884                         return (error);
  885                 }
  886                 newvp = NFSTOV(np);
  887                 if (lockparent && (flags & ISLASTCN)) {
  888                         error = vn_lock(dvp, LK_EXCLUSIVE, td);
  889                         if (error) {
  890                                 cnp->cn_flags |= PDIRUNLOCK;
  891                                 vput(newvp);
  892                                 return (error);
  893                         }
  894                 } else
  895                         cnp->cn_flags |= PDIRUNLOCK;
  896         } else if (NFS_CMPFH(np, fhp, fhsize)) {
  897                 VREF(dvp);
  898                 newvp = dvp;
  899         } else {
  900                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
  901                 if (error) {
  902                         m_freem(mrep);
  903                         return (error);
  904                 }
  905                 if (!lockparent || !(flags & ISLASTCN)) {
  906                         cnp->cn_flags |= PDIRUNLOCK;
  907                         VOP_UNLOCK(dvp, 0, td);
  908                 }
  909                 newvp = NFSTOV(np);
  910         }
  911         if (v3) {
  912                 nfsm_postop_attr(newvp, attrflag);
  913                 nfsm_postop_attr(dvp, attrflag);
  914         } else
  915                 nfsm_loadattr(newvp, NULL);
  916         if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
  917                 cnp->cn_flags |= SAVENAME;
  918         if ((cnp->cn_flags & MAKEENTRY) &&
  919             (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
  920                 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
  921                 cache_enter(dvp, newvp, cnp);
  922         }
  923         *vpp = newvp;
  924         m_freem(mrep);
  925 nfsmout:
  926         if (error) {
  927                 if (newvp != NULLVP) {
  928                         vrele(newvp);
  929                         *vpp = NULLVP;
  930                 }
  931                 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
  932                     (flags & ISLASTCN) && error == ENOENT) {
  933                         if (!lockparent) {
  934                                 VOP_UNLOCK(dvp, 0, td);
  935                                 cnp->cn_flags |= PDIRUNLOCK;
  936                         }
  937                         if (dvp->v_mount->mnt_flag & MNT_RDONLY)
  938                                 error = EROFS;
  939                         else
  940                                 error = EJUSTRETURN;
  941                 }
  942                 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
  943                         cnp->cn_flags |= SAVENAME;
  944         }
  945         return (error);
  946 }
  947 
  948 /*
  949  * nfs read call.
  950  * Just call nfs_bioread() to do the work.
  951  */
  952 static int
  953 nfs_read(struct vop_read_args *ap)
  954 {
  955         struct vnode *vp = ap->a_vp;
  956 
  957         switch (vp->v_type) {
  958         case VREG:
  959                 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
  960         case VDIR:
  961                 return (EISDIR);
  962         default:
  963                 return (EOPNOTSUPP);
  964         }
  965 }
  966 
  967 /*
  968  * nfs readlink call
  969  */
  970 static int
  971 nfs_readlink(struct vop_readlink_args *ap)
  972 {
  973         struct vnode *vp = ap->a_vp;
  974 
  975         if (vp->v_type != VLNK)
  976                 return (EINVAL);
  977         return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
  978 }
  979 
  980 /*
  981  * Do a readlink rpc.
  982  * Called by nfs_doio() from below the buffer cache.
  983  */
  984 int
  985 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
  986 {
  987         caddr_t bpos, dpos;
  988         int error = 0, len, attrflag;
  989         struct mbuf *mreq, *mrep, *md, *mb;
  990         int v3 = NFS_ISV3(vp);
  991 
  992         nfsstats.rpccnt[NFSPROC_READLINK]++;
  993         mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
  994         mb = mreq;
  995         bpos = mtod(mb, caddr_t);
  996         nfsm_fhtom(vp, v3);
  997         nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
  998         if (v3)
  999                 nfsm_postop_attr(vp, attrflag);
 1000         if (!error) {
 1001                 nfsm_strsiz(len, NFS_MAXPATHLEN);
 1002                 if (len == NFS_MAXPATHLEN) {
 1003                         struct nfsnode *np = VTONFS(vp);
 1004                         if (np->n_size && np->n_size < NFS_MAXPATHLEN)
 1005                                 len = np->n_size;
 1006                 }
 1007                 nfsm_mtouio(uiop, len);
 1008         }
 1009         m_freem(mrep);
 1010 nfsmout:
 1011         return (error);
 1012 }
 1013 
 1014 /*
 1015  * nfs read rpc call
 1016  * Ditto above
 1017  */
 1018 int
 1019 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 1020 {
 1021         u_int32_t *tl;
 1022         caddr_t bpos, dpos;
 1023         struct mbuf *mreq, *mrep, *md, *mb;
 1024         struct nfsmount *nmp;
 1025         int error = 0, len, retlen, tsiz, eof, attrflag;
 1026         int v3 = NFS_ISV3(vp);
 1027 
 1028 #ifndef nolint
 1029         eof = 0;
 1030 #endif
 1031         nmp = VFSTONFS(vp->v_mount);
 1032         tsiz = uiop->uio_resid;
 1033         if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
 1034                 return (EFBIG);
 1035         while (tsiz > 0) {
 1036                 nfsstats.rpccnt[NFSPROC_READ]++;
 1037                 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
 1038                 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
 1039                 mb = mreq;
 1040                 bpos = mtod(mb, caddr_t);
 1041                 nfsm_fhtom(vp, v3);
 1042                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
 1043                 if (v3) {
 1044                         txdr_hyper(uiop->uio_offset, tl);
 1045                         *(tl + 2) = txdr_unsigned(len);
 1046                 } else {
 1047                         *tl++ = txdr_unsigned(uiop->uio_offset);
 1048                         *tl++ = txdr_unsigned(len);
 1049                         *tl = 0;
 1050                 }
 1051                 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
 1052                 if (v3) {
 1053                         nfsm_postop_attr(vp, attrflag);
 1054                         if (error) {
 1055                                 m_freem(mrep);
 1056                                 goto nfsmout;
 1057                         }
 1058                         tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
 1059                         eof = fxdr_unsigned(int, *(tl + 1));
 1060                 } else
 1061                         nfsm_loadattr(vp, NULL);
 1062                 nfsm_strsiz(retlen, nmp->nm_rsize);
 1063                 nfsm_mtouio(uiop, retlen);
 1064                 m_freem(mrep);
 1065                 tsiz -= retlen;
 1066                 if (v3) {
 1067                         if (eof || retlen == 0) {
 1068                                 tsiz = 0;
 1069                         }
 1070                 } else if (retlen < len) {
 1071                         tsiz = 0;
 1072                 }
 1073         }
 1074 nfsmout:
 1075         return (error);
 1076 }
 1077 
 1078 /*
 1079  * nfs write call
 1080  */
 1081 int
 1082 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
 1083     int *iomode, int *must_commit)
 1084 {
 1085         u_int32_t *tl;
 1086         int32_t backup;
 1087         caddr_t bpos, dpos;
 1088         struct mbuf *mreq, *mrep, *md, *mb;
 1089         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 1090         int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
 1091         int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
 1092 
 1093 #ifndef DIAGNOSTIC
 1094         if (uiop->uio_iovcnt != 1)
 1095                 panic("nfs: writerpc iovcnt > 1");
 1096 #endif
 1097         *must_commit = 0;
 1098         tsiz = uiop->uio_resid;
 1099         if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
 1100                 return (EFBIG);
 1101         while (tsiz > 0) {
 1102                 nfsstats.rpccnt[NFSPROC_WRITE]++;
 1103                 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
 1104                 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
 1105                         NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
 1106                 mb = mreq;
 1107                 bpos = mtod(mb, caddr_t);
 1108                 nfsm_fhtom(vp, v3);
 1109                 if (v3) {
 1110                         tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
 1111                         txdr_hyper(uiop->uio_offset, tl);
 1112                         tl += 2;
 1113                         *tl++ = txdr_unsigned(len);
 1114                         *tl++ = txdr_unsigned(*iomode);
 1115                         *tl = txdr_unsigned(len);
 1116                 } else {
 1117                         u_int32_t x;
 1118 
 1119                         tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
 1120                         /* Set both "begin" and "current" to non-garbage. */
 1121                         x = txdr_unsigned((u_int32_t)uiop->uio_offset);
 1122                         *tl++ = x;      /* "begin offset" */
 1123                         *tl++ = x;      /* "current offset" */
 1124                         x = txdr_unsigned(len);
 1125                         *tl++ = x;      /* total to this offset */
 1126                         *tl = x;        /* size of this write */
 1127                 }
 1128                 nfsm_uiotom(uiop, len);
 1129                 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
 1130                 if (v3) {
 1131                         wccflag = NFSV3_WCCCHK;
 1132                         nfsm_wcc_data(vp, wccflag);
 1133                         if (!error) {
 1134                                 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
 1135                                         + NFSX_V3WRITEVERF);
 1136                                 rlen = fxdr_unsigned(int, *tl++);
 1137                                 if (rlen == 0) {
 1138                                         error = NFSERR_IO;
 1139                                         m_freem(mrep);
 1140                                         break;
 1141                                 } else if (rlen < len) {
 1142                                         backup = len - rlen;
 1143                                         uiop->uio_iov->iov_base =
 1144                                             (char *)uiop->uio_iov->iov_base -
 1145                                             backup;
 1146                                         uiop->uio_iov->iov_len += backup;
 1147                                         uiop->uio_offset -= backup;
 1148                                         uiop->uio_resid += backup;
 1149                                         len = rlen;
 1150                                 }
 1151                                 commit = fxdr_unsigned(int, *tl++);
 1152 
 1153                                 /*
 1154                                  * Return the lowest committment level
 1155                                  * obtained by any of the RPCs.
 1156                                  */
 1157                                 if (committed == NFSV3WRITE_FILESYNC)
 1158                                         committed = commit;
 1159                                 else if (committed == NFSV3WRITE_DATASYNC &&
 1160                                         commit == NFSV3WRITE_UNSTABLE)
 1161                                         committed = commit;
 1162                                 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
 1163                                     bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 1164                                         NFSX_V3WRITEVERF);
 1165                                     nmp->nm_state |= NFSSTA_HASWRITEVERF;
 1166                                 } else if (bcmp((caddr_t)tl,
 1167                                     (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
 1168                                     *must_commit = 1;
 1169                                     bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 1170                                         NFSX_V3WRITEVERF);
 1171                                 }
 1172                         }
 1173                 } else
 1174                     nfsm_loadattr(vp, NULL);
 1175                 if (wccflag)
 1176                     VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime.tv_sec;
 1177                 m_freem(mrep);
 1178                 if (error)
 1179                         break;
 1180                 tsiz -= len;
 1181         }
 1182 nfsmout:
 1183         if (vp->v_mount->mnt_flag & MNT_ASYNC)
 1184                 committed = NFSV3WRITE_FILESYNC;
 1185         *iomode = committed;
 1186         if (error)
 1187                 uiop->uio_resid = tsiz;
 1188         return (error);
 1189 }
 1190 
 1191 /*
 1192  * nfs mknod rpc
 1193  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
 1194  * mode set to specify the file type and the size field for rdev.
 1195  */
 1196 static int
 1197 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
 1198     struct vattr *vap)
 1199 {
 1200         struct nfsv2_sattr *sp;
 1201         u_int32_t *tl;
 1202         struct vnode *newvp = NULL;
 1203         struct nfsnode *np = NULL;
 1204         struct vattr vattr;
 1205         caddr_t bpos, dpos;
 1206         int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
 1207         struct mbuf *mreq, *mrep, *md, *mb;
 1208         u_int32_t rdev;
 1209         int v3 = NFS_ISV3(dvp);
 1210 
 1211         if (vap->va_type == VCHR || vap->va_type == VBLK)
 1212                 rdev = txdr_unsigned(vap->va_rdev);
 1213         else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
 1214                 rdev = nfs_xdrneg1;
 1215         else {
 1216                 return (EOPNOTSUPP);
 1217         }
 1218         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1219                 return (error);
 1220         }
 1221         nfsstats.rpccnt[NFSPROC_MKNOD]++;
 1222         mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
 1223                 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
 1224         mb = mreq;
 1225         bpos = mtod(mb, caddr_t);
 1226         nfsm_fhtom(dvp, v3);
 1227         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1228         if (v3) {
 1229                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
 1230                 *tl++ = vtonfsv3_type(vap->va_type);
 1231                 nfsm_v3attrbuild(vap, FALSE);
 1232                 if (vap->va_type == VCHR || vap->va_type == VBLK) {
 1233                         tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
 1234                         *tl++ = txdr_unsigned(umajor(vap->va_rdev));
 1235                         *tl = txdr_unsigned(uminor(vap->va_rdev));
 1236                 }
 1237         } else {
 1238                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1239                 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
 1240                 sp->sa_uid = nfs_xdrneg1;
 1241                 sp->sa_gid = nfs_xdrneg1;
 1242                 sp->sa_size = rdev;
 1243                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1244                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1245         }
 1246         nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
 1247         if (!error) {
 1248                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1249                 if (!gotvp) {
 1250                         if (newvp) {
 1251                                 vput(newvp);
 1252                                 newvp = NULL;
 1253                         }
 1254                         error = nfs_lookitup(dvp, cnp->cn_nameptr,
 1255                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
 1256                         if (!error)
 1257                                 newvp = NFSTOV(np);
 1258                 }
 1259         }
 1260         if (v3)
 1261                 nfsm_wcc_data(dvp, wccflag);
 1262         m_freem(mrep);
 1263 nfsmout:
 1264         if (error) {
 1265                 if (newvp)
 1266                         vput(newvp);
 1267         } else {
 1268                 if (cnp->cn_flags & MAKEENTRY)
 1269                         cache_enter(dvp, newvp, cnp);
 1270                 *vpp = newvp;
 1271         }
 1272         VTONFS(dvp)->n_flag |= NMODIFIED;
 1273         if (!wccflag)
 1274                 VTONFS(dvp)->n_attrstamp = 0;
 1275         return (error);
 1276 }
 1277 
 1278 /*
 1279  * nfs mknod vop
 1280  * just call nfs_mknodrpc() to do the work.
 1281  */
 1282 /* ARGSUSED */
 1283 static int
 1284 nfs_mknod(struct vop_mknod_args *ap)
 1285 {
 1286 
 1287         return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
 1288 }
 1289 
 1290 static u_long create_verf;
 1291 /*
 1292  * nfs file create call
 1293  */
 1294 static int
 1295 nfs_create(struct vop_create_args *ap)
 1296 {
 1297         struct vnode *dvp = ap->a_dvp;
 1298         struct vattr *vap = ap->a_vap;
 1299         struct componentname *cnp = ap->a_cnp;
 1300         struct nfsv2_sattr *sp;
 1301         u_int32_t *tl;
 1302         struct nfsnode *np = NULL;
 1303         struct vnode *newvp = NULL;
 1304         caddr_t bpos, dpos;
 1305         int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
 1306         struct mbuf *mreq, *mrep, *md, *mb;
 1307         struct vattr vattr;
 1308         int v3 = NFS_ISV3(dvp);
 1309 
 1310         /*
 1311          * Oops, not for me..
 1312          */
 1313         if (vap->va_type == VSOCK)
 1314                 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
 1315 
 1316         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1317                 return (error);
 1318         }
 1319         if (vap->va_vaflags & VA_EXCLUSIVE)
 1320                 fmode |= O_EXCL;
 1321 again:
 1322         nfsstats.rpccnt[NFSPROC_CREATE]++;
 1323         mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
 1324                 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
 1325         mb = mreq;
 1326         bpos = mtod(mb, caddr_t);
 1327         nfsm_fhtom(dvp, v3);
 1328         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1329         if (v3) {
 1330                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
 1331                 if (fmode & O_EXCL) {
 1332                         *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
 1333                         tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
 1334 #ifdef INET
 1335                         if (!TAILQ_EMPTY(&in_ifaddrhead))
 1336                                 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
 1337                         else
 1338 #endif
 1339                                 *tl++ = create_verf;
 1340                         *tl = ++create_verf;
 1341                 } else {
 1342                         *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
 1343                         nfsm_v3attrbuild(vap, FALSE);
 1344                 }
 1345         } else {
 1346                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1347                 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
 1348                 sp->sa_uid = nfs_xdrneg1;
 1349                 sp->sa_gid = nfs_xdrneg1;
 1350                 sp->sa_size = 0;
 1351                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1352                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1353         }
 1354         nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
 1355         if (!error) {
 1356                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1357                 if (!gotvp) {
 1358                         if (newvp) {
 1359                                 vput(newvp);
 1360                                 newvp = NULL;
 1361                         }
 1362                         error = nfs_lookitup(dvp, cnp->cn_nameptr,
 1363                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
 1364                         if (!error)
 1365                                 newvp = NFSTOV(np);
 1366                 }
 1367         }
 1368         if (v3)
 1369                 nfsm_wcc_data(dvp, wccflag);
 1370         m_freem(mrep);
 1371 nfsmout:
 1372         if (error) {
 1373                 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
 1374                         fmode &= ~O_EXCL;
 1375                         goto again;
 1376                 }
 1377                 if (newvp)
 1378                         vput(newvp);
 1379         } else if (v3 && (fmode & O_EXCL)) {
 1380                 /*
 1381                  * We are normally called with only a partially initialized
 1382                  * VAP.  Since the NFSv3 spec says that server may use the
 1383                  * file attributes to store the verifier, the spec requires
 1384                  * us to do a SETATTR RPC. FreeBSD servers store the verifier
 1385                  * in atime, but we can't really assume that all servers will
 1386                  * so we ensure that our SETATTR sets both atime and mtime.
 1387                  */
 1388                 if (vap->va_mtime.tv_sec == VNOVAL)
 1389                         vfs_timestamp(&vap->va_mtime);
 1390                 if (vap->va_atime.tv_sec == VNOVAL)
 1391                         vap->va_atime = vap->va_mtime;
 1392                 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_thread);
 1393         }
 1394         if (!error) {
 1395                 if (cnp->cn_flags & MAKEENTRY)
 1396                         cache_enter(dvp, newvp, cnp);
 1397                 *ap->a_vpp = newvp;
 1398         }
 1399         VTONFS(dvp)->n_flag |= NMODIFIED;
 1400         if (!wccflag)
 1401                 VTONFS(dvp)->n_attrstamp = 0;
 1402         return (error);
 1403 }
 1404 
 1405 /*
 1406  * nfs file remove call
 1407  * To try and make nfs semantics closer to ufs semantics, a file that has
 1408  * other processes using the vnode is renamed instead of removed and then
 1409  * removed later on the last close.
 1410  * - If v_usecount > 1
 1411  *        If a rename is not already in the works
 1412  *           call nfs_sillyrename() to set it up
 1413  *     else
 1414  *        do the remove rpc
 1415  */
 1416 static int
 1417 nfs_remove(struct vop_remove_args *ap)
 1418 {
 1419         struct vnode *vp = ap->a_vp;
 1420         struct vnode *dvp = ap->a_dvp;
 1421         struct componentname *cnp = ap->a_cnp;
 1422         struct nfsnode *np = VTONFS(vp);
 1423         int error = 0;
 1424         struct vattr vattr;
 1425 
 1426 #ifndef DIAGNOSTIC
 1427         if ((cnp->cn_flags & HASBUF) == 0)
 1428                 panic("nfs_remove: no name");
 1429         if (vrefcnt(vp) < 1)
 1430                 panic("nfs_remove: bad v_usecount");
 1431 #endif
 1432         if (vp->v_type == VDIR)
 1433                 error = EPERM;
 1434         else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
 1435             VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_thread) == 0 &&
 1436             vattr.va_nlink > 1)) {
 1437                 /*
 1438                  * Purge the name cache so that the chance of a lookup for
 1439                  * the name succeeding while the remove is in progress is
 1440                  * minimized. Without node locking it can still happen, such
 1441                  * that an I/O op returns ESTALE, but since you get this if
 1442                  * another host removes the file..
 1443                  */
 1444                 cache_purge(vp);
 1445                 /*
 1446                  * throw away biocache buffers, mainly to avoid
 1447                  * unnecessary delayed writes later.
 1448                  */
 1449                 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_thread, 1);
 1450                 /* Do the rpc */
 1451                 if (error != EINTR && error != EIO)
 1452                         error = nfs_removerpc(dvp, cnp->cn_nameptr,
 1453                                 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
 1454                 /*
 1455                  * Kludge City: If the first reply to the remove rpc is lost..
 1456                  *   the reply to the retransmitted request will be ENOENT
 1457                  *   since the file was in fact removed
 1458                  *   Therefore, we cheat and return success.
 1459                  */
 1460                 if (error == ENOENT)
 1461                         error = 0;
 1462         } else if (!np->n_sillyrename)
 1463                 error = nfs_sillyrename(dvp, vp, cnp);
 1464         np->n_attrstamp = 0;
 1465         return (error);
 1466 }
 1467 
 1468 /*
 1469  * nfs file remove rpc called from nfs_inactive
 1470  */
 1471 int
 1472 nfs_removeit(struct sillyrename *sp)
 1473 {
 1474 
 1475         /*
 1476          * Make sure that the directory vnode is still valid.
 1477          * XXX we should lock sp->s_dvp here.
 1478          */
 1479         if (sp->s_dvp->v_type == VBAD)
 1480                 return (0);
 1481         return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 1482                 NULL));
 1483 }
 1484 
 1485 /*
 1486  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
 1487  */
 1488 static int
 1489 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
 1490     struct ucred *cred, struct thread *td)
 1491 {
 1492         caddr_t bpos, dpos;
 1493         int error = 0, wccflag = NFSV3_WCCRATTR;
 1494         struct mbuf *mreq, *mrep, *md, *mb;
 1495         int v3 = NFS_ISV3(dvp);
 1496 
 1497         nfsstats.rpccnt[NFSPROC_REMOVE]++;
 1498         mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
 1499                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
 1500         mb = mreq;
 1501         bpos = mtod(mb, caddr_t);
 1502         nfsm_fhtom(dvp, v3);
 1503         nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
 1504         nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
 1505         if (v3)
 1506                 nfsm_wcc_data(dvp, wccflag);
 1507         m_freem(mrep);
 1508 nfsmout:
 1509         VTONFS(dvp)->n_flag |= NMODIFIED;
 1510         if (!wccflag)
 1511                 VTONFS(dvp)->n_attrstamp = 0;
 1512         return (error);
 1513 }
 1514 
 1515 /*
 1516  * nfs file rename call
 1517  */
 1518 static int
 1519 nfs_rename(struct vop_rename_args *ap)
 1520 {
 1521         struct vnode *fvp = ap->a_fvp;
 1522         struct vnode *tvp = ap->a_tvp;
 1523         struct vnode *fdvp = ap->a_fdvp;
 1524         struct vnode *tdvp = ap->a_tdvp;
 1525         struct componentname *tcnp = ap->a_tcnp;
 1526         struct componentname *fcnp = ap->a_fcnp;
 1527         int error;
 1528 
 1529 #ifndef DIAGNOSTIC
 1530         if ((tcnp->cn_flags & HASBUF) == 0 ||
 1531             (fcnp->cn_flags & HASBUF) == 0)
 1532                 panic("nfs_rename: no name");
 1533 #endif
 1534         /* Check for cross-device rename */
 1535         if ((fvp->v_mount != tdvp->v_mount) ||
 1536             (tvp && (fvp->v_mount != tvp->v_mount))) {
 1537                 error = EXDEV;
 1538                 goto out;
 1539         }
 1540 
 1541         if (fvp == tvp) {
 1542                 printf("nfs_rename: fvp == tvp (can't happen)\n");
 1543                 error = 0;
 1544                 goto out;
 1545         }
 1546         if ((error = vn_lock(fvp, LK_EXCLUSIVE, fcnp->cn_thread)) != 0)
 1547                 goto out;
 1548 
 1549         /*
 1550          * We have to flush B_DELWRI data prior to renaming
 1551          * the file.  If we don't, the delayed-write buffers
 1552          * can be flushed out later after the file has gone stale
 1553          * under NFSV3.  NFSV2 does not have this problem because
 1554          * ( as far as I can tell ) it flushes dirty buffers more
 1555          * often.
 1556          */
 1557         VOP_FSYNC(fvp, fcnp->cn_cred, MNT_WAIT, fcnp->cn_thread);
 1558         VOP_UNLOCK(fvp, 0, fcnp->cn_thread);
 1559         if (tvp)
 1560             VOP_FSYNC(tvp, tcnp->cn_cred, MNT_WAIT, tcnp->cn_thread);
 1561 
 1562         /*
 1563          * If the tvp exists and is in use, sillyrename it before doing the
 1564          * rename of the new file over it.
 1565          * XXX Can't sillyrename a directory.
 1566          */
 1567         if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
 1568                 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
 1569                 vput(tvp);
 1570                 tvp = NULL;
 1571         }
 1572 
 1573         error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
 1574                 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
 1575                 tcnp->cn_thread);
 1576 
 1577         if (fvp->v_type == VDIR) {
 1578                 if (tvp != NULL && tvp->v_type == VDIR)
 1579                         cache_purge(tdvp);
 1580                 cache_purge(fdvp);
 1581         }
 1582 
 1583 out:
 1584         if (tdvp == tvp)
 1585                 vrele(tdvp);
 1586         else
 1587                 vput(tdvp);
 1588         if (tvp)
 1589                 vput(tvp);
 1590         vrele(fdvp);
 1591         vrele(fvp);
 1592         /*
 1593          * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
 1594          */
 1595         if (error == ENOENT)
 1596                 error = 0;
 1597         return (error);
 1598 }
 1599 
 1600 /*
 1601  * nfs file rename rpc called from nfs_remove() above
 1602  */
 1603 static int
 1604 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
 1605     struct sillyrename *sp)
 1606 {
 1607 
 1608         return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
 1609             sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
 1610 }
 1611 
 1612 /*
 1613  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
 1614  */
 1615 static int
 1616 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
 1617     struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
 1618     struct thread *td)
 1619 {
 1620         caddr_t bpos, dpos;
 1621         int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
 1622         struct mbuf *mreq, *mrep, *md, *mb;
 1623         int v3 = NFS_ISV3(fdvp);
 1624 
 1625         nfsstats.rpccnt[NFSPROC_RENAME]++;
 1626         mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
 1627                 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
 1628                 nfsm_rndup(tnamelen));
 1629         mb = mreq;
 1630         bpos = mtod(mb, caddr_t);
 1631         nfsm_fhtom(fdvp, v3);
 1632         nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
 1633         nfsm_fhtom(tdvp, v3);
 1634         nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
 1635         nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
 1636         if (v3) {
 1637                 nfsm_wcc_data(fdvp, fwccflag);
 1638                 nfsm_wcc_data(tdvp, twccflag);
 1639         }
 1640         m_freem(mrep);
 1641 nfsmout:
 1642         VTONFS(fdvp)->n_flag |= NMODIFIED;
 1643         VTONFS(tdvp)->n_flag |= NMODIFIED;
 1644         if (!fwccflag)
 1645                 VTONFS(fdvp)->n_attrstamp = 0;
 1646         if (!twccflag)
 1647                 VTONFS(tdvp)->n_attrstamp = 0;
 1648         return (error);
 1649 }
 1650 
 1651 /*
 1652  * nfs hard link create call
 1653  */
 1654 static int
 1655 nfs_link(struct vop_link_args *ap)
 1656 {
 1657         struct vnode *vp = ap->a_vp;
 1658         struct vnode *tdvp = ap->a_tdvp;
 1659         struct componentname *cnp = ap->a_cnp;
 1660         caddr_t bpos, dpos;
 1661         int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
 1662         struct mbuf *mreq, *mrep, *md, *mb;
 1663         int v3;
 1664 
 1665         if (vp->v_mount != tdvp->v_mount) {
 1666                 return (EXDEV);
 1667         }
 1668 
 1669         /*
 1670          * Push all writes to the server, so that the attribute cache
 1671          * doesn't get "out of sync" with the server.
 1672          * XXX There should be a better way!
 1673          */
 1674         VOP_FSYNC(vp, cnp->cn_cred, MNT_WAIT, cnp->cn_thread);
 1675 
 1676         v3 = NFS_ISV3(vp);
 1677         nfsstats.rpccnt[NFSPROC_LINK]++;
 1678         mreq = nfsm_reqhead(vp, NFSPROC_LINK,
 1679                 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
 1680         mb = mreq;
 1681         bpos = mtod(mb, caddr_t);
 1682         nfsm_fhtom(vp, v3);
 1683         nfsm_fhtom(tdvp, v3);
 1684         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1685         nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
 1686         if (v3) {
 1687                 nfsm_postop_attr(vp, attrflag);
 1688                 nfsm_wcc_data(tdvp, wccflag);
 1689         }
 1690         m_freem(mrep);
 1691 nfsmout:
 1692         VTONFS(tdvp)->n_flag |= NMODIFIED;
 1693         if (!attrflag)
 1694                 VTONFS(vp)->n_attrstamp = 0;
 1695         if (!wccflag)
 1696                 VTONFS(tdvp)->n_attrstamp = 0;
 1697         /*
 1698          * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
 1699          */
 1700         if (error == EEXIST)
 1701                 error = 0;
 1702         return (error);
 1703 }
 1704 
 1705 /*
 1706  * nfs symbolic link create call
 1707  */
 1708 static int
 1709 nfs_symlink(struct vop_symlink_args *ap)
 1710 {
 1711         struct vnode *dvp = ap->a_dvp;
 1712         struct vattr *vap = ap->a_vap;
 1713         struct componentname *cnp = ap->a_cnp;
 1714         struct nfsv2_sattr *sp;
 1715         caddr_t bpos, dpos;
 1716         int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
 1717         struct mbuf *mreq, *mrep, *md, *mb;
 1718         struct vnode *newvp = NULL;
 1719         int v3 = NFS_ISV3(dvp);
 1720 
 1721         nfsstats.rpccnt[NFSPROC_SYMLINK]++;
 1722         slen = strlen(ap->a_target);
 1723         mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
 1724             nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
 1725         mb = mreq;
 1726         bpos = mtod(mb, caddr_t);
 1727         nfsm_fhtom(dvp, v3);
 1728         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1729         if (v3) {
 1730                 nfsm_v3attrbuild(vap, FALSE);
 1731         }
 1732         nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
 1733         if (!v3) {
 1734                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1735                 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
 1736                 sp->sa_uid = nfs_xdrneg1;
 1737                 sp->sa_gid = nfs_xdrneg1;
 1738                 sp->sa_size = nfs_xdrneg1;
 1739                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1740                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1741         }
 1742 
 1743         /*
 1744          * Issue the NFS request and get the rpc response.
 1745          *
 1746          * Only NFSv3 responses returning an error of 0 actually return
 1747          * a file handle that can be converted into newvp without having
 1748          * to do an extra lookup rpc.
 1749          */
 1750         nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
 1751         if (v3) {
 1752                 if (error == 0)
 1753                         nfsm_mtofh(dvp, newvp, v3, gotvp);
 1754                 nfsm_wcc_data(dvp, wccflag);
 1755         }
 1756 
 1757         /*
 1758          * out code jumps -> here, mrep is also freed.
 1759          */
 1760 
 1761         m_freem(mrep);
 1762 nfsmout:
 1763 
 1764         /*
 1765          * If we get an EEXIST error, silently convert it to no-error
 1766          * in case of an NFS retry.
 1767          */
 1768         if (error == EEXIST)
 1769                 error = 0;
 1770 
 1771         /*
 1772          * If we do not have (or no longer have) an error, and we could
 1773          * not extract the newvp from the response due to the request being
 1774          * NFSv2 or the error being EEXIST.  We have to do a lookup in order
 1775          * to obtain a newvp to return.
 1776          */
 1777         if (error == 0 && newvp == NULL) {
 1778                 struct nfsnode *np = NULL;
 1779 
 1780                 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 1781                     cnp->cn_cred, cnp->cn_thread, &np);
 1782                 if (!error)
 1783                         newvp = NFSTOV(np);
 1784         }
 1785         if (error) {
 1786                 if (newvp)
 1787                         vput(newvp);
 1788         } else {
 1789                 *ap->a_vpp = newvp;
 1790         }
 1791         VTONFS(dvp)->n_flag |= NMODIFIED;
 1792         if (!wccflag)
 1793                 VTONFS(dvp)->n_attrstamp = 0;
 1794         return (error);
 1795 }
 1796 
 1797 /*
 1798  * nfs make dir call
 1799  */
 1800 static int
 1801 nfs_mkdir(struct vop_mkdir_args *ap)
 1802 {
 1803         struct vnode *dvp = ap->a_dvp;
 1804         struct vattr *vap = ap->a_vap;
 1805         struct componentname *cnp = ap->a_cnp;
 1806         struct nfsv2_sattr *sp;
 1807         int len;
 1808         struct nfsnode *np = NULL;
 1809         struct vnode *newvp = NULL;
 1810         caddr_t bpos, dpos;
 1811         int error = 0, wccflag = NFSV3_WCCRATTR;
 1812         int gotvp = 0;
 1813         struct mbuf *mreq, *mrep, *md, *mb;
 1814         struct vattr vattr;
 1815         int v3 = NFS_ISV3(dvp);
 1816 
 1817         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1818                 return (error);
 1819         }
 1820         len = cnp->cn_namelen;
 1821         nfsstats.rpccnt[NFSPROC_MKDIR]++;
 1822         mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
 1823           NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
 1824         mb = mreq;
 1825         bpos = mtod(mb, caddr_t);
 1826         nfsm_fhtom(dvp, v3);
 1827         nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
 1828         if (v3) {
 1829                 nfsm_v3attrbuild(vap, FALSE);
 1830         } else {
 1831                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1832                 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
 1833                 sp->sa_uid = nfs_xdrneg1;
 1834                 sp->sa_gid = nfs_xdrneg1;
 1835                 sp->sa_size = nfs_xdrneg1;
 1836                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1837                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1838         }
 1839         nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
 1840         if (!error)
 1841                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1842         if (v3)
 1843                 nfsm_wcc_data(dvp, wccflag);
 1844         m_freem(mrep);
 1845 nfsmout:
 1846         VTONFS(dvp)->n_flag |= NMODIFIED;
 1847         if (!wccflag)
 1848                 VTONFS(dvp)->n_attrstamp = 0;
 1849         /*
 1850          * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
 1851          * if we can succeed in looking up the directory.
 1852          */
 1853         if (error == EEXIST || (!error && !gotvp)) {
 1854                 if (newvp) {
 1855                         vrele(newvp);
 1856                         newvp = NULL;
 1857                 }
 1858                 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
 1859                         cnp->cn_thread, &np);
 1860                 if (!error) {
 1861                         newvp = NFSTOV(np);
 1862                         if (newvp->v_type != VDIR)
 1863                                 error = EEXIST;
 1864                 }
 1865         }
 1866         if (error) {
 1867                 if (newvp)
 1868                         vrele(newvp);
 1869         } else
 1870                 *ap->a_vpp = newvp;
 1871         return (error);
 1872 }
 1873 
 1874 /*
 1875  * nfs remove directory call
 1876  */
 1877 static int
 1878 nfs_rmdir(struct vop_rmdir_args *ap)
 1879 {
 1880         struct vnode *vp = ap->a_vp;
 1881         struct vnode *dvp = ap->a_dvp;
 1882         struct componentname *cnp = ap->a_cnp;
 1883         caddr_t bpos, dpos;
 1884         int error = 0, wccflag = NFSV3_WCCRATTR;
 1885         struct mbuf *mreq, *mrep, *md, *mb;
 1886         int v3 = NFS_ISV3(dvp);
 1887 
 1888         if (dvp == vp)
 1889                 return (EINVAL);
 1890         nfsstats.rpccnt[NFSPROC_RMDIR]++;
 1891         mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
 1892                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
 1893         mb = mreq;
 1894         bpos = mtod(mb, caddr_t);
 1895         nfsm_fhtom(dvp, v3);
 1896         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1897         nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
 1898         if (v3)
 1899                 nfsm_wcc_data(dvp, wccflag);
 1900         m_freem(mrep);
 1901 nfsmout:
 1902         VTONFS(dvp)->n_flag |= NMODIFIED;
 1903         if (!wccflag)
 1904                 VTONFS(dvp)->n_attrstamp = 0;
 1905         cache_purge(dvp);
 1906         cache_purge(vp);
 1907         /*
 1908          * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
 1909          */
 1910         if (error == ENOENT)
 1911                 error = 0;
 1912         return (error);
 1913 }
 1914 
 1915 /*
 1916  * nfs readdir call
 1917  */
 1918 static int
 1919 nfs_readdir(struct vop_readdir_args *ap)
 1920 {
 1921         struct vnode *vp = ap->a_vp;
 1922         struct nfsnode *np = VTONFS(vp);
 1923         struct uio *uio = ap->a_uio;
 1924         int tresid, error;
 1925         struct vattr vattr;
 1926 
 1927         if (vp->v_type != VDIR)
 1928                 return (EPERM);
 1929         /*
 1930          * First, check for hit on the EOF offset cache
 1931          */
 1932         if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
 1933             (np->n_flag & NMODIFIED) == 0) {
 1934                 if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_td) == 0 &&
 1935                         np->n_mtime == vattr.va_mtime.tv_sec) {
 1936                         nfsstats.direofcache_hits++;
 1937                         return (0);
 1938                 }
 1939         }
 1940 
 1941         /*
 1942          * Call nfs_bioread() to do the real work.
 1943          */
 1944         tresid = uio->uio_resid;
 1945         error = nfs_bioread(vp, uio, 0, ap->a_cred);
 1946 
 1947         if (!error && uio->uio_resid == tresid)
 1948                 nfsstats.direofcache_misses++;
 1949         return (error);
 1950 }
 1951 
 1952 /*
 1953  * Readdir rpc call.
 1954  * Called from below the buffer cache by nfs_doio().
 1955  */
 1956 int
 1957 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 1958 {
 1959         int len, left;
 1960         struct dirent *dp = NULL;
 1961         u_int32_t *tl;
 1962         caddr_t cp;
 1963         nfsuint64 *cookiep;
 1964         caddr_t bpos, dpos;
 1965         struct mbuf *mreq, *mrep, *md, *mb;
 1966         nfsuint64 cookie;
 1967         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 1968         struct nfsnode *dnp = VTONFS(vp);
 1969         u_quad_t fileno;
 1970         int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
 1971         int attrflag;
 1972         int v3 = NFS_ISV3(vp);
 1973 
 1974 #ifndef DIAGNOSTIC
 1975         if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
 1976                 (uiop->uio_resid & (DIRBLKSIZ - 1)))
 1977                 panic("nfs readdirrpc bad uio");
 1978 #endif
 1979 
 1980         /*
 1981          * If there is no cookie, assume directory was stale.
 1982          */
 1983         cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
 1984         if (cookiep)
 1985                 cookie = *cookiep;
 1986         else
 1987                 return (NFSERR_BAD_COOKIE);
 1988         /*
 1989          * Loop around doing readdir rpc's of size nm_readdirsize
 1990          * truncated to a multiple of DIRBLKSIZ.
 1991          * The stopping criteria is EOF or buffer full.
 1992          */
 1993         while (more_dirs && bigenough) {
 1994                 nfsstats.rpccnt[NFSPROC_READDIR]++;
 1995                 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
 1996                         NFSX_READDIR(v3));
 1997                 mb = mreq;
 1998                 bpos = mtod(mb, caddr_t);
 1999                 nfsm_fhtom(vp, v3);
 2000                 if (v3) {
 2001                         tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
 2002                         *tl++ = cookie.nfsuquad[0];
 2003                         *tl++ = cookie.nfsuquad[1];
 2004                         *tl++ = dnp->n_cookieverf.nfsuquad[0];
 2005                         *tl++ = dnp->n_cookieverf.nfsuquad[1];
 2006                 } else {
 2007                         tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
 2008                         *tl++ = cookie.nfsuquad[0];
 2009                 }
 2010                 *tl = txdr_unsigned(nmp->nm_readdirsize);
 2011                 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
 2012                 if (v3) {
 2013                         nfsm_postop_attr(vp, attrflag);
 2014                         if (!error) {
 2015                                 tl = nfsm_dissect(u_int32_t *,
 2016                                     2 * NFSX_UNSIGNED);
 2017                                 dnp->n_cookieverf.nfsuquad[0] = *tl++;
 2018                                 dnp->n_cookieverf.nfsuquad[1] = *tl;
 2019                         } else {
 2020                                 m_freem(mrep);
 2021                                 goto nfsmout;
 2022                         }
 2023                 }
 2024                 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2025                 more_dirs = fxdr_unsigned(int, *tl);
 2026 
 2027                 /* loop thru the dir entries, doctoring them to 4bsd form */
 2028                 while (more_dirs && bigenough) {
 2029                         if (v3) {
 2030                                 tl = nfsm_dissect(u_int32_t *,
 2031                                     3 * NFSX_UNSIGNED);
 2032                                 fileno = fxdr_hyper(tl);
 2033                                 len = fxdr_unsigned(int, *(tl + 2));
 2034                         } else {
 2035                                 tl = nfsm_dissect(u_int32_t *,
 2036                                     2 * NFSX_UNSIGNED);
 2037                                 fileno = fxdr_unsigned(u_quad_t, *tl++);
 2038                                 len = fxdr_unsigned(int, *tl);
 2039                         }
 2040                         if (len <= 0 || len > NFS_MAXNAMLEN) {
 2041                                 error = EBADRPC;
 2042                                 m_freem(mrep);
 2043                                 goto nfsmout;
 2044                         }
 2045                         tlen = nfsm_rndup(len);
 2046                         if (tlen == len)
 2047                                 tlen += 4;      /* To ensure null termination */
 2048                         left = DIRBLKSIZ - blksiz;
 2049                         if ((tlen + DIRHDSIZ) > left) {
 2050                                 dp->d_reclen += left;
 2051                                 uiop->uio_iov->iov_base =
 2052                                     (char *)uiop->uio_iov->iov_base + left;
 2053                                 uiop->uio_iov->iov_len -= left;
 2054                                 uiop->uio_offset += left;
 2055                                 uiop->uio_resid -= left;
 2056                                 blksiz = 0;
 2057                         }
 2058                         if ((tlen + DIRHDSIZ) > uiop->uio_resid)
 2059                                 bigenough = 0;
 2060                         if (bigenough) {
 2061                                 dp = (struct dirent *)uiop->uio_iov->iov_base;
 2062                                 dp->d_fileno = (int)fileno;
 2063                                 dp->d_namlen = len;
 2064                                 dp->d_reclen = tlen + DIRHDSIZ;
 2065                                 dp->d_type = DT_UNKNOWN;
 2066                                 blksiz += dp->d_reclen;
 2067                                 if (blksiz == DIRBLKSIZ)
 2068                                         blksiz = 0;
 2069                                 uiop->uio_offset += DIRHDSIZ;
 2070                                 uiop->uio_resid -= DIRHDSIZ;
 2071                                 uiop->uio_iov->iov_base =
 2072                                     (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
 2073                                 uiop->uio_iov->iov_len -= DIRHDSIZ;
 2074                                 nfsm_mtouio(uiop, len);
 2075                                 cp = uiop->uio_iov->iov_base;
 2076                                 tlen -= len;
 2077                                 *cp = '\0';     /* null terminate */
 2078                                 uiop->uio_iov->iov_base =
 2079                                     (char *)uiop->uio_iov->iov_base + tlen;
 2080                                 uiop->uio_iov->iov_len -= tlen;
 2081                                 uiop->uio_offset += tlen;
 2082                                 uiop->uio_resid -= tlen;
 2083                         } else
 2084                                 nfsm_adv(nfsm_rndup(len));
 2085                         if (v3) {
 2086                                 tl = nfsm_dissect(u_int32_t *,
 2087                                     3 * NFSX_UNSIGNED);
 2088                         } else {
 2089                                 tl = nfsm_dissect(u_int32_t *,
 2090                                     2 * NFSX_UNSIGNED);
 2091                         }
 2092                         if (bigenough) {
 2093                                 cookie.nfsuquad[0] = *tl++;
 2094                                 if (v3)
 2095                                         cookie.nfsuquad[1] = *tl++;
 2096                         } else if (v3)
 2097                                 tl += 2;
 2098                         else
 2099                                 tl++;
 2100                         more_dirs = fxdr_unsigned(int, *tl);
 2101                 }
 2102                 /*
 2103                  * If at end of rpc data, get the eof boolean
 2104                  */
 2105                 if (!more_dirs) {
 2106                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2107                         more_dirs = (fxdr_unsigned(int, *tl) == 0);
 2108                 }
 2109                 m_freem(mrep);
 2110         }
 2111         /*
 2112          * Fill last record, iff any, out to a multiple of DIRBLKSIZ
 2113          * by increasing d_reclen for the last record.
 2114          */
 2115         if (blksiz > 0) {
 2116                 left = DIRBLKSIZ - blksiz;
 2117                 dp->d_reclen += left;
 2118                 uiop->uio_iov->iov_base =
 2119                     (char *)uiop->uio_iov->iov_base + left;
 2120                 uiop->uio_iov->iov_len -= left;
 2121                 uiop->uio_offset += left;
 2122                 uiop->uio_resid -= left;
 2123         }
 2124 
 2125         /*
 2126          * We are now either at the end of the directory or have filled the
 2127          * block.
 2128          */
 2129         if (bigenough)
 2130                 dnp->n_direofoffset = uiop->uio_offset;
 2131         else {
 2132                 if (uiop->uio_resid > 0)
 2133                         printf("EEK! readdirrpc resid > 0\n");
 2134                 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
 2135                 *cookiep = cookie;
 2136         }
 2137 nfsmout:
 2138         return (error);
 2139 }
 2140 
 2141 /*
 2142  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
 2143  */
 2144 int
 2145 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 2146 {
 2147         int len, left;
 2148         struct dirent *dp;
 2149         u_int32_t *tl;
 2150         caddr_t cp;
 2151         struct vnode *newvp;
 2152         nfsuint64 *cookiep;
 2153         caddr_t bpos, dpos, dpossav1, dpossav2;
 2154         struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
 2155         struct nameidata nami, *ndp = &nami;
 2156         struct componentname *cnp = &ndp->ni_cnd;
 2157         nfsuint64 cookie;
 2158         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2159         struct nfsnode *dnp = VTONFS(vp), *np;
 2160         nfsfh_t *fhp;
 2161         u_quad_t fileno;
 2162         int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
 2163         int attrflag, fhsize;
 2164 
 2165 #ifndef nolint
 2166         dp = NULL;
 2167 #endif
 2168 #ifndef DIAGNOSTIC
 2169         if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
 2170                 (uiop->uio_resid & (DIRBLKSIZ - 1)))
 2171                 panic("nfs readdirplusrpc bad uio");
 2172 #endif
 2173         ndp->ni_dvp = vp;
 2174         newvp = NULLVP;
 2175 
 2176         /*
 2177          * If there is no cookie, assume directory was stale.
 2178          */
 2179         cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
 2180         if (cookiep)
 2181                 cookie = *cookiep;
 2182         else
 2183                 return (NFSERR_BAD_COOKIE);
 2184         /*
 2185          * Loop around doing readdir rpc's of size nm_readdirsize
 2186          * truncated to a multiple of DIRBLKSIZ.
 2187          * The stopping criteria is EOF or buffer full.
 2188          */
 2189         while (more_dirs && bigenough) {
 2190                 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
 2191                 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
 2192                         NFSX_FH(1) + 6 * NFSX_UNSIGNED);
 2193                 mb = mreq;
 2194                 bpos = mtod(mb, caddr_t);
 2195                 nfsm_fhtom(vp, 1);
 2196                 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
 2197                 *tl++ = cookie.nfsuquad[0];
 2198                 *tl++ = cookie.nfsuquad[1];
 2199                 *tl++ = dnp->n_cookieverf.nfsuquad[0];
 2200                 *tl++ = dnp->n_cookieverf.nfsuquad[1];
 2201                 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
 2202                 *tl = txdr_unsigned(nmp->nm_rsize);
 2203                 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
 2204                 nfsm_postop_attr(vp, attrflag);
 2205                 if (error) {
 2206                         m_freem(mrep);
 2207                         goto nfsmout;
 2208                 }
 2209                 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2210                 dnp->n_cookieverf.nfsuquad[0] = *tl++;
 2211                 dnp->n_cookieverf.nfsuquad[1] = *tl++;
 2212                 more_dirs = fxdr_unsigned(int, *tl);
 2213 
 2214                 /* loop thru the dir entries, doctoring them to 4bsd form */
 2215                 while (more_dirs && bigenough) {
 2216                         tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2217                         fileno = fxdr_hyper(tl);
 2218                         len = fxdr_unsigned(int, *(tl + 2));
 2219                         if (len <= 0 || len > NFS_MAXNAMLEN) {
 2220                                 error = EBADRPC;
 2221                                 m_freem(mrep);
 2222                                 goto nfsmout;
 2223                         }
 2224                         tlen = nfsm_rndup(len);
 2225                         if (tlen == len)
 2226                                 tlen += 4;      /* To ensure null termination*/
 2227                         left = DIRBLKSIZ - blksiz;
 2228                         if ((tlen + DIRHDSIZ) > left) {
 2229                                 dp->d_reclen += left;
 2230                                 uiop->uio_iov->iov_base =
 2231                                     (char *)uiop->uio_iov->iov_base + left;
 2232                                 uiop->uio_iov->iov_len -= left;
 2233                                 uiop->uio_offset += left;
 2234                                 uiop->uio_resid -= left;
 2235                                 blksiz = 0;
 2236                         }
 2237                         if ((tlen + DIRHDSIZ) > uiop->uio_resid)
 2238                                 bigenough = 0;
 2239                         if (bigenough) {
 2240                                 dp = (struct dirent *)uiop->uio_iov->iov_base;
 2241                                 dp->d_fileno = (int)fileno;
 2242                                 dp->d_namlen = len;
 2243                                 dp->d_reclen = tlen + DIRHDSIZ;
 2244                                 dp->d_type = DT_UNKNOWN;
 2245                                 blksiz += dp->d_reclen;
 2246                                 if (blksiz == DIRBLKSIZ)
 2247                                         blksiz = 0;
 2248                                 uiop->uio_offset += DIRHDSIZ;
 2249                                 uiop->uio_resid -= DIRHDSIZ;
 2250                                 uiop->uio_iov->iov_base =
 2251                                     (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
 2252                                 uiop->uio_iov->iov_len -= DIRHDSIZ;
 2253                                 cnp->cn_nameptr = uiop->uio_iov->iov_base;
 2254                                 cnp->cn_namelen = len;
 2255                                 nfsm_mtouio(uiop, len);
 2256                                 cp = uiop->uio_iov->iov_base;
 2257                                 tlen -= len;
 2258                                 *cp = '\0';
 2259                                 uiop->uio_iov->iov_base =
 2260                                     (char *)uiop->uio_iov->iov_base + tlen;
 2261                                 uiop->uio_iov->iov_len -= tlen;
 2262                                 uiop->uio_offset += tlen;
 2263                                 uiop->uio_resid -= tlen;
 2264                         } else
 2265                                 nfsm_adv(nfsm_rndup(len));
 2266                         tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2267                         if (bigenough) {
 2268                                 cookie.nfsuquad[0] = *tl++;
 2269                                 cookie.nfsuquad[1] = *tl++;
 2270                         } else
 2271                                 tl += 2;
 2272 
 2273                         /*
 2274                          * Since the attributes are before the file handle
 2275                          * (sigh), we must skip over the attributes and then
 2276                          * come back and get them.
 2277                          */
 2278                         attrflag = fxdr_unsigned(int, *tl);
 2279                         if (attrflag) {
 2280                             dpossav1 = dpos;
 2281                             mdsav1 = md;
 2282                             nfsm_adv(NFSX_V3FATTR);
 2283                             tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2284                             doit = fxdr_unsigned(int, *tl);
 2285                             if (doit) {
 2286                                 nfsm_getfh(fhp, fhsize, 1);
 2287                                 if (NFS_CMPFH(dnp, fhp, fhsize)) {
 2288                                     VREF(vp);
 2289                                     newvp = vp;
 2290                                     np = dnp;
 2291                                 } else {
 2292                                     error = nfs_nget(vp->v_mount, fhp,
 2293                                         fhsize, &np);
 2294                                     if (error)
 2295                                         doit = 0;
 2296                                     else
 2297                                         newvp = NFSTOV(np);
 2298                                 }
 2299                             }
 2300                             if (doit && bigenough) {
 2301                                 dpossav2 = dpos;
 2302                                 dpos = dpossav1;
 2303                                 mdsav2 = md;
 2304                                 md = mdsav1;
 2305                                 nfsm_loadattr(newvp, NULL);
 2306                                 dpos = dpossav2;
 2307                                 md = mdsav2;
 2308                                 dp->d_type =
 2309                                     IFTODT(VTTOIF(np->n_vattr.va_type));
 2310                                 ndp->ni_vp = newvp;
 2311                                 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
 2312                             }
 2313                         } else {
 2314                             /* Just skip over the file handle */
 2315                             tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2316                             i = fxdr_unsigned(int, *tl);
 2317                             nfsm_adv(nfsm_rndup(i));
 2318                         }
 2319                         if (newvp != NULLVP) {
 2320                             if (newvp == vp)
 2321                                 vrele(newvp);
 2322                             else
 2323                                 vput(newvp);
 2324                             newvp = NULLVP;
 2325                         }
 2326                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2327                         more_dirs = fxdr_unsigned(int, *tl);
 2328                 }
 2329                 /*
 2330                  * If at end of rpc data, get the eof boolean
 2331                  */
 2332                 if (!more_dirs) {
 2333                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2334                         more_dirs = (fxdr_unsigned(int, *tl) == 0);
 2335                 }
 2336                 m_freem(mrep);
 2337         }
 2338         /*
 2339          * Fill last record, iff any, out to a multiple of DIRBLKSIZ
 2340          * by increasing d_reclen for the last record.
 2341          */
 2342         if (blksiz > 0) {
 2343                 left = DIRBLKSIZ - blksiz;
 2344                 dp->d_reclen += left;
 2345                 uiop->uio_iov->iov_base =
 2346                     (char *)uiop->uio_iov->iov_base + left;
 2347                 uiop->uio_iov->iov_len -= left;
 2348                 uiop->uio_offset += left;
 2349                 uiop->uio_resid -= left;
 2350         }
 2351 
 2352         /*
 2353          * We are now either at the end of the directory or have filled the
 2354          * block.
 2355          */
 2356         if (bigenough)
 2357                 dnp->n_direofoffset = uiop->uio_offset;
 2358         else {
 2359                 if (uiop->uio_resid > 0)
 2360                         printf("EEK! readdirplusrpc resid > 0\n");
 2361                 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
 2362                 *cookiep = cookie;
 2363         }
 2364 nfsmout:
 2365         if (newvp != NULLVP) {
 2366                 if (newvp == vp)
 2367                         vrele(newvp);
 2368                 else
 2369                         vput(newvp);
 2370                 newvp = NULLVP;
 2371         }
 2372         return (error);
 2373 }
 2374 
 2375 /*
 2376  * Silly rename. To make the NFS filesystem that is stateless look a little
 2377  * more like the "ufs" a remove of an active vnode is translated to a rename
 2378  * to a funny looking filename that is removed by nfs_inactive on the
 2379  * nfsnode. There is the potential for another process on a different client
 2380  * to create the same funny name between the nfs_lookitup() fails and the
 2381  * nfs_rename() completes, but...
 2382  */
 2383 static int
 2384 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
 2385 {
 2386         struct sillyrename *sp;
 2387         struct nfsnode *np;
 2388         int error;
 2389         short pid;
 2390 
 2391         cache_purge(dvp);
 2392         np = VTONFS(vp);
 2393 #ifndef DIAGNOSTIC
 2394         if (vp->v_type == VDIR)
 2395                 panic("nfs: sillyrename dir");
 2396 #endif
 2397         MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
 2398                 M_NFSREQ, M_WAITOK);
 2399         sp->s_cred = crhold(cnp->cn_cred);
 2400         sp->s_dvp = dvp;
 2401         sp->s_removeit = nfs_removeit;
 2402         VREF(dvp);
 2403 
 2404         /* Fudge together a funny name */
 2405         pid = cnp->cn_thread->td_proc->p_pid;
 2406         sp->s_namlen = sprintf(sp->s_name, ".nfsA%04x4.4", pid);
 2407 
 2408         /* Try lookitups until we get one that isn't there */
 2409         while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2410                 cnp->cn_thread, NULL) == 0) {
 2411                 sp->s_name[4]++;
 2412                 if (sp->s_name[4] > 'z') {
 2413                         error = EINVAL;
 2414                         goto bad;
 2415                 }
 2416         }
 2417         error = nfs_renameit(dvp, cnp, sp);
 2418         if (error)
 2419                 goto bad;
 2420         error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2421                 cnp->cn_thread, &np);
 2422         np->n_sillyrename = sp;
 2423         return (0);
 2424 bad:
 2425         vrele(sp->s_dvp);
 2426         crfree(sp->s_cred);
 2427         free((caddr_t)sp, M_NFSREQ);
 2428         return (error);
 2429 }
 2430 
 2431 /*
 2432  * Look up a file name and optionally either update the file handle or
 2433  * allocate an nfsnode, depending on the value of npp.
 2434  * npp == NULL  --> just do the lookup
 2435  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
 2436  *                      handled too
 2437  * *npp != NULL --> update the file handle in the vnode
 2438  */
 2439 static int
 2440 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
 2441     struct thread *td, struct nfsnode **npp)
 2442 {
 2443         struct vnode *newvp = NULL;
 2444         struct nfsnode *np, *dnp = VTONFS(dvp);
 2445         caddr_t bpos, dpos;
 2446         int error = 0, fhlen, attrflag;
 2447         struct mbuf *mreq, *mrep, *md, *mb;
 2448         nfsfh_t *nfhp;
 2449         int v3 = NFS_ISV3(dvp);
 2450 
 2451         nfsstats.rpccnt[NFSPROC_LOOKUP]++;
 2452         mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
 2453                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
 2454         mb = mreq;
 2455         bpos = mtod(mb, caddr_t);
 2456         nfsm_fhtom(dvp, v3);
 2457         nfsm_strtom(name, len, NFS_MAXNAMLEN);
 2458         nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
 2459         if (npp && !error) {
 2460                 nfsm_getfh(nfhp, fhlen, v3);
 2461                 if (*npp) {
 2462                     np = *npp;
 2463                     if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
 2464                         free((caddr_t)np->n_fhp, M_NFSBIGFH);
 2465                         np->n_fhp = &np->n_fh;
 2466                     } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
 2467                         np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
 2468                     bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
 2469                     np->n_fhsize = fhlen;
 2470                     newvp = NFSTOV(np);
 2471                 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
 2472                     VREF(dvp);
 2473                     newvp = dvp;
 2474                 } else {
 2475                     error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
 2476                     if (error) {
 2477                         m_freem(mrep);
 2478                         return (error);
 2479                     }
 2480                     newvp = NFSTOV(np);
 2481                 }
 2482                 if (v3) {
 2483                         nfsm_postop_attr(newvp, attrflag);
 2484                         if (!attrflag && *npp == NULL) {
 2485                                 m_freem(mrep);
 2486                                 if (newvp == dvp)
 2487                                         vrele(newvp);
 2488                                 else
 2489                                         vput(newvp);
 2490                                 return (ENOENT);
 2491                         }
 2492                 } else
 2493                         nfsm_loadattr(newvp, NULL);
 2494         }
 2495         m_freem(mrep);
 2496 nfsmout:
 2497         if (npp && *npp == NULL) {
 2498                 if (error) {
 2499                         if (newvp) {
 2500                                 if (newvp == dvp)
 2501                                         vrele(newvp);
 2502                                 else
 2503                                         vput(newvp);
 2504                         }
 2505                 } else
 2506                         *npp = np;
 2507         }
 2508         return (error);
 2509 }
 2510 
 2511 /*
 2512  * Nfs Version 3 commit rpc
 2513  */
 2514 int
 2515 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
 2516     struct thread *td)
 2517 {
 2518         u_int32_t *tl;
 2519         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2520         caddr_t bpos, dpos;
 2521         int error = 0, wccflag = NFSV3_WCCRATTR;
 2522         struct mbuf *mreq, *mrep, *md, *mb;
 2523 
 2524         if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
 2525                 return (0);
 2526         nfsstats.rpccnt[NFSPROC_COMMIT]++;
 2527         mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
 2528         mb = mreq;
 2529         bpos = mtod(mb, caddr_t);
 2530         nfsm_fhtom(vp, 1);
 2531         tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
 2532         txdr_hyper(offset, tl);
 2533         tl += 2;
 2534         *tl = txdr_unsigned(cnt);
 2535         nfsm_request(vp, NFSPROC_COMMIT, td, cred);
 2536         nfsm_wcc_data(vp, wccflag);
 2537         if (!error) {
 2538                 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
 2539                 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
 2540                         NFSX_V3WRITEVERF)) {
 2541                         bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 2542                                 NFSX_V3WRITEVERF);
 2543                         error = NFSERR_STALEWRITEVERF;
 2544                 }
 2545         }
 2546         m_freem(mrep);
 2547 nfsmout:
 2548         return (error);
 2549 }
 2550 
 2551 /*
 2552  * Strategy routine.
 2553  * For async requests when nfsiod(s) are running, queue the request by
 2554  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
 2555  * request.
 2556  */
 2557 static int
 2558 nfs_strategy(struct vop_strategy_args *ap)
 2559 {
 2560         struct buf *bp = ap->a_bp;
 2561         struct ucred *cr;
 2562         struct thread *td;
 2563         int error = 0;
 2564 
 2565         KASSERT(ap->a_vp == ap->a_bp->b_vp, ("%s(%p != %p)",
 2566             __func__, ap->a_vp, ap->a_bp->b_vp));
 2567         KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
 2568         KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
 2569 
 2570         if (bp->b_flags & B_ASYNC)
 2571                 td = NULL;
 2572         else
 2573                 td = curthread; /* XXX */
 2574 
 2575         if (bp->b_iocmd == BIO_READ)
 2576                 cr = bp->b_rcred;
 2577         else
 2578                 cr = bp->b_wcred;
 2579 
 2580         /*
 2581          * If the op is asynchronous and an i/o daemon is waiting
 2582          * queue the request, wake it up and wait for completion
 2583          * otherwise just do it ourselves.
 2584          */
 2585         if ((bp->b_flags & B_ASYNC) == 0 ||
 2586                 nfs_asyncio(bp, NOCRED, td))
 2587                 error = nfs_doio(bp, cr, td);
 2588         return (error);
 2589 }
 2590 
 2591 /*
 2592  * fsync vnode op. Just call nfs_flush() with commit == 1.
 2593  */
 2594 /* ARGSUSED */
 2595 static int
 2596 nfs_fsync(struct vop_fsync_args *ap)
 2597 {
 2598 
 2599         return (nfs_flush(ap->a_vp, ap->a_cred, ap->a_waitfor, ap->a_td, 1));
 2600 }
 2601 
 2602 /*
 2603  * Flush all the blocks associated with a vnode.
 2604  *      Walk through the buffer pool and push any dirty pages
 2605  *      associated with the vnode.
 2606  */
 2607 static int
 2608 nfs_flush(struct vnode *vp, struct ucred *cred, int waitfor, struct thread *td,
 2609     int commit)
 2610 {
 2611         struct nfsnode *np = VTONFS(vp);
 2612         struct buf *bp;
 2613         int i;
 2614         struct buf *nbp;
 2615         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2616         int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
 2617         int passone = 1;
 2618         u_quad_t off, endoff, toff;
 2619         struct ucred* wcred = NULL;
 2620         struct buf **bvec = NULL;
 2621 #ifndef NFS_COMMITBVECSIZ
 2622 #define NFS_COMMITBVECSIZ       20
 2623 #endif
 2624         struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
 2625         int bvecsize = 0, bveccount;
 2626 
 2627         if (nmp->nm_flag & NFSMNT_INT)
 2628                 slpflag = PCATCH;
 2629         if (!commit)
 2630                 passone = 0;
 2631         /*
 2632          * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
 2633          * server, but nas not been committed to stable storage on the server
 2634          * yet. On the first pass, the byte range is worked out and the commit
 2635          * rpc is done. On the second pass, nfs_writebp() is called to do the
 2636          * job.
 2637          */
 2638 again:
 2639         off = (u_quad_t)-1;
 2640         endoff = 0;
 2641         bvecpos = 0;
 2642         if (NFS_ISV3(vp) && commit) {
 2643                 s = splbio();
 2644                 if (bvec != NULL && bvec != bvec_on_stack)
 2645                         free(bvec, M_TEMP);
 2646                 /*
 2647                  * Count up how many buffers waiting for a commit.
 2648                  */
 2649                 bveccount = 0;
 2650                 VI_LOCK(vp);
 2651                 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
 2652                         nbp = TAILQ_NEXT(bp, b_vnbufs);
 2653                         if (BUF_REFCNT(bp) == 0 &&
 2654                             (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
 2655                                 == (B_DELWRI | B_NEEDCOMMIT))
 2656                                 bveccount++;
 2657                 }
 2658                 /*
 2659                  * Allocate space to remember the list of bufs to commit.  It is
 2660                  * important to use M_NOWAIT here to avoid a race with nfs_write.
 2661                  * If we can't get memory (for whatever reason), we will end up
 2662                  * committing the buffers one-by-one in the loop below.
 2663                  */
 2664                 if (bveccount > NFS_COMMITBVECSIZ) {
 2665                         /*
 2666                          * Release the vnode interlock to avoid a lock
 2667                          * order reversal.
 2668                          */
 2669                         VI_UNLOCK(vp);
 2670                         bvec = (struct buf **)
 2671                                 malloc(bveccount * sizeof(struct buf *),
 2672                                        M_TEMP, M_NOWAIT);
 2673                         VI_LOCK(vp);
 2674                         if (bvec == NULL) {
 2675                                 bvec = bvec_on_stack;
 2676                                 bvecsize = NFS_COMMITBVECSIZ;
 2677                         } else
 2678                                 bvecsize = bveccount;
 2679                 } else {
 2680                         bvec = bvec_on_stack;
 2681                         bvecsize = NFS_COMMITBVECSIZ;
 2682                 }
 2683                 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
 2684                         if (bvecpos >= bvecsize)
 2685                                 break;
 2686                         if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
 2687                                 nbp = TAILQ_NEXT(bp, b_vnbufs);
 2688                                 continue;
 2689                         }
 2690                         if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
 2691                             (B_DELWRI | B_NEEDCOMMIT)) {
 2692                                 BUF_UNLOCK(bp);
 2693                                 nbp = TAILQ_NEXT(bp, b_vnbufs);
 2694                                 continue;
 2695                         }
 2696                         VI_UNLOCK(vp);
 2697                         bremfree(bp);
 2698                         /*
 2699                          * Work out if all buffers are using the same cred
 2700                          * so we can deal with them all with one commit.
 2701                          *
 2702                          * NOTE: we are not clearing B_DONE here, so we have
 2703                          * to do it later on in this routine if we intend to
 2704                          * initiate I/O on the bp.
 2705                          *
 2706                          * Note: to avoid loopback deadlocks, we do not
 2707                          * assign b_runningbufspace.
 2708                          */
 2709                         if (wcred == NULL)
 2710                                 wcred = bp->b_wcred;
 2711                         else if (wcred != bp->b_wcred)
 2712                                 wcred = NOCRED;
 2713                         bp->b_flags |= B_WRITEINPROG;
 2714                         vfs_busy_pages(bp, 1);
 2715 
 2716                         VI_LOCK(vp);
 2717                         /*
 2718                          * bp is protected by being locked, but nbp is not
 2719                          * and vfs_busy_pages() may sleep.  We have to
 2720                          * recalculate nbp.
 2721                          */
 2722                         nbp = TAILQ_NEXT(bp, b_vnbufs);
 2723 
 2724                         /*
 2725                          * A list of these buffers is kept so that the
 2726                          * second loop knows which buffers have actually
 2727                          * been committed. This is necessary, since there
 2728                          * may be a race between the commit rpc and new
 2729                          * uncommitted writes on the file.
 2730                          */
 2731                         bvec[bvecpos++] = bp;
 2732                         toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2733                                 bp->b_dirtyoff;
 2734                         if (toff < off)
 2735                                 off = toff;
 2736                         toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
 2737                         if (toff > endoff)
 2738                                 endoff = toff;
 2739                 }
 2740                 splx(s);
 2741                 VI_UNLOCK(vp);
 2742         }
 2743         if (bvecpos > 0) {
 2744                 /*
 2745                  * Commit data on the server, as required.
 2746                  * If all bufs are using the same wcred, then use that with
 2747                  * one call for all of them, otherwise commit each one
 2748                  * separately.
 2749                  */
 2750                 if (wcred != NOCRED)
 2751                         retv = nfs_commit(vp, off, (int)(endoff - off),
 2752                                           wcred, td);
 2753                 else {
 2754                         retv = 0;
 2755                         for (i = 0; i < bvecpos; i++) {
 2756                                 off_t off, size;
 2757                                 bp = bvec[i];
 2758                                 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2759                                         bp->b_dirtyoff;
 2760                                 size = (u_quad_t)(bp->b_dirtyend
 2761                                                   - bp->b_dirtyoff);
 2762                                 retv = nfs_commit(vp, off, (int)size,
 2763                                                   bp->b_wcred, td);
 2764                                 if (retv) break;
 2765                         }
 2766                 }
 2767 
 2768                 if (retv == NFSERR_STALEWRITEVERF)
 2769                         nfs_clearcommit(vp->v_mount);
 2770 
 2771                 /*
 2772                  * Now, either mark the blocks I/O done or mark the
 2773                  * blocks dirty, depending on whether the commit
 2774                  * succeeded.
 2775                  */
 2776                 for (i = 0; i < bvecpos; i++) {
 2777                         bp = bvec[i];
 2778                         bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG | B_CLUSTEROK);
 2779                         if (retv) {
 2780                                 /*
 2781                                  * Error, leave B_DELWRI intact
 2782                                  */
 2783                                 vfs_unbusy_pages(bp);
 2784                                 brelse(bp);
 2785                         } else {
 2786                                 /*
 2787                                  * Success, remove B_DELWRI ( bundirty() ).
 2788                                  *
 2789                                  * b_dirtyoff/b_dirtyend seem to be NFS
 2790                                  * specific.  We should probably move that
 2791                                  * into bundirty(). XXX
 2792                                  */
 2793                                 s = splbio();
 2794                                 VI_LOCK(vp);
 2795                                 vp->v_numoutput++;
 2796                                 VI_UNLOCK(vp);
 2797                                 bp->b_flags |= B_ASYNC;
 2798                                 bundirty(bp);
 2799                                 bp->b_flags &= ~B_DONE;
 2800                                 bp->b_ioflags &= ~BIO_ERROR;
 2801                                 bp->b_dirtyoff = bp->b_dirtyend = 0;
 2802                                 splx(s);
 2803                                 bufdone(bp);
 2804                         }
 2805                 }
 2806         }
 2807 
 2808         /*
 2809          * Start/do any write(s) that are required.
 2810          */
 2811 loop:
 2812         s = splbio();
 2813         VI_LOCK(vp);
 2814         for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
 2815                 nbp = TAILQ_NEXT(bp, b_vnbufs);
 2816                 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
 2817                         if (waitfor != MNT_WAIT || passone)
 2818                                 continue;
 2819 
 2820                         error = BUF_TIMELOCK(bp,
 2821                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 2822                             VI_MTX(vp), "nfsfsync", slpflag, slptimeo);
 2823                         splx(s);
 2824                         if (error == 0)
 2825                                 panic("nfs_fsync: inconsistent lock");
 2826                         if (error == ENOLCK)
 2827                                 goto loop;
 2828                         error = nfs_sigintr(nmp, NULL, td);
 2829                         if (error)
 2830                                 goto done;
 2831                         if (slpflag == PCATCH) {
 2832                                 slpflag = 0;
 2833                                 slptimeo = 2 * hz;
 2834                         }
 2835                         goto loop;
 2836                 }
 2837                 if ((bp->b_flags & B_DELWRI) == 0)
 2838                         panic("nfs_fsync: not dirty");
 2839                 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
 2840                         BUF_UNLOCK(bp);
 2841                         continue;
 2842                 }
 2843                 VI_UNLOCK(vp);
 2844                 bremfree(bp);
 2845                 if (passone || !commit)
 2846                     bp->b_flags |= B_ASYNC;
 2847                 else
 2848                     bp->b_flags |= B_ASYNC | B_WRITEINPROG;
 2849                 splx(s);
 2850                 bwrite(bp);
 2851                 goto loop;
 2852         }
 2853         splx(s);
 2854         if (passone) {
 2855                 passone = 0;
 2856                 VI_UNLOCK(vp);
 2857                 goto again;
 2858         }
 2859         if (waitfor == MNT_WAIT) {
 2860                 while (vp->v_numoutput) {
 2861                         vp->v_iflag |= VI_BWAIT;
 2862                         error = msleep((caddr_t)&vp->v_numoutput, VI_MTX(vp),
 2863                                 slpflag | (PRIBIO + 1), "nfsfsync", slptimeo);
 2864                         if (error) {
 2865                             VI_UNLOCK(vp);
 2866                             error = nfs_sigintr(nmp, NULL, td);
 2867                             if (error)
 2868                                 goto done;
 2869                             if (slpflag == PCATCH) {
 2870                                 slpflag = 0;
 2871                                 slptimeo = 2 * hz;
 2872                             }
 2873                             VI_LOCK(vp);
 2874                         }
 2875                 }
 2876                 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) && commit) {
 2877                         VI_UNLOCK(vp);
 2878                         goto loop;
 2879                 }
 2880         }
 2881         VI_UNLOCK(vp);
 2882         if (np->n_flag & NWRITEERR) {
 2883                 error = np->n_error;
 2884                 np->n_flag &= ~NWRITEERR;
 2885         }
 2886 done:
 2887         if (bvec != NULL && bvec != bvec_on_stack)
 2888                 free(bvec, M_TEMP);
 2889         return (error);
 2890 }
 2891 
 2892 /*
 2893  * NFS advisory byte-level locks.
 2894  */
 2895 static int
 2896 nfs_advlock(struct vop_advlock_args *ap)
 2897 {
 2898 
 2899         if ((VFSTONFS(ap->a_vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
 2900                 struct nfsnode *np = VTONFS(ap->a_vp);
 2901 
 2902                 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
 2903         }
 2904         return (nfs_dolock(ap));
 2905 }
 2906 
 2907 /*
 2908  * Print out the contents of an nfsnode.
 2909  */
 2910 static int
 2911 nfs_print(struct vop_print_args *ap)
 2912 {
 2913         struct vnode *vp = ap->a_vp;
 2914         struct nfsnode *np = VTONFS(vp);
 2915 
 2916         printf("\tfileid %ld fsid 0x%x",
 2917            np->n_vattr.va_fileid, np->n_vattr.va_fsid);
 2918         if (vp->v_type == VFIFO)
 2919                 fifo_printinfo(vp);
 2920         printf("\n");
 2921         return (0);
 2922 }
 2923 
 2924 /*
 2925  * This is the "real" nfs::bwrite(struct buf*).
 2926  * B_WRITEINPROG isn't set unless the force flag is one and it
 2927  * handles the B_NEEDCOMMIT flag.
 2928  * We set B_CACHE if this is a VMIO buffer.
 2929  */
 2930 int
 2931 nfs_writebp(struct buf *bp, int force, struct thread *td)
 2932 {
 2933         int s;
 2934         int oldflags = bp->b_flags;
 2935 #if 0
 2936         int retv = 1;
 2937         off_t off;
 2938 #endif
 2939 
 2940         if (BUF_REFCNT(bp) == 0)
 2941                 panic("bwrite: buffer is not locked???");
 2942 
 2943         if (bp->b_flags & B_INVAL) {
 2944                 brelse(bp);
 2945                 return(0);
 2946         }
 2947 
 2948         bp->b_flags |= B_CACHE;
 2949 
 2950         /*
 2951          * Undirty the bp.  We will redirty it later if the I/O fails.
 2952          */
 2953 
 2954         s = splbio();
 2955         bundirty(bp);
 2956         bp->b_flags &= ~B_DONE;
 2957         bp->b_ioflags &= ~BIO_ERROR;
 2958         bp->b_iocmd = BIO_WRITE;
 2959 
 2960         VI_LOCK(bp->b_vp);
 2961         bp->b_vp->v_numoutput++;
 2962         VI_UNLOCK(bp->b_vp);
 2963         curthread->td_proc->p_stats->p_ru.ru_oublock++;
 2964         splx(s);
 2965 
 2966         /*
 2967          * Note: to avoid loopback deadlocks, we do not
 2968          * assign b_runningbufspace.
 2969          */
 2970         vfs_busy_pages(bp, 1);
 2971 
 2972         if (force)
 2973                 bp->b_flags |= B_WRITEINPROG;
 2974         BUF_KERNPROC(bp);
 2975         bp->b_iooffset = dbtob(bp->b_blkno);
 2976         VOP_STRATEGY(bp->b_vp, bp);
 2977 
 2978         if( (oldflags & B_ASYNC) == 0) {
 2979                 int rtval = bufwait(bp);
 2980 
 2981                 if (oldflags & B_DELWRI) {
 2982                         s = splbio();
 2983                         reassignbuf(bp);
 2984                         splx(s);
 2985                 }
 2986 
 2987                 brelse(bp);
 2988                 return (rtval);
 2989         }
 2990 
 2991         return (0);
 2992 }
 2993 
 2994 /*
 2995  * nfs special file access vnode op.
 2996  * Essentially just get vattr and then imitate iaccess() since the device is
 2997  * local to the client.
 2998  */
 2999 static int
 3000 nfsspec_access(struct vop_access_args *ap)
 3001 {
 3002         struct vattr *vap;
 3003         struct ucred *cred = ap->a_cred;
 3004         struct vnode *vp = ap->a_vp;
 3005         mode_t mode = ap->a_mode;
 3006         struct vattr vattr;
 3007         int error;
 3008 
 3009         /*
 3010          * Disallow write attempts on filesystems mounted read-only;
 3011          * unless the file is a socket, fifo, or a block or character
 3012          * device resident on the filesystem.
 3013          */
 3014         if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
 3015                 switch (vp->v_type) {
 3016                 case VREG:
 3017                 case VDIR:
 3018                 case VLNK:
 3019                         return (EROFS);
 3020                 default:
 3021                         break;
 3022                 }
 3023         }
 3024         vap = &vattr;
 3025         error = VOP_GETATTR(vp, vap, cred, ap->a_td);
 3026         if (error)
 3027                 return (error);
 3028         return (vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
 3029             mode, cred, NULL));
 3030 }
 3031 
 3032 /*
 3033  * Read wrapper for special devices.
 3034  */
 3035 static int
 3036 nfsspec_read(struct vop_read_args *ap)
 3037 {
 3038         struct nfsnode *np = VTONFS(ap->a_vp);
 3039 
 3040         /*
 3041          * Set access flag.
 3042          */
 3043         np->n_flag |= NACC;
 3044         getnanotime(&np->n_atim);
 3045         return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
 3046 }
 3047 
 3048 /*
 3049  * Write wrapper for special devices.
 3050  */
 3051 static int
 3052 nfsspec_write(struct vop_write_args *ap)
 3053 {
 3054         struct nfsnode *np = VTONFS(ap->a_vp);
 3055 
 3056         /*
 3057          * Set update flag.
 3058          */
 3059         np->n_flag |= NUPD;
 3060         getnanotime(&np->n_mtim);
 3061         return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
 3062 }
 3063 
 3064 /*
 3065  * Close wrapper for special devices.
 3066  *
 3067  * Update the times on the nfsnode then do device close.
 3068  */
 3069 static int
 3070 nfsspec_close(struct vop_close_args *ap)
 3071 {
 3072         struct vnode *vp = ap->a_vp;
 3073         struct nfsnode *np = VTONFS(vp);
 3074         struct vattr vattr;
 3075 
 3076         if (np->n_flag & (NACC | NUPD)) {
 3077                 np->n_flag |= NCHG;
 3078                 if (vrefcnt(vp) == 1 &&
 3079                     (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
 3080                         VATTR_NULL(&vattr);
 3081                         if (np->n_flag & NACC)
 3082                                 vattr.va_atime = np->n_atim;
 3083                         if (np->n_flag & NUPD)
 3084                                 vattr.va_mtime = np->n_mtim;
 3085                         (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
 3086                 }
 3087         }
 3088         return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
 3089 }
 3090 
 3091 /*
 3092  * Read wrapper for fifos.
 3093  */
 3094 static int
 3095 nfsfifo_read(struct vop_read_args *ap)
 3096 {
 3097         struct nfsnode *np = VTONFS(ap->a_vp);
 3098 
 3099         /*
 3100          * Set access flag.
 3101          */
 3102         np->n_flag |= NACC;
 3103         getnanotime(&np->n_atim);
 3104         return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
 3105 }
 3106 
 3107 /*
 3108  * Write wrapper for fifos.
 3109  */
 3110 static int
 3111 nfsfifo_write(struct vop_write_args *ap)
 3112 {
 3113         struct nfsnode *np = VTONFS(ap->a_vp);
 3114 
 3115         /*
 3116          * Set update flag.
 3117          */
 3118         np->n_flag |= NUPD;
 3119         getnanotime(&np->n_mtim);
 3120         return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
 3121 }
 3122 
 3123 /*
 3124  * Close wrapper for fifos.
 3125  *
 3126  * Update the times on the nfsnode then do fifo close.
 3127  */
 3128 static int
 3129 nfsfifo_close(struct vop_close_args *ap)
 3130 {
 3131         struct vnode *vp = ap->a_vp;
 3132         struct nfsnode *np = VTONFS(vp);
 3133         struct vattr vattr;
 3134         struct timespec ts;
 3135 
 3136         if (np->n_flag & (NACC | NUPD)) {
 3137                 getnanotime(&ts);
 3138                 if (np->n_flag & NACC)
 3139                         np->n_atim = ts;
 3140                 if (np->n_flag & NUPD)
 3141                         np->n_mtim = ts;
 3142                 np->n_flag |= NCHG;
 3143                 if (vrefcnt(vp) == 1 &&
 3144                     (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
 3145                         VATTR_NULL(&vattr);
 3146                         if (np->n_flag & NACC)
 3147                                 vattr.va_atime = np->n_atim;
 3148                         if (np->n_flag & NUPD)
 3149                                 vattr.va_mtime = np->n_mtim;
 3150                         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, ap->a_td);
 3151                         (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
 3152                         VOP_UNLOCK(vp, 0, ap->a_td);
 3153                 }
 3154         }
 3155         return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
 3156 }
 3157 

Cache object: eb3a09f6ec6cb85f106696ed54ee63c2


[ 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.