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

    1 /*-
    2  * Copyright (c) 1989, 1993
    3  *      The Regents of the University of California.  All rights reserved.
    4  *
    5  * This code is derived from software contributed to Berkeley by
    6  * Rick Macklem at The University of Guelph.
    7  *
    8  * Redistribution and use in source and binary forms, with or without
    9  * modification, are permitted provided that the following conditions
   10  * are met:
   11  * 1. Redistributions of source code must retain the above copyright
   12  *    notice, this list of conditions and the following disclaimer.
   13  * 2. Redistributions in binary form must reproduce the above copyright
   14  *    notice, this list of conditions and the following disclaimer in the
   15  *    documentation and/or other materials provided with the distribution.
   16  * 4. Neither the name of the University nor the names of its contributors
   17  *    may be used to endorse or promote products derived from this software
   18  *    without specific prior written permission.
   19  *
   20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   30  * SUCH DAMAGE.
   31  *
   32  *      @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
   33  */
   34 
   35 #include <sys/cdefs.h>
   36 __FBSDID("$FreeBSD$");
   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 #include <sys/signalvar.h>
   63 
   64 #include <vm/vm.h>
   65 #include <vm/vm_object.h>
   66 #include <vm/vm_extern.h>
   67 #include <vm/vm_object.h>
   68 
   69 #include <fs/fifofs/fifo.h>
   70 
   71 #include <rpc/rpcclnt.h>
   72 
   73 #include <nfs/rpcv2.h>
   74 #include <nfs/nfsproto.h>
   75 #include <nfsclient/nfs.h>
   76 #include <nfsclient/nfsnode.h>
   77 #include <nfsclient/nfsmount.h>
   78 #include <nfsclient/nfs_lock.h>
   79 #include <nfs/xdr_subs.h>
   80 #include <nfsclient/nfsm_subs.h>
   81 
   82 #include <net/if.h>
   83 #include <netinet/in.h>
   84 #include <netinet/in_var.h>
   85 
   86 /* Defs */
   87 #define TRUE    1
   88 #define FALSE   0
   89 
   90 /*
   91  * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
   92  * calls are not in getblk() and brelse() so that they would not be necessary
   93  * here.
   94  */
   95 #ifndef B_VMIO
   96 #define vfs_busy_pages(bp, f)
   97 #endif
   98 
   99 static vop_read_t       nfsfifo_read;
  100 static vop_write_t      nfsfifo_write;
  101 static vop_close_t      nfsfifo_close;
  102 static int      nfs_flush(struct vnode *, int, struct thread *,
  103                     int);
  104 static int      nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
  105                     struct thread *);
  106 static vop_lookup_t     nfs_lookup;
  107 static vop_create_t     nfs_create;
  108 static vop_mknod_t      nfs_mknod;
  109 static vop_open_t       nfs_open;
  110 static vop_close_t      nfs_close;
  111 static vop_access_t     nfs_access;
  112 static vop_getattr_t    nfs_getattr;
  113 static vop_setattr_t    nfs_setattr;
  114 static vop_read_t       nfs_read;
  115 static vop_fsync_t      nfs_fsync;
  116 static vop_remove_t     nfs_remove;
  117 static vop_link_t       nfs_link;
  118 static vop_rename_t     nfs_rename;
  119 static vop_mkdir_t      nfs_mkdir;
  120 static vop_rmdir_t      nfs_rmdir;
  121 static vop_symlink_t    nfs_symlink;
  122 static vop_readdir_t    nfs_readdir;
  123 static vop_strategy_t   nfs_strategy;
  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 vop_access_t     nfsspec_access;
  129 static vop_readlink_t   nfs_readlink;
  130 static vop_print_t      nfs_print;
  131 static vop_advlock_t    nfs_advlock;
  132 static vop_advlockasync_t nfs_advlockasync;
  133 
  134 /*
  135  * Global vfs data structures for nfs
  136  */
  137 struct vop_vector nfs_vnodeops = {
  138         .vop_default =          &default_vnodeops,
  139         .vop_access =           nfs_access,
  140         .vop_advlock =          nfs_advlock,
  141         .vop_advlockasync =     nfs_advlockasync,
  142         .vop_close =            nfs_close,
  143         .vop_create =           nfs_create,
  144         .vop_fsync =            nfs_fsync,
  145         .vop_getattr =          nfs_getattr,
  146         .vop_getpages =         nfs_getpages,
  147         .vop_putpages =         nfs_putpages,
  148         .vop_inactive =         nfs_inactive,
  149         .vop_lease =            VOP_NULL,
  150         .vop_link =             nfs_link,
  151         .vop_lookup =           nfs_lookup,
  152         .vop_mkdir =            nfs_mkdir,
  153         .vop_mknod =            nfs_mknod,
  154         .vop_open =             nfs_open,
  155         .vop_print =            nfs_print,
  156         .vop_read =             nfs_read,
  157         .vop_readdir =          nfs_readdir,
  158         .vop_readlink =         nfs_readlink,
  159         .vop_reclaim =          nfs_reclaim,
  160         .vop_remove =           nfs_remove,
  161         .vop_rename =           nfs_rename,
  162         .vop_rmdir =            nfs_rmdir,
  163         .vop_setattr =          nfs_setattr,
  164         .vop_strategy =         nfs_strategy,
  165         .vop_symlink =          nfs_symlink,
  166         .vop_write =            nfs_write,
  167 };
  168 
  169 struct vop_vector nfs_fifoops = {
  170         .vop_default =          &fifo_specops,
  171         .vop_access =           nfsspec_access,
  172         .vop_close =            nfsfifo_close,
  173         .vop_fsync =            nfs_fsync,
  174         .vop_getattr =          nfs_getattr,
  175         .vop_inactive =         nfs_inactive,
  176         .vop_print =            nfs_print,
  177         .vop_read =             nfsfifo_read,
  178         .vop_reclaim =          nfs_reclaim,
  179         .vop_setattr =          nfs_setattr,
  180         .vop_write =            nfsfifo_write,
  181 };
  182 
  183 static int      nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
  184                              struct componentname *cnp, struct vattr *vap);
  185 static int      nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
  186                               struct ucred *cred, struct thread *td);
  187 static int      nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
  188                               int fnamelen, struct vnode *tdvp,
  189                               const char *tnameptr, int tnamelen,
  190                               struct ucred *cred, struct thread *td);
  191 static int      nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
  192                              struct sillyrename *sp);
  193 
  194 /*
  195  * Global variables
  196  */
  197 struct proc     *nfs_iodwant[NFS_MAXASYNCDAEMON];
  198 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
  199 int              nfs_numasync = 0;
  200 #define DIRHDSIZ        (sizeof (struct dirent) - (MAXNAMLEN + 1))
  201 
  202 SYSCTL_DECL(_vfs_nfs);
  203 
  204 static int      nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
  205 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
  206            &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
  207 
  208 static int      nfs_prime_access_cache = 1;
  209 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
  210            &nfs_prime_access_cache, 0,
  211            "Prime NFS ACCESS cache when fetching attributes");
  212 
  213 static int      nfsv3_commit_on_close = 0;
  214 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
  215            &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
  216 
  217 static int      nfs_clean_pages_on_close = 1;
  218 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
  219            &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
  220 
  221 int nfs_directio_enable = 0;
  222 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
  223            &nfs_directio_enable, 0, "Enable NFS directio");
  224 
  225 /*
  226  * This sysctl allows other processes to mmap a file that has been opened
  227  * O_DIRECT by a process.  In general, having processes mmap the file while
  228  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
  229  * this by default to prevent DoS attacks - to prevent a malicious user from
  230  * opening up files O_DIRECT preventing other users from mmap'ing these
  231  * files.  "Protected" environments where stricter consistency guarantees are
  232  * required can disable this knob.  The process that opened the file O_DIRECT
  233  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
  234  * meaningful.
  235  */
  236 int nfs_directio_allow_mmap = 1;
  237 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
  238            &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
  239 
  240 #if 0
  241 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
  242            &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
  243 
  244 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
  245            &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
  246 #endif
  247 
  248 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY          \
  249                          | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE     \
  250                          | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
  251 static int
  252 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
  253     struct ucred *cred)
  254 {
  255         const int v3 = 1;
  256         u_int32_t *tl;
  257         int error = 0, attrflag;
  258 
  259         struct mbuf *mreq, *mrep, *md, *mb;
  260         caddr_t bpos, dpos;
  261         u_int32_t rmode;
  262         struct nfsnode *np = VTONFS(vp);
  263 
  264         nfsstats.rpccnt[NFSPROC_ACCESS]++;
  265         mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
  266         mb = mreq;
  267         bpos = mtod(mb, caddr_t);
  268         nfsm_fhtom(vp, v3);
  269         tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
  270         *tl = txdr_unsigned(wmode);
  271         nfsm_request(vp, NFSPROC_ACCESS, td, cred);
  272         nfsm_postop_attr(vp, attrflag);
  273         if (!error) {
  274                 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
  275                 rmode = fxdr_unsigned(u_int32_t, *tl);
  276                 np->n_mode = rmode;
  277                 np->n_modeuid = cred->cr_uid;
  278                 np->n_modestamp = time_second;
  279         }
  280         m_freem(mrep);
  281 nfsmout:
  282         return (error);
  283 }
  284 
  285 /*
  286  * nfs access vnode op.
  287  * For nfs version 2, just return ok. File accesses may fail later.
  288  * For nfs version 3, use the access rpc to check accessibility. If file modes
  289  * are changed on the server, accesses might still fail later.
  290  */
  291 static int
  292 nfs_access(struct vop_access_args *ap)
  293 {
  294         struct vnode *vp = ap->a_vp;
  295         int error = 0;
  296         u_int32_t mode, wmode;
  297         int v3 = NFS_ISV3(vp);
  298         struct nfsnode *np = VTONFS(vp);
  299 
  300         /*
  301          * Disallow write attempts on filesystems mounted read-only;
  302          * unless the file is a socket, fifo, or a block or character
  303          * device resident on the filesystem.
  304          */
  305         if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
  306                 switch (vp->v_type) {
  307                 case VREG:
  308                 case VDIR:
  309                 case VLNK:
  310                         return (EROFS);
  311                 default:
  312                         break;
  313                 }
  314         }
  315         /*
  316          * For nfs v3, check to see if we have done this recently, and if
  317          * so return our cached result instead of making an ACCESS call.
  318          * If not, do an access rpc, otherwise you are stuck emulating
  319          * ufs_access() locally using the vattr. This may not be correct,
  320          * since the server may apply other access criteria such as
  321          * client uid-->server uid mapping that we do not know about.
  322          */
  323         if (v3) {
  324                 if (ap->a_mode & VREAD)
  325                         mode = NFSV3ACCESS_READ;
  326                 else
  327                         mode = 0;
  328                 if (vp->v_type != VDIR) {
  329                         if (ap->a_mode & VWRITE)
  330                                 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
  331                         if (ap->a_mode & VEXEC)
  332                                 mode |= NFSV3ACCESS_EXECUTE;
  333                 } else {
  334                         if (ap->a_mode & VWRITE)
  335                                 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
  336                                          NFSV3ACCESS_DELETE);
  337                         if (ap->a_mode & VEXEC)
  338                                 mode |= NFSV3ACCESS_LOOKUP;
  339                 }
  340                 /* XXX safety belt, only make blanket request if caching */
  341                 if (nfsaccess_cache_timeout > 0) {
  342                         wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
  343                                 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
  344                                 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
  345                 } else {
  346                         wmode = mode;
  347                 }
  348 
  349                 /*
  350                  * Does our cached result allow us to give a definite yes to
  351                  * this request?
  352                  */
  353                 if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
  354                     (ap->a_cred->cr_uid == np->n_modeuid) &&
  355                     ((np->n_mode & mode) == mode)) {
  356                         nfsstats.accesscache_hits++;
  357                 } else {
  358                         /*
  359                          * Either a no, or a don't know.  Go to the wire.
  360                          */
  361                         nfsstats.accesscache_misses++;
  362                         error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
  363                         if (!error) {
  364                                 if ((np->n_mode & mode) != mode) {
  365                                         error = EACCES;
  366                                 }
  367                         }
  368                 }
  369                 return (error);
  370         } else {
  371                 if ((error = nfsspec_access(ap)) != 0)
  372                         return (error);
  373 
  374                 /*
  375                  * Attempt to prevent a mapped root from accessing a file
  376                  * which it shouldn't.  We try to read a byte from the file
  377                  * if the user is root and the file is not zero length.
  378                  * After calling nfsspec_access, we should have the correct
  379                  * file size cached.
  380                  */
  381                 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
  382                     && VTONFS(vp)->n_size > 0) {
  383                         struct iovec aiov;
  384                         struct uio auio;
  385                         char buf[1];
  386 
  387                         aiov.iov_base = buf;
  388                         aiov.iov_len = 1;
  389                         auio.uio_iov = &aiov;
  390                         auio.uio_iovcnt = 1;
  391                         auio.uio_offset = 0;
  392                         auio.uio_resid = 1;
  393                         auio.uio_segflg = UIO_SYSSPACE;
  394                         auio.uio_rw = UIO_READ;
  395                         auio.uio_td = ap->a_td;
  396 
  397                         if (vp->v_type == VREG)
  398                                 error = nfs_readrpc(vp, &auio, ap->a_cred);
  399                         else if (vp->v_type == VDIR) {
  400                                 char* bp;
  401                                 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
  402                                 aiov.iov_base = bp;
  403                                 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
  404                                 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
  405                                 free(bp, M_TEMP);
  406                         } else if (vp->v_type == VLNK)
  407                                 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
  408                         else
  409                                 error = EACCES;
  410                 }
  411                 return (error);
  412         }
  413 }
  414 
  415 /*
  416  * nfs open vnode op
  417  * Check to see if the type is ok
  418  * and that deletion is not in progress.
  419  * For paged in text files, you will need to flush the page cache
  420  * if consistency is lost.
  421  */
  422 /* ARGSUSED */
  423 static int
  424 nfs_open(struct vop_open_args *ap)
  425 {
  426         struct vnode *vp = ap->a_vp;
  427         struct nfsnode *np = VTONFS(vp);
  428         struct vattr vattr;
  429         int error;
  430         int fmode = ap->a_mode;
  431 
  432         if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
  433                 return (EOPNOTSUPP);
  434 
  435         /*
  436          * Get a valid lease. If cached data is stale, flush it.
  437          */
  438         if (np->n_flag & NMODIFIED) {
  439                 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  440                 if (error == EINTR || error == EIO)
  441                         return (error);
  442                 np->n_attrstamp = 0;
  443                 if (vp->v_type == VDIR)
  444                         np->n_direofoffset = 0;
  445                 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
  446                 if (error)
  447                         return (error);
  448                 np->n_mtime = vattr.va_mtime;
  449         } else {
  450                 np->n_attrstamp = 0;
  451                 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
  452                 if (error)
  453                         return (error);
  454                 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
  455                         if (vp->v_type == VDIR)
  456                                 np->n_direofoffset = 0;
  457                         error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  458                         if (error == EINTR || error == EIO)
  459                                 return (error);
  460                         np->n_mtime = vattr.va_mtime;
  461                 }
  462         }
  463         /*
  464          * If the object has >= 1 O_DIRECT active opens, we disable caching.
  465          */
  466         if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
  467                 if (np->n_directio_opens == 0) {
  468                         error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  469                         if (error)
  470                                 return (error);
  471                         np->n_flag |= NNONCACHE;
  472                 }
  473                 np->n_directio_opens++;
  474         }
  475         np->ra_expect_lbn = 0;
  476         vnode_create_vobject_off(vp, vattr.va_size, ap->a_td);
  477         return (0);
  478 }
  479 
  480 /*
  481  * nfs close vnode op
  482  * What an NFS client should do upon close after writing is a debatable issue.
  483  * Most NFS clients push delayed writes to the server upon close, basically for
  484  * two reasons:
  485  * 1 - So that any write errors may be reported back to the client process
  486  *     doing the close system call. By far the two most likely errors are
  487  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
  488  * 2 - To put a worst case upper bound on cache inconsistency between
  489  *     multiple clients for the file.
  490  * There is also a consistency problem for Version 2 of the protocol w.r.t.
  491  * not being able to tell if other clients are writing a file concurrently,
  492  * since there is no way of knowing if the changed modify time in the reply
  493  * is only due to the write for this client.
  494  * (NFS Version 3 provides weak cache consistency data in the reply that
  495  *  should be sufficient to detect and handle this case.)
  496  *
  497  * The current code does the following:
  498  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
  499  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
  500  *                     or commit them (this satisfies 1 and 2 except for the
  501  *                     case where the server crashes after this close but
  502  *                     before the commit RPC, which is felt to be "good
  503  *                     enough". Changing the last argument to nfs_flush() to
  504  *                     a 1 would force a commit operation, if it is felt a
  505  *                     commit is necessary now.
  506  */
  507 /* ARGSUSED */
  508 static int
  509 nfs_close(struct vop_close_args *ap)
  510 {
  511         struct vnode *vp = ap->a_vp;
  512         struct nfsnode *np = VTONFS(vp);
  513         int error = 0;
  514         int fmode = ap->a_fflag;
  515 
  516         if (vp->v_type == VREG) {
  517             /*
  518              * Examine and clean dirty pages, regardless of NMODIFIED.
  519              * This closes a major hole in close-to-open consistency.
  520              * We want to push out all dirty pages (and buffers) on
  521              * close, regardless of whether they were dirtied by
  522              * mmap'ed writes or via write().
  523              */
  524             if (nfs_clean_pages_on_close && vp->v_object) {
  525                 VM_OBJECT_LOCK(vp->v_object);
  526                 vm_object_page_clean(vp->v_object, 0, 0, 0);
  527                 VM_OBJECT_UNLOCK(vp->v_object);
  528             }
  529             if (np->n_flag & NMODIFIED) {
  530                 if (NFS_ISV3(vp)) {
  531                     /*
  532                      * Under NFSv3 we have dirty buffers to dispose of.  We
  533                      * must flush them to the NFS server.  We have the option
  534                      * of waiting all the way through the commit rpc or just
  535                      * waiting for the initial write.  The default is to only
  536                      * wait through the initial write so the data is in the
  537                      * server's cache, which is roughly similar to the state
  538                      * a standard disk subsystem leaves the file in on close().
  539                      *
  540                      * We cannot clear the NMODIFIED bit in np->n_flag due to
  541                      * potential races with other processes, and certainly
  542                      * cannot clear it if we don't commit.
  543                      */
  544                     int cm = nfsv3_commit_on_close ? 1 : 0;
  545                     error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm);
  546                     /* np->n_flag &= ~NMODIFIED; */
  547                 } else
  548                     error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  549             }
  550             if (np->n_flag & NWRITEERR) {
  551                 np->n_flag &= ~NWRITEERR;
  552                 error = np->n_error;
  553             }
  554         }
  555         if (nfs_directio_enable)
  556                 KASSERT((np->n_directio_asyncwr == 0),
  557                         ("nfs_close: dirty unflushed (%d) directio buffers\n",
  558                          np->n_directio_asyncwr));
  559         if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
  560                 KASSERT((np->n_directio_opens > 0), 
  561                         ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));           
  562                 np->n_directio_opens--;
  563                 if (np->n_directio_opens == 0)
  564                         np->n_flag &= ~NNONCACHE;
  565         }
  566         return (error);
  567 }
  568 
  569 /*
  570  * nfs getattr call from vfs.
  571  */
  572 static int
  573 nfs_getattr(struct vop_getattr_args *ap)
  574 {
  575         struct vnode *vp = ap->a_vp;
  576         struct nfsnode *np = VTONFS(vp);
  577         caddr_t bpos, dpos;
  578         int error = 0;
  579         struct mbuf *mreq, *mrep, *md, *mb;
  580         int v3 = NFS_ISV3(vp);
  581 
  582         /*
  583          * Update local times for special files.
  584          */
  585         if (np->n_flag & (NACC | NUPD))
  586                 np->n_flag |= NCHG;
  587         /*
  588          * First look in the cache.
  589          */
  590         if (nfs_getattrcache(vp, ap->a_vap) == 0)
  591                 return (0);
  592 
  593         if (v3 && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) {
  594                 nfsstats.accesscache_misses++;
  595                 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, ap->a_cred);
  596                 if (nfs_getattrcache(vp, ap->a_vap) == 0)
  597                         return (0);
  598         }
  599 
  600         nfsstats.rpccnt[NFSPROC_GETATTR]++;
  601         mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
  602         mb = mreq;
  603         bpos = mtod(mb, caddr_t);
  604         nfsm_fhtom(vp, v3);
  605         nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, ap->a_cred);
  606         if (!error) {
  607                 nfsm_loadattr(vp, ap->a_vap);
  608         }
  609         m_freem(mrep);
  610 nfsmout:
  611         return (error);
  612 }
  613 
  614 /*
  615  * nfs setattr call.
  616  */
  617 static int
  618 nfs_setattr(struct vop_setattr_args *ap)
  619 {
  620         struct vnode *vp = ap->a_vp;
  621         struct nfsnode *np = VTONFS(vp);
  622         struct vattr *vap = ap->a_vap;
  623         int error = 0;
  624         u_quad_t tsize;
  625 
  626 #ifndef nolint
  627         tsize = (u_quad_t)0;
  628 #endif
  629 
  630         /*
  631          * Setting of flags and marking of atimes are not supported.
  632          */
  633         if (vap->va_flags != VNOVAL || (vap->va_vaflags & VA_MARK_ATIME))
  634                 return (EOPNOTSUPP);
  635 
  636         /*
  637          * Disallow write attempts if the filesystem is mounted read-only.
  638          */
  639         if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
  640             vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
  641             vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
  642             (vp->v_mount->mnt_flag & MNT_RDONLY))
  643                 return (EROFS);
  644         if (vap->va_size != VNOVAL) {
  645                 switch (vp->v_type) {
  646                 case VDIR:
  647                         return (EISDIR);
  648                 case VCHR:
  649                 case VBLK:
  650                 case VSOCK:
  651                 case VFIFO:
  652                         if (vap->va_mtime.tv_sec == VNOVAL &&
  653                             vap->va_atime.tv_sec == VNOVAL &&
  654                             vap->va_mode == (mode_t)VNOVAL &&
  655                             vap->va_uid == (uid_t)VNOVAL &&
  656                             vap->va_gid == (gid_t)VNOVAL)
  657                                 return (0);
  658                         vap->va_size = VNOVAL;
  659                         break;
  660                 default:
  661                         /*
  662                          * Disallow write attempts if the filesystem is
  663                          * mounted read-only.
  664                          */
  665                         if (vp->v_mount->mnt_flag & MNT_RDONLY)
  666                                 return (EROFS);
  667 
  668                         /*
  669                          *  We run vnode_pager_setsize() early (why?),
  670                          * we must set np->n_size now to avoid vinvalbuf
  671                          * V_SAVE races that might setsize a lower
  672                          * value.
  673                          */
  674 
  675                         tsize = np->n_size;
  676                         error = nfs_meta_setsize(vp, ap->a_cred, 
  677                                                 ap->a_td, vap->va_size);
  678 
  679                         if (np->n_flag & NMODIFIED) {
  680                             if (vap->va_size == 0)
  681                                 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1);
  682                             else
  683                                 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  684                             if (error) {
  685                                 vnode_pager_setsize(vp, np->n_size);
  686                                 return (error);
  687                             }
  688                         }
  689                         /*
  690                          * np->n_size has already been set to vap->va_size
  691                          * in nfs_meta_setsize(). We must set it again since
  692                          * nfs_loadattrcache() could be called through
  693                          * nfs_meta_setsize() and could modify np->n_size.
  694                          */
  695                         np->n_vattr.va_size = np->n_size = vap->va_size;
  696                 };
  697         } else if ((vap->va_mtime.tv_sec != VNOVAL ||
  698                 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) &&
  699                 vp->v_type == VREG &&
  700                 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) != 0 &&
  701                     (error == EINTR || error == EIO))
  702                 return (error);
  703         error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
  704         if (error && vap->va_size != VNOVAL) {
  705                 np->n_size = np->n_vattr.va_size = tsize;
  706                 vnode_pager_setsize(vp, np->n_size);
  707         }
  708         return (error);
  709 }
  710 
  711 /*
  712  * Do an nfs setattr rpc.
  713  */
  714 static int
  715 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
  716     struct thread *td)
  717 {
  718         struct nfsv2_sattr *sp;
  719         struct nfsnode *np = VTONFS(vp);
  720         caddr_t bpos, dpos;
  721         u_int32_t *tl;
  722         int error = 0, wccflag = NFSV3_WCCRATTR;
  723         struct mbuf *mreq, *mrep, *md, *mb;
  724         int v3 = NFS_ISV3(vp);
  725 
  726         nfsstats.rpccnt[NFSPROC_SETATTR]++;
  727         mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
  728         mb = mreq;
  729         bpos = mtod(mb, caddr_t);
  730         nfsm_fhtom(vp, v3);
  731         if (v3) {
  732                 nfsm_v3attrbuild(vap, TRUE);
  733                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
  734                 *tl = nfs_false;
  735         } else {
  736                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
  737                 if (vap->va_mode == (mode_t)VNOVAL)
  738                         sp->sa_mode = nfs_xdrneg1;
  739                 else
  740                         sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
  741                 if (vap->va_uid == (uid_t)VNOVAL)
  742                         sp->sa_uid = nfs_xdrneg1;
  743                 else
  744                         sp->sa_uid = txdr_unsigned(vap->va_uid);
  745                 if (vap->va_gid == (gid_t)VNOVAL)
  746                         sp->sa_gid = nfs_xdrneg1;
  747                 else
  748                         sp->sa_gid = txdr_unsigned(vap->va_gid);
  749                 sp->sa_size = txdr_unsigned(vap->va_size);
  750                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
  751                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
  752         }
  753         nfsm_request(vp, NFSPROC_SETATTR, td, cred);
  754         if (v3) {
  755                 np->n_modestamp = 0;
  756                 nfsm_wcc_data(vp, wccflag);
  757         } else
  758                 nfsm_loadattr(vp, NULL);
  759         m_freem(mrep);
  760 nfsmout:
  761         return (error);
  762 }
  763 
  764 /*
  765  * nfs lookup call, one step at a time...
  766  * First look in cache
  767  * If not found, unlock the directory nfsnode and do the rpc
  768  */
  769 static int
  770 nfs_lookup(struct vop_lookup_args *ap)
  771 {
  772         struct componentname *cnp = ap->a_cnp;
  773         struct vnode *dvp = ap->a_dvp;
  774         struct vnode **vpp = ap->a_vpp;
  775         int flags = cnp->cn_flags;
  776         struct vnode *newvp;
  777         struct nfsmount *nmp;
  778         caddr_t bpos, dpos;
  779         struct mbuf *mreq, *mrep, *md, *mb;
  780         long len;
  781         nfsfh_t *fhp;
  782         struct nfsnode *np;
  783         int error = 0, attrflag, fhsize;
  784         int v3 = NFS_ISV3(dvp);
  785         struct thread *td = cnp->cn_thread;
  786 
  787         *vpp = NULLVP;
  788         if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
  789             (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
  790                 return (EROFS);
  791         if (dvp->v_type != VDIR)
  792                 return (ENOTDIR);
  793         nmp = VFSTONFS(dvp->v_mount);
  794         np = VTONFS(dvp);
  795         if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
  796                 *vpp = NULLVP;
  797                 return (error);
  798         }
  799         if ((error = cache_lookup(dvp, vpp, cnp)) && error != ENOENT) {
  800                 struct vattr vattr;
  801 
  802                 newvp = *vpp;
  803                 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, td)
  804                  && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
  805                      nfsstats.lookupcache_hits++;
  806                      if (cnp->cn_nameiop != LOOKUP &&
  807                          (flags & ISLASTCN))
  808                              cnp->cn_flags |= SAVENAME;
  809                      return (0);
  810                 }
  811                 cache_purge(newvp);
  812                 if (dvp != newvp)
  813                         vput(newvp);
  814                 else 
  815                         vrele(newvp);
  816                 *vpp = NULLVP;
  817         }
  818         error = 0;
  819         newvp = NULLVP;
  820         nfsstats.lookupcache_misses++;
  821         nfsstats.rpccnt[NFSPROC_LOOKUP]++;
  822         len = cnp->cn_namelen;
  823         mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
  824                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
  825         mb = mreq;
  826         bpos = mtod(mb, caddr_t);
  827         nfsm_fhtom(dvp, v3);
  828         nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
  829         nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
  830         if (error) {
  831                 if (v3) {
  832                         nfsm_postop_attr(dvp, attrflag);
  833                         m_freem(mrep);
  834                 }
  835                 goto nfsmout;
  836         }
  837         nfsm_getfh(fhp, fhsize, v3);
  838 
  839         /*
  840          * Handle RENAME case...
  841          */
  842         if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
  843                 if (NFS_CMPFH(np, fhp, fhsize)) {
  844                         m_freem(mrep);
  845                         return (EISDIR);
  846                 }
  847                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, LK_EXCLUSIVE);
  848                 if (error) {
  849                         m_freem(mrep);
  850                         return (error);
  851                 }
  852                 newvp = NFSTOV(np);
  853                 if (v3) {
  854                         nfsm_postop_attr(newvp, attrflag);
  855                         nfsm_postop_attr(dvp, attrflag);
  856                 } else
  857                         nfsm_loadattr(newvp, NULL);
  858                 *vpp = newvp;
  859                 m_freem(mrep);
  860                 cnp->cn_flags |= SAVENAME;
  861                 return (0);
  862         }
  863 
  864         if (flags & ISDOTDOT) {
  865                 VOP_UNLOCK(dvp, 0, td);
  866                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, cnp->cn_lkflags);
  867                 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
  868                 if (error)
  869                         return (error);
  870                 newvp = NFSTOV(np);
  871         } else if (NFS_CMPFH(np, fhp, fhsize)) {
  872                 VREF(dvp);
  873                 newvp = dvp;
  874         } else {
  875                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, cnp->cn_lkflags);
  876                 if (error) {
  877                         m_freem(mrep);
  878                         return (error);
  879                 }
  880                 newvp = NFSTOV(np);
  881         }
  882         if (v3) {
  883                 nfsm_postop_attr(newvp, attrflag);
  884                 nfsm_postop_attr(dvp, attrflag);
  885         } else
  886                 nfsm_loadattr(newvp, NULL);
  887         if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
  888                 cnp->cn_flags |= SAVENAME;
  889         if ((cnp->cn_flags & MAKEENTRY) &&
  890             (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
  891                 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
  892                 cache_enter(dvp, newvp, cnp);
  893         }
  894         *vpp = newvp;
  895         m_freem(mrep);
  896 nfsmout:
  897         if (error) {
  898                 if (newvp != NULLVP) {
  899                         vput(newvp);
  900                         *vpp = NULLVP;
  901                 }
  902                 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
  903                     (flags & ISLASTCN) && error == ENOENT) {
  904                         if (dvp->v_mount->mnt_flag & MNT_RDONLY)
  905                                 error = EROFS;
  906                         else
  907                                 error = EJUSTRETURN;
  908                 }
  909                 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
  910                         cnp->cn_flags |= SAVENAME;
  911         }
  912         return (error);
  913 }
  914 
  915 /*
  916  * nfs read call.
  917  * Just call nfs_bioread() to do the work.
  918  */
  919 static int
  920 nfs_read(struct vop_read_args *ap)
  921 {
  922         struct vnode *vp = ap->a_vp;
  923 
  924         switch (vp->v_type) {
  925         case VREG:
  926                 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
  927         case VDIR:
  928                 return (EISDIR);
  929         default:
  930                 return (EOPNOTSUPP);
  931         }
  932 }
  933 
  934 /*
  935  * nfs readlink call
  936  */
  937 static int
  938 nfs_readlink(struct vop_readlink_args *ap)
  939 {
  940         struct vnode *vp = ap->a_vp;
  941 
  942         if (vp->v_type != VLNK)
  943                 return (EINVAL);
  944         return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
  945 }
  946 
  947 /*
  948  * Do a readlink rpc.
  949  * Called by nfs_doio() from below the buffer cache.
  950  */
  951 int
  952 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
  953 {
  954         caddr_t bpos, dpos;
  955         int error = 0, len, attrflag;
  956         struct mbuf *mreq, *mrep, *md, *mb;
  957         int v3 = NFS_ISV3(vp);
  958 
  959         nfsstats.rpccnt[NFSPROC_READLINK]++;
  960         mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
  961         mb = mreq;
  962         bpos = mtod(mb, caddr_t);
  963         nfsm_fhtom(vp, v3);
  964         nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
  965         if (v3)
  966                 nfsm_postop_attr(vp, attrflag);
  967         if (!error) {
  968                 nfsm_strsiz(len, NFS_MAXPATHLEN);
  969                 if (len == NFS_MAXPATHLEN) {
  970                         struct nfsnode *np = VTONFS(vp);
  971                         if (np->n_size && np->n_size < NFS_MAXPATHLEN)
  972                                 len = np->n_size;
  973                 }
  974                 nfsm_mtouio(uiop, len);
  975         }
  976         m_freem(mrep);
  977 nfsmout:
  978         return (error);
  979 }
  980 
  981 /*
  982  * nfs read rpc call
  983  * Ditto above
  984  */
  985 int
  986 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
  987 {
  988         u_int32_t *tl;
  989         caddr_t bpos, dpos;
  990         struct mbuf *mreq, *mrep, *md, *mb;
  991         struct nfsmount *nmp;
  992         int error = 0, len, retlen, tsiz, eof, attrflag;
  993         int v3 = NFS_ISV3(vp);
  994 
  995 #ifndef nolint
  996         eof = 0;
  997 #endif
  998         nmp = VFSTONFS(vp->v_mount);
  999         tsiz = uiop->uio_resid;
 1000         if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
 1001                 return (EFBIG);
 1002         while (tsiz > 0) {
 1003                 nfsstats.rpccnt[NFSPROC_READ]++;
 1004                 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
 1005                 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
 1006                 mb = mreq;
 1007                 bpos = mtod(mb, caddr_t);
 1008                 nfsm_fhtom(vp, v3);
 1009                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
 1010                 if (v3) {
 1011                         txdr_hyper(uiop->uio_offset, tl);
 1012                         *(tl + 2) = txdr_unsigned(len);
 1013                 } else {
 1014                         *tl++ = txdr_unsigned(uiop->uio_offset);
 1015                         *tl++ = txdr_unsigned(len);
 1016                         *tl = 0;
 1017                 }
 1018                 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
 1019                 if (v3) {
 1020                         nfsm_postop_attr(vp, attrflag);
 1021                         if (error) {
 1022                                 m_freem(mrep);
 1023                                 goto nfsmout;
 1024                         }
 1025                         tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
 1026                         eof = fxdr_unsigned(int, *(tl + 1));
 1027                 } else
 1028                         nfsm_loadattr(vp, NULL);
 1029                 nfsm_strsiz(retlen, nmp->nm_rsize);
 1030                 nfsm_mtouio(uiop, retlen);
 1031                 m_freem(mrep);
 1032                 tsiz -= retlen;
 1033                 if (v3) {
 1034                         if (eof || retlen == 0) {
 1035                                 tsiz = 0;
 1036                         }
 1037                 } else if (retlen < len) {
 1038                         tsiz = 0;
 1039                 }
 1040         }
 1041 nfsmout:
 1042         return (error);
 1043 }
 1044 
 1045 /*
 1046  * nfs write call
 1047  */
 1048 int
 1049 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
 1050     int *iomode, int *must_commit)
 1051 {
 1052         u_int32_t *tl;
 1053         int32_t backup;
 1054         caddr_t bpos, dpos;
 1055         struct mbuf *mreq, *mrep, *md, *mb;
 1056         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 1057         int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
 1058         int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
 1059 
 1060 #ifndef DIAGNOSTIC
 1061         if (uiop->uio_iovcnt != 1)
 1062                 panic("nfs: writerpc iovcnt > 1");
 1063 #endif
 1064         *must_commit = 0;
 1065         tsiz = uiop->uio_resid;
 1066         if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
 1067                 return (EFBIG);
 1068         while (tsiz > 0) {
 1069                 nfsstats.rpccnt[NFSPROC_WRITE]++;
 1070                 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
 1071                 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
 1072                         NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
 1073                 mb = mreq;
 1074                 bpos = mtod(mb, caddr_t);
 1075                 nfsm_fhtom(vp, v3);
 1076                 if (v3) {
 1077                         tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
 1078                         txdr_hyper(uiop->uio_offset, tl);
 1079                         tl += 2;
 1080                         *tl++ = txdr_unsigned(len);
 1081                         *tl++ = txdr_unsigned(*iomode);
 1082                         *tl = txdr_unsigned(len);
 1083                 } else {
 1084                         u_int32_t x;
 1085 
 1086                         tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
 1087                         /* Set both "begin" and "current" to non-garbage. */
 1088                         x = txdr_unsigned((u_int32_t)uiop->uio_offset);
 1089                         *tl++ = x;      /* "begin offset" */
 1090                         *tl++ = x;      /* "current offset" */
 1091                         x = txdr_unsigned(len);
 1092                         *tl++ = x;      /* total to this offset */
 1093                         *tl = x;        /* size of this write */
 1094                 }
 1095                 nfsm_uiotom(uiop, len);
 1096                 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
 1097                 if (v3) {
 1098                         wccflag = NFSV3_WCCCHK;
 1099                         nfsm_wcc_data(vp, wccflag);
 1100                         if (!error) {
 1101                                 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
 1102                                         + NFSX_V3WRITEVERF);
 1103                                 rlen = fxdr_unsigned(int, *tl++);
 1104                                 if (rlen == 0) {
 1105                                         error = NFSERR_IO;
 1106                                         m_freem(mrep);
 1107                                         break;
 1108                                 } else if (rlen < len) {
 1109                                         backup = len - rlen;
 1110                                         uiop->uio_iov->iov_base =
 1111                                             (char *)uiop->uio_iov->iov_base -
 1112                                             backup;
 1113                                         uiop->uio_iov->iov_len += backup;
 1114                                         uiop->uio_offset -= backup;
 1115                                         uiop->uio_resid += backup;
 1116                                         len = rlen;
 1117                                 }
 1118                                 commit = fxdr_unsigned(int, *tl++);
 1119 
 1120                                 /*
 1121                                  * Return the lowest committment level
 1122                                  * obtained by any of the RPCs.
 1123                                  */
 1124                                 if (committed == NFSV3WRITE_FILESYNC)
 1125                                         committed = commit;
 1126                                 else if (committed == NFSV3WRITE_DATASYNC &&
 1127                                         commit == NFSV3WRITE_UNSTABLE)
 1128                                         committed = commit;
 1129                                 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
 1130                                     bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 1131                                         NFSX_V3WRITEVERF);
 1132                                     nmp->nm_state |= NFSSTA_HASWRITEVERF;
 1133                                 } else if (bcmp((caddr_t)tl,
 1134                                     (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
 1135                                     *must_commit = 1;
 1136                                     bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 1137                                         NFSX_V3WRITEVERF);
 1138                                 }
 1139                         }
 1140                 } else
 1141                     nfsm_loadattr(vp, NULL);
 1142                 if (wccflag)
 1143                     VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
 1144                 m_freem(mrep);
 1145                 if (error)
 1146                         break;
 1147                 tsiz -= len;
 1148         }
 1149 nfsmout:
 1150         if (vp->v_mount->mnt_flag & MNT_ASYNC)
 1151                 committed = NFSV3WRITE_FILESYNC;
 1152         *iomode = committed;
 1153         if (error)
 1154                 uiop->uio_resid = tsiz;
 1155         return (error);
 1156 }
 1157 
 1158 /*
 1159  * nfs mknod rpc
 1160  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
 1161  * mode set to specify the file type and the size field for rdev.
 1162  */
 1163 static int
 1164 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
 1165     struct vattr *vap)
 1166 {
 1167         struct nfsv2_sattr *sp;
 1168         u_int32_t *tl;
 1169         struct vnode *newvp = NULL;
 1170         struct nfsnode *np = NULL;
 1171         struct vattr vattr;
 1172         caddr_t bpos, dpos;
 1173         int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
 1174         struct mbuf *mreq, *mrep, *md, *mb;
 1175         u_int32_t rdev;
 1176         int v3 = NFS_ISV3(dvp);
 1177 
 1178         if (vap->va_type == VCHR || vap->va_type == VBLK)
 1179                 rdev = txdr_unsigned(vap->va_rdev);
 1180         else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
 1181                 rdev = nfs_xdrneg1;
 1182         else {
 1183                 return (EOPNOTSUPP);
 1184         }
 1185         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1186                 return (error);
 1187         }
 1188         nfsstats.rpccnt[NFSPROC_MKNOD]++;
 1189         mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
 1190                 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
 1191         mb = mreq;
 1192         bpos = mtod(mb, caddr_t);
 1193         nfsm_fhtom(dvp, v3);
 1194         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1195         if (v3) {
 1196                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
 1197                 *tl++ = vtonfsv3_type(vap->va_type);
 1198                 nfsm_v3attrbuild(vap, FALSE);
 1199                 if (vap->va_type == VCHR || vap->va_type == VBLK) {
 1200                         tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
 1201                         *tl++ = txdr_unsigned(umajor(vap->va_rdev));
 1202                         *tl = txdr_unsigned(uminor(vap->va_rdev));
 1203                 }
 1204         } else {
 1205                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1206                 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
 1207                 sp->sa_uid = nfs_xdrneg1;
 1208                 sp->sa_gid = nfs_xdrneg1;
 1209                 sp->sa_size = rdev;
 1210                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1211                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1212         }
 1213         nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
 1214         if (!error) {
 1215                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1216                 if (!gotvp) {
 1217                         if (newvp) {
 1218                                 vput(newvp);
 1219                                 newvp = NULL;
 1220                         }
 1221                         error = nfs_lookitup(dvp, cnp->cn_nameptr,
 1222                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
 1223                         if (!error)
 1224                                 newvp = NFSTOV(np);
 1225                 }
 1226         }
 1227         if (v3)
 1228                 nfsm_wcc_data(dvp, wccflag);
 1229         m_freem(mrep);
 1230 nfsmout:
 1231         if (error) {
 1232                 if (newvp)
 1233                         vput(newvp);
 1234         } else {
 1235                 if (cnp->cn_flags & MAKEENTRY)
 1236                         cache_enter(dvp, newvp, cnp);
 1237                 *vpp = newvp;
 1238         }
 1239         VTONFS(dvp)->n_flag |= NMODIFIED;
 1240         if (!wccflag)
 1241                 VTONFS(dvp)->n_attrstamp = 0;
 1242         return (error);
 1243 }
 1244 
 1245 /*
 1246  * nfs mknod vop
 1247  * just call nfs_mknodrpc() to do the work.
 1248  */
 1249 /* ARGSUSED */
 1250 static int
 1251 nfs_mknod(struct vop_mknod_args *ap)
 1252 {
 1253 
 1254         return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
 1255 }
 1256 
 1257 static u_long create_verf;
 1258 /*
 1259  * nfs file create call
 1260  */
 1261 static int
 1262 nfs_create(struct vop_create_args *ap)
 1263 {
 1264         struct vnode *dvp = ap->a_dvp;
 1265         struct vattr *vap = ap->a_vap;
 1266         struct componentname *cnp = ap->a_cnp;
 1267         struct nfsv2_sattr *sp;
 1268         u_int32_t *tl;
 1269         struct nfsnode *np = NULL;
 1270         struct vnode *newvp = NULL;
 1271         caddr_t bpos, dpos;
 1272         int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
 1273         struct mbuf *mreq, *mrep, *md, *mb;
 1274         struct vattr vattr;
 1275         int v3 = NFS_ISV3(dvp);
 1276 
 1277         /*
 1278          * Oops, not for me..
 1279          */
 1280         if (vap->va_type == VSOCK)
 1281                 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
 1282 
 1283         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1284                 return (error);
 1285         }
 1286         if (vap->va_vaflags & VA_EXCLUSIVE)
 1287                 fmode |= O_EXCL;
 1288 again:
 1289         nfsstats.rpccnt[NFSPROC_CREATE]++;
 1290         mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
 1291                 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
 1292         mb = mreq;
 1293         bpos = mtod(mb, caddr_t);
 1294         nfsm_fhtom(dvp, v3);
 1295         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1296         if (v3) {
 1297                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
 1298                 if (fmode & O_EXCL) {
 1299                         *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
 1300                         tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
 1301 #ifdef INET
 1302                         if (!TAILQ_EMPTY(&in_ifaddrhead))
 1303                                 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
 1304                         else
 1305 #endif
 1306                                 *tl++ = create_verf;
 1307                         *tl = ++create_verf;
 1308                 } else {
 1309                         *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
 1310                         nfsm_v3attrbuild(vap, FALSE);
 1311                 }
 1312         } else {
 1313                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1314                 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
 1315                 sp->sa_uid = nfs_xdrneg1;
 1316                 sp->sa_gid = nfs_xdrneg1;
 1317                 sp->sa_size = 0;
 1318                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1319                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1320         }
 1321         nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
 1322         if (!error) {
 1323                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1324                 if (!gotvp) {
 1325                         if (newvp) {
 1326                                 vput(newvp);
 1327                                 newvp = NULL;
 1328                         }
 1329                         error = nfs_lookitup(dvp, cnp->cn_nameptr,
 1330                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
 1331                         if (!error)
 1332                                 newvp = NFSTOV(np);
 1333                 }
 1334         }
 1335         if (v3)
 1336                 nfsm_wcc_data(dvp, wccflag);
 1337         m_freem(mrep);
 1338 nfsmout:
 1339         if (error) {
 1340                 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
 1341                         fmode &= ~O_EXCL;
 1342                         goto again;
 1343                 }
 1344                 if (newvp)
 1345                         vput(newvp);
 1346         } else if (v3 && (fmode & O_EXCL)) {
 1347                 /*
 1348                  * We are normally called with only a partially initialized
 1349                  * VAP.  Since the NFSv3 spec says that server may use the
 1350                  * file attributes to store the verifier, the spec requires
 1351                  * us to do a SETATTR RPC. FreeBSD servers store the verifier
 1352                  * in atime, but we can't really assume that all servers will
 1353                  * so we ensure that our SETATTR sets both atime and mtime.
 1354                  */
 1355                 if (vap->va_mtime.tv_sec == VNOVAL)
 1356                         vfs_timestamp(&vap->va_mtime);
 1357                 if (vap->va_atime.tv_sec == VNOVAL)
 1358                         vap->va_atime = vap->va_mtime;
 1359                 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_thread);
 1360                 if (error)
 1361                         vput(newvp);
 1362         }
 1363         if (!error) {
 1364                 if (cnp->cn_flags & MAKEENTRY)
 1365                         cache_enter(dvp, newvp, cnp);
 1366                 *ap->a_vpp = newvp;
 1367         }
 1368         VTONFS(dvp)->n_flag |= NMODIFIED;
 1369         if (!wccflag)
 1370                 VTONFS(dvp)->n_attrstamp = 0;
 1371         return (error);
 1372 }
 1373 
 1374 /*
 1375  * nfs file remove call
 1376  * To try and make nfs semantics closer to ufs semantics, a file that has
 1377  * other processes using the vnode is renamed instead of removed and then
 1378  * removed later on the last close.
 1379  * - If v_usecount > 1
 1380  *        If a rename is not already in the works
 1381  *           call nfs_sillyrename() to set it up
 1382  *     else
 1383  *        do the remove rpc
 1384  */
 1385 static int
 1386 nfs_remove(struct vop_remove_args *ap)
 1387 {
 1388         struct vnode *vp = ap->a_vp;
 1389         struct vnode *dvp = ap->a_dvp;
 1390         struct componentname *cnp = ap->a_cnp;
 1391         struct nfsnode *np = VTONFS(vp);
 1392         int error = 0;
 1393         struct vattr vattr;
 1394 
 1395 #ifndef DIAGNOSTIC
 1396         if ((cnp->cn_flags & HASBUF) == 0)
 1397                 panic("nfs_remove: no name");
 1398         if (vrefcnt(vp) < 1)
 1399                 panic("nfs_remove: bad v_usecount");
 1400 #endif
 1401         if (vp->v_type == VDIR)
 1402                 error = EPERM;
 1403         else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
 1404             VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_thread) == 0 &&
 1405             vattr.va_nlink > 1)) {
 1406                 /*
 1407                  * Purge the name cache so that the chance of a lookup for
 1408                  * the name succeeding while the remove is in progress is
 1409                  * minimized. Without node locking it can still happen, such
 1410                  * that an I/O op returns ESTALE, but since you get this if
 1411                  * another host removes the file..
 1412                  */
 1413                 cache_purge(vp);
 1414                 /*
 1415                  * throw away biocache buffers, mainly to avoid
 1416                  * unnecessary delayed writes later.
 1417                  */
 1418                 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
 1419                 /* Do the rpc */
 1420                 if (error != EINTR && error != EIO)
 1421                         error = nfs_removerpc(dvp, cnp->cn_nameptr,
 1422                                 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
 1423                 /*
 1424                  * Kludge City: If the first reply to the remove rpc is lost..
 1425                  *   the reply to the retransmitted request will be ENOENT
 1426                  *   since the file was in fact removed
 1427                  *   Therefore, we cheat and return success.
 1428                  */
 1429                 if (error == ENOENT)
 1430                         error = 0;
 1431         } else if (!np->n_sillyrename)
 1432                 error = nfs_sillyrename(dvp, vp, cnp);
 1433         np->n_attrstamp = 0;
 1434         return (error);
 1435 }
 1436 
 1437 /*
 1438  * nfs file remove rpc called from nfs_inactive
 1439  */
 1440 int
 1441 nfs_removeit(struct sillyrename *sp)
 1442 {
 1443 
 1444         /*
 1445          * Make sure that the directory vnode is still valid.
 1446          * XXX we should lock sp->s_dvp here.
 1447          */
 1448         if (sp->s_dvp->v_type == VBAD)
 1449                 return (0);
 1450         return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 1451                 NULL));
 1452 }
 1453 
 1454 /*
 1455  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
 1456  */
 1457 static int
 1458 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
 1459     struct ucred *cred, struct thread *td)
 1460 {
 1461         caddr_t bpos, dpos;
 1462         int error = 0, wccflag = NFSV3_WCCRATTR;
 1463         struct mbuf *mreq, *mrep, *md, *mb;
 1464         int v3 = NFS_ISV3(dvp);
 1465 
 1466         nfsstats.rpccnt[NFSPROC_REMOVE]++;
 1467         mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
 1468                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
 1469         mb = mreq;
 1470         bpos = mtod(mb, caddr_t);
 1471         nfsm_fhtom(dvp, v3);
 1472         nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
 1473         nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
 1474         if (v3)
 1475                 nfsm_wcc_data(dvp, wccflag);
 1476         m_freem(mrep);
 1477 nfsmout:
 1478         VTONFS(dvp)->n_flag |= NMODIFIED;
 1479         if (!wccflag)
 1480                 VTONFS(dvp)->n_attrstamp = 0;
 1481         return (error);
 1482 }
 1483 
 1484 /*
 1485  * nfs file rename call
 1486  */
 1487 static int
 1488 nfs_rename(struct vop_rename_args *ap)
 1489 {
 1490         struct vnode *fvp = ap->a_fvp;
 1491         struct vnode *tvp = ap->a_tvp;
 1492         struct vnode *fdvp = ap->a_fdvp;
 1493         struct vnode *tdvp = ap->a_tdvp;
 1494         struct componentname *tcnp = ap->a_tcnp;
 1495         struct componentname *fcnp = ap->a_fcnp;
 1496         int error;
 1497 
 1498 #ifndef DIAGNOSTIC
 1499         if ((tcnp->cn_flags & HASBUF) == 0 ||
 1500             (fcnp->cn_flags & HASBUF) == 0)
 1501                 panic("nfs_rename: no name");
 1502 #endif
 1503         /* Check for cross-device rename */
 1504         if ((fvp->v_mount != tdvp->v_mount) ||
 1505             (tvp && (fvp->v_mount != tvp->v_mount))) {
 1506                 error = EXDEV;
 1507                 goto out;
 1508         }
 1509 
 1510         if (fvp == tvp) {
 1511                 printf("nfs_rename: fvp == tvp (can't happen)\n");
 1512                 error = 0;
 1513                 goto out;
 1514         }
 1515         if ((error = vn_lock(fvp, LK_EXCLUSIVE, fcnp->cn_thread)) != 0)
 1516                 goto out;
 1517 
 1518         /*
 1519          * We have to flush B_DELWRI data prior to renaming
 1520          * the file.  If we don't, the delayed-write buffers
 1521          * can be flushed out later after the file has gone stale
 1522          * under NFSV3.  NFSV2 does not have this problem because
 1523          * ( as far as I can tell ) it flushes dirty buffers more
 1524          * often.
 1525          * 
 1526          * Skip the rename operation if the fsync fails, this can happen
 1527          * due to the server's volume being full, when we pushed out data
 1528          * that was written back to our cache earlier. Not checking for
 1529          * this condition can result in potential (silent) data loss.
 1530          */
 1531         error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
 1532         VOP_UNLOCK(fvp, 0, fcnp->cn_thread);
 1533         if (!error && tvp)
 1534                 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
 1535         if (error)
 1536                 goto out;
 1537 
 1538         /*
 1539          * If the tvp exists and is in use, sillyrename it before doing the
 1540          * rename of the new file over it.
 1541          * XXX Can't sillyrename a directory.
 1542          */
 1543         if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
 1544                 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
 1545                 vput(tvp);
 1546                 tvp = NULL;
 1547         }
 1548 
 1549         error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
 1550                 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
 1551                 tcnp->cn_thread);
 1552 
 1553         if (fvp->v_type == VDIR) {
 1554                 if (tvp != NULL && tvp->v_type == VDIR)
 1555                         cache_purge(tdvp);
 1556                 cache_purge(fdvp);
 1557         }
 1558 
 1559 out:
 1560         if (tdvp == tvp)
 1561                 vrele(tdvp);
 1562         else
 1563                 vput(tdvp);
 1564         if (tvp)
 1565                 vput(tvp);
 1566         vrele(fdvp);
 1567         vrele(fvp);
 1568         /*
 1569          * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
 1570          */
 1571         if (error == ENOENT)
 1572                 error = 0;
 1573         return (error);
 1574 }
 1575 
 1576 /*
 1577  * nfs file rename rpc called from nfs_remove() above
 1578  */
 1579 static int
 1580 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
 1581     struct sillyrename *sp)
 1582 {
 1583 
 1584         return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
 1585             sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
 1586 }
 1587 
 1588 /*
 1589  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
 1590  */
 1591 static int
 1592 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
 1593     struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
 1594     struct thread *td)
 1595 {
 1596         caddr_t bpos, dpos;
 1597         int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
 1598         struct mbuf *mreq, *mrep, *md, *mb;
 1599         int v3 = NFS_ISV3(fdvp);
 1600 
 1601         nfsstats.rpccnt[NFSPROC_RENAME]++;
 1602         mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
 1603                 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
 1604                 nfsm_rndup(tnamelen));
 1605         mb = mreq;
 1606         bpos = mtod(mb, caddr_t);
 1607         nfsm_fhtom(fdvp, v3);
 1608         nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
 1609         nfsm_fhtom(tdvp, v3);
 1610         nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
 1611         nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
 1612         if (v3) {
 1613                 nfsm_wcc_data(fdvp, fwccflag);
 1614                 nfsm_wcc_data(tdvp, twccflag);
 1615         }
 1616         m_freem(mrep);
 1617 nfsmout:
 1618         VTONFS(fdvp)->n_flag |= NMODIFIED;
 1619         VTONFS(tdvp)->n_flag |= NMODIFIED;
 1620         if (!fwccflag)
 1621                 VTONFS(fdvp)->n_attrstamp = 0;
 1622         if (!twccflag)
 1623                 VTONFS(tdvp)->n_attrstamp = 0;
 1624         return (error);
 1625 }
 1626 
 1627 /*
 1628  * nfs hard link create call
 1629  */
 1630 static int
 1631 nfs_link(struct vop_link_args *ap)
 1632 {
 1633         struct vnode *vp = ap->a_vp;
 1634         struct vnode *tdvp = ap->a_tdvp;
 1635         struct componentname *cnp = ap->a_cnp;
 1636         caddr_t bpos, dpos;
 1637         int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
 1638         struct mbuf *mreq, *mrep, *md, *mb;
 1639         int v3;
 1640 
 1641         if (vp->v_mount != tdvp->v_mount) {
 1642                 return (EXDEV);
 1643         }
 1644 
 1645         /*
 1646          * Push all writes to the server, so that the attribute cache
 1647          * doesn't get "out of sync" with the server.
 1648          * XXX There should be a better way!
 1649          */
 1650         VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
 1651 
 1652         v3 = NFS_ISV3(vp);
 1653         nfsstats.rpccnt[NFSPROC_LINK]++;
 1654         mreq = nfsm_reqhead(vp, NFSPROC_LINK,
 1655                 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
 1656         mb = mreq;
 1657         bpos = mtod(mb, caddr_t);
 1658         nfsm_fhtom(vp, v3);
 1659         nfsm_fhtom(tdvp, v3);
 1660         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1661         nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
 1662         if (v3) {
 1663                 nfsm_postop_attr(vp, attrflag);
 1664                 nfsm_wcc_data(tdvp, wccflag);
 1665         }
 1666         m_freem(mrep);
 1667 nfsmout:
 1668         VTONFS(tdvp)->n_flag |= NMODIFIED;
 1669         if (!attrflag)
 1670                 VTONFS(vp)->n_attrstamp = 0;
 1671         if (!wccflag)
 1672                 VTONFS(tdvp)->n_attrstamp = 0;
 1673         return (error);
 1674 }
 1675 
 1676 /*
 1677  * nfs symbolic link create call
 1678  */
 1679 static int
 1680 nfs_symlink(struct vop_symlink_args *ap)
 1681 {
 1682         struct vnode *dvp = ap->a_dvp;
 1683         struct vattr *vap = ap->a_vap;
 1684         struct componentname *cnp = ap->a_cnp;
 1685         struct nfsv2_sattr *sp;
 1686         caddr_t bpos, dpos;
 1687         int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
 1688         struct mbuf *mreq, *mrep, *md, *mb;
 1689         struct vnode *newvp = NULL;
 1690         int v3 = NFS_ISV3(dvp);
 1691 
 1692         nfsstats.rpccnt[NFSPROC_SYMLINK]++;
 1693         slen = strlen(ap->a_target);
 1694         mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
 1695             nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
 1696         mb = mreq;
 1697         bpos = mtod(mb, caddr_t);
 1698         nfsm_fhtom(dvp, v3);
 1699         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1700         if (v3) {
 1701                 nfsm_v3attrbuild(vap, FALSE);
 1702         }
 1703         nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
 1704         if (!v3) {
 1705                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1706                 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
 1707                 sp->sa_uid = nfs_xdrneg1;
 1708                 sp->sa_gid = nfs_xdrneg1;
 1709                 sp->sa_size = nfs_xdrneg1;
 1710                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1711                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1712         }
 1713 
 1714         /*
 1715          * Issue the NFS request and get the rpc response.
 1716          *
 1717          * Only NFSv3 responses returning an error of 0 actually return
 1718          * a file handle that can be converted into newvp without having
 1719          * to do an extra lookup rpc.
 1720          */
 1721         nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
 1722         if (v3) {
 1723                 if (error == 0)
 1724                         nfsm_mtofh(dvp, newvp, v3, gotvp);
 1725                 nfsm_wcc_data(dvp, wccflag);
 1726         }
 1727 
 1728         /*
 1729          * out code jumps -> here, mrep is also freed.
 1730          */
 1731 
 1732         m_freem(mrep);
 1733 nfsmout:
 1734 
 1735         /*
 1736          * If we do not have an error and we could not extract the newvp from
 1737          * the response due to the request being NFSv2, we have to do a
 1738          * lookup in order to obtain a newvp to return.
 1739          */
 1740         if (error == 0 && newvp == NULL) {
 1741                 struct nfsnode *np = NULL;
 1742 
 1743                 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 1744                     cnp->cn_cred, cnp->cn_thread, &np);
 1745                 if (!error)
 1746                         newvp = NFSTOV(np);
 1747         }
 1748         if (error) {
 1749                 if (newvp)
 1750                         vput(newvp);
 1751         } else {
 1752                 *ap->a_vpp = newvp;
 1753         }
 1754         VTONFS(dvp)->n_flag |= NMODIFIED;
 1755         if (!wccflag)
 1756                 VTONFS(dvp)->n_attrstamp = 0;
 1757         return (error);
 1758 }
 1759 
 1760 /*
 1761  * nfs make dir call
 1762  */
 1763 static int
 1764 nfs_mkdir(struct vop_mkdir_args *ap)
 1765 {
 1766         struct vnode *dvp = ap->a_dvp;
 1767         struct vattr *vap = ap->a_vap;
 1768         struct componentname *cnp = ap->a_cnp;
 1769         struct nfsv2_sattr *sp;
 1770         int len;
 1771         struct nfsnode *np = NULL;
 1772         struct vnode *newvp = NULL;
 1773         caddr_t bpos, dpos;
 1774         int error = 0, wccflag = NFSV3_WCCRATTR;
 1775         int gotvp = 0;
 1776         struct mbuf *mreq, *mrep, *md, *mb;
 1777         struct vattr vattr;
 1778         int v3 = NFS_ISV3(dvp);
 1779 
 1780         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1781                 return (error);
 1782         }
 1783         len = cnp->cn_namelen;
 1784         nfsstats.rpccnt[NFSPROC_MKDIR]++;
 1785         mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
 1786           NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
 1787         mb = mreq;
 1788         bpos = mtod(mb, caddr_t);
 1789         nfsm_fhtom(dvp, v3);
 1790         nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
 1791         if (v3) {
 1792                 nfsm_v3attrbuild(vap, FALSE);
 1793         } else {
 1794                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1795                 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
 1796                 sp->sa_uid = nfs_xdrneg1;
 1797                 sp->sa_gid = nfs_xdrneg1;
 1798                 sp->sa_size = nfs_xdrneg1;
 1799                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1800                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1801         }
 1802         nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
 1803         if (!error)
 1804                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1805         if (v3)
 1806                 nfsm_wcc_data(dvp, wccflag);
 1807         m_freem(mrep);
 1808 nfsmout:
 1809         VTONFS(dvp)->n_flag |= NMODIFIED;
 1810         if (!wccflag)
 1811                 VTONFS(dvp)->n_attrstamp = 0;
 1812         if (error == 0 && newvp == NULL) {
 1813                 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
 1814                         cnp->cn_thread, &np);
 1815                 if (!error) {
 1816                         newvp = NFSTOV(np);
 1817                         if (newvp->v_type != VDIR)
 1818                                 error = EEXIST;
 1819                 }
 1820         }
 1821         if (error) {
 1822                 if (newvp)
 1823                         vput(newvp);
 1824         } else
 1825                 *ap->a_vpp = newvp;
 1826         return (error);
 1827 }
 1828 
 1829 /*
 1830  * nfs remove directory call
 1831  */
 1832 static int
 1833 nfs_rmdir(struct vop_rmdir_args *ap)
 1834 {
 1835         struct vnode *vp = ap->a_vp;
 1836         struct vnode *dvp = ap->a_dvp;
 1837         struct componentname *cnp = ap->a_cnp;
 1838         caddr_t bpos, dpos;
 1839         int error = 0, wccflag = NFSV3_WCCRATTR;
 1840         struct mbuf *mreq, *mrep, *md, *mb;
 1841         int v3 = NFS_ISV3(dvp);
 1842 
 1843         if (dvp == vp)
 1844                 return (EINVAL);
 1845         nfsstats.rpccnt[NFSPROC_RMDIR]++;
 1846         mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
 1847                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
 1848         mb = mreq;
 1849         bpos = mtod(mb, caddr_t);
 1850         nfsm_fhtom(dvp, v3);
 1851         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1852         nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
 1853         if (v3)
 1854                 nfsm_wcc_data(dvp, wccflag);
 1855         m_freem(mrep);
 1856 nfsmout:
 1857         VTONFS(dvp)->n_flag |= NMODIFIED;
 1858         if (!wccflag)
 1859                 VTONFS(dvp)->n_attrstamp = 0;
 1860         cache_purge(dvp);
 1861         cache_purge(vp);
 1862         /*
 1863          * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
 1864          */
 1865         if (error == ENOENT)
 1866                 error = 0;
 1867         return (error);
 1868 }
 1869 
 1870 /*
 1871  * nfs readdir call
 1872  */
 1873 static int
 1874 nfs_readdir(struct vop_readdir_args *ap)
 1875 {
 1876         struct vnode *vp = ap->a_vp;
 1877         struct nfsnode *np = VTONFS(vp);
 1878         struct uio *uio = ap->a_uio;
 1879         int tresid, error;
 1880         struct vattr vattr;
 1881 
 1882         if (vp->v_type != VDIR)
 1883                 return (EPERM);
 1884         /*
 1885          * First, check for hit on the EOF offset cache
 1886          */
 1887         if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
 1888             (np->n_flag & NMODIFIED) == 0) {
 1889                 if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_td) == 0 &&
 1890                     !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
 1891                         nfsstats.direofcache_hits++;
 1892                         return (0);
 1893                 }
 1894         }
 1895 
 1896         /*
 1897          * Call nfs_bioread() to do the real work.
 1898          */
 1899         tresid = uio->uio_resid;
 1900         error = nfs_bioread(vp, uio, 0, ap->a_cred);
 1901 
 1902         if (!error && uio->uio_resid == tresid)
 1903                 nfsstats.direofcache_misses++;
 1904         return (error);
 1905 }
 1906 
 1907 /*
 1908  * Readdir rpc call.
 1909  * Called from below the buffer cache by nfs_doio().
 1910  */
 1911 int
 1912 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 1913 {
 1914         int len, left;
 1915         struct dirent *dp = NULL;
 1916         u_int32_t *tl;
 1917         caddr_t cp;
 1918         nfsuint64 *cookiep;
 1919         caddr_t bpos, dpos;
 1920         struct mbuf *mreq, *mrep, *md, *mb;
 1921         nfsuint64 cookie;
 1922         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 1923         struct nfsnode *dnp = VTONFS(vp);
 1924         u_quad_t fileno;
 1925         int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
 1926         int attrflag;
 1927         int v3 = NFS_ISV3(vp);
 1928 
 1929 #ifndef DIAGNOSTIC
 1930         if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
 1931                 (uiop->uio_resid & (DIRBLKSIZ - 1)))
 1932                 panic("nfs readdirrpc bad uio");
 1933 #endif
 1934 
 1935         /*
 1936          * If there is no cookie, assume directory was stale.
 1937          */
 1938         cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
 1939         if (cookiep)
 1940                 cookie = *cookiep;
 1941         else
 1942                 return (NFSERR_BAD_COOKIE);
 1943         /*
 1944          * Loop around doing readdir rpc's of size nm_readdirsize
 1945          * truncated to a multiple of DIRBLKSIZ.
 1946          * The stopping criteria is EOF or buffer full.
 1947          */
 1948         while (more_dirs && bigenough) {
 1949                 nfsstats.rpccnt[NFSPROC_READDIR]++;
 1950                 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
 1951                         NFSX_READDIR(v3));
 1952                 mb = mreq;
 1953                 bpos = mtod(mb, caddr_t);
 1954                 nfsm_fhtom(vp, v3);
 1955                 if (v3) {
 1956                         tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
 1957                         *tl++ = cookie.nfsuquad[0];
 1958                         *tl++ = cookie.nfsuquad[1];
 1959                         *tl++ = dnp->n_cookieverf.nfsuquad[0];
 1960                         *tl++ = dnp->n_cookieverf.nfsuquad[1];
 1961                 } else {
 1962                         tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
 1963                         *tl++ = cookie.nfsuquad[0];
 1964                 }
 1965                 *tl = txdr_unsigned(nmp->nm_readdirsize);
 1966                 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
 1967                 if (v3) {
 1968                         nfsm_postop_attr(vp, attrflag);
 1969                         if (!error) {
 1970                                 tl = nfsm_dissect(u_int32_t *,
 1971                                     2 * NFSX_UNSIGNED);
 1972                                 dnp->n_cookieverf.nfsuquad[0] = *tl++;
 1973                                 dnp->n_cookieverf.nfsuquad[1] = *tl;
 1974                         } else {
 1975                                 m_freem(mrep);
 1976                                 goto nfsmout;
 1977                         }
 1978                 }
 1979                 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 1980                 more_dirs = fxdr_unsigned(int, *tl);
 1981 
 1982                 /* loop thru the dir entries, doctoring them to 4bsd form */
 1983                 while (more_dirs && bigenough) {
 1984                         if (v3) {
 1985                                 tl = nfsm_dissect(u_int32_t *,
 1986                                     3 * NFSX_UNSIGNED);
 1987                                 fileno = fxdr_hyper(tl);
 1988                                 len = fxdr_unsigned(int, *(tl + 2));
 1989                         } else {
 1990                                 tl = nfsm_dissect(u_int32_t *,
 1991                                     2 * NFSX_UNSIGNED);
 1992                                 fileno = fxdr_unsigned(u_quad_t, *tl++);
 1993                                 len = fxdr_unsigned(int, *tl);
 1994                         }
 1995                         if (len <= 0 || len > NFS_MAXNAMLEN) {
 1996                                 error = EBADRPC;
 1997                                 m_freem(mrep);
 1998                                 goto nfsmout;
 1999                         }
 2000                         tlen = nfsm_rndup(len);
 2001                         if (tlen == len)
 2002                                 tlen += 4;      /* To ensure null termination */
 2003                         left = DIRBLKSIZ - blksiz;
 2004                         if ((tlen + DIRHDSIZ) > left) {
 2005                                 dp->d_reclen += left;
 2006                                 uiop->uio_iov->iov_base =
 2007                                     (char *)uiop->uio_iov->iov_base + left;
 2008                                 uiop->uio_iov->iov_len -= left;
 2009                                 uiop->uio_offset += left;
 2010                                 uiop->uio_resid -= left;
 2011                                 blksiz = 0;
 2012                         }
 2013                         if ((tlen + DIRHDSIZ) > uiop->uio_resid)
 2014                                 bigenough = 0;
 2015                         if (bigenough) {
 2016                                 dp = (struct dirent *)uiop->uio_iov->iov_base;
 2017                                 dp->d_fileno = (int)fileno;
 2018                                 dp->d_namlen = len;
 2019                                 dp->d_reclen = tlen + DIRHDSIZ;
 2020                                 dp->d_type = DT_UNKNOWN;
 2021                                 blksiz += dp->d_reclen;
 2022                                 if (blksiz == DIRBLKSIZ)
 2023                                         blksiz = 0;
 2024                                 uiop->uio_offset += DIRHDSIZ;
 2025                                 uiop->uio_resid -= DIRHDSIZ;
 2026                                 uiop->uio_iov->iov_base =
 2027                                     (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
 2028                                 uiop->uio_iov->iov_len -= DIRHDSIZ;
 2029                                 nfsm_mtouio(uiop, len);
 2030                                 cp = uiop->uio_iov->iov_base;
 2031                                 tlen -= len;
 2032                                 *cp = '\0';     /* null terminate */
 2033                                 uiop->uio_iov->iov_base =
 2034                                     (char *)uiop->uio_iov->iov_base + tlen;
 2035                                 uiop->uio_iov->iov_len -= tlen;
 2036                                 uiop->uio_offset += tlen;
 2037                                 uiop->uio_resid -= tlen;
 2038                         } else
 2039                                 nfsm_adv(nfsm_rndup(len));
 2040                         if (v3) {
 2041                                 tl = nfsm_dissect(u_int32_t *,
 2042                                     3 * NFSX_UNSIGNED);
 2043                         } else {
 2044                                 tl = nfsm_dissect(u_int32_t *,
 2045                                     2 * NFSX_UNSIGNED);
 2046                         }
 2047                         if (bigenough) {
 2048                                 cookie.nfsuquad[0] = *tl++;
 2049                                 if (v3)
 2050                                         cookie.nfsuquad[1] = *tl++;
 2051                         } else if (v3)
 2052                                 tl += 2;
 2053                         else
 2054                                 tl++;
 2055                         more_dirs = fxdr_unsigned(int, *tl);
 2056                 }
 2057                 /*
 2058                  * If at end of rpc data, get the eof boolean
 2059                  */
 2060                 if (!more_dirs) {
 2061                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2062                         more_dirs = (fxdr_unsigned(int, *tl) == 0);
 2063                 }
 2064                 m_freem(mrep);
 2065         }
 2066         /*
 2067          * Fill last record, iff any, out to a multiple of DIRBLKSIZ
 2068          * by increasing d_reclen for the last record.
 2069          */
 2070         if (blksiz > 0) {
 2071                 left = DIRBLKSIZ - blksiz;
 2072                 dp->d_reclen += left;
 2073                 uiop->uio_iov->iov_base =
 2074                     (char *)uiop->uio_iov->iov_base + left;
 2075                 uiop->uio_iov->iov_len -= left;
 2076                 uiop->uio_offset += left;
 2077                 uiop->uio_resid -= left;
 2078         }
 2079 
 2080         /*
 2081          * We are now either at the end of the directory or have filled the
 2082          * block.
 2083          */
 2084         if (bigenough)
 2085                 dnp->n_direofoffset = uiop->uio_offset;
 2086         else {
 2087                 if (uiop->uio_resid > 0)
 2088                         printf("EEK! readdirrpc resid > 0\n");
 2089                 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
 2090                 *cookiep = cookie;
 2091         }
 2092 nfsmout:
 2093         return (error);
 2094 }
 2095 
 2096 /*
 2097  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
 2098  */
 2099 int
 2100 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 2101 {
 2102         int len, left;
 2103         struct dirent *dp;
 2104         u_int32_t *tl;
 2105         caddr_t cp;
 2106         struct vnode *newvp;
 2107         nfsuint64 *cookiep;
 2108         caddr_t bpos, dpos, dpossav1, dpossav2;
 2109         struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
 2110         struct nameidata nami, *ndp = &nami;
 2111         struct componentname *cnp = &ndp->ni_cnd;
 2112         nfsuint64 cookie;
 2113         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2114         struct nfsnode *dnp = VTONFS(vp), *np;
 2115         nfsfh_t *fhp;
 2116         u_quad_t fileno;
 2117         int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
 2118         int attrflag, fhsize;
 2119 
 2120 #ifndef nolint
 2121         dp = NULL;
 2122 #endif
 2123 #ifndef DIAGNOSTIC
 2124         if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
 2125                 (uiop->uio_resid & (DIRBLKSIZ - 1)))
 2126                 panic("nfs readdirplusrpc bad uio");
 2127 #endif
 2128         ndp->ni_dvp = vp;
 2129         newvp = NULLVP;
 2130 
 2131         /*
 2132          * If there is no cookie, assume directory was stale.
 2133          */
 2134         cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
 2135         if (cookiep)
 2136                 cookie = *cookiep;
 2137         else
 2138                 return (NFSERR_BAD_COOKIE);
 2139         /*
 2140          * Loop around doing readdir rpc's of size nm_readdirsize
 2141          * truncated to a multiple of DIRBLKSIZ.
 2142          * The stopping criteria is EOF or buffer full.
 2143          */
 2144         while (more_dirs && bigenough) {
 2145                 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
 2146                 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
 2147                         NFSX_FH(1) + 6 * NFSX_UNSIGNED);
 2148                 mb = mreq;
 2149                 bpos = mtod(mb, caddr_t);
 2150                 nfsm_fhtom(vp, 1);
 2151                 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
 2152                 *tl++ = cookie.nfsuquad[0];
 2153                 *tl++ = cookie.nfsuquad[1];
 2154                 *tl++ = dnp->n_cookieverf.nfsuquad[0];
 2155                 *tl++ = dnp->n_cookieverf.nfsuquad[1];
 2156                 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
 2157                 *tl = txdr_unsigned(nmp->nm_rsize);
 2158                 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
 2159                 nfsm_postop_attr(vp, attrflag);
 2160                 if (error) {
 2161                         m_freem(mrep);
 2162                         goto nfsmout;
 2163                 }
 2164                 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2165                 dnp->n_cookieverf.nfsuquad[0] = *tl++;
 2166                 dnp->n_cookieverf.nfsuquad[1] = *tl++;
 2167                 more_dirs = fxdr_unsigned(int, *tl);
 2168 
 2169                 /* loop thru the dir entries, doctoring them to 4bsd form */
 2170                 while (more_dirs && bigenough) {
 2171                         tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2172                         fileno = fxdr_hyper(tl);
 2173                         len = fxdr_unsigned(int, *(tl + 2));
 2174                         if (len <= 0 || len > NFS_MAXNAMLEN) {
 2175                                 error = EBADRPC;
 2176                                 m_freem(mrep);
 2177                                 goto nfsmout;
 2178                         }
 2179                         tlen = nfsm_rndup(len);
 2180                         if (tlen == len)
 2181                                 tlen += 4;      /* To ensure null termination*/
 2182                         left = DIRBLKSIZ - blksiz;
 2183                         if ((tlen + DIRHDSIZ) > left) {
 2184                                 dp->d_reclen += left;
 2185                                 uiop->uio_iov->iov_base =
 2186                                     (char *)uiop->uio_iov->iov_base + left;
 2187                                 uiop->uio_iov->iov_len -= left;
 2188                                 uiop->uio_offset += left;
 2189                                 uiop->uio_resid -= left;
 2190                                 blksiz = 0;
 2191                         }
 2192                         if ((tlen + DIRHDSIZ) > uiop->uio_resid)
 2193                                 bigenough = 0;
 2194                         if (bigenough) {
 2195                                 dp = (struct dirent *)uiop->uio_iov->iov_base;
 2196                                 dp->d_fileno = (int)fileno;
 2197                                 dp->d_namlen = len;
 2198                                 dp->d_reclen = tlen + DIRHDSIZ;
 2199                                 dp->d_type = DT_UNKNOWN;
 2200                                 blksiz += dp->d_reclen;
 2201                                 if (blksiz == DIRBLKSIZ)
 2202                                         blksiz = 0;
 2203                                 uiop->uio_offset += DIRHDSIZ;
 2204                                 uiop->uio_resid -= DIRHDSIZ;
 2205                                 uiop->uio_iov->iov_base =
 2206                                     (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
 2207                                 uiop->uio_iov->iov_len -= DIRHDSIZ;
 2208                                 cnp->cn_nameptr = uiop->uio_iov->iov_base;
 2209                                 cnp->cn_namelen = len;
 2210                                 nfsm_mtouio(uiop, len);
 2211                                 cp = uiop->uio_iov->iov_base;
 2212                                 tlen -= len;
 2213                                 *cp = '\0';
 2214                                 uiop->uio_iov->iov_base =
 2215                                     (char *)uiop->uio_iov->iov_base + tlen;
 2216                                 uiop->uio_iov->iov_len -= tlen;
 2217                                 uiop->uio_offset += tlen;
 2218                                 uiop->uio_resid -= tlen;
 2219                         } else
 2220                                 nfsm_adv(nfsm_rndup(len));
 2221                         tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2222                         if (bigenough) {
 2223                                 cookie.nfsuquad[0] = *tl++;
 2224                                 cookie.nfsuquad[1] = *tl++;
 2225                         } else
 2226                                 tl += 2;
 2227 
 2228                         /*
 2229                          * Since the attributes are before the file handle
 2230                          * (sigh), we must skip over the attributes and then
 2231                          * come back and get them.
 2232                          */
 2233                         attrflag = fxdr_unsigned(int, *tl);
 2234                         if (attrflag) {
 2235                             dpossav1 = dpos;
 2236                             mdsav1 = md;
 2237                             nfsm_adv(NFSX_V3FATTR);
 2238                             tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2239                             doit = fxdr_unsigned(int, *tl);
 2240                             /*
 2241                              * Skip loading the attrs for "..". There's a 
 2242                              * race between loading the attrs here and 
 2243                              * lookups that look for the directory currently
 2244                              * being read (in the parent). We try to acquire
 2245                              * the exclusive lock on ".." here, owning the 
 2246                              * lock on the directory being read. Lookup will
 2247                              * hold the lock on ".." and try to acquire the 
 2248                              * lock on the directory being read.
 2249                              * 
 2250                              * There are other ways of fixing this, one would
 2251                              * be to do a trylock on the ".." vnode and skip
 2252                              * loading the attrs on ".." if it happens to be 
 2253                              * locked by another process. But skipping the
 2254                              * attrload on ".." seems the easiest option.
 2255                              */
 2256                             if (strcmp(dp->d_name, "..") == 0) {
 2257                                     doit = 0;
 2258                                     /*
 2259                                      * We've already skipped over the attrs, 
 2260                                      * skip over the filehandle. And store d_type
 2261                                      * as VDIR.
 2262                                      */
 2263                                     tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2264                                     i = fxdr_unsigned(int, *tl);
 2265                                     nfsm_adv(nfsm_rndup(i));
 2266                                     dp->d_type = IFTODT(VTTOIF(VDIR));
 2267                             }       
 2268                             if (doit) {
 2269                                 nfsm_getfh(fhp, fhsize, 1);
 2270                                 if (NFS_CMPFH(dnp, fhp, fhsize)) {
 2271                                     VREF(vp);
 2272                                     newvp = vp;
 2273                                     np = dnp;
 2274                                 } else {
 2275                                     error = nfs_nget(vp->v_mount, fhp,
 2276                                         fhsize, &np, LK_EXCLUSIVE);
 2277                                     if (error)
 2278                                         doit = 0;
 2279                                     else
 2280                                         newvp = NFSTOV(np);
 2281                                 }
 2282                             }
 2283                             if (doit && bigenough) {
 2284                                 dpossav2 = dpos;
 2285                                 dpos = dpossav1;
 2286                                 mdsav2 = md;
 2287                                 md = mdsav1;
 2288                                 nfsm_loadattr(newvp, NULL);
 2289                                 dpos = dpossav2;
 2290                                 md = mdsav2;
 2291                                 dp->d_type =
 2292                                     IFTODT(VTTOIF(np->n_vattr.va_type));
 2293                                 ndp->ni_vp = newvp;
 2294                                 /* Update n_ctime, so subsequent lookup doesn't purge entry */
 2295                                 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
 2296                                 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
 2297                             }
 2298                         } else {
 2299                             /* Just skip over the file handle */
 2300                             tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2301                             i = fxdr_unsigned(int, *tl);
 2302                             if (i) {
 2303                                 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2304                                 fhsize = fxdr_unsigned(int, *tl);
 2305                                 nfsm_adv(nfsm_rndup(fhsize));
 2306                             }
 2307                         }
 2308                         if (newvp != NULLVP) {
 2309                             if (newvp == vp)
 2310                                 vrele(newvp);
 2311                             else
 2312                                 vput(newvp);
 2313                             newvp = NULLVP;
 2314                         }
 2315                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2316                         more_dirs = fxdr_unsigned(int, *tl);
 2317                 }
 2318                 /*
 2319                  * If at end of rpc data, get the eof boolean
 2320                  */
 2321                 if (!more_dirs) {
 2322                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2323                         more_dirs = (fxdr_unsigned(int, *tl) == 0);
 2324                 }
 2325                 m_freem(mrep);
 2326         }
 2327         /*
 2328          * Fill last record, iff any, out to a multiple of DIRBLKSIZ
 2329          * by increasing d_reclen for the last record.
 2330          */
 2331         if (blksiz > 0) {
 2332                 left = DIRBLKSIZ - blksiz;
 2333                 dp->d_reclen += left;
 2334                 uiop->uio_iov->iov_base =
 2335                     (char *)uiop->uio_iov->iov_base + left;
 2336                 uiop->uio_iov->iov_len -= left;
 2337                 uiop->uio_offset += left;
 2338                 uiop->uio_resid -= left;
 2339         }
 2340 
 2341         /*
 2342          * We are now either at the end of the directory or have filled the
 2343          * block.
 2344          */
 2345         if (bigenough)
 2346                 dnp->n_direofoffset = uiop->uio_offset;
 2347         else {
 2348                 if (uiop->uio_resid > 0)
 2349                         printf("EEK! readdirplusrpc resid > 0\n");
 2350                 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
 2351                 *cookiep = cookie;
 2352         }
 2353 nfsmout:
 2354         if (newvp != NULLVP) {
 2355                 if (newvp == vp)
 2356                         vrele(newvp);
 2357                 else
 2358                         vput(newvp);
 2359                 newvp = NULLVP;
 2360         }
 2361         return (error);
 2362 }
 2363 
 2364 /*
 2365  * Silly rename. To make the NFS filesystem that is stateless look a little
 2366  * more like the "ufs" a remove of an active vnode is translated to a rename
 2367  * to a funny looking filename that is removed by nfs_inactive on the
 2368  * nfsnode. There is the potential for another process on a different client
 2369  * to create the same funny name between the nfs_lookitup() fails and the
 2370  * nfs_rename() completes, but...
 2371  */
 2372 static int
 2373 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
 2374 {
 2375         struct sillyrename *sp;
 2376         struct nfsnode *np;
 2377         int error;
 2378         short pid;
 2379         unsigned int lticks;
 2380 
 2381         cache_purge(dvp);
 2382         np = VTONFS(vp);
 2383 #ifndef DIAGNOSTIC
 2384         if (vp->v_type == VDIR)
 2385                 panic("nfs: sillyrename dir");
 2386 #endif
 2387         MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
 2388                 M_NFSREQ, M_WAITOK);
 2389         sp->s_cred = crhold(cnp->cn_cred);
 2390         sp->s_dvp = dvp;
 2391         sp->s_removeit = nfs_removeit;
 2392         VREF(dvp);
 2393 
 2394         /* 
 2395          * Fudge together a funny name.
 2396          * Changing the format of the funny name to accomodate more 
 2397          * sillynames per directory.
 2398          * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 
 2399          * CPU ticks since boot.
 2400          */
 2401         pid = cnp->cn_thread->td_proc->p_pid;
 2402         lticks = (unsigned int)ticks;
 2403         for ( ; ; ) {
 2404                 sp->s_namlen = sprintf(sp->s_name, 
 2405                                        ".nfs.%08x.%04x4.4", lticks, 
 2406                                        pid);
 2407                 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2408                                  cnp->cn_thread, NULL))
 2409                         break;
 2410                 lticks++;
 2411         }
 2412         error = nfs_renameit(dvp, cnp, sp);
 2413         if (error)
 2414                 goto bad;
 2415         error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2416                 cnp->cn_thread, &np);
 2417         np->n_sillyrename = sp;
 2418         return (0);
 2419 bad:
 2420         vrele(sp->s_dvp);
 2421         crfree(sp->s_cred);
 2422         free((caddr_t)sp, M_NFSREQ);
 2423         return (error);
 2424 }
 2425 
 2426 /*
 2427  * Look up a file name and optionally either update the file handle or
 2428  * allocate an nfsnode, depending on the value of npp.
 2429  * npp == NULL  --> just do the lookup
 2430  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
 2431  *                      handled too
 2432  * *npp != NULL --> update the file handle in the vnode
 2433  */
 2434 static int
 2435 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
 2436     struct thread *td, struct nfsnode **npp)
 2437 {
 2438         struct vnode *newvp = NULL;
 2439         struct nfsnode *np, *dnp = VTONFS(dvp);
 2440         caddr_t bpos, dpos;
 2441         int error = 0, fhlen, attrflag;
 2442         struct mbuf *mreq, *mrep, *md, *mb;
 2443         nfsfh_t *nfhp;
 2444         int v3 = NFS_ISV3(dvp);
 2445 
 2446         nfsstats.rpccnt[NFSPROC_LOOKUP]++;
 2447         mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
 2448                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
 2449         mb = mreq;
 2450         bpos = mtod(mb, caddr_t);
 2451         nfsm_fhtom(dvp, v3);
 2452         nfsm_strtom(name, len, NFS_MAXNAMLEN);
 2453         nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
 2454         if (npp && !error) {
 2455                 nfsm_getfh(nfhp, fhlen, v3);
 2456                 if (*npp) {
 2457                     np = *npp;
 2458                     if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
 2459                         free((caddr_t)np->n_fhp, M_NFSBIGFH);
 2460                         np->n_fhp = &np->n_fh;
 2461                     } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
 2462                         np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
 2463                     bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
 2464                     np->n_fhsize = fhlen;
 2465                     newvp = NFSTOV(np);
 2466                 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
 2467                     VREF(dvp);
 2468                     newvp = dvp;
 2469                 } else {
 2470                     error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
 2471                     if (error) {
 2472                         m_freem(mrep);
 2473                         return (error);
 2474                     }
 2475                     newvp = NFSTOV(np);
 2476                 }
 2477                 if (v3) {
 2478                         nfsm_postop_attr(newvp, attrflag);
 2479                         if (!attrflag && *npp == NULL) {
 2480                                 m_freem(mrep);
 2481                                 if (newvp == dvp)
 2482                                         vrele(newvp);
 2483                                 else
 2484                                         vput(newvp);
 2485                                 return (ENOENT);
 2486                         }
 2487                 } else
 2488                         nfsm_loadattr(newvp, NULL);
 2489         }
 2490         m_freem(mrep);
 2491 nfsmout:
 2492         if (npp && *npp == NULL) {
 2493                 if (error) {
 2494                         if (newvp) {
 2495                                 if (newvp == dvp)
 2496                                         vrele(newvp);
 2497                                 else
 2498                                         vput(newvp);
 2499                         }
 2500                 } else
 2501                         *npp = np;
 2502         }
 2503         return (error);
 2504 }
 2505 
 2506 /*
 2507  * Nfs Version 3 commit rpc
 2508  */
 2509 int
 2510 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
 2511     struct thread *td)
 2512 {
 2513         u_int32_t *tl;
 2514         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2515         caddr_t bpos, dpos;
 2516         int error = 0, wccflag = NFSV3_WCCRATTR;
 2517         struct mbuf *mreq, *mrep, *md, *mb;
 2518 
 2519         if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
 2520                 return (0);
 2521         nfsstats.rpccnt[NFSPROC_COMMIT]++;
 2522         mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
 2523         mb = mreq;
 2524         bpos = mtod(mb, caddr_t);
 2525         nfsm_fhtom(vp, 1);
 2526         tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
 2527         txdr_hyper(offset, tl);
 2528         tl += 2;
 2529         *tl = txdr_unsigned(cnt);
 2530         nfsm_request(vp, NFSPROC_COMMIT, td, cred);
 2531         nfsm_wcc_data(vp, wccflag);
 2532         if (!error) {
 2533                 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
 2534                 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
 2535                         NFSX_V3WRITEVERF)) {
 2536                         bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 2537                                 NFSX_V3WRITEVERF);
 2538                         error = NFSERR_STALEWRITEVERF;
 2539                 }
 2540         }
 2541         m_freem(mrep);
 2542 nfsmout:
 2543         return (error);
 2544 }
 2545 
 2546 /*
 2547  * Strategy routine.
 2548  * For async requests when nfsiod(s) are running, queue the request by
 2549  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
 2550  * request.
 2551  */
 2552 static int
 2553 nfs_strategy(struct vop_strategy_args *ap)
 2554 {
 2555         struct buf *bp = ap->a_bp;
 2556         struct ucred *cr;
 2557 
 2558         KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
 2559         KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
 2560 
 2561         if (bp->b_iocmd == BIO_READ)
 2562                 cr = bp->b_rcred;
 2563         else
 2564                 cr = bp->b_wcred;
 2565 
 2566         /*
 2567          * If the op is asynchronous and an i/o daemon is waiting
 2568          * queue the request, wake it up and wait for completion
 2569          * otherwise just do it ourselves.
 2570          */
 2571         if ((bp->b_flags & B_ASYNC) == 0 ||
 2572             nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
 2573                 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
 2574         return (0);
 2575 }
 2576 
 2577 /*
 2578  * fsync vnode op. Just call nfs_flush() with commit == 1.
 2579  */
 2580 /* ARGSUSED */
 2581 static int
 2582 nfs_fsync(struct vop_fsync_args *ap)
 2583 {
 2584 
 2585         return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
 2586 }
 2587 
 2588 /*
 2589  * Flush all the blocks associated with a vnode.
 2590  *      Walk through the buffer pool and push any dirty pages
 2591  *      associated with the vnode.
 2592  */
 2593 static int
 2594 nfs_flush(struct vnode *vp, int waitfor, struct thread *td,
 2595     int commit)
 2596 {
 2597         struct nfsnode *np = VTONFS(vp);
 2598         struct buf *bp;
 2599         int i;
 2600         struct buf *nbp;
 2601         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2602         int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
 2603         int passone = 1;
 2604         u_quad_t off, endoff, toff;
 2605         struct ucred* wcred = NULL;
 2606         struct buf **bvec = NULL;
 2607 #ifndef NFS_COMMITBVECSIZ
 2608 #define NFS_COMMITBVECSIZ       20
 2609 #endif
 2610         struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
 2611         int bvecsize = 0, bveccount;
 2612 
 2613         if (nmp->nm_flag & NFSMNT_INT)
 2614                 slpflag = PCATCH;
 2615         if (!commit)
 2616                 passone = 0;
 2617         /*
 2618          * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
 2619          * server, but has not been committed to stable storage on the server
 2620          * yet. On the first pass, the byte range is worked out and the commit
 2621          * rpc is done. On the second pass, nfs_writebp() is called to do the
 2622          * job.
 2623          */
 2624 again:
 2625         off = (u_quad_t)-1;
 2626         endoff = 0;
 2627         bvecpos = 0;
 2628         if (NFS_ISV3(vp) && commit) {
 2629                 s = splbio();
 2630                 if (bvec != NULL && bvec != bvec_on_stack)
 2631                         free(bvec, M_TEMP);
 2632                 /*
 2633                  * Count up how many buffers waiting for a commit.
 2634                  */
 2635                 bveccount = 0;
 2636                 VI_LOCK(vp);
 2637                 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
 2638                         if (BUF_REFCNT(bp) == 0 &&
 2639                             (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
 2640                                 == (B_DELWRI | B_NEEDCOMMIT))
 2641                                 bveccount++;
 2642                 }
 2643                 /*
 2644                  * Allocate space to remember the list of bufs to commit.  It is
 2645                  * important to use M_NOWAIT here to avoid a race with nfs_write.
 2646                  * If we can't get memory (for whatever reason), we will end up
 2647                  * committing the buffers one-by-one in the loop below.
 2648                  */
 2649                 if (bveccount > NFS_COMMITBVECSIZ) {
 2650                         /*
 2651                          * Release the vnode interlock to avoid a lock
 2652                          * order reversal.
 2653                          */
 2654                         VI_UNLOCK(vp);
 2655                         bvec = (struct buf **)
 2656                                 malloc(bveccount * sizeof(struct buf *),
 2657                                        M_TEMP, M_NOWAIT);
 2658                         VI_LOCK(vp);
 2659                         if (bvec == NULL) {
 2660                                 bvec = bvec_on_stack;
 2661                                 bvecsize = NFS_COMMITBVECSIZ;
 2662                         } else
 2663                                 bvecsize = bveccount;
 2664                 } else {
 2665                         bvec = bvec_on_stack;
 2666                         bvecsize = NFS_COMMITBVECSIZ;
 2667                 }
 2668                 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
 2669                         if (bvecpos >= bvecsize)
 2670                                 break;
 2671                         if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
 2672                                 nbp = TAILQ_NEXT(bp, b_bobufs);
 2673                                 continue;
 2674                         }
 2675                         if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
 2676                             (B_DELWRI | B_NEEDCOMMIT)) {
 2677                                 BUF_UNLOCK(bp);
 2678                                 nbp = TAILQ_NEXT(bp, b_bobufs);
 2679                                 continue;
 2680                         }
 2681                         VI_UNLOCK(vp);
 2682                         bremfree(bp);
 2683                         /*
 2684                          * Work out if all buffers are using the same cred
 2685                          * so we can deal with them all with one commit.
 2686                          *
 2687                          * NOTE: we are not clearing B_DONE here, so we have
 2688                          * to do it later on in this routine if we intend to
 2689                          * initiate I/O on the bp.
 2690                          *
 2691                          * Note: to avoid loopback deadlocks, we do not
 2692                          * assign b_runningbufspace.
 2693                          */
 2694                         if (wcred == NULL)
 2695                                 wcred = bp->b_wcred;
 2696                         else if (wcred != bp->b_wcred)
 2697                                 wcred = NOCRED;
 2698                         vfs_busy_pages(bp, 1);
 2699 
 2700                         VI_LOCK(vp);
 2701                         /*
 2702                          * bp is protected by being locked, but nbp is not
 2703                          * and vfs_busy_pages() may sleep.  We have to
 2704                          * recalculate nbp.
 2705                          */
 2706                         nbp = TAILQ_NEXT(bp, b_bobufs);
 2707 
 2708                         /*
 2709                          * A list of these buffers is kept so that the
 2710                          * second loop knows which buffers have actually
 2711                          * been committed. This is necessary, since there
 2712                          * may be a race between the commit rpc and new
 2713                          * uncommitted writes on the file.
 2714                          */
 2715                         bvec[bvecpos++] = bp;
 2716                         toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2717                                 bp->b_dirtyoff;
 2718                         if (toff < off)
 2719                                 off = toff;
 2720                         toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
 2721                         if (toff > endoff)
 2722                                 endoff = toff;
 2723                 }
 2724                 splx(s);
 2725                 VI_UNLOCK(vp);
 2726         }
 2727         if (bvecpos > 0) {
 2728                 /*
 2729                  * Commit data on the server, as required.
 2730                  * If all bufs are using the same wcred, then use that with
 2731                  * one call for all of them, otherwise commit each one
 2732                  * separately.
 2733                  */
 2734                 if (wcred != NOCRED)
 2735                         retv = nfs_commit(vp, off, (int)(endoff - off),
 2736                                           wcred, td);
 2737                 else {
 2738                         retv = 0;
 2739                         for (i = 0; i < bvecpos; i++) {
 2740                                 off_t off, size;
 2741                                 bp = bvec[i];
 2742                                 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2743                                         bp->b_dirtyoff;
 2744                                 size = (u_quad_t)(bp->b_dirtyend
 2745                                                   - bp->b_dirtyoff);
 2746                                 retv = nfs_commit(vp, off, (int)size,
 2747                                                   bp->b_wcred, td);
 2748                                 if (retv) break;
 2749                         }
 2750                 }
 2751 
 2752                 if (retv == NFSERR_STALEWRITEVERF)
 2753                         nfs_clearcommit(vp->v_mount);
 2754 
 2755                 /*
 2756                  * Now, either mark the blocks I/O done or mark the
 2757                  * blocks dirty, depending on whether the commit
 2758                  * succeeded.
 2759                  */
 2760                 for (i = 0; i < bvecpos; i++) {
 2761                         bp = bvec[i];
 2762                         bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
 2763                         if (retv) {
 2764                                 /*
 2765                                  * Error, leave B_DELWRI intact
 2766                                  */
 2767                                 vfs_unbusy_pages(bp);
 2768                                 brelse(bp);
 2769                         } else {
 2770                                 /*
 2771                                  * Success, remove B_DELWRI ( bundirty() ).
 2772                                  *
 2773                                  * b_dirtyoff/b_dirtyend seem to be NFS
 2774                                  * specific.  We should probably move that
 2775                                  * into bundirty(). XXX
 2776                                  */
 2777                                 s = splbio();
 2778                                 bufobj_wref(&vp->v_bufobj);
 2779                                 bp->b_flags |= B_ASYNC;
 2780                                 bundirty(bp);
 2781                                 bp->b_flags &= ~B_DONE;
 2782                                 bp->b_ioflags &= ~BIO_ERROR;
 2783                                 bp->b_dirtyoff = bp->b_dirtyend = 0;
 2784                                 splx(s);
 2785                                 bufdone(bp);
 2786                         }
 2787                 }
 2788         }
 2789 
 2790         /*
 2791          * Start/do any write(s) that are required.
 2792          */
 2793 loop:
 2794         s = splbio();
 2795         VI_LOCK(vp);
 2796         TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
 2797                 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
 2798                         if (waitfor != MNT_WAIT || passone)
 2799                                 continue;
 2800 
 2801                         error = BUF_TIMELOCK(bp,
 2802                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 2803                             VI_MTX(vp), "nfsfsync", slpflag, slptimeo);
 2804                         splx(s);
 2805                         if (error == 0)
 2806                                 panic("nfs_fsync: inconsistent lock");
 2807                         if (error == ENOLCK)
 2808                                 goto loop;
 2809                         if (nfs_sigintr(nmp, NULL, td)) {
 2810                                 error = EINTR;
 2811                                 goto done;
 2812                         }
 2813                         if (slpflag == PCATCH) {
 2814                                 slpflag = 0;
 2815                                 slptimeo = 2 * hz;
 2816                         }
 2817                         goto loop;
 2818                 }
 2819                 if ((bp->b_flags & B_DELWRI) == 0)
 2820                         panic("nfs_fsync: not dirty");
 2821                 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
 2822                         BUF_UNLOCK(bp);
 2823                         continue;
 2824                 }
 2825                 VI_UNLOCK(vp);
 2826                 bremfree(bp);
 2827                 if (passone || !commit)
 2828                     bp->b_flags |= B_ASYNC;
 2829                 else
 2830                     bp->b_flags |= B_ASYNC;
 2831                 splx(s);
 2832                 bwrite(bp);
 2833                 if (nfs_sigintr(nmp, NULL, td)) {
 2834                         error = EINTR;
 2835                         goto done;
 2836                 }
 2837                 goto loop;
 2838         }
 2839         splx(s);
 2840         if (passone) {
 2841                 passone = 0;
 2842                 VI_UNLOCK(vp);
 2843                 goto again;
 2844         }
 2845         if (waitfor == MNT_WAIT) {
 2846                 while (vp->v_bufobj.bo_numoutput) {
 2847                         error = bufobj_wwait(&vp->v_bufobj, slpflag, slptimeo);
 2848                         if (error) {
 2849                             VI_UNLOCK(vp);
 2850                             error = nfs_sigintr(nmp, NULL, td);
 2851                             if (error)
 2852                                 goto done;
 2853                             if (slpflag == PCATCH) {
 2854                                 slpflag = 0;
 2855                                 slptimeo = 2 * hz;
 2856                             }
 2857                             VI_LOCK(vp);
 2858                         }
 2859                 }
 2860                 if (vp->v_bufobj.bo_dirty.bv_cnt != 0 && commit) {
 2861                         VI_UNLOCK(vp);
 2862                         goto loop;
 2863                 }
 2864                 VI_UNLOCK(vp);
 2865                 /* 
 2866                  * Wait for all the async IO requests to drain 
 2867                  */
 2868                 while (np->n_directio_asyncwr > 0) {
 2869                         np->n_flag |= NFSYNCWAIT;
 2870                         error = nfs_tsleep(td, (caddr_t)&np->n_directio_asyncwr,
 2871                                            slpflag | (PRIBIO + 1), "nfsfsync", 0);
 2872                         if (error) {
 2873                                 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
 2874                                         error = EINTR;
 2875                                         goto done;
 2876                                 }
 2877                         }
 2878                 }
 2879 
 2880         } else
 2881                 VI_UNLOCK(vp);
 2882         if (np->n_flag & NWRITEERR) {
 2883                 error = np->n_error;
 2884                 np->n_flag &= ~NWRITEERR;
 2885         }
 2886         if (commit && vp->v_bufobj.bo_dirty.bv_cnt == 0 &&
 2887             vp->v_bufobj.bo_numoutput == 0 && np->n_directio_asyncwr == 0)
 2888                 np->n_flag &= ~NMODIFIED;
 2889 done:
 2890         if (bvec != NULL && bvec != bvec_on_stack)
 2891                 free(bvec, M_TEMP);
 2892         return (error);
 2893 }
 2894 
 2895 /*
 2896  * NFS advisory byte-level locks.
 2897  */
 2898 static int
 2899 nfs_advlock(struct vop_advlock_args *ap)
 2900 {
 2901 
 2902         if ((VFSTONFS(ap->a_vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
 2903                 struct nfsnode *np = VTONFS(ap->a_vp);
 2904 
 2905                 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
 2906         }
 2907         return (nfs_dolock(ap));
 2908 }
 2909 
 2910 /*
 2911  * NFS advisory byte-level locks.
 2912  */
 2913 static int
 2914 nfs_advlockasync(struct vop_advlockasync_args *ap)
 2915 {
 2916         int error;
 2917         
 2918         mtx_lock(&Giant);
 2919         if ((VFSTONFS(ap->a_vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
 2920                 struct nfsnode *np = VTONFS(ap->a_vp);
 2921 
 2922                 error = lf_advlockasync(ap, &(np->n_lockf), np->n_size);
 2923                 goto out;
 2924         }
 2925         error = EOPNOTSUPP;
 2926 out:    
 2927         mtx_unlock(&Giant);
 2928         return (error);
 2929 }
 2930 
 2931 /*
 2932  * Print out the contents of an nfsnode.
 2933  */
 2934 static int
 2935 nfs_print(struct vop_print_args *ap)
 2936 {
 2937         struct vnode *vp = ap->a_vp;
 2938         struct nfsnode *np = VTONFS(vp);
 2939 
 2940         printf("\tfileid %ld fsid 0x%x",
 2941            np->n_vattr.va_fileid, np->n_vattr.va_fsid);
 2942         if (vp->v_type == VFIFO)
 2943                 fifo_printinfo(vp);
 2944         printf("\n");
 2945         return (0);
 2946 }
 2947 
 2948 /*
 2949  * This is the "real" nfs::bwrite(struct buf*).
 2950  * We set B_CACHE if this is a VMIO buffer.
 2951  */
 2952 int
 2953 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
 2954 {
 2955         int s;
 2956         int oldflags = bp->b_flags;
 2957 #if 0
 2958         int retv = 1;
 2959         off_t off;
 2960 #endif
 2961 
 2962         if (BUF_REFCNT(bp) == 0)
 2963                 panic("bwrite: buffer is not locked???");
 2964 
 2965         if (bp->b_flags & B_INVAL) {
 2966                 brelse(bp);
 2967                 return(0);
 2968         }
 2969 
 2970         bp->b_flags |= B_CACHE;
 2971 
 2972         /*
 2973          * Undirty the bp.  We will redirty it later if the I/O fails.
 2974          */
 2975 
 2976         s = splbio();
 2977         bundirty(bp);
 2978         bp->b_flags &= ~B_DONE;
 2979         bp->b_ioflags &= ~BIO_ERROR;
 2980         bp->b_iocmd = BIO_WRITE;
 2981 
 2982         bufobj_wref(bp->b_bufobj);
 2983         curthread->td_proc->p_stats->p_ru.ru_oublock++;
 2984         splx(s);
 2985 
 2986         /*
 2987          * Note: to avoid loopback deadlocks, we do not
 2988          * assign b_runningbufspace.
 2989          */
 2990         vfs_busy_pages(bp, 1);
 2991 
 2992         BUF_KERNPROC(bp);
 2993         bp->b_iooffset = dbtob(bp->b_blkno);
 2994         bstrategy(bp);
 2995 
 2996         if( (oldflags & B_ASYNC) == 0) {
 2997                 int rtval = bufwait(bp);
 2998 
 2999                 if (oldflags & B_DELWRI) {
 3000                         s = splbio();
 3001                         reassignbuf(bp);
 3002                         splx(s);
 3003                 }
 3004 
 3005                 brelse(bp);
 3006                 return (rtval);
 3007         }
 3008 
 3009         return (0);
 3010 }
 3011 
 3012 /*
 3013  * nfs special file access vnode op.
 3014  * Essentially just get vattr and then imitate iaccess() since the device is
 3015  * local to the client.
 3016  */
 3017 static int
 3018 nfsspec_access(struct vop_access_args *ap)
 3019 {
 3020         struct vattr *vap;
 3021         struct ucred *cred = ap->a_cred;
 3022         struct vnode *vp = ap->a_vp;
 3023         mode_t mode = ap->a_mode;
 3024         struct vattr vattr;
 3025         int error;
 3026 
 3027         /*
 3028          * Disallow write attempts on filesystems mounted read-only;
 3029          * unless the file is a socket, fifo, or a block or character
 3030          * device resident on the filesystem.
 3031          */
 3032         if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
 3033                 switch (vp->v_type) {
 3034                 case VREG:
 3035                 case VDIR:
 3036                 case VLNK:
 3037                         return (EROFS);
 3038                 default:
 3039                         break;
 3040                 }
 3041         }
 3042         vap = &vattr;
 3043         error = VOP_GETATTR(vp, vap, cred, ap->a_td);
 3044         if (error)
 3045                 return (error);
 3046         return (vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
 3047             mode, cred, NULL));
 3048 }
 3049 
 3050 /*
 3051  * Read wrapper for fifos.
 3052  */
 3053 static int
 3054 nfsfifo_read(struct vop_read_args *ap)
 3055 {
 3056         struct nfsnode *np = VTONFS(ap->a_vp);
 3057 
 3058         /*
 3059          * Set access flag.
 3060          */
 3061         np->n_flag |= NACC;
 3062         getnanotime(&np->n_atim);
 3063         return (fifo_specops.vop_read(ap));
 3064 }
 3065 
 3066 /*
 3067  * Write wrapper for fifos.
 3068  */
 3069 static int
 3070 nfsfifo_write(struct vop_write_args *ap)
 3071 {
 3072         struct nfsnode *np = VTONFS(ap->a_vp);
 3073 
 3074         /*
 3075          * Set update flag.
 3076          */
 3077         np->n_flag |= NUPD;
 3078         getnanotime(&np->n_mtim);
 3079         return (fifo_specops.vop_write(ap));
 3080 }
 3081 
 3082 /*
 3083  * Close wrapper for fifos.
 3084  *
 3085  * Update the times on the nfsnode then do fifo close.
 3086  */
 3087 static int
 3088 nfsfifo_close(struct vop_close_args *ap)
 3089 {
 3090         struct vnode *vp = ap->a_vp;
 3091         struct nfsnode *np = VTONFS(vp);
 3092         struct vattr vattr;
 3093         struct timespec ts;
 3094 
 3095         if (np->n_flag & (NACC | NUPD)) {
 3096                 getnanotime(&ts);
 3097                 if (np->n_flag & NACC)
 3098                         np->n_atim = ts;
 3099                 if (np->n_flag & NUPD)
 3100                         np->n_mtim = ts;
 3101                 np->n_flag |= NCHG;
 3102                 if (vrefcnt(vp) == 1 &&
 3103                     (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
 3104                         VATTR_NULL(&vattr);
 3105                         if (np->n_flag & NACC)
 3106                                 vattr.va_atime = np->n_atim;
 3107                         if (np->n_flag & NUPD)
 3108                                 vattr.va_mtime = np->n_mtim;
 3109                         (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
 3110                 }
 3111         }
 3112         return (fifo_specops.vop_close(ap));
 3113 }
 3114 
 3115 /*
 3116  * Just call nfs_writebp() with the force argument set to 1.
 3117  *
 3118  * NOTE: B_DONE may or may not be set in a_bp on call.
 3119  */
 3120 static int
 3121 nfs_bwrite(struct buf *bp)
 3122 {
 3123 
 3124         return (nfs_writebp(bp, 1, curthread));
 3125 }
 3126 
 3127 struct buf_ops buf_ops_nfs = {
 3128         .bop_name       =       "buf_ops_nfs",
 3129         .bop_write      =       nfs_bwrite,
 3130         .bop_strategy   =       bufstrategy,
 3131         .bop_sync       =       bufsync,
 3132         .bop_bdflush    =       bufbdflush,
 3133 };

Cache object: 82d8e9194f103d0c88a41c26651ddfd9


[ 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.